Datasets:

---

LLMDH

/

OpenScience

like 0

Follow

LLM - Digital Humanities 16

ERROR: type should be string, got "https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Correspondence: Maurus Borne (maurus.borne@kit.edu) Correspondence: Maurus Borne (maurus.borne@kit.edu) Abstract. Since its launch by the European Space Agency in 2018, the Aeolus satellite has been using the first Doppler\nwind lidar in space to acquire three-dimensional atmospheric wind profiles around the globe. Especially in the tropics, these\nmeasurements compensate for the currently limited number of other wind observations, making an assessment of the quality of\nAeolus wind products in this region crucial for numerical weather prediction. To evaluate the quality of the Aeolus L2B wind products across the tropical Atlantic Ocean, 20 radiosondes corresponding to Aeolus overpasses were launched from the islands\n5\nof Sal, Saint Croix and Puerto Rico during August-September 2021 as part of the Joint Aeolus Tropical Atlantic Campaign. During this period, Aeolus sampled winds within a complex environment with a variety of cloud types in the vicinity of the\nInter-tropical Convergence Zone and aerosol particles from Saharan dust outbreaks. On average, the validation for Aeolus\nRaleigh-clear revealed a random error of 3.8 – 4.3 ms–1 between 2–16 km and 4.3 – 4.8 ms–1 between 16–20 km, with a 5 systematic error of -0.5±0.2 ms–1. For Mie-cloudy, the random error between 2–16 km is 1.1 – 2.3 ms–1 and the systematic\n10\nerror is -0.9 ±0.3 ms–1. Below clouds or within dust layers, the quality of Rayleigh-clear measurements can be degraded when\nthe useful signal is reduced. In these conditions, we also noticed an underestimation of the L2B estimated error. Gross outliers\nwhich we define with large deviations from the radiosonde but low error estimates account for less than 5% of the data. These\noutliers appear at all altitudes and under all environmental conditions; however, their root-cause remains unknown. Finally, systematic error of -0.5±0.2 ms–1. For Mie-cloudy, the random error between 2–16 km is 1.1 – 2.3 ms–1 and the systematic\n10\nerror is -0.9 ±0.3 ms–1. Below clouds or within dust layers, the quality of Rayleigh-clear measurements can be degraded when\nthe useful signal is reduced. In these conditions, we also noticed an underestimation of the L2B estimated error. Gross outliers\nwhich we define with large deviations from the radiosonde but low error estimates account for less than 5% of the data. These\noutliers appear at all altitudes and under all environmental conditions; however, their root-cause remains unknown. Validation of Aeolus L2B products over the tropical Atlantic using\nradiosondes Maurus Borne 1, Peter Knippertz 1, Martin Weissmann 2, Benjamin Witschas 3, Cyrille Flamant 4,\nRosimar Rios-Berrios 5, and Peter Veals 6 1Institute of Meteorology and Climate Research, Karlsruhe Institute of Technology (KIT), 76131 Karlsruhe, Germany\n2Universität Wien, Institut für Meteorologie und Geophysik, Wien, Austria g\np y\n3Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Institut für Physik der Atmosphäre, 82234 Oberpfaffenhofen,\nGermany\n4 3Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Institut für Physik der Atmosphäre, 82234 Oberpfaffenhofen,\nGermany 4Laboratoire Atmosphères, Milieux, Observations Spatiales (LATMOS), UMR 8190, CNRS, Sorbonne Université and\nUniversité Paris Saclay, Paris, France\n5N i\nl C\nf\nA\nh i R\nh B\nld\nCO USA 4Laboratoire Atmosphères, Milieux, Observations Spatiales (LATMOS), UMR 8190, CNRS, Sorbonne Université and\nUniversité Paris Saclay, Paris, France\n5 5National Center for Atmospheric Research, Boulder, CO, USA 5National Center for Atmospheric Research, Boulder, CO, USA\n6 p\n6Department of Atmospheric Sciences, University of Utah, Salt Lake City, Utah, USA 6Department of Atmospheric Sciences, University of Utah, Salt Lake City, Utah, USA https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 1\nIntroduction In situ measurements derived from aircraft reports, ground stations or radiosondes\nare not globally distributed and lead to a lack of observations in the aforementioned regions. To address these deficiencies, the European Space Agency (ESA) deployed the Atmospheric Dynamics Mission Aeolus in To address these deficiencies, the European Space Agency (ESA) deployed the Atmospheric Dynamics Mission Aeolus in\n2018, the first satellite capable of measuring atmospheric winds around the globe from space with a homogeneous space-time\n30\nwind coverage and altitude-resolved profiles up to 30 km height (Reitebuch, 2012). The instrument carries a direct detection\nDoppler wind lidar called ALADIN (Atmospheric LAser Doppler INstrument) that emits short ultraviolet (UV) pulses at 355\nnm along the Line Of Sight (LOS) of the instrument. The Doppler shift of the backscatter signal is detected by a dual-channel\nreceiver consisting of the following elements: a Fizeau interferometer analysing the Doppler shift of the narrowband particle 2018, the first satellite capable of measuring atmospheric winds around the globe from space with a homogeneous space-time\n30\nwind coverage and altitude-resolved profiles up to 30 km height (Reitebuch, 2012). The instrument carries a direct detection\nDoppler wind lidar called ALADIN (Atmospheric LAser Doppler INstrument) that emits short ultraviolet (UV) pulses at 355\nnm along the Line Of Sight (LOS) of the instrument. The Doppler shift of the backscatter signal is detected by a dual-channel\nreceiver consisting of the following elements: a Fizeau interferometer analysing the Doppler shift of the narrowband particle backscatter signal (cloud droplets and aerosols or ice crystals) using the fringe imaging technique (McKay, 2002), referred to\n35\nas the Mie channel, and a dual Fabry-Pérot interferometer detecting the Doppler-shifted frequency of the Rayleigh-Brillouin\nbackscatter spectrum (air molecules) using the double-edge technique (Flesia and Korb, 1999), called the Rayleigh channel. The processing algorithm also distinguishes between retrievals originating from \"cloudy\" or \"clear\" atmospheric conditions,\nresulting in Rayleigh-clear and Mie-cloudy observation types. The two channels complement each other, as Mie-cloudy winds backscatter signal (cloud droplets and aerosols or ice crystals) using the fringe imaging technique (McKay, 2002), referred to\n35\nas the Mie channel, and a dual Fabry-Pérot interferometer detecting the Doppler-shifted frequency of the Rayleigh-Brillouin\nbackscatter spectrum (air molecules) using the double-edge technique (Flesia and Korb, 1999), called the Rayleigh channel. The processing algorithm also distinguishes between retrievals originating from \"cloudy\" or \"clear\" atmospheric conditions,\nresulting in Rayleigh-clear and Mie-cloudy observation types. Correspondence: Maurus Borne (maurus.borne@kit.edu) Finally, we confirm the presence of an orbital-dependent bias of up to 2.5 ms–1 observed with both radiosondes and European Centre\n15\nfor Medium-Range Weather Forecasts model equivalents. The results of this study contribute to a better characterization of\nthe Aeolus wind product in different atmospheric conditions and provide valuable information for further improvement of the\nwind retrieval algorithm. we confirm the presence of an orbital-dependent bias of up to 2.5 ms–1 observed with both radiosondes and European Centre\n15\nfor Medium-Range Weather Forecasts model equivalents. The results of this study contribute to a better characterization of\nthe Aeolus wind product in different atmospheric conditions and provide valuable information for further improvement of the\nwind retrieval algorithm. 1 1 1\nIntroduction According to the World Meteorological Organisation (WMO), wind is the most critical atmospheric variable lacking in the\n20\ncurrent Global Observing System (GOS) (Baker et al., 2014). Especially in the Southern Hemisphere (SH), over the oceans\nand near equatorial regions, numerical weather prediction (NWP) models require additional wind observations with sufficient\ncoverage in time and space to identify key atmospheric dynamics (Stoffelen et al., 2005; Straume et al., 2020). Before the\nlaunch of Aeolus in 2018, satellite wind observations in these regions were only available for a limited number of tropospheric According to the World Meteorological Organisation (WMO), wind is the most critical atmospheric variable lacking in the\n20\ncurrent Global Observing System (GOS) (Baker et al., 2014). Especially in the Southern Hemisphere (SH), over the oceans\nand near equatorial regions, numerical weather prediction (NWP) models require additional wind observations with sufficient\ncoverage in time and space to identify key atmospheric dynamics (Stoffelen et al., 2005; Straume et al., 2020). Before the\nlaunch of Aeolus in 2018, satellite wind observations in these regions were only available for a limited number of tropospheric layers and were mainly provided by atmospheric motion vectors (AMVs) estimated from tracking cloud and water vapour\n25\nfeatures (Bormann et al., 2003; Folger and Weissmann, 2014), or by scatterometer measurements of surface winds (Naderi\net al., 1991; Portabella and Stoffelen, 2009). In situ measurements derived from aircraft reports, ground stations or radiosondes\nare not globally distributed and lead to a lack of observations in the aforementioned regions. dd\nh\nd fi i\ni\nh\nd\nl\nd h\nh i\ni\ni i\nl\ni layers and were mainly provided by atmospheric motion vectors (AMVs) estimated from tracking cloud and water vapour\n25\nfeatures (Bormann et al., 2003; Folger and Weissmann, 2014), or by scatterometer measurements of surface winds (Naderi\net al., 1991; Portabella and Stoffelen, 2009). In situ measurements derived from aircraft reports, ground stations or radiosondes\nare not globally distributed and lead to a lack of observations in the aforementioned regions. To address these deficiencies the European Space Agency (ESA) deployed the Atmospheric Dynamics Mission Aeolus in layers and were mainly provided by atmospheric motion vectors (AMVs) estimated from tracking cloud and water vapour\n25\nfeatures (Bormann et al., 2003; Folger and Weissmann, 2014), or by scatterometer measurements of surface winds (Naderi\net al., 1991; Portabella and Stoffelen, 2009). 1\nIntroduction The two channels complement each other, as Mie-cloudy winds can compensate for gaps in Rayleigh-clear measurements, especially in cloudy and aerosol-loaded regions. Various NWP\n40\ncentres have demonstrated the added value of assimilating Aeolus winds through significant improvements in model fields\nand model background information, especially in tropical regions, the upper tropical troposphere and the lower stratosphere\n(Rennie et al., 2021; Martin et al., 2022a, b; Garrett et al., 2022). can compensate for gaps in Rayleigh-clear measurements, especially in cloudy and aerosol-loaded regions. Various NWP\n40\ncentres have demonstrated the added value of assimilating Aeolus winds through significant improvements in model fields\nand model background information, especially in tropical regions, the upper tropical troposphere and the lower stratosphere\n(Rennie et al., 2021; Martin et al., 2022a, b; Garrett et al., 2022). For an optimal use of the Aeolus wind observations in NWP models, an assessment of the data quality is essential. To achieve\nthis, several scientific and technical studies are carried out in the framework of Calibration/Validation (Cal/Val) activities\n45\norganised by ESA. For wind validation, several reference products have been used such as ground-based remote sensing\nobservations (Belova et al., 2021; Guo et al., 2021; Iwai et al., 2021; Abril-Gago et al., 2022), in situ measurements (Baars\net al., 2020; Chen et al., 2021; Ratynski et al., 2022), airborne measurements (Lux et al., 2020; Witschas et al., 2020; Bedka\net al., 2021; Witschas et al., 2022) or NWP model equivalents (Martin et al., 2021; Zuo et al., 2022). For an optimal use of the Aeolus wind observations in NWP models, an assessment of the data quality is essential. To achieve\nthis, several scientific and technical studies are carried out in the framework of Calibration/Validation (Cal/Val) activities\n45\norganised by ESA. For wind validation, several reference products have been used such as ground-based remote sensing\nobservations (Belova et al., 2021; Guo et al., 2021; Iwai et al., 2021; Abril-Gago et al., 2022), in situ measurements (Baars\net al., 2020; Chen et al., 2021; Ratynski et al., 2022), airborne measurements (Lux et al., 2020; Witschas et al., 2020; Bedka\net al., 2021; Witschas et al., 2022) or NWP model equivalents (Martin et al., 2021; Zuo et al., 2022). this, several scientific and technical studies are carried out in the framework of Calibration/Validation (Cal/Val) activities\n45\norganised by ESA. The contribution\n70\nof the radiosondes in JATAC is complementary to other instruments as they provide accurate wind measurements throughout\nthe troposphere up to the lower stratosphere, which is not probed by many other instruments and provides an almost unique\ndata set for validating the Aeolus winds at this altitude. This article is structured as follows: Section 2 describes the instruments and data while section 3 details the quality con the warm and moist Caribbean, where heavy rainfall events and tropical cyclones frequently affect the area. The contribution\n70\nof the radiosondes in JATAC is complementary to other instruments as they provide accurate wind measurements throughout\nthe troposphere up to the lower stratosphere, which is not probed by many other instruments and provides an almost unique\ndata set for validating the Aeolus winds at this altitude. This article is structured as follows: Section 2 describes the instruments and data while section 3 details the quality con-\ntrol and co-location criteria used for the validation study. Section 4 deals with the quantification of errors, their dependency\n75\non temporal and spatial distance between the compared observations as well as on the presence of clouds and dust. For this\npurpose, we use the Satellite Application Facility for supporting NoWCasting and very short range forecasting (SAFNWC,\nAlonso Lasheras et al. (2005)) satellite-based meteorological Cloud Type (CT) product and the Copernicus Atmosphere Mon-\nitoring Service (CAMS) dust mixing ratio reanalysis. Furthermore, the section includes a case study illustrating the different\nbehaviour of Rayleigh-clear and Mie-cloudy winds under different environmental conditions Finally, section 5 summarises\n80 trol and co-location criteria used for the validation study. Section 4 deals with the quantification of errors, their dependency\n75\non temporal and spatial distance between the compared observations as well as on the presence of clouds and dust. For this\npurpose, we use the Satellite Application Facility for supporting NoWCasting and very short range forecasting (SAFNWC,\nAlonso Lasheras et al. (2005)) satellite-based meteorological Cloud Type (CT) product and the Copernicus Atmosphere Mon-\nitoring Service (CAMS) dust mixing ratio reanalysis. Furthermore, the section includes a case study illustrating the different behaviour of Rayleigh-clear and Mie-cloudy winds under different environmental conditions. Finally, section 5 summarises\n80\nthe main results and provides recommendations for improving the Aeolus wind retrieval algorithm. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. lies\" (Weiler et al., 2021a), which reduced the systematic and random errors in the Rayleigh channel. One phenomenon that\nstill needs to be explored is the sensitivity of Aeolus wind quality to the presence of aerosols and clouds, potentially affecting lies\" (Weiler et al., 2021a), which reduced the systematic and random errors in the Rayleigh channel. One phenomenon that\nstill needs to be explored is the sensitivity of Aeolus wind quality to the presence of aerosols and clouds, potentially affecting\nkey parameters such as the signal levels or scattering ratio (SR) used to calculate the Horizontal Line of Sight (HLOS) winds\n55\nand the associated error estimate (EE). The tropical Atlantic during the boreal summer, spanning from West Africa to the\nCaribbean, is the ideal place to explore these dependencies, with a wide range of atmospheric aerosols (Saharan dust aerosols,\nsea salt aerosols, biomass combustion aerosols) and convective cloud types associated with the West African Monsoon (WAM) lies\" (Weiler et al., 2021a), which reduced the systematic and random errors in the Rayleigh channel. One phenomenon that\nstill needs to be explored is the sensitivity of Aeolus wind quality to the presence of aerosols and clouds, potentially affecting key parameters such as the signal levels or scattering ratio (SR) used to calculate the Horizontal Line of Sight (HLOS) winds\n55\nand the associated error estimate (EE). The tropical Atlantic during the boreal summer, spanning from West Africa to the\nCaribbean, is the ideal place to explore these dependencies, with a wide range of atmospheric aerosols (Saharan dust aerosols,\nsea salt aerosols, biomass combustion aerosols) and convective cloud types associated with the West African Monsoon (WAM)\ncirculation and the Inter Tropical Convergence Zone (ITCZ). key parameters such as the signal levels or scattering ratio (SR) used to calculate the Horizontal Line of Sight (HLOS) winds\n55\nand the associated error estimate (EE). The tropical Atlantic during the boreal summer, spanning from West Africa to the\nCaribbean, is the ideal place to explore these dependencies, with a wide range of atmospheric aerosols (Saharan dust aerosols,\nsea salt aerosols, biomass combustion aerosols) and convective cloud types associated with the West African Monsoon (WAM)\ncirculation and the Inter Tropical Convergence Zone (ITCZ). 1\nIntroduction For wind validation, several reference products have been used such as ground-based remote sensing\nobservations (Belova et al., 2021; Guo et al., 2021; Iwai et al., 2021; Abril-Gago et al., 2022), in situ measurements (Baars\net al., 2020; Chen et al., 2021; Ratynski et al., 2022), airborne measurements (Lux et al., 2020; Witschas et al., 2020; Bedka\net al., 2021; Witschas et al., 2022) or NWP model equivalents (Martin et al., 2021; Zuo et al., 2022). Several anomalies in the Aeolus data have already been detected and improvements in the processing chain and the in-\n50\nstrument have been made accordingly. These include the implementation of a bias correction in both channels related to the\norbital-dependent temperature variations of ALADIN’s M1 mirror (Weiler et al., 2021b) and the correction of \"pixel anoma- Several anomalies in the Aeolus data have already been detected and improvements in the processing chain and the in-\n50\nstrument have been made accordingly. These include the implementation of a bias correction in both channels related to the\norbital-dependent temperature variations of ALADIN’s M1 mirror (Weiler et al., 2021b) and the correction of \"pixel anoma- Several anomalies in the Aeolus data have already been detected and improvements in the processing chain and the in-\n50\nstrument have been made accordingly. These include the implementation of a bias correction in both channels related to the\norbital-dependent temperature variations of ALADIN’s M1 mirror (Weiler et al., 2021b) and the correction of \"pixel anoma- 2 For this purpose, ESA organized the Joint Aeolus Tropical Atlantic Campaign (JATAC) in the period July to September\n60\n2021, which deployed sophisticated airborne lidar instruments over Cabo Verde (German Aerospace Center (DLR), Labo-\nratoire ATmosphères, Milieux, Observations Spatiales (LATMOS)) and the Virgin Islands (National Aeronautics and Space\nAdministration (NASA)) but also ground-based instruments such as radiosondes (Karlsruhe Institute of Technology (KIT),\nUniversity of Oklahoma, University of Utah) and Doppler lidar systems (Leibniz Institute for Tropospheric Research (TRO- For this purpose, ESA organized the Joint Aeolus Tropical Atlantic Campaign (JATAC) in the period July to September\n60\n2021, which deployed sophisticated airborne lidar instruments over Cabo Verde (German Aerospace Center (DLR), Labo-\nratoire ATmosphères, Milieux, Observations Spatiales (LATMOS)) and the Virgin Islands (National Aeronautics and Space\nAdministration (NASA)) but also ground-based instruments such as radiosondes (Karlsruhe Institute of Technology (KIT),\nUniversity of Oklahoma, University of Utah) and Doppler lidar systems (Leibniz Institute for Tropospheric Research (TRO- POS), National Observatory of Athens (NOA)). In this study, we validate Aeolus wind products using radiosondes launched\n65\nfrom western Puerto Rico, northern St. Croix and Sal airport on Cabo Verde. The semi-arid island of Sal is located over the\ntropical East Atlantic off the West African coast, near the northern boundary of the WAM. Rain events are relatively sporadic\nthere, as most synoptic and mesoscale precipitation systems propagate south of the island. The region is exposed to mineral\ndust plumes emanating from Saharan dust outbreaks. In contrast, the islands of St. Croix and Puerto Rico are located within POS), National Observatory of Athens (NOA)). In this study, we validate Aeolus wind products using radiosondes launched\n65\nfrom western Puerto Rico, northern St. Croix and Sal airport on Cabo Verde. The semi-arid island of Sal is located over the\ntropical East Atlantic off the West African coast, near the northern boundary of the WAM. Rain events are relatively sporadic\nthere, as most synoptic and mesoscale precipitation systems propagate south of the island. The region is exposed to mineral\ndust plumes emanating from Saharan dust outbreaks. In contrast, the islands of St. Croix and Puerto Rico are located within the warm and moist Caribbean, where heavy rainfall events and tropical cyclones frequently affect the area. 2.1\nALADIN and Aeolus wind products Aeolus is the second Earth Explorer Core mission and measures global atmospheric wind profiles from a 320 km high sun-\nsynchronous dusk-dawn orbit. It carries the ALADIN instrument (Schillinger et al. (2003)), which is a direct-detection high-\n85 85 3 3 The L1B product comprises the geolocated and observation data as well as optical information (SNR, useful signal, scattering ratio , etc.). The wind product called L2B contains the final horizontal projection of the LOS wind speed profiles of\n105\nthe Rayleigh and Mie channels, where all necessary calibration and instrument corrections have been performed (Dabas et al.,\n2008). This product is suitable for the assimilation in NWP models and scientific research. The L2B product also provides scene\nclassification based upon the backscatter ratio corresponding to the wind originating from a ’cloudy’ or ’clear’ atmospheric\nregion, resulting in Rayleigh-clear, Rayleigh-cloudy, Mie-clear and Mie-cloudy observation types. Throughout the processing scattering ratio , etc.). The wind product called L2B contains the final horizontal projection of the LOS wind speed profiles of\n105\nthe Rayleigh and Mie channels, where all necessary calibration and instrument corrections have been performed (Dabas et al.,\n2008). This product is suitable for the assimilation in NWP models and scientific research. The L2B product also provides scene\nclassification based upon the backscatter ratio corresponding to the wind originating from a ’cloudy’ or ’clear’ atmospheric\nregion, resulting in Rayleigh-clear, Rayleigh-cloudy, Mie-clear and Mie-cloudy observation types. Throughout the processing\nchain, the L1B and L2B processors are continuously updated into different baseline versions to account for revisions and\n110 chain, the L1B and L2B processors are continuously updated into different baseline versions to account for revisions and\n110\nidentified problems. This leads to different HLOS wind observations and quality in different time periods. In this study, data from the near-real-time version Baseline product 12 (L2bP 3.50) are used. We evaluate all observation\ntypes and corresponding Error Estimates (EEs) of the L2B product except Mie-clear observations as the Mie signal should is\nusually weak in clear sky conditions. Additionally, two L1B products are used, namely the scattering ratio (SR) and the useful chain, the L1B and L2B processors are continuously updated into different baseline versions to account for revisions and\n110\nidentified problems. This leads to different HLOS wind observations and quality in different time periods. In this study, data from the near-real-time version Baseline product 12 (L2bP 3.50) are used. We evaluate all observation\ntypes and corresponding Error Estimates (EEs) of the L2B product except Mie-clear observations as the Mie signal should is\nusually weak in clear sky conditions. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. spectral-resolution wind lidar with a Nd:YAG laser transmitter that operates at an ultraviolet wavelength of 354.8 nm. It points\nat 35 ◦with and angle of ∼10° from the zonal direction. ALADIN consists of a two-channel receiver that allows the instrument to measure wind speed from molecular backscatter\n(Rayleigh channel) and particle backscatter (Mie channel). The Rayleigh channel relies on the double-edge technique (Flesia\nand Korb, 1999) using a sequential Fabry-Perot interferometer, where the Doppler shift of the backscattered molecular spectrum\n90\nis retrieved from the signal intensities that are transmitted through two band-pass filters A and B. The final Rayleigh response\nis computed from a contrast-function between both filter signals. For Mie winds, the computation is based on a fringe-imaging\ntechnique (McKay, 2002), in which the Fizeau interferometer forms a linear interference fringe on the detector from the\nnarrowband particle backscatter signal. The lateral displacement of the interference fringe is then used to calculate the Doppler 90 shift. 95\nTo ensure a sufficient Signal-to-Noise Ratio (SNR), the wind measurements are averaged vertically and horizontally into\nsingle observations. Vertical sampling is performed within 24 vertical elevation bins with a resolution that can vary from\n0.25 km at lower elevations to 2 km at higher elevations. They are defined by the Range Bin Settings (RBS) and can vary\ngeographically and between the respective detection channel (Rayleigh and Mie). Horizontally, the measurements are averaged shift. 95\nTo ensure a sufficient Signal-to-Noise Ratio (SNR), the wind measurements are averaged vertically and horizontally into\nsingle observations. Vertical sampling is performed within 24 vertical elevation bins with a resolution that can vary from\n0.25 km at lower elevations to 2 km at higher elevations. They are defined by the Range Bin Settings (RBS) and can vary\ngeographically and between the respective detection channel (Rayleigh and Mie). Horizontally, the measurements are averaged over 87 km and 10 km integration lengths for Rayleigh and Mie channels, respectively, owing to the lower signal levels of the\n100\nRayleigh measurements. The data products are processed through a multi-stage processing chain, with each level containing different information\n(Reitebuch et al., 2018; Tan et al., 2008). In this study, the Level1B (L1B) and Level2B (L2B) products are of particular\ninterest. Additionally, two L1B products are used, namely the scattering ratio (SR) and the useful signal. The SR represents the ratio between the total (molecular and particulate) and the molecular backscatter coefficients. 115\nIt is strictly equal to or greater than one and describes the contribution of the particles to the backscattered signal. Note that\nthe SRs of the L2B products are not used, as some SRs were manually set to one during the processor baseline to eliminate a\ncross-talk correction, which had detrimental effects on the wind quality. The useful signal represents the returned signal levels\nper observation and comprises corrections for the solar background, the dark current and the detection chain offset (DCO). signal. The SR represents the ratio between the total (molecular and particulate) and the molecular backscatter coefficients. 115\nIt is strictly equal to or greater than one and describes the contribution of the particles to the backscattered signal. Note that\nthe SRs of the L2B products are not used, as some SRs were manually set to one during the processor baseline to eliminate a\ncross-talk correction, which had detrimental effects on the wind quality. The useful signal represents the returned signal levels\nper observation and comprises corrections for the solar background, the dark current and the detection chain offset (DCO). signal. The SR represents the ratio between the total (molecular and particulate) and the molecular backscatter coefficients. 115\nIt is strictly equal to or greater than one and describes the contribution of the particles to the backscattered signal. Note that\nthe SRs of the L2B products are not used, as some SRs were manually set to one during the processor baseline to eliminate a\ncross-talk correction, which had detrimental effects on the wind quality. The useful signal represents the returned signal levels\nper observation and comprises corrections for the solar background, the dark current and the detection chain offset (DCO). We apply an additional range correction and signal normalization that takes into account the different range bin thickness and\n120 4 4 At the end of each assimilation cycle, the feedback files with the Aeolus winds\nand their model equivalents can be retrieved from the Meteorological Archival and Retrieval System (MARS). These reports\ncontain information on the assimilated observations, their model background (short-range forecast) and analysis equivalents as\nwell as various quality control flags.. In this study, background equivalents of Aeolus observations are used as an additional\nreference to validate Aeolus HLOS winds. Note that only Rayleigh-clear and the Mie-cloudy winds are in operational use for\n130\nNWP. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. distances between the instruments and the height bins. Due to the sequential implementation of the Fizeau and the Fabry-Perot\ninterometers, signal from Mie scattering can leak into the Rayleigh channel signal. This optical \"cross-talk\" can cause biases,\nespecially in the case of strong Mie returns, as the Rayleigh-channel assumes pure molecular signal in the processing chain. distances between the instruments and the height bins. Due to the sequential implementation of the Fizeau and the Fabry-Perot\ninterometers, signal from Mie scattering can leak into the Rayleigh channel signal. This optical \"cross-talk\" can cause biases,\nespecially in the case of strong Mie returns, as the Rayleigh-channel assumes pure molecular signal in the processing chain. Along with many other NWP centers, the data were assimilated in the European Centre for Medium-Range Weather Fore-\ncasts (ECMWF) Integrated Forecasting System (IFS) by means of the operational four-dimensional ensemble-variational (4D-\n125\nEnVar) data assimilation scheme (4D-EnVar). At the end of each assimilation cycle, the feedback files with the Aeolus winds\nand their model equivalents can be retrieved from the Meteorological Archival and Retrieval System (MARS). These reports\ncontain information on the assimilated observations, their model background (short-range forecast) and analysis equivalents as\nwell as various quality control flags.. In this study, background equivalents of Aeolus observations are used as an additional\nreference to validate Aeolus HLOS winds. Note that only Rayleigh-clear and the Mie-cloudy winds are in operational use for\n130\nNWP. distances between the instruments and the height bins. Due to the sequential implementation of the Fizeau and the Fabry-Perot\ninterometers, signal from Mie scattering can leak into the Rayleigh channel signal. This optical \"cross-talk\" can cause biases,\nespecially in the case of strong Mie returns, as the Rayleigh-channel assumes pure molecular signal in the processing chain. interometers, signal from Mie scattering can leak into the Rayleigh channel signal. This optical \"cross-talk\" can cause biases,\nespecially in the case of strong Mie returns, as the Rayleigh-channel assumes pure molecular signal in the processing chain. Along with many other NWP centers, the data were assimilated in the European Centre for Medium-Range Weather Fore-\ncasts (ECMWF) Integrated Forecasting System (IFS) by means of the operational four-dimensional ensemble-variational (4D-\n125\nEnVar) data assimilation scheme (4D-EnVar). different research components of JATAC. Between the 7 and 28th of September 2021, a total of 37 radiosondes were launched\nfrom Sal airport in Cape Verde, 9 of them corresponding to Aeolus overflights. The launches were coordinated by the Karlsruhe\n135\nInstitute of Technology (KIT) with local support from the JATAC team. This was accomplished using the DFM-09 (GRAW)\nlight weather radiosondes, which measure air pressure, air temperature, relative humidity, wind speed and wind direction. The\nvertical resolution depends on the ascent speed, which varies with the amount of helium in the balloon, but can generally be\nestimated at about 5 ms–1. Most of the radiosondes launched at Sal were ingested into the Global Telecommunication System\n(GTS).\n140 As for Sal, these\n145\nmeasurements were performed with the radiosonde instrument DFM-09 (GRAW). Lastly, 32 launches were conducted from\nthe University of Puerto Rico at Mayagüez (UPRM) campus between 26 August and 14 September 2021, 7 of which could be\nused for the validation of Aeolus. All launches were performed with iMet-4 radiosondes from the International Met System. As with DFM-09, the iMet-4 radiosondes provide measurements of wind speed, wind direction, temperature, humidity and air pressure. The radiosonde data also underwent a quality control check using the Atmospheric Sounding Processing Envi-\n150\nronment (ASPEN) software (Martin and Suhr, 2021) developed by the Earth Observing Laboratory at the National Center for\nAtmospheric Research (NCAR). A summary of the radiosonde launches and weather events sampled at UPRM was provided\nby Rios-Berrios et al. (2023). The total number of radiosonde profiles corresponding to Aeolus overpasses thus amounts to 20, of which 12 correspond to air pressure. The radiosonde data also underwent a quality control check using the Atmospheric Sounding Processing Envi-\n150\nronment (ASPEN) software (Martin and Suhr, 2021) developed by the Earth Observing Laboratory at the National Center for\nAtmospheric Research (NCAR). A summary of the radiosonde launches and weather events sampled at UPRM was provided\nby Rios-Berrios et al. (2023). The total number of radiosonde profiles corresponding to Aeolus overpasses thus amounts to 20 of which 12 correspond to The total number of radiosonde profiles corresponding to Aeolus overpasses thus amounts to 20, of which 12 correspond to\nascending and 8 to descending orbits of Aeolus. An overview of the launches from the different sites can be found in Table 1,\n155\nalong with other co-location parameters fully discussed in Section 3.1. ascending and 8 to descending orbits of Aeolus. An overview of the launches from the different sites can be found in Table 1,\n155\nalong with other co-location parameters fully discussed in Section 3.1. ascending and 8 to descending orbits of Aeolus. An overview of the launches from the different sites can be found in Table 1,\n155\nalong with other co-location parameters fully discussed in Section 3.1. 2.2\nRadiosondes Table 1. Overview of Aeolus overflights and associated radiosonde profiles. Table 1. Overview of Aeolus overflights and associated radiosonde profiles. Week day\nStart and stop time\nOrbit node\nCo-location radius\nNumber of profiles\nSal\nTuesday\n07:28 – 07:29 UTC\nDescending\n50 km\n3\nThursday\n19:23 – 19:24 UTC\nAscending\n180 km\n3\nFriday\n19:36 – 19:37 UTC\nAscending\n280 km\n3\nSaint Croix\nMonday\n10:17 - 10:18 UTC\nDescending\n90 km\n3\nWednesday\n22:12 – 22:13 UTC\nAscending\n160 km\n3\nThursday\n22:25 – 22:26 UTC\nAscending\n340 km\n1\nPuerto Rico\nTuesday\n10:29 – 10:30 UTC\nDescending\n160 km\n2\nThursday\n22:25 – 22:26 UTC\nAscending\n100 km\n2 Week day\nStart and stop time\nOrbit node\nCo-location radius\nNumber of profiles During the campaign, radiosondes were launched from three different locations over the tropical Atlantic and coordinated by\ndifferent research components of JATAC. Between the 7 and 28th of September 2021, a total of 37 radiosondes were launched from Sal airport in Cape Verde, 9 of them corresponding to Aeolus overflights. The launches were coordinated by the Karlsruhe\n135\nInstitute of Technology (KIT) with local support from the JATAC team. This was accomplished using the DFM-09 (GRAW)\nlight weather radiosondes, which measure air pressure, air temperature, relative humidity, wind speed and wind direction. The\nvertical resolution depends on the ascent speed, which varies with the amount of helium in the balloon, but can generally be\nestimated at about 5 ms–1. Most of the radiosondes launched at Sal were ingested into the Global Telecommunication System (GTS). 140 5 5 2.3\nEUMETSAT SAFNWC Cloud type product The Satellite Application Facility for supporting NoWCasting and very short range forecasting (SAFNWC; Alonso Lasheras\net al., 2005) developed a number of satellite-based meteorological products distributed by the European Organisation for the\nExploitation of Meteorological Satellites (EUMETSAT). Among others, they provide the Cloud Type (CT) product (Derrien\n160\nand Le Gléau, 2005), which is a detailed scenery classification of clouds based on different main classes. The Satellite Application Facility for supporting NoWCasting and very short range forecasting (SAFNWC; Alonso Lasheras\net al., 2005) developed a number of satellite-based meteorological products distributed by the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT). Among others, they provide the Cloud Type (CT) product (Derrien\n160\nand Le Gléau, 2005), which is a detailed scenery classification of clouds based on different main classes. The baseline data originate from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) operated onboard the second\ngeneration METEOSAT geostationary satellites (MSG). Multispectral thresholding techniques (Saunders and Kriebel, 1988;\nDerrien et al., 1993; Stowe et al., 1999) are subsequently applied in the NWCSAF software to process the SEVIRI/MSG images Exploitation of Meteorological Satellites (EUMETSAT). Among others, they provide the Cloud Type (CT) product (Derrien\n160\nand Le Gléau, 2005), which is a detailed scenery classification of clouds based on different main classes. The baseline data originate from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) operated onboard the second\ngeneration METEOSAT geostationary satellites (MSG). Multispectral thresholding techniques (Saunders and Kriebel, 1988;\nDerrien et al., 1993; Stowe et al., 1999) are subsequently applied in the NWCSAF software to process the SEVIRI/MSG images into the various NWC products. The product is available with a temporal resolution of 15 minutes and a nadir spatial resolution\n165\nof 3 km, compared to 11 km at the edge of the field of view. In this study, CT is used to identify the cloud type and cloud cover along the Aeolus tracks and to assess the quality of\nthe Aeolus wind products relative to the presence of clouds. More specifically, we identify the pixels closest to each track of\nAeolus and determine the average percentage of cloud cover at each altitude based on a cloud classification. According to this classification, a measurement bin is considered as cloudy, if it is situated within or below a cloud. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. The radiosondes launched on the Virgin Islands were organised by National Aeronautics and Space Administration (NASA)’s\nConvective Processes Experiment-Aerosols and Winds (CPEX-AW) campaign component of JATAC, with the University of\nUtah conducting the launches on Saint Croix and the University of Oklahoma conducting the launches from Puerto Rico. On Saint Croix, launches were conducted from Carambola between 19 August 2021 and 14 September 2021. Altogether 73\nlaunches were conducted, of which a total of seven radiosondes were used to validate Aeolus in this study. As for Sal, these\n145\nmeasurements were performed with the radiosonde instrument DFM-09 (GRAW). Lastly, 32 launches were conducted from The radiosondes launched on the Virgin Islands were organised by National Aeronautics and Space Administration (NASA)’s\nConvective Processes Experiment-Aerosols and Winds (CPEX-AW) campaign component of JATAC, with the University of\nUtah conducting the launches on Saint Croix and the University of Oklahoma conducting the launches from Puerto Rico. On Saint Croix, launches were conducted from Carambola between 19 August 2021 and 14 September 2021. Altogether 73\nlaunches were conducted, of which a total of seven radiosondes were used to validate Aeolus in this study. As for Sal, these\n145\nmeasurements were performed with the radiosonde instrument DFM-09 (GRAW). Lastly, 32 launches were conducted from\nthe University of Puerto Rico at Mayagüez (UPRM) campus between 26 August and 14 September 2021, 7 of which could be\nused for the validation of Aeolus. All launches were performed with iMet-4 radiosondes from the International Met System. As with DFM-09, the iMet-4 radiosondes provide measurements of wind speed, wind direction, temperature, humidity and ,\ng\np\ng\nlaunches were conducted, of which a total of seven radiosondes were used to validate Aeolus in this study. As for Sal, these\n145\nmeasurements were performed with the radiosonde instrument DFM-09 (GRAW). Lastly, 32 launches were conducted from\nthe University of Puerto Rico at Mayagüez (UPRM) campus between 26 August and 14 September 2021, 7 of which could be\nused for the validation of Aeolus. All launches were performed with iMet-4 radiosondes from the International Met System. As with DFM-09, the iMet-4 radiosondes provide measurements of wind speed, wind direction, temperature, humidity and launches were conducted, of which a total of seven radiosondes were used to validate Aeolus in this study. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 3.1\nCo-location criteria For the comparison of Aeolus against radiosonde profiles, several steps are required to fit the radiosonde wind measurements\nto the Aeolus measurement grid and to co-locate them in time and space. To ensure vertical consistency, the high-resolution radiosonde measurements are vertically averaged within the 24 range bins\n190\nas specified in the Aeolus L2B product. Subsequently, the radiosondes total horizontal wind speed VRS and direction φRS are\nprojected to the Aeolus HLOS (HLOSRS) using the azimuth angle φAEOLUS also specified in the L2B product, in accordance\nto\nHLOSRS = VRS × cos(φAEOLUS – φRS)\n(1) To ensure vertical consistency, the high-resolution radiosonde measurements are vertically averaged within the 24 range bins\n190\nas specified in the Aeolus L2B product. Subsequently, the radiosondes total horizontal wind speed VRS and direction φRS are\nprojected to the Aeolus HLOS (HLOSRS) using the azimuth angle φAEOLUS also specified in the L2B product, in accordance\nto\nHLOSRS = VRS × cos(φAEOLUS – φRS)\n(1) (1) Moreover, we have chosen co-location radii of up to 340 km, as we assume typical variations in zonal wind to be of a larger\n195\nscale. In fact, during boreal summer, African Easterly Waves (AEWs) and tropical disturbances dominate the tropospheric zonal\nwind variability over the tropical Atlantic, which generally have a horizontal wavelength of 2000-5000 km with a periodicity\nof 2-7 days (Belanger et al., 2016). Section 4.3.2 discusses the error dependencies related to co-location aspects in more detail. Moreover, we have chosen co-location radii of up to 340 km, as we assume typical variations in zonal wind to be of a larger\n195\nscale. In fact, during boreal summer, African Easterly Waves (AEWs) and tropical disturbances dominate the tropospheric zonal\nwind variability over the tropical Atlantic, which generally have a horizontal wavelength of 2000-5000 km with a periodicity\nof 2-7 days (Belanger et al., 2016). Section 4.3.2 discusses the error dependencies related to co-location aspects in more detail. Moreover, we have chosen co-location radii of up to 340 km, as we assume typical variations in zonal wind to be of a larger\n195\nscale. In fact, during boreal summer, African Easterly Waves (AEWs) and tropical disturbances dominate the tropospheric zonal\nwind variability over the tropical Atlantic, which generally have a horizontal wavelength of 2000-5000 km with a periodicity\nof 2-7 days (Belanger et al., 2016). Section 4.3.2 discusses the error dependencies related to co-location aspects in more detail. 2.4\nCAMS Dust products The dust-aerosol mixing ratio is thereby averaged along each track and projected onto Rayleigh-clear and\nMie-cloudy measurement bins to obtain an estimate of the dust concentration for each observation. 185 Mie-cloudy measurement bins to obtain an estimate of the dust concentration for each observation. 185 Mie-cloudy measurement bins to obtain an estimate of the dust concentration for each observation. 185 2.3\nEUMETSAT SAFNWC Cloud type product This refers to following\n170\nclasses for altitudes above 16 km (very high clouds), between 7 and 16 km (very high and high cloud types), between 3 and 7\nkm (very high, high, mid-level, low and fractional cloud types) and finally below 3 km (very high, high, mid-level, low, very\nlow and fractional cloud types). classification, a measurement bin is considered as cloudy, if it is situated within or below a cloud. This refers to following\n170\nclasses for altitudes above 16 km (very high clouds), between 7 and 16 km (very high and high cloud types), between 3 and 7\nkm (very high, high, mid-level, low and fractional cloud types) and finally below 3 km (very high, high, mid-level, low, very\nlow and fractional cloud types). classification, a measurement bin is considered as cloudy, if it is situated within or below a cloud. This refers to following\n170\nclasses for altitudes above 16 km (very high clouds), between 7 and 16 km (very high and high cloud types), between 3 and 7\nkm (very high, high, mid-level, low and fractional cloud types) and finally below 3 km (very high, high, mid-level, low, very\nlow and fractional cloud types). 6 6 2.4\nCAMS Dust products The fourth generation of ECMWF Global Atmospheric Composition Reanalysis (EAC4) (Inness et al., 2019) is produced by\n175\nthe Copernicus Atmosphere Monitoring Service (CAMS) with the main objective of global aerosol monitoring. EAC4 relies on\nECMWF’s IFS, which has been extended to predict and assimilate aerosols (Rémy et al., 2019), trace gases (Flemming et al.,\n2015; Huijnen et al., 2019) and greenhouse gases. The IFS meteorological and atmospheric composition models are combined\nwith data assimilation from satellite products using the 4D-Var data assimilation scheme in CY42R1. In particular, CAMS assimilates the Aerosol Optical Depth (AOD) at 550 nm derived from MODIS and the Polar Multi-Sensor Aerosol Optical\n180\nProperties (PMAp). Reanalysis outputs are provided on three-dimensional time-consistent fields interpolated on 25 pressure\nlevels, a horizontal resolution of about 80 km and a sub-daily time resolution of 6 hours. assimilates the Aerosol Optical Depth (AOD) at 550 nm derived from MODIS and the Polar Multi-Sensor Aerosol Optical\n180\nProperties (PMAp). Reanalysis outputs are provided on three-dimensional time-consistent fields interpolated on 25 pressure\nlevels, a horizontal resolution of about 80 km and a sub-daily time resolution of 6 hours. assimilates the Aerosol Optical Depth (AOD) at 550 nm derived from MODIS and the Polar Multi-Sensor Aerosol Optical\n180\nProperties (PMAp). Reanalysis outputs are provided on three-dimensional time-consistent fields interpolated on 25 pressure\nlevels, a horizontal resolution of about 80 km and a sub-daily time resolution of 6 hours. Similar to the SAFNWC CT, the dust-aerosol mixing ratio is used to assess the quality of the Aeolus wind products in\npresence of dust. The dust-aerosol mixing ratio is thereby averaged along each track and projected onto Rayleigh-clear and Properties (PMAp). Reanalysis outputs are provided on three-dimensional time-consistent fields interpolated on 25 pressure\nlevels, a horizontal resolution of about 80 km and a sub-daily time resolution of 6 hours. Similar to the SAFNWC CT, the dust-aerosol mixing ratio is used to assess the quality of the Aeolus wind products in\npresence of dust. The dust-aerosol mixing ratio is thereby averaged along each track and projected onto Rayleigh-clear and\nMie-cloudy measurement bins to obtain an estimate of the dust concentration for each observation. 185 Similar to the SAFNWC CT, the dust-aerosol mixing ratio is used to assess the quality of the Aeolus wind products in\npresence of dust. 3.2\nStatistical metrics Different metrics were used to validate and estimate the systematic and random error of Aeolus wind products. The bin-to-bin\n200\nwind speed difference between Aeolus and radiosonde along the HLOS is defined as Different metrics were used to validate and estimate the systematic and random error of Aeolus wind products. The bin-to-bin\n200\nwind speed difference between Aeolus and radiosonde along the HLOS is defined as\nΔdiffHLOS = (HLOSAEOLUS – HLOSRS)\n(2) Different metrics were used to validate and estimate the systematic and random error of Aeolus wind products. The bin-to-bin\n200\nwind speed difference between Aeolus and radiosonde along the HLOS is defined as\nΔdiffHLOS = (HLOSAEOLUS – HLOSRS)\n(2) wind speed difference between Aeolus and radiosonde along the HLOS is defined as\nΔdiffHLOS = (HLOSAEOLUS – HLOSRS)\n(2) (2) 7 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Thus, the bias μ is defined as the total mean difference μ = 1\nN\nN\nX\ni=1\nΔdiffHLOS (3) with the Mean Absolute Difference (MADI) yielding\n205 MADI = 1\nN\nN\nX\ni=1\n|ΔdiffHLOS| MADI = 1\nN\nN\nX\ni=1\n|ΔdiffHLOS| (4) and N the total number of data points. and N the total number of data points. and N the total number of data points. Additionally, we calculated the standard deviation of the difference STD =\nv\nu\nu\nt 1\ni – 1\nN\nX\ni=1\n(HLOSAEOLUS – HLOSRS)2 (5) and the scaled median absolute deviation (SMAD)\n210 SMAD = 1.4826 × median\n\u0000\f\fΔdiffHLOS – median(ΔdiffHLOS)\n\f\f\u0001 (6) The SMAD is equivalent to the standard deviation for a normal distribution of errors, but is often used in Aeolus validation\nstudies as it is less sensitive to individual outliers with very large differences than the standard deviation. Since the number of data points varies greatly depending on the measurement channel and height, we define the uncertainty\nof the mean bias εμ as\n215 (7) (7) https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. study is comparably homogeneous, we estimate the representativeness error for our comparison to be in the range of 1.5 ms–1\nto 2.5 ms–1. The radiosonde observation error σRS is estimated to be 0.7 ms–1 based on Dirksen et al. (2014). study is comparably homogeneous, we estimate the representativeness error for our comparison to be in the range of 1.5 ms–1\nto 2.5 ms–1. The radiosonde observation error σRS is estimated to be 0.7 ms–1 based on Dirksen et al. (2014). The representativeness and radiosonde observations errors also need to be considered when comparing the differences be-\ntween Aeolus and radiosonde observations with the expected error provided in the Aeolus data product (EEAeolus). To account\nfor this, we add the the radiosonde observation error and an estimated representativeness error of 2 ms–1 to achieve the total\nexpected error for the comparison (EEtot) as follows: The representativeness and radiosonde observations errors also need to be considered when comparing the differences be-\ntween Aeolus and radiosonde observations with the expected error provided in the Aeolus data product (EEAeolus). To account\nfor this, we add the the radiosonde observation error and an estimated representativeness error of 2 ms–1 to achieve the total\nexpected error for the comparison (EEtot) as follows: 230 EEtot =\nq\nEE2\nAeolus + σ2\nRS + σ2rep (9) 3.3\nRepresentativeness The difference between Aeolus and radiosonde observations is the sum of the Aeolus observation error, the radiosonde ob-\nservation error and the error arising from spatial and temporal displacement of the observations and different observation\ngeometries. The latter is usually referred to as representativeness error (Weissmann et al., 2005). As the three error components\n220\ncan be assumed to be uncorrelated, the standard deviation of the Aeolus HLOS winds observation error (σAeolus) can therefore\nbe calculated as The difference between Aeolus and radiosonde observations is the sum of the Aeolus observation error, the radiosonde ob-\nservation error and the error arising from spatial and temporal displacement of the observations and different observation servation error and the error arising from spatial and temporal displacement of the observations and different observation\ngeometries. The latter is usually referred to as representativeness error (Weissmann et al., 2005). As the three error components\n220\ncan be assumed to be uncorrelated, the standard deviation of the Aeolus HLOS winds observation error (σAeolus) can therefore\nbe calculated as geometries. The latter is usually referred to as representativeness error (Weissmann et al., 2005). As the three error components\n220\ncan be assumed to be uncorrelated, the standard deviation of the Aeolus HLOS winds observation error (σAeolus) can therefore\nbe calculated as σAeolus =\nq\nσ2tot – σ2\nRS – σ2rep (8) where σtot is the standard deviation of the total difference between Aeolus and radiosonde observations (STD), σRS is the standard deviation of the radiosonde observation error and σrep is the standard deviation of the representativeness error. Martin\n225\net al. (2021) estimated that the representativeness error for the comparison of Aeolus and radiosonde observations in mit-\nlatitudes is about 2.5 ms–1 based on high-resolution model simulations. As the wind fields in the area of the present validation standard deviation of the radiosonde observation error and σrep is the standard deviation of the representativeness error. Martin\n225\net al. (2021) estimated that the representativeness error for the comparison of Aeolus and radiosonde observations in mit-\nlatitudes is about 2.5 ms–1 based on high-resolution model simulations. As the wind fields in the area of the present validation 8 8 3.4\nQuality control\n235 Quality control (QC) is an important step in the evaluation of Aeolus wind errors. The aim is to check for the validity of the\nmeasurements and discard nonphysical wind results from the analysis process. The QC we apply here is based on the existing\nquality control recommendations (Rennie et al., 2020) from the Aeolus Data Science and Innovation Cluster (DISC), and\nprimarily rely on the HLOS wind error estimate (EE) in the L2B product and the validity flags. The Rayleigh channel EE is based on the uncertainty of the SNR spectrometer response and takes into account error propaga-\n240\ntion arising from the sensitivity of the Fabry-Perot interferometer, Poisson noise in the useful signal and the solar background. Ultimately, the Rayleigh EE is proportional to the inverse squared root of the useful signal on the detector. Future baseline\nversions will include additional noise terms, such as noise related to atmospheric temperature and pressure, or cross-talk con-\ntamination. In contrary, the Mie EE is determined from the accuracy of the fringe peak position using the solution covariance of the Lorentzian fitting algorithm based on four characteristics of the signal shape, i.e the peak position, height, width and\n245\noffset. Following the default QC flags, all Aeolus wind products with a validity flag of 0, EE above 8 ms–1 for Rayleigh and 4\nms–1 for Mie, are omitted. Nevertheless, the used QC might not be enough and the data algorithm may contain gross errors in\nthe wind estimate that have not been flagged as invalid. These errors are usually due to non-Gaussian error sources, such as of the Lorentzian fitting algorithm based on four characteristics of the signal shape, i.e the peak position, height, width and\n245\noffset. Following the default QC flags, all Aeolus wind products with a validity flag of 0, EE above 8 ms–1 for Rayleigh and 4\nms–1 for Mie, are omitted. Nevertheless, the used QC might not be enough and the data algorithm may contain gross errors in\nthe wind estimate that have not been flagged as invalid. These errors are usually due to non-Gaussian error sources, such as instrument/transmission failure, or to a misrepresentation of the measurements in space and time. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 4.1\nStatistical comparison of Aeolus with radiosonde observations and model winds In this section, the L2B HLOS winds (L2bP 3.50) from Aeolus are compared statistically with radiosonde observations and\n260\nmodel winds. This includes a comparison with the ECMWF model equivalents (subsection 4.1.1), an overview of systematic\nand random differences with respect to Cal/Val sites and orbital nodes (subsection 4.1.2), and finally the identification of an\norbital- and altitude-dependent bias in the Rayleigh-clear channel (subsection 4.1.3). In this section, the L2B HLOS winds (L2bP 3.50) from Aeolus are compared statistically with radiosonde observations and\n260\nmodel winds. This includes a comparison with the ECMWF model equivalents (subsection 4.1.1), an overview of systematic\nand random differences with respect to Cal/Val sites and orbital nodes (subsection 4.1.2), and finally the identification of an\norbital- and altitude-dependent bias in the Rayleigh-clear channel (subsection 4.1.3). The present study relies on a total of 384 Rayleigh-clear and 59 Mie-cloudy bin pairs, of which ∼60% and ∼53% are\nfrom ascending orbits, respectively, with the majority of observations obtained from the Caribbean launch sites (∼56% for\n265\nRayleigh-clear and ∼64% for Mie-cloudy). Rayleigh-cloudy bin pairs are also available, but only in a very small number (16\ncounts), which makes a statistical analysis difficult. 3.4\nQuality control\n235 Since the two aforementioned\n250\nQC are not sufficient to remove these gross errors, an additional QC parameter is used, namely the modified Z-score (Lux et al.,\n2022b; Witschas et al., 2022; Iglewicz and Hoaglin, 1993). The modified Z-score Zm,i is defined as instrument/transmission failure, or to a misrepresentation of the measurements in space and time. Since the two aforementioned\n250\nQC are not sufficient to remove these gross errors, an additional QC parameter is used, namely the modified Z-score (Lux et al.,\n2022b; Witschas et al., 2022; Iglewicz and Hoaglin, 1993). The modified Z-score Zm,i is defined as (10) and describes the median deviations between each wind speed difference normalized with the SMAD. The modified Z-score\nsignificantly influences small data sets, such as those used in this study. Following literature recommendations (Lux et al.,\n255\n2022b; Witschas et al., 2022; Sandbhor and Chaphalkar, 2019; Tripathy et al., 2013), we discard wind observations with a\nmodified Z-score greater than 3 as a final QC. significantly influences small data sets, such as those used in this study. Following literature recommendations (Lux et al.,\n255\n2022b; Witschas et al., 2022; Sandbhor and Chaphalkar, 2019; Tripathy et al., 2013), we discard wind observations with a\nmodified Z-score greater than 3 as a final QC. 9 The generally good agreement between radiosonde and model equivalent\nshows that the co-location parameters used in this study are reliable, as most of the systematic and random errors seem to be\nspecific to the Aeolus Rayleigh-clear data. This stresses the need to identify the underlying potential error sources of Rayleigh- in the model equivalent is smaller than for Aeolus Mie-cloudy winds, with biases of 0.4 ± 0.3 ms–1 and –0.9 ± 0.3 ms–1,\n280\nrespectively. For Rayleigh-cloudy, the STD is larger at 6.6 ms–1 with a bias of 1.0 ± 1.4 ms–1, but given the small statistical\nsample size, there is a risk of a large margin of error. The generally good agreement between radiosonde and model equivalent\nshows that the co-location parameters used in this study are reliable, as most of the systematic and random errors seem to be\nspecific to the Aeolus Rayleigh-clear data. This stresses the need to identify the underlying potential error sources of Rayleigh- clear observations with respect to the presence of clouds and dust aerosols, which are frequent in the region of interest. It is\n285\nalso worth noting that this good agreement indicates that the model equivalent is a robust reference for validating the Aeolus\nwinds in the tropical Atlantic. clear observations with respect to the presence of clouds and dust aerosols, which are frequent in the region of interest. It is\n285\nalso worth noting that this good agreement indicates that the model equivalent is a robust reference for validating the Aeolus\nwinds in the tropical Atlantic. 4.1.1\nComparative analysis with the ECMWF model equivalents 4.1.1\nComparative analysis with the ECMWF model equivalents Figure 1. (a) Aeolus HLOS Rayleigh-clear (blue), Mie-cloudy (red) and Rayleigh-cloudy (orange) wind products plotted against radiosonde\nmeasurements projected along the HLOS for the 20 radiosonde profiles. The gross errors (crosses) are determined using the modified Z-score\nwith a threshold of 3. (b) Aeolus HLOS model equivalents from the ECMWF feedback files plotted against radiosonde measurements. The\ndashed lines are located at the ±10 ms–1 and ±20 ms–1 wind speed difference between two measurements. Figure 1. (a) Aeolus HLOS Rayleigh-clear (blue), Mie-cloudy (red) and Rayleigh-cloudy (orange) wind products plotted against radiosonde\nmeasurements projected along the HLOS for the 20 radiosonde profiles. The gross errors (crosses) are determined using the modified Z-score\nwith a threshold of 3. (b) Aeolus HLOS model equivalents from the ECMWF feedback files plotted against radiosonde measurements. The\ndashed lines are located at the ±10 ms–1 and ±20 ms–1 wind speed difference between two measurements. 10 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 1 shows a scatter plot of the radiosonde HLOS (HLOSRS) against Aeolus L2B (HLOSAEOLUS) Rayleigh-clear (blue), Figure 1 shows a scatter plot of the radiosonde HLOS (HLOSRS) against Aeolus L2B (HLOSAEOLUS) Rayleigh-clear (blue),\nMie-cloudy (red) and Rayleigh-cloudy (orange) wind products (a) as well as against Aeolus ECMWF model equivalents\n270\n(HLOSECMWF) (b). Since Rayleigh-cloudy wind observations are not assimilated at ECMWF, they are not displayed in Fig. 1b. The × symbol represent the gross errors rejected with a Z-score threshold of 3 (∼3.5%, ∼4.8% and ∼6.7% of the\ntotal Rayleigh-clear, Mie-cloudy and Rayleigh-cloudy data points, respectively). The dashed lines represent the ±10 ms–1\nand ±20 ms–1 difference between two measurements. The Aeolus model equivalent HLOSECMWF for Rayleigh-clear shows a Mie-cloudy (red) and Rayleigh-cloudy (orange) wind products (a) as well as against Aeolus ECMWF model equivalents\n270\n(HLOSECMWF) (b). Since Rayleigh-cloudy wind observations are not assimilated at ECMWF, they are not displayed in Fig. 1b. The × symbol represent the gross errors rejected with a Z-score threshold of 3 (∼3.5%, ∼4.8% and ∼6.7% of the\ntotal Rayleigh-clear, Mie-cloudy and Rayleigh-cloudy data points, respectively). The dashed lines represent the ±10 ms–1\nand ±20 ms–1 difference between two measurements. The Aeolus model equivalent HLOSECMWF for Rayleigh-clear shows a much better agreement with the radiosonde measurements HLOSRS with a STD of 2.1 ms–1 (Fig. 1b) compared to the Aeolus\n275\nHLOSAEOLUS Rayleigh-clear observations, which have a larger spread and a STD of 4.8 ms–1 (Fig. 1a). The systematic\ndifference of the model equivalent is also smaller with a bias of 0.1 ± 0.1 ms–1 compared to –0.5 ± 0.2 ms–1 for the Aeolus\nobservations. In contrast, the Mie-cloudy winds of both Aeolus model equivalents and HLOSAEOLUS behave similarly with\nrespect to the radiosonde measurements, with STD of 2.93 ms–1 and 2.9 ms–1, respectively. Again, the systematic difference in the model equivalent is smaller than for Aeolus Mie-cloudy winds, with biases of 0.4 ± 0.3 ms–1 and –0.9 ± 0.3 ms–1,\n280\nrespectively. For Rayleigh-cloudy, the STD is larger at 6.6 ms–1 with a bias of 1.0 ± 1.4 ms–1, but given the small statistical\nsample size, there is a risk of a large margin of error. 4.1.2\nSystematic and random errors using radiosondes An overview of the bias and random differences of both channels can be found in Table 2. In terms of systematic errors,\nRayleigh-clear shows a relatively small negative bias of –0.5 ± 0.2 ms–1, on average, which is below ESA’s specification of\n290\n0.7 ms–1 (Ingmann and Straume, 2016). This bias is, however, the result of a large heterogeneity with respect to the Cal/Val\nsites and orbital nodes, with compensating biases of –1.5±0.6 ms–1 and 0.6±0.4 ms–1 for the descending and ascending nodes\non Sal, respectively, compared to negative biases of –1.0±0.3 ms–1 (ascending) and –0.6±0.4 ms–1 (descending) in the Virgin\nIsland. As for random differences, Rayleigh-clear has an average STD of 4.8 ms–1, which varies only marginally between the Rayleigh-clear shows a relatively small negative bias of –0.5 ± 0.2 ms–1, on average, which is below ESA’s specification of\n290\n0.7 ms–1 (Ingmann and Straume, 2016). This bias is, however, the result of a large heterogeneity with respect to the Cal/Val\nsites and orbital nodes, with compensating biases of –1.5±0.6 ms–1 and 0.6±0.4 ms–1 for the descending and ascending nodes\non Sal, respectively, compared to negative biases of –1.0±0.3 ms–1 (ascending) and –0.6±0.4 ms–1 (descending) in the Virgin\nIsland. As for random differences, Rayleigh-clear has an average STD of 4.8 ms–1, which varies only marginally between the Cal/Val sites and orbital nodes, ranging from 4.1 ms–1 to 5.3 ms–1. The overall SMAD is found to be slightly below at 4.3 ms–1. 295\nFor comparison with the ESA recommendation for random errors, we derived the random errors for Aeolus observations\nconsidering also the representativeness errors for the comparison and radiosonde observation errors according to Eq. 8 (table\n3). The random error at 2–16 km altitude of 3.8 – 4.3 ms–1 exceeds the threshold of 2.5 ms–1, while at 16–20 km altitude it\namounts to 4.3 – 4.8 ms–1, also exceeding the ESA threshold of 3 ms–1. The quality of Rayleigh-clear measurements primarily depends on the signal accumulation, which can vary with the thickness of the RBS and the horizontal accumulation length\n300\nas well as with the atmospheric path signal. The latter has been decreasing in recent years as a result of initial instrumental depends on the signal accumulation, which can vary with the thickness of the RBS and the horizontal accumulation length\n300\nas well as with the atmospheric path signal. 4.1.2\nSystematic and random errors using radiosondes The latter has been decreasing in recent years as a result of initial instrumental 11 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Table 2. Overview of the mean bias and uncertainty (μ, σμ; ms–1), Standard deviation (STD; ms–1), Scaled Median Absolute Deviation\n(SMAD; ms–1) and counts (COUNT) for the Rayleigh-clear and Mie-cloudy channels, orbital nodes and the different radiosonde locations. Due to the small amount of available data, Rayleigh-cloudy is not shown here. Table 2. Overview of the mean bias and uncertainty (μ, σμ; ms–1), Standard deviation (STD; ms–1), Scaled Median Absolute Deviation\n(SMAD; ms–1) and counts (COUNT) for the Rayleigh-clear and Mie-cloudy channels, orbital nodes and the different radiosonde locations. Due to the small amount of available data, Rayleigh-cloudy is not shown here. Region\nOrbital node\nRayleigh-clear\nMie-cloudy\nμ\nSTD\nSMAD\nCOUNT\nμ\nSTD\nSMAD\nCOUNT\nAscending\n0.6±0.4\n4.9\n4.4\n112\n-1±0.9\n2.9\n3.5\n15\nSal\nDescending\n-1.5±0.6\n4.6\n4.8\n55\n-1.6±0.8\n2.2\n2.1\n6\nAll\n-0.1±0.3\n4.9\n4.5\n167\n-1.2±0.7\n2.7\n3.2\n21\nAscending\n-1.0±0.3\n4.1\n3.7\n119\n-0.6±0.7\n2.9\n3.7\n16\nSCRX/PR\nDescending\n-0.6±0.4\n5.3\n4.3\n98\n-1.0±0.5\n2.9\n2.5\n22\nAll\n-0.8±0.3\n4.7\n4.3\n217\n-0.8±0.4\n2.9\n2.5\n38\nAscending\n-0.2±0.3\n4.6\n4.2\n231\n-0.8±0.6\n2.9\n3.3\n31\nSal/SCRX/PR\nDescending\n-0.9±0.4\n5.0\n4.6\n153\n-1.1±0.4\n2.8\n2.2\n28\nAll\n-0.5±0.2\n4.8\n4.3\n384\n-0.9±0.3\n2.9\n2.6\n59 misalignment, laser-induced contamination, as well as the wavefront error of the 1.5 m telescope. The solar background noise,\nwhich varies along the orbit and season, can also affect the quality of the Rayleigh-clear measurements. ESA’s specification and more uniform across regions and orbital nodes with a slightly larger bias in the descending orbits\n305\nand over Sal. Concerning the random differences, the measurements exhibit a total random error of 1.1 – 2.3 ms–1, which is\nbelow ESA’s 2–16 km recommendation, as most Mie-cloudy measurements are located underneath 16 km altitude. As with the\nbias, the STD and SMAD of Mie-cloudy are also quite independent of orbital and regional dependence. The overall accuracy\nof Mie-cloudy depends on the signal accumulation, the classification algorithm and the quality of the calibration data. The accuracy of Mie-cloudy winds is higher than that of Rayleigh-clear winds as particle backscatter is usually stronger than that\n310\nof clear air in addition to the fact that Mie backscatter is not subject to broadening induced by Rayleigh-Brillouin scattering\n(Witschas et al., 2012). https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Table 3. Overview of the total systematic (μ, σμ; ms–1) and random (σAeolus; ms–1) errors derived according to Eq. 8 for Rayleigh-clear\nand Mie-cloudy winds for altitudes ranges 2–16km and 16–20km, as well as the corresponding ESA’s error recommendations. The random\nerror σAeolus was computed for a representativeness error σrep ranging from 1.5 ms–1 to 2.5 ms–1. For Mie-cloudy, only the altitude range\n2–16km is shown for the random error, as Mie-cloudy does not sample sufficiently above 16km. Rayleigh-clear\nMie-cloudy\nσAeolus 2–16km\nσAeolus 16–20km\nμ\nσAeolus 2–16km\nμ\nAscending\n3.4 – 3.9\n4.0 – 4.4\n-0.2±0.3\n1.1 – 2.3\n-0.8±0.6\nDescending\n4.3 – 4.7\n4.4 – 4.9\n-0.9±0.4\n0.5 – 2.1\n-1.1±0.4\nAll\n3.8 – 4.3\n4.3 – 4.8\n-0.5±0.2\n1.1 – 2.3\n-0.9±0.3\nESA\n2.5\n3\n0.7\n2.5\n0.7 The difference in results is caused by the different altitudes at which the data are sampled, as the The difference in results is caused by the different altitudes at which the data are sampled, as the aircraft only samples the lower\n10 km portion of the troposphere, which is shown to be more noisy owing to the abundance of dust aerosols in this region. 320\nFor Mie-cloudy, the random error gives 2.9 ± 0.3 ms–1, which is similar to our radiosonde-based results as most Mie-cloudy\nscattering occurs at lower levels. 320 Comparing the results of different Cal/Val studies is tricky as the influence of geographical regions, atmospheric conditions,\ndecreasing laser energy, product baseline and quality control procedures on the result can be significant and must be considered. accuracy of Mie-cloudy winds is higher than that of Rayleigh-clear winds as particle backscatter is usually stronger than that\n310\nof clear air in addition to the fact that Mie backscatter is not subject to broadening induced by Rayleigh-Brillouin scattering\n(Witschas et al., 2012). Comparing the results of different Cal/Val studies is tricky as the influence of geographical regions, atmospheric conditions,\ndecreasing laser energy, product baseline and quality control procedures on the result can be significant and must be considered. In this analysis, comparisons are only made with statistics derived from AVATAR-T airborne-based measurements (Witschas\n315\net al., 2022; Lux et al., 2022b), as these were carried out in the framework of the same JATAC campaign. The statistical analysis\nof AVATAR-T shows systematic errors of –0.1 ± 0.3 ms–1 for Rayleigh-clear and –0.7 ± 0.2 ms–1 for Mie-cloudy, which are\nslightly smaller than for radiosondes. However, the random error of 7.1 ± 0.3 ms–1 for Rayleigh-clear is significantly higher. In this analysis, comparisons are only made with statistics derived from AVATAR-T airborne-based measurements (Witschas\n315\net al., 2022; Lux et al., 2022b), as these were carried out in the framework of the same JATAC campaign. The statistical analysis\nof AVATAR-T shows systematic errors of –0.1 ± 0.3 ms–1 for Rayleigh-clear and –0.7 ± 0.2 ms–1 for Mie-cloudy, which are\nslightly smaller than for radiosondes. However, the random error of 7.1 ± 0.3 ms–1 for Rayleigh-clear is significantly higher. 12 4.1.3\nOrbital bias in the Rayleigh-clear channel Figure 2 shows vertical profiles of the differences between Aeolus Rayleigh-clear observations and radiosonde measurements\nprojected along HLOS (O-RS; solid lines), and the corresponding ECMWF model equivalents (O-B; dotted lines) for both\n325\nascending (red) and descending (blue) orbits over Sal (a), PR and SCRX (b). The shading represents the bias uncertainty σμ. HLOS winds from the descending track are multiplied by -1 to conform with the sign convention of the model coordinate\nsystem. The vertical profiles illustrate the presence of an ascending/descending bias visible in both the O-B and O-RS profiles,\nreaching up to 2.5 ms–1 around 8 km altitude in both regions. The differences below 5 km altitude could be related to the greater amount of dust in Cabo Verde during this period, while above 17 km the differences could partly be related to the lack\n330\nof descending orbit data over Sal (Fig. 2a). This altitude- and orbit-dependent bias was already described by Borne et al. (2023)\nusing first-guess departure statistics over West Africa. greater amount of dust in Cabo Verde during this period, while above 17 km the differences could partly be related to the lack\n330\nof descending orbit data over Sal (Fig. 2a). This altitude- and orbit-dependent bias was already described by Borne et al. (2023)\nusing first-guess departure statistics over West Africa. using first-guess departure statistics over West Africa. This latitude consistent bias caused the zonal winds in the ECMWF analysis to accelerate in the morning and weaken in\nthe evening, affecting the African Easterly Jet (AEJ) and Tropical Easterly Jet (TEJ) in particular. Correcting this bias with This latitude consistent bias caused the zonal winds in the ECMWF analysis to accelerate in the morning and weaken in\nthe evening, affecting the African Easterly Jet (AEJ) and Tropical Easterly Jet (TEJ) in particular. Correcting this bias with a temperature-dependent approach helped to improve the representation of winds in the analysis and forecast fields (Borne\n335\net al., 2023). However, the cause of this bias remains unknown, as it has not been proven to be related to temperature, nor has\nany dependence on wind speed, SNR or useful signal been found (not shown here). Here, as both the O-B and O-RS profiles\nare very close to each other, with deviations below 0.5 ms–1, the existence of this bias can be confirmed observationally with\nradiosondes. As highlighted by Horányi et al. 4.1.3\nOrbital bias in the Rayleigh-clear channel Differences (dots) and average differences (lines) between ascending (red) and descending (blue) winds between Aeolus observa-\ntions (O) and radiosonde HLOS wind measurements (RS, solid line) along with ECMWF model equivalents (B, dotted line) over Sal (a) and\nPuerto Rico - Saint Croix (PR/SCRX, (b). The shadow represents the bias uncertainty σμ. To comply with the sign convention of the model\ncoordinate system, the HLOS winds from the descending orbit are multiplied by -1. Figure 2. Differences (dots) and average differences (lines) between ascending (red) and descending (blue) w Figure 2. Differences (dots) and average differences (lines) between ascending (red) and descending (blue) winds between Aeolus observa-\ntions (O) and radiosonde HLOS wind measurements (RS, solid line) along with ECMWF model equivalents (B, dotted line) over Sal (a) and\nPuerto Rico - Saint Croix (PR/SCRX, (b). The shadow represents the bias uncertainty σμ. To comply with the sign convention of the model\ncoordinate system, the HLOS winds from the descending orbit are multiplied by -1. 4.1.3\nOrbital bias in the Rayleigh-clear channel (2015), biases of the order of 1 ms–1 can already deteriorate forecast quality. a temperature-dependent approach helped to improve the representation of winds in the analysis and forecast fields (Borne\n335\net al., 2023). However, the cause of this bias remains unknown, as it has not been proven to be related to temperature, nor has\nany dependence on wind speed, SNR or useful signal been found (not shown here). Here, as both the O-B and O-RS profiles\nare very close to each other, with deviations below 0.5 ms–1, the existence of this bias can be confirmed observationally with\nradiosondes. As highlighted by Horányi et al. (2015), biases of the order of 1 ms–1 can already deteriorate forecast quality. a temperature-dependent approach helped to improve the representation of winds in the analysis and forecast fields (Borne\n335\net al., 2023). However, the cause of this bias remains unknown, as it has not been proven to be related to temperature, nor has\nany dependence on wind speed, SNR or useful signal been found (not shown here). Here, as both the O-B and O-RS profiles\nare very close to each other, with deviations below 0.5 ms–1, the existence of this bias can be confirmed observationally with\nradiosondes. As highlighted by Horányi et al. (2015), biases of the order of 1 ms–1 can already deteriorate forecast quality. 13 Figure 2. Differences (dots) and average differences (lines) between ascending (red) and descending (blue) winds between Aeolus observa-\ntions (O) and radiosonde HLOS wind measurements (RS, solid line) along with ECMWF model equivalents (B, dotted line) over Sal (a) and\nPuerto Rico\nSaint Croix (PR/SCRX (b) The shadow represents the bias uncertainty\nTo comply with the sign convention of the model Figure 2. Differences (dots) and average differences (lines) between ascending (red) and descending (blue) winds between Aeolus observa-\ntions (O) and radiosonde HLOS wind measurements (RS, solid line) along with ECMWF model equivalents (B, dotted line) over Sal (a) and\nPuerto Rico - Saint Croix (PR/SCRX, (b). The shadow represents the bias uncertainty σμ. To comply with the sign convention of the model\ncoordinate system, the HLOS winds from the descending orbit are multiplied by -1. Figure 2. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Rayleigh-clear and Rayleigh-cloudy Figure 3 shows the absolute difference between Aeolus and radiosonde measurement points |ΔdiffHLOS| as a function of EEtot\n(a), altitude (b), co-location radius (c) and co-location time (d) for the Rayleigh-clear (blue) and Rayleigh-cloudy (orange)\nobservation types. The solid and dashed blue lines show the Rayleigh-clear MADI and SMAD, respectively, with each value\ncalculated using a minimum sample size of 40 data points for panels a, b and d. Also shown are outliers (cross symbol +),\n350\nthat we define in this study as values with low EE (< 5 ms–1) and large absolute difference (> 10 ms–1), which are of particular\ninterest as they contribute the most to the wind quality degradation. The Rayleigh-clear outliers account for 13 observations,\ni.e. ∼3.4% of the data points. For Rayleigh-cloudy, no MADI and SMAD are computed due to the lack of data. Figure 3 shows the absolute difference between Aeolus and radiosonde measurement points |ΔdiffHLOS| as a function of EEtot\n(a), altitude (b), co-location radius (c) and co-location time (d) for the Rayleigh-clear (blue) and Rayleigh-cloudy (orange)\nobservation types. The solid and dashed blue lines show the Rayleigh-clear MADI and SMAD, respectively, with each value observation types. The solid and dashed blue lines show the Rayleigh-clear MADI and SMAD, respectively, with each value\ncalculated using a minimum sample size of 40 data points for panels a, b and d. Also shown are outliers (cross symbol +),\n350\nthat we define in this study as values with low EE (< 5 ms–1) and large absolute difference (> 10 ms–1), which are of particular\ninterest as they contribute the most to the wind quality degradation. The Rayleigh-clear outliers account for 13 observations,\ni.e. ∼3.4% of the data points. For Rayleigh-cloudy, no MADI and SMAD are computed due to the lack of data. In general, the MADI and SMAD between Rayleigh-clear and radiosonde wind measurements appear to be proportional to In general, the MADI and SMAD between Rayleigh-clear and radiosonde wind measurements appear to be proportional to\nthe Aeolus EEtot (Fig. 3a), with larger deviations associated with larger EEtots, as expected. However, on average, the mean\n355\nEEtot overestimates the MADI by 1 ms–1 for EEtot values below 6 ms–1 (see grey line). This discrepancy can be attributed to\nthe relatively small amount of data used in the study, as the EE is based on the Gaussian assumption of a large data set. 4.2\nError dependency\n340 In this section we examine the error dependency and associated error sources of the different Aeolus wind products. Firstly,\nwe investigate the error dependency as a function of co-location parameters, such as radius and time difference between two\nmeasurement points, to account for representativeness. Secondly, we explore the error dependency in relation to the presence\nof clouds and dust, as these supposedly influence the quality of Aeolus wind products. 14 Rayleigh-clear and Rayleigh-cloudy For\nRayleigh-cloudy measurements, it is difficult to establish a dependency although the absolute difference appears to be generally\nlarger owing to the large STD of 6.6 ms–1 for this observation type. Considering the altitude error dependency of Rayleigh- the Aeolus EEtot (Fig. 3a), with larger deviations associated with larger EEtots, as expected. However, on average, the mean\n355\nEEtot overestimates the MADI by 1 ms–1 for EEtot values below 6 ms–1 (see grey line). This discrepancy can be attributed to\nthe relatively small amount of data used in the study, as the EE is based on the Gaussian assumption of a large data set. For\nRayleigh-cloudy measurements, it is difficult to establish a dependency although the absolute difference appears to be generally\nlarger owing to the large STD of 6.6 ms–1 for this observation type. Considering the altitude error dependency of Rayleigh- clear (Fig. 3b), a general pattern emerges with MADI and SMAD reaching a minimum of 3 ms–1 and 2 ms–1 respectively on\n360\naverage in the middle troposphere at 10 km, while increasing above and below, with MADIs of 4–5 ms–1 and SMADs of almost\n6 ms–1 at 2.5 km and 19 km altitude. As we will see in the next subsection 4.2.2, this error pattern is inversely proportional\nto the Rayleigh backscattered useful signal, as it directly affects the SNR and thereby the quality of the measurement points. Rayleigh-clear outliers seem to occur at all altitudes and Rayleigh-cloudy measurements are primarily found in the lower clear (Fig. 3b), a general pattern emerges with MADI and SMAD reaching a minimum of 3 ms–1 and 2 ms–1 respectively on\n360\naverage in the middle troposphere at 10 km, while increasing above and below, with MADIs of 4–5 ms–1 and SMADs of almost\n6 ms–1 at 2.5 km and 19 km altitude. As we will see in the next subsection 4.2.2, this error pattern is inversely proportional\nto the Rayleigh backscattered useful signal, as it directly affects the SNR and thereby the quality of the measurement points. Rayleigh-clear outliers seem to occur at all altitudes and Rayleigh-cloudy measurements are primarily found in the lower troposphere, below 6 km. 365\nIn Fig. 3c we examine the error dependency with respect to the co-location radius, which extends up to 340 km, a distance\nthat is large relative to the 100 km specified in ESA’s recommendations. Rayleigh-clear and Rayleigh-cloudy However, the MADI and SMAD for Rayleigh-clear\ndo not increase with radius, but stagnate at an average of 3–4 ms–1 for radii above 100 km, while they are slightly higher\nbelow 100 km, reaching 4–5 ms–1. Furthermore, outliers appear across all co-location radii. This indicates that the use of a troposphere, below 6 km. 365\nIn Fig. 3c we examine the error dependency with respect to the co-location radius, which extends up to 340 km, a distance\nthat is large relative to the 100 km specified in ESA’s recommendations. However, the MADI and SMAD for Rayleigh-clear\ndo not increase with radius, but stagnate at an average of 3–4 ms–1 for radii above 100 km, while they are slightly higher\nbelow 100 km, reaching 4–5 ms–1. Furthermore, outliers appear across all co-location radii. This indicates that the use of a troposphere, below 6 km. 365\nIn Fig. 3c we examine the error dependency with respect to the co-location radius, which extends up to 340 km, a distance\nthat is large relative to the 100 km specified in ESA’s recommendations. However, the MADI and SMAD for Rayleigh-clear\ndo not increase with radius, but stagnate at an average of 3–4 ms–1 for radii above 100 km, while they are slightly higher\nbelow 100 km, reaching 4–5 ms–1. Furthermore, outliers appear across all co-location radii. This indicates that the use of a co-location distance up to 340 km is acceptable for the statistical comparison. Exploring the error dependency with respect to\n370\nthe time difference between the observations (Fig. 3d), there is indication for increasing difference for larger time-differences,\ngoing from 3–4 ms–1 at 0 minutes to 4–6 ms–1 above 30 minutes. There is also an asymmetry of the error dependence, with a\nlarger error magnitude for radiosonde observations preceding the Aeolus passage. Since most radiosondes were launched with\nthe objective of reaching the mid-troposphere during the satellite’s passage, the measurements preceding/following Aeolus of co-location distance up to 340 km is acceptable for the statistical comparison. Exploring the error dependency with respect to\n370\nthe time difference between the observations (Fig. 3d), there is indication for increasing difference for larger time-differences,\ngoing from 3–4 ms–1 at 0 minutes to 4–6 ms–1 above 30 minutes. There is also an asymmetry of the error dependence, with a\nlarger error magnitude for radiosonde observations preceding the Aeolus passage. Rayleigh-clear and Rayleigh-cloudy EEtot (a), altitude (b), co-location radius (c) and co-location time quantities expressed as a function of the absolute difference\nbetween radiosonde HLOS winds (HLOSRS) and Aeolus (HLOSAEOLUS) Rayleigh-clear (blue) and Rayleigh-cloudy (orange) observations. Outliers are defined as values with an EE below 5 ms–1 and absolute difference larger than 10 ms–1 and are represented by the cross symbol\n+. The solid blue lines indicate the MADI while the dotted blue lines represent the SMAD of Rayleigh-clear and each value is computed\nusing a minimum sample size of 40 data points. The grey line in panel a represents the diagonal at intercept 0 with slope 1. Due to the limited\namount of data, no MADI and SMAD are shown for Rayleigh-cloudy. Figure 3. EEtot (a), altitude (b), co-location radius (c) and co-location time quantities expressed as a function of the absolute difference\nbetween radiosonde HLOS winds (HLOSRS) and Aeolus (HLOSAEOLUS) Rayleigh-clear (blue) and Rayleigh-cloudy (orange) observations. Outliers are defined as values with an EE below 5 ms–1 and absolute difference larger than 10 ms–1 and are represented by the cross symbol\n+. The solid blue lines indicate the MADI while the dotted blue lines represent the SMAD of Rayleigh-clear and each value is computed\nusing a minimum sample size of 40 data points. The grey line in panel a represents the diagonal at intercept 0 with slope 1. Due to the limited\namount of data, no MADI and SMAD are shown for Rayleigh-cloudy. +. The solid blue lines indicate the MADI while the dotted blue lines represent the SMAD of Rayleigh-cle\nusing a minimum sample size of 40 data points. The grey line in panel a represents the diagonal at intercept 0\namount of data, no MADI and SMAD are shown for Rayleigh-cloudy. Rayleigh-clear and Rayleigh-cloudy Since most radiosondes were launched with\nthe objective of reaching the mid-troposphere during the satellite’s passage, the measurements preceding/following Aeolus of co-location distance up to 340 km is acceptable for the statistical comparison. Exploring the error dependency with respect to\n370\nthe time difference between the observations (Fig. 3d), there is indication for increasing difference for larger time-differences,\ngoing from 3–4 ms–1 at 0 minutes to 4–6 ms–1 above 30 minutes. There is also an asymmetry of the error dependence, with a\nlarger error magnitude for radiosonde observations preceding the Aeolus passage. Since most radiosondes were launched with\nthe objective of reaching the mid-troposphere during the satellite’s passage, the measurements preceding/following Aeolus of more than 30 minutes correspond mainly to measurements at lower/higher altitudes. The larger MADI and SMADI values for\n375\nthese time differences could hence be an indirect effect of the larger errors found at those altitudes (Fig. 3b). Again, no error\ndependency is observed for outliers, with most occurring below ±40 minutes time differences. more than 30 minutes correspond mainly to measurements at lower/higher altitudes. The larger MADI and SMADI values for\n375\nthese time differences could hence be an indirect effect of the larger errors found at those altitudes (Fig. 3b). Again, no error\ndependency is observed for outliers, with most occurring below ±40 minutes time differences. 15 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 3. EEtot (a), altitude (b), co-location radius (c) and co-location time quantities expressed as a function of the absolute difference\nbetween radiosonde HLOS winds (HLOSRS) and Aeolus (HLOSAEOLUS) Rayleigh-clear (blue) and Rayleigh-cloudy (orange) observations. Outliers are defined as values with an EE below 5 ms–1 and absolute difference larger than 10 ms–1 and are represented by the cross symbol\n+. The solid blue lines indicate the MADI while the dotted blue lines represent the SMAD of Rayleigh-clear and each value is computed\nusing a minimum sample size of 40 data points. The grey line in panel a represents the diagonal at intercept 0 with slope 1. Due to the limited\namount of data, no MADI and SMAD are shown for Rayleigh-cloudy. Figure 3. Mie-cloudy Figure 4 shows the same error dependencies as in Fig. 3, but for the Mie-cloudy observation type. For Mie-cloudy, we define\noutliers as values exceeding an absolute error of 6 ms–1 along with EEs inferior to 3 ms–1. With a total of 3 data points, they\n380\naccount for ∼5% of the total Mie-cloudy observations. In panels a, b and d, each MADI and SMAD value is calculated using\na minimum sample size of 15 data points As shown in Fig. 4a, the absolute differences for Mie-cloudy measurements are generally smaller than for Rayleigh-clear,\nwith the largest deviations around 7–8 ms–1, while attaining 13–14 ms–1 for Rayleigh-clear. The MADI and SMAD remain\nbetween 2 and 3 ms–1, indicating an overestimation of the EEtot, especially for increasing EEtot. Regarding the altitude error\n385 16 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. dependency (Fig. 4b), most of the data are found within the 10-15 km layer, which is probably related to the presence of high-\nlevel clouds, and below 7 km, where low- and mid-level clouds and dust layers are found. Due to the sparseness of Mie-cloudy\ndata, both MADI and SMADI do not show a specific vertical error trend. While MADI and SMAD remain between 2.3 and 2.7\nms–1, respectively, they decrease to 1.8 and 2.3 between 1.5 and 3 km altitude before increasing to almost 3 ms–1 in the lowest\n1 km. Fig. 4c shows that similarly to Rayleigh-clear, Mie-cloudy reveals no error dependency with respect to co-location radii,\n390\nwith the mean absolute error and SMAD mainly ranging from 1.7 to 3.2 ms–1, and outliers found at all radii. Regarding the\nerror dependence on time difference (Fig. 4d), we find that most of the measurement differences occur at time intervals of\nless than ±40 minutes. MADIs and SMADs are generally higher for negative co-location times, corresponding to cases where\nradiosonde observations are sampled before those from Aeolus. Nevertheless, we do not notice a strong relationship between\nco location time and errors\n395 data, both MADI and SMADI do not show a specific vertical error trend. While MADI and SMAD remain between 2.3 and 2.7\nms–1, respectively, they decrease to 1.8 and 2.3 between 1.5 and 3 km altitude before increasing to almost 3 ms–1 in the lowest\n1 km. Fig. 4c shows that similarly to Rayleigh-clear, Mie-cloudy reveals no error dependency with respect to co-location radii,\n390\nwith the mean absolute error and SMAD mainly ranging from 1.7 to 3.2 ms–1, and outliers found at all radii. Regarding the\nerror dependence on time difference (Fig. 4d), we find that most of the measurement differences occur at time intervals of\nless than ±40 minutes. MADIs and SMADs are generally higher for negative co-location times, corresponding to cases where\nradiosonde observations are sampled before those from Aeolus. Nevertheless, we do not notice a strong relationship between\nl\nti\nti\nd\n395 1 km. Fig. 4c shows that similarly to Rayleigh-clear, Mie-cloudy reveals no error dependency with respect to co-location radii,\n390\nwith the mean absolute error and SMAD mainly ranging from 1.7 to 3.2 ms–1, and outliers found at all radii. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 4.2.2\nCloud type and dust As already mentioned, the accuracy of Rayleigh-clear and, to a lesser extent, Mie-cloudy depends on the signal level and\nSNR. In general, the signal level depends on the range bin thickness, the horizontal accumulation length, the atmospheric\npath signal and the overall signal background level. In addition, Rayleigh-clear winds are sensitive to signal attenuation due\nto atmospheric conditions, with weaker signal return under optically thick clouds and dust-aerosol layers. Mie-cloudy is less\n400\nconcerned as backscatter from particles is stronger, although it is sensitive to weak backscatter, e.g. from dust layers. Because\nof its strong sensitivity on signal levels, the EE of Rayleigh-clear only considers Poisson noise and is therefore inversely\nproportional to the squared root of the useful signal. For Mie-cloudy, this rule of thumb is not true. In this context, we aim to\ninvestigate the quality of the Rayleigh-clear and Mie-cloudy winds and the reliability of the corresponding EE with respect to the presence of clouds and dust. 405 Rayleigh-clear Table 4 describes the error dependency of the Rayleigh-clear observations with respect to the presence of clouds and dust,\nwith cases below 50%, above 50% and above 75% of cloudiness, as well as sub-categories distinguishing the dust mixing\nratio above (Dust) and below (DustNO) 10–8 kgkg–1. Note that SMAD is not used for this analysis as this reliably removes Table 4 describes the error dependency of the Rayleigh clear observations with respect to the presence of clouds and dust,\nwith cases below 50%, above 50% and above 75% of cloudiness, as well as sub-categories distinguishing the dust mixing\nratio above (Dust) and below (DustNO) 10–8 kgkg–1. Note that SMAD is not used for this analysis as this reliably removes\noutliers, which ought to be quantified here. We note that the MADI, the STD, and the EEtot all increase with the amount of\n410\nclouds and dust along the track, presumably due to the reduced return signal. In non-dusty conditions (DustNO), we observe\nthat for low cloud cover (<50%), the MADI (3.3 ± 0.2 ms–1) is significantly lower than the EEtot (4.8 ms–1) with a difference\nof 1.5 ms–1, while for higher cloud cover, the difference between MADI and EEtot is much smaller (1.1 ms–1 and 1.0 ms–1\nfor above 50% and 75% of cloudiness, respectively). This phenomenon is further enhanced at higher dust concentrations, with\n1\n1 NO\ng g\ny\ny\noutliers, which ought to be quantified here. We note that the MADI, the STD, and the EEtot all increase with the amount of\n410\nclouds and dust along the track, presumably due to the reduced return signal. In non-dusty conditions (DustNO), we observe\nthat for low cloud cover (<50%), the MADI (3.3 ± 0.2 ms–1) is significantly lower than the EEtot (4.8 ms–1) with a difference\nof 1.5 ms–1, while for higher cloud cover, the difference between MADI and EEtot is much smaller (1.1 ms–1 and 1.0 ms–1\nfor above 50% and 75% of cloudiness, respectively). This phenomenon is further enhanced at higher dust concentrations, with the MADI reaching even higher values (5.7 ± 0.8 ms–1) than the EEtot (5.8 ms–1) for cloud cover above 75%. This highlights\n415\nhow the EEtot in clear sky conditions is well calibrated, while it is becoming gradually too low with the increasing presence of\nclouds and dust. Regarding the\nerror dependence on time difference (Fig. 4d), we find that most of the measurement differences occur at time intervals of\nless than ±40 minutes. MADIs and SMADs are generally higher for negative co-location times, corresponding to cases where\nradiosonde observations are sampled before those from Aeolus. Nevertheless, we do not notice a strong relationship between Figure 4. Same as for Fig. 3, but for Mie-cloudy. For Mie-cloudy (red), outliers are defined as values having an absolute error above 6 ms–1\nand an EE inferior to 3 ms–1. The MADI and the SMAD values are computed using a minimum sample size of 15 data points. Figure 4. Same as for Fig. 3, but for Mie-cloudy. For Mie-cloudy (red), outliers are defined as values having an absolute error above 6 ms–1\nand an EE inferior to 3 ms–1. The MADI and the SMAD values are computed using a minimum sample size of 15 data points. 17 17 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Table 4. Overview of the total Error Estimate (EEtot; ms–1), mean absolute difference and uncertainty (MADI, σμ; ms–1), Standard deviation\n(STD; ms–1) and counts (COUNT) for the Rayleigh-clear measurements under different cloud and dust conditions. This includes three\ncategories of cloud cover (< 25 %, > 50 %, > 75 %) for dust mixing ratios above (Dust) and below (DustNO) 109 kgkg–1 along the track. Table 4. Overview of the total Error Estimate (EEtot; ms–1), mean absolute difference and uncertainty (MADI, σμ; ms–1), Standard deviation\n(STD; ms–1) and counts (COUNT) for the Rayleigh-clear measurements under different cloud and dust conditions. This includes three\ncategories of cloud cover (< 25 %, > 50 %, > 75 %) for dust mixing ratios above (Dust) and below (DustNO) 109 kgkg–1 along the track. Cloud < 50 %\nCloud > 50 %\nCloud > 75 %\nDustNO\nDust\nDustNO\nDust\nDustNO\nDust\nEEtot\n4.8\n5.4\n5.0\n5.6\n5.3\n5.8\nMADI\n3.3±0.2\n4.4±0.6\n3.9±0.5\n5.0±0.5\n4.3±0.7\n5.7±0.8\nSTD\n4.3\n5.0\n5.1\n5.9\n5.6\n6.4\nCOUNT\n234\n28\n64\n52\n38\n24\nFigure 5. Altitude as a function of Rayleigh-clear absolute difference |ΔdiffHLOS| (a,e), EEtot (b,f), normalized useful signal (c,g) and SR\n(d,h), where the colouring is dependent on the percentage of SAF clouds (upper row) and CAMS dust mixing ratio (lower row) along the\ntrack. The cross symbol + stands for outliers and defines values with an EE below 5 ms–1 and an absolute difference of more than 10 ms–1. Panel (a) includes the MADI for each cloud cover percentage, with a minimum sample size of 10 data points used to compute each value. Cloud < 50 %\nCloud > 50 %\nCloud > 75 %\nDustNO\nDust\nDustNO\nDust\nDustNO\nDust\nEEtot\n4.8\n5.4\n5.0\n5.6\n5.3\n5.8\nMADI\n3.3±0.2\n4.4±0.6\n3.9±0.5\n5.0±0.5\n4.3±0.7\n5.7±0.8\nSTD\n4.3\n5.0\n5.1\n5.9\n5.6\n6.4\nCOUNT\n234\n28\n64\n52\n38\n24 Cloud < 50 %\nCloud > 50 %\nCloud > 75 %\nDustNO\nDust\nDustNO\nDust\nDustNO\nDust\nEEtot\n4.8\n5.4\n5.0\n5.6\n5.3\n5.8\nMADI\n3.3±0.2\n4.4±0.6\n3.9±0.5\n5.0±0.5\n4.3±0.7\n5.7±0.8\nSTD\n4.3\n5.0\n5.1\n5.9\n5.6\n6.4\nCOUNT\n234\n28\n64\n52\n38\n24 Figure 5. Rayleigh-clear The larger STD with increasing cloudiness and dust concentration suggests an increasingly perturbed pattern\nof Rayleigh-clear measurements,possibly owing to the lower signal levels or to a cross-talk. Figure 5 puts this phenomenon into perspective, by showing the altitude-dependent absolute difference |ΔdiffHLOS| (a,e), the EEtot (b,f), the normalized useful signal (c,g) and the SR (d,h), where the colouring depends on the percentage of SAF clouds\n420\n(upper row) and the CAMS dust mixing ratio (lower row) along the track. For reference, the values that did not pass the QC are\nshown transparently. In addition, panel 5a includes the MADI of four cloud cover percentage categories, where each MADI is\ncomputed with a minimum sample size of 10 values. The colourings in Fig. 5 are illustrative of the results summarised in Table\n4, with measurements showing generally greater MADI under high cloud cover (red, orange, Fig. 5a) than under lower cloud cover (blue, blue-green). Measurements in the lower troposphere are naturally more strongly affected by cloud cover compared\n425\nto higher levels. The same applies to dust (Fig. 5e), which also occurs mainly in the lower 5 km of the troposphere. As we have already shown in Fig. 3b, the absolute error is higher in the upper and lower troposphere and minimised in\nthe middle troposphere around 10 km altitude. This trend is well reflected in the EEtot in Fig. 5b, which is an indication of As we have already shown in Fig. 3b, the absolute error is higher in the upper and lower troposphere and minimised in\nthe middle troposphere around 10 km altitude. This trend is well reflected in the EEtot in Fig. 5b, which is an indication of 18 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. density. In the lower troposphere, the return signal is lower due to strong attenuation under clouds and dust layers. Interestingly,\nthe values with high EEtot and smaller useful signal in the mid-troposphere between 5 and 12.5 km in red likely correspond to\nmeasurements sampled under thick clouds, resulting in a strongly attenuated signal. They account for most of the measurements\nwith cloud cover greater than 75 % in this altitude range, while the cloud tops appear to be located between 12.5 and 15 km, as\n435\nthey exhibit a larger normalized useful signal and a SR greater than 1 (Fig. 5d,h). Finally, outliers are found under all types of\ncloud and dust conditions and affect different altitude ranges. They also occur for regular normalized useful signals, with most\nSRs lying around 1, which rules out a cause related to atmospheric particles. Altitude as a function of Rayleigh-clear absolute difference |ΔdiffHLOS| (a,e), EEtot (b,f), normalized useful signal (c,g) and SR\n(d,h), where the colouring is dependent on the percentage of SAF clouds (upper row) and CAMS dust mixing ratio (lower row) along the\ntrack. The cross symbol + stands for outliers and defines values with an EE below 5 ms–1 and an absolute difference of more than 10 ms–1. Panel (a) includes the MADI for each cloud cover percentage, with a minimum sample size of 10 data points used to compute each value. the generally good consistency between the EEtot and the absolute differences. As expected, this tendency fits inversely with\nthe normalized useful signal shown in Fig. 5c, with lower signal in the upper and lower troposphere. Indeed, in the higher\n430\ntroposphere the air is less dense and the thickness of the RB’s is not sufficient to compensate for the decrease in air molecule the generally good consistency between the EEtot and the absolute differences. As expected, this tendency fits inversely with\nthe normalized useful signal shown in Fig. 5c, with lower signal in the upper and lower troposphere. Indeed, in the higher\n430\ntroposphere the air is less dense and the thickness of the RB’s is not sufficient to compensate for the decrease in air molecule i\nthe normalized useful signal shown in Fig. 5c, with lower signal in the upper and lower troposphere. Indeed, in the higher\n430\ntroposphere the air is less dense and the thickness of the RB’s is not sufficient to compensate for the decrease in air molecule 19 Mie-cloudy 4b,\nmost backscatter occurs in two layers, i.e. within 10–15 km and below 7 km altitude. The majority of measurements have\nnormalized useful signals above 5e13 a.u. (Fig. 6c,g), which is overall above the normalized useful signal of the rejected\n450\nmeasurements shown in transparent. Furthermore, the SRs are generally above 1 (Fig. 6d,h), which is characteristic of Mie-\ncloudy measurements. More specifically, measurements sampled above 12.5 km have a cloud cover of more than 75 % along\nthe track and probably correspond to cloud tops, as they have stronger SRs between 1.5 and 3 (Fig. 6d,h). They exhibit good\nquality as well, with an average MADI of 1.5 ms–1 (Fig. 4b). Between 7.5 and 12.5 km altitude, most of the measurements\noccur with cloud cover less than 50 %, with SRs falling below 1.3. In this altitude range, there are also 2 outliers, which\n455\ninterestingly have SRs around 1 and a normalized useful signal in the same order of magnitude as the discarded ones. Their\npresence is unusual, as Mie-cloudy measurements are only obtainable for SRs above 1. Finally, below 7.5 km, the cloud cover\nis mainly above 50 %, while the dust concentration is mainly below 5 × 10–8 kgkg–1, showing that most of the Mie-cloudy\nbackscatter results from clouds and not from dust. As can be seen in Fig. 6g, measurements with high dust concentration\n(brown) are discarded (transparent) with normalized useful signals below 5e13 a.u. Surprisingly, measurements sampled at the\n460\nlower 1 km have the lowest normalized useful signals, mostly below 5e13 a.u. and are not discarded. These, however, tend to\nhave larger SRs between 1 and 2, which can compensate for the low normalized useful signal in the calculation of the EE. They\nalso correspond to the largest MADI scaling up to 4 ms–1 on average (i.e. Fig. 4d) in addition to relatively high EEtots (i.e. Fig. 6b,f). Two measurements also show negative SRs, which is an artifact, due to insufficient background signal corrections. The third outlier in the lower 1 km does not have abnormal characteristics compared to other measurements at this altitude. 465 normalized useful signals above 5e13 a.u. (Fig. 6c,g), which is overall above the normalized useful signal of the rejected\n450\nmeasurements shown in transparent. Furthermore, the SRs are generally above 1 (Fig. 6d,h), which is characteristic of Mie-\ncloudy measurements. Mie-cloudy More specifically, measurements sampled above 12.5 km have a cloud cover of more than 75 % along\nthe track and probably correspond to cloud tops, as they have stronger SRs between 1.5 and 3 (Fig. 6d,h). They exhibit good\nquality as well, with an average MADI of 1.5 ms–1 (Fig. 4b). Between 7.5 and 12.5 km altitude, most of the measurements normalized useful signals above 5e13 a.u. (Fig. 6c,g), which is overall above the normalized useful signal of the rejected\n450\nmeasurements shown in transparent. Furthermore, the SRs are generally above 1 (Fig. 6d,h), which is characteristic of Mie-\ncloudy measurements. More specifically, measurements sampled above 12.5 km have a cloud cover of more than 75 % along\nthe track and probably correspond to cloud tops, as they have stronger SRs between 1.5 and 3 (Fig. 6d,h). They exhibit good\nquality as well, with an average MADI of 1.5 ms–1 (Fig. 4b). Between 7.5 and 12.5 km altitude, most of the measurements normalized useful signals above 5e13 a.u. (Fig. 6c,g), which is overall above the normalized useful signal of the rejected\n450\nmeasurements shown in transparent. Furthermore, the SRs are generally above 1 (Fig. 6d,h), which is characteristic of Mie-\ncloudy measurements. More specifically, measurements sampled above 12.5 km have a cloud cover of more than 75 % along\nthe track and probably correspond to cloud tops, as they have stronger SRs between 1.5 and 3 (Fig. 6d,h). They exhibit good\nquality as well, with an average MADI of 1.5 ms–1 (Fig. 4b). Between 7.5 and 12.5 km altitude, most of the measurements occur with cloud cover less than 50 %, with SRs falling below 1.3. In this altitude range, there are also 2 outliers, which\n455\ninterestingly have SRs around 1 and a normalized useful signal in the same order of magnitude as the discarded ones. Their\npresence is unusual, as Mie-cloudy measurements are only obtainable for SRs above 1. Finally, below 7.5 km, the cloud cover\nis mainly above 50 %, while the dust concentration is mainly below 5 × 10–8 kgkg–1, showing that most of the Mie-cloudy\nbackscatter results from clouds and not from dust. As can be seen in Fig. 6g, measurements with high dust concentration occur with cloud cover less than 50 %, with SRs falling below 1.3. Mie-cloudy Table 5 shows the same as Table 4, but for Mie-cloudy. Due to the limited amount of data for Mie-cloudy winds, the inter-\n440\npretation of the results should be treated with caution. We find that, in contrast to Rayleigh-clear, the EE, MADI and STD\ndecrease with the percentage of cloud cover along the path. This is understandable as clouds provide the strongest backscatter\nsignal required for high quality Mie-cloudy measurements. However, the presence of dust for cloud cover below 50 % leads to\na decrease in EEtot, MADI and STD, while conversely there is an increase of these quantities in more dense cloudy conditions Table 5 shows the same as Table 4, but for Mie-cloudy. Due to the limited amount of data for Mie-cloudy winds, the inter-\n440\npretation of the results should be treated with caution. We find that, in contrast to Rayleigh-clear, the EE, MADI and STD\ndecrease with the percentage of cloud cover along the path. This is understandable as clouds provide the strongest backscatter\nsignal required for high quality Mie-cloudy measurements. However, the presence of dust for cloud cover below 50 % leads to\na decrease in EEtot, MADI and STD, while conversely there is an increase of these quantities in more dense cloudy conditions (>50 %, >75 %). A possible explanation is that in clear-sky conditions, the backscatter from dust layers is strong enough to\n445\nobtain high quality measurements, whereas in cloudy conditions, the attenuation by clouds weakens the backscatter return from\nthe dust. Figure 6 depicts the same as Fig. 5, but for Mie-cloudy. As mentioned in the previous section when discussing in Fig. 4b,\nmost backscatter occurs in two layers, i.e. within 10–15 km and below 7 km altitude. The majority of measurements have (>50 %, >75 %). A possible explanation is that in clear-sky conditions, the backscatter from dust layers is strong enough to\n445\nobtain high quality measurements, whereas in cloudy conditions, the attenuation by clouds weakens the backscatter return from\nthe dust. Figure 6 depicts the same as Fig. 5, but for Mie-cloudy. As mentioned in the previous section when discussing in Fig. 4b,\nmost backscatter occurs in two layers, i.e. within 10–15 km and below 7 km altitude. The majority of measurements have Figure 6 depicts the same as Fig. 5, but for Mie-cloudy. As mentioned in the previous section when discussing in Fig. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Table 5. Same as table 4, but for Mie-cloudy. Cloud < 50 %\nCloud > 50 %\nCloud > 75 %\nDustNO\nDust\nDustNO\nDust\nDustNO\nDust\nEEtot\n3.7\n3.6\n3.4\n3.5\n3.2\n3.4\nMADI\n2.8±0.5\n2.4±0.2\n1.8±0.3\n2.5±0.4\n1.6±0.3\n2.6±0.6\nSTD\n2.96\n1.53\n1.89\n2.95\n1.68\n3.18\nCOUNT\n11\n9\n16\n23\n8\n13\nFigure 6. Same as Fig. 5 but for Mie-cloudy. Here the cross symbol + defines values with an EE below 3 ms–1 and an absolute difference\nabove 6 ms–1. Table 5. Same as table 4, but for Mie-cloudy. Cloud < 50 %\nCloud > 50 %\nCloud > 75 %\nDustNO\nDust\nDustNO\nDust\nDustNO\nDust\nEEtot\n3.7\n3.6\n3.4\n3.5\n3.2\n3.4\nMADI\n2.8±0.5\n2.4±0.2\n1.8±0.3\n2.5±0.4\n1.6±0.3\n2.6±0.6\nSTD\n2.96\n1.53\n1.89\n2.95\n1.68\n3.18\nCOUNT\n11\n9\n16\n23\n8\n13 Figure 6 Same as Fig 5 but for Mie cloudy Here the cross symbol + defines values with an EE below 3 ms–1 and an absolute difference Figure 6. Same as Fig. 5 but for Mie-cloudy. Here the cross symbol + defines values with an EE below 3 ms–1 and an absolute difference\nabove 6 ms–1. Mie-cloudy In this altitude range, there are also 2 outliers, which\n455\ninterestingly have SRs around 1 and a normalized useful signal in the same order of magnitude as the discarded ones. Their\npresence is unusual, as Mie-cloudy measurements are only obtainable for SRs above 1. Finally, below 7.5 km, the cloud cover\nis mainly above 50 %, while the dust concentration is mainly below 5 × 10–8 kgkg–1, showing that most of the Mie-cloudy\nbackscatter results from clouds and not from dust. As can be seen in Fig. 6g, measurements with high dust concentration occur with cloud cover less than 50 %, with SRs falling below 1.3. In this altitude range, there are also 2 outliers, which\n455\ninterestingly have SRs around 1 and a normalized useful signal in the same order of magnitude as the discarded ones. Their\npresence is unusual, as Mie-cloudy measurements are only obtainable for SRs above 1. Finally, below 7.5 km, the cloud cover\nis mainly above 50 %, while the dust concentration is mainly below 5 × 10–8 kgkg–1, showing that most of the Mie-cloudy\nbackscatter results from clouds and not from dust. As can be seen in Fig. 6g, measurements with high dust concentration (brown) are discarded (transparent) with normalized useful signals below 5e13 a.u. Surprisingly, measurements sampled at the\n460\nlower 1 km have the lowest normalized useful signals, mostly below 5e13 a.u. and are not discarded. These, however, tend to\nhave larger SRs between 1 and 2, which can compensate for the low normalized useful signal in the calculation of the EE. They\nalso correspond to the largest MADI scaling up to 4 ms–1 on average (i.e. Fig. 4d) in addition to relatively high EEtots (i.e. Fig. 6b,f). Two measurements also show negative SRs, which is an artifact, due to insufficient background signal corrections. The third outlier in the lower 1 km does not have abnormal characteristics compared to other measurements at this altitude. 465 20 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 7. Overview of the cloud-free case study for a radiosonde launched from Sal airport on 9th September 2021 at 18:45 UTC and the\nascending orbit of Aeolus between 19:23:56-19:24:31 UTC for a co-location radius of 180km. (a) Vertical radiosonde HLOS wind profile\n(black solid line) and projected onto Rayleigh-clear RBS (black stepped line), as well as averaged Rayleigh-clear observations (blue dots),\ncorresponding EE (error bars) and ECMWF model equivalents (Meq, stepped lines). (b) Vertical profile of the Rayleigh-clear EEtot (blue\nline), together with the EEtot of all profiles (grey solid lines) and their average (black solid line). (c), (d) Same as (b), but for normalised\nuseful signal and CAMS dust mixing ratio, respectively. (e) Horizontal map showing the SAFNWC CT at 19:00 UTC and the co-location\nperimeter (white solid line), the Aeolus track (red solid line) and the radiosonde launch site (red cross). Figure 7. Overview of the cloud-free case study for a radiosonde launched from Sal airport on 9th September 2021 at 18:45 UTC and the\nascending orbit of Aeolus between 19:23:56-19:24:31 UTC for a co-location radius of 180km. (a) Vertical radiosonde HLOS wind profile\n(black solid line) and projected onto Rayleigh-clear RBS (black stepped line), as well as averaged Rayleigh-clear observations (blue dots),\ncorresponding EE (error bars) and ECMWF model equivalents (Meq, stepped lines). (b) Vertical profile of the Rayleigh-clear EEtot (blue\nline), together with the EEtot of all profiles (grey solid lines) and their average (black solid line). (c), (d) Same as (b), but for normalised\nuseful signal and CAMS dust mixing ratio, respectively. (e) Horizontal map showing the SAFNWC CT at 19:00 UTC and the co-location\nperimeter (white solid line), the Aeolus track (red solid line) and the radiosonde launch site (red cross). passed over on an ascending orbit between 19:23:56 UTC and 19:24:31 UTC within a co-location radius of 180 km around the\nlaunch site. Figure 7a depicts the corresponding sampled radiosonde HLOS wind profile (black lines) as well as Rayleigh-clear\n(blue) wind measurement points with associated EEtot shown as error bars and ECMWF model equivalents shown as stepped\nlines. The corresponding Rayleigh-clear EEtot, normalized useful signal and CAMS dust mixing ratio profiles are shown in\n475\nblue in Figs. 4.2.3\nCase studies To further investigate the properties of the Aeolus wind errors, this section presents three case studies comparing Aeolus and\nradiosonde wind measurements under three different atmospheric conditions, namely clear sky, high cloud cover and high dust\nconcentration. 470 The first case-study illustrated in Fig. 7 presents a comparison between Aeolus and radiosonde wind measurements collected\nunder clear sky conditions. The radiosonde was launched over Sal Airport at 18:45 UTC on 9 September 2021, and Aeolus 21 7b, 7c and 7d, respectively, along with all other profiles in grey and the average of all profiles in black. Figure (blue) wind measurement points with associated EEtot shown as error bars and ECMWF model equivalents shown as stepped\nlines. The corresponding Rayleigh-clear EEtot, normalized useful signal and CAMS dust mixing ratio profiles are shown in\n475\nblue in Figs. 7b, 7c and 7d, respectively, along with all other profiles in grey and the average of all profiles in black. Figure 22 Beneath the cloud\nbase at 13 km altitude, however, it appears that the Rayleigh-clear measurements follow an irregular pattern, with most of the\nmeasurements and error bars not matching the radiosonde observations, reaching deviations higher than 10 ms–1. Accordingly,\nwe find that the EEtot (Fig. 8b) is larger in this altitude range mainly varying between 5 and 6 ms–1, which also corresponds to a sharp decrease of the normalized useful signal well below the average (Fig. 8c). Nonetheless, the ECMWF model-equivalents\n500\nin Fig. 8a remain fairly accurate relative to the radiosonde measurements. This result mirrors the findings presented in the\nprevious section, namely that the Rayleigh-clear EEtot is systematically underestimated when the normalized useful signal\nis strongly attenuated. It appears that the normalized useful signal further decreases below 2.5 km, presumably as a result\nof the increasing dust concentration at this height (Fig. 8d), which most likely leads to a QC rejection of the Rayleigh-clear sharp decrease of the normalized useful signal well below the average (Fig. 8c). Nonetheless, the ECMWF model-equivalents\n500\nin Fig. 8a remain fairly accurate relative to the radiosonde measurements. This result mirrors the findings presented in the\nprevious section, namely that the Rayleigh-clear EEtot is systematically underestimated when the normalized useful signal\nis strongly attenuated. It appears that the normalized useful signal further decreases below 2.5 km, presumably as a result\nof the increasing dust concentration at this height (Fig. 8d), which most likely leads to a QC rejection of the Rayleigh-clear measurements. 505\nLastly, Fig. 9 examines the influence of dust on the quality of Aeolus. In this case, the radiosonde was launched on 21\nSeptember 2021 at 06:50 UTC for a descending orbit of Aeolus, which passed over a co-location perimeter with a radius of 60\nkm between 07:28:44 UTC and 07:29:07 UTC. As can be seen in Fig. 9e, the atmospheric conditions in the co-location area\nwere completely cloud free at 07:30, with some low level cloud further south of the island. The radiosonde profile shown in Fig. measurements. 505\nLastly, Fig. 9 examines the influence of dust on the quality of Aeolus. In this case, the radiosonde was launched on 21\nSeptember 2021 at 06:50 UTC for a descending orbit of Aeolus, which passed over a co-location perimeter with a radius of 60\nkm between 07:28:44 UTC and 07:29:07 UTC. As can be seen in Fig. These high-clouds appear to be located between 13 km and 16 km altitude, as three Mie-cloudy (red) and two Rayleigh-\ncloudy (orange) measurements are found in this range, and where the normalized useful signal is found to have a maximum. In\nthis altitude range, all Rayleigh-clear, Rayleigh-cloudy and Mie-cloudy measurements exhibit good quality, with radiosonde co-location region between 07:28:32 UTC and 07:28:55 UTC, i.e. during the descending node. As can been seen in panel 8e,\n490\nwhich corresponds to SAFNWC CT at 07:30 UTC, Aeolus overpasses a variety of high clouds, mainly high semitransparent\nclouds. These high-clouds appear to be located between 13 km and 16 km altitude, as three Mie-cloudy (red) and two Rayleigh-\ncloudy (orange) measurements are found in this range, and where the normalized useful signal is found to have a maximum. In\nthis altitude range, all Rayleigh-clear, Rayleigh-cloudy and Mie-cloudy measurements exhibit good quality, with radiosonde co-location region between 07:28:32 UTC and 07:28:55 UTC, i.e. during the descending node. As can been seen in panel 8e,\n490\nwhich corresponds to SAFNWC CT at 07:30 UTC, Aeolus overpasses a variety of high clouds, mainly high semitransparent\nclouds. These high-clouds appear to be located between 13 km and 16 km altitude, as three Mie-cloudy (red) and two Rayleigh-\ncloudy (orange) measurements are found in this range, and where the normalized useful signal is found to have a maximum. In\nthis altitude range, all Rayleigh-clear, Rayleigh-cloudy and Mie-cloudy measurements exhibit good quality, with radiosonde measurements generally within the error bars. Above this cloud cover at 16 km, we only find Rayleigh-clear measurements\n495\nthat also perform well, with an EEtot (Fig. 8b) and normalized useful signal (Fig. 8c) close to average. Beneath the cloud\nbase at 13 km altitude, however, it appears that the Rayleigh-clear measurements follow an irregular pattern, with most of the\nmeasurements and error bars not matching the radiosonde observations, reaching deviations higher than 10 ms–1. Accordingly,\nwe find that the EEtot (Fig. 8b) is larger in this altitude range mainly varying between 5 and 6 ms–1, which also corresponds to a measurements generally within the error bars. Above this cloud cover at 16 km, we only find Rayleigh-clear measurements\n495\nthat also perform well, with an EEtot (Fig. 8b) and normalized useful signal (Fig. 8c) close to average. In this case study, the radiosonde was also launched from\nSal airport, this time at 07:00 UTC on 14 September 2021, with a co-location radius of 60 km. Aeolus passed across the\nco-location region between 07:28:32 UTC and 07:28:55 UTC, i.e. during the descending node. As can been seen in panel 8e,\n490\nwhich corresponds to SAFNWC CT at 07:30 UTC, Aeolus overpasses a variety of high clouds, mainly high semitransparent\nclouds. These high-clouds appear to be located between 13 km and 16 km altitude, as three Mie-cloudy (red) and two Rayleigh-\ncloudy (orange) measurements are found in this range, and where the normalized useful signal is found to have a maximum. In\nthis altitude range, all Rayleigh-clear, Rayleigh-cloudy and Mie-cloudy measurements exhibit good quality, with radiosonde\nmeasurements generally within the error bars. Above this cloud cover at 16 km, we only find Rayleigh-clear measurements\n495\nthat also perform well, with an EEtot (Fig. 8b) and normalized useful signal (Fig. 8c) close to average. Beneath the cloud\nbase at 13 km altitude, however, it appears that the Rayleigh-clear measurements follow an irregular pattern, with most of the\nmeasurements and error bars not matching the radiosonde observations, reaching deviations higher than 10 ms–1. Accordingly,\nwe find that the EEtot (Fig. 8b) is larger in this altitude range mainly varying between 5 and 6 ms–1, which also corresponds to a\nsharp decrease of the normalized useful signal well below the average (Fig. 8c). Nonetheless, the ECMWF model-equivalents\n500\nin Fig. 8a remain fairly accurate relative to the radiosonde measurements. This result mirrors the findings presented in the\nprevious section, namely that the Rayleigh-clear EEtot is systematically underestimated when the normalized useful signal\nis strongly attenuated. It appears that the normalized useful signal further decreases below 2.5 km, presumably as a result\nof the increasing dust concentration at this height (Fig. 8d), which most likely leads to a QC rejection of the Rayleigh-clear\nmeasurements. 505 co-location region between 07:28:32 UTC and 07:28:55 UTC, i.e. during the descending node. As can been seen in panel 8e,\n490\nwhich corresponds to SAFNWC CT at 07:30 UTC, Aeolus overpasses a variety of high clouds, mainly high semitransparent\nclouds. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. 7e shows the SAFNWC CT over the Cape Verde region at 19:00 UTC. In the latter panel, it can be seen that conditions were\npredominantly cloud-free along the Aeolus track (red solid line) and within the co-location radius (white solid line), while\nsome low clouds can be found in the surrounding area. In these clear-sky conditions, it is not surprising to find that most of the\nmeasurements are of the Rayleigh-clear observation type, with no Mie-cloudy and Rayleigh-cloudy measurements (Fig. 7a). 480\nThroughout the atmosphere above 2.5 km, the quality of Rayleigh-clear is very good, with most error bars overlapping with\nradiosonde measurements and ECMWF model equivalents. In general, we found that the EEtot estimate (Fig. 7b) is below\naverage throughout the atmosphere, with a minimum of 3.5 ms–1 at 8 km altitude and a maximum above 5 ms–1 at 17.5 km\nand 2.5 km altitude. This is consistent with to a normalized useful signal (Fig. 7c) close to the average, except between 2.5 and 12.5 km, where it is higher, most likely due to the absence of cloud attenuation. In general, EEtot and normalized useful signal\n485\ndecrease below 5 km, which is accompanied by an increase in the dust mixing ratio. This increase reaches 1.2 kgkg–1 at about\n2 km altitude, below which no measurements are found, presumably filtered out during the QC procedure. Figure 8 shows the same as Fig. 7, but for cloudy conditions. In this case study, the radiosonde was also launched from\nSal airport, this time at 07:00 UTC on 14 September 2021, with a co-location radius of 60 km. Aeolus passed across the 12.5 km, where it is higher, most likely due to the absence of cloud attenuation. In general, EEtot and normalized useful signal\n485\ndecrease below 5 km, which is accompanied by an increase in the dust mixing ratio. This increase reaches 1.2 kgkg–1 at about\n2 km altitude, below which no measurements are found, presumably filtered out during the QC procedure. Figure 8 shows the same as Fig. 7, but for cloudy conditions. In this case study, the radiosonde was also launched from\nSal airport, this time at 07:00 UTC on 14 September 2021, with a co-location radius of 60 km. Aeolus passed across the Figure 8 shows the same as Fig. 7, but for cloudy conditions. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 8. Same as Fig. 7, but for the case study with high cloud cover. Here, the radiosonde was launched from the Sal airport at 07:00 UTC\non the 14th September 2021, while Aeolus overpassed the co-location area, with radius of 60km, on a descending node between 07:28:32\nand 07:28:55 UTC. In panel a, the red and orange colours represent the averaged Mie-cloudy and Rayleigh-cloudy observations (points),\nrespectively, with the corresponding EE shown as error bars and ECMWF model equivalents (Meq) shown as stepped lines. The SAFNWC\nCT shown in (e) corresponds to 07:30 UTC. Figure 8. Same as Fig. 7, but for the case study with high cloud cover. Here, the radiosonde was launched from the Sal airport at 07:00 UTC\non the 14th September 2021, while Aeolus overpassed the co-location area, with radius of 60km, on a descending node between 07:28:32\nand 07:28:55 UTC. In panel a, the red and orange colours represent the averaged Mie-cloudy and Rayleigh-cloudy observations (points),\nrespectively, with the corresponding EE shown as error bars and ECMWF model equivalents (Meq) shown as stepped lines. The SAFNWC\nCT shown in (e) corresponds to 07:30 UTC. while outliers with EEs of less than 5 ms–1 (Fig. 9b) can be spotted above 15km and below 5km. This error structure is\nsurprising, as both the normalized useful signal and error estimation curves are similar to the one of the cloud-free case study\nin Fig. 7b and 7c. However, in panel 9c, we see that the Rayleigh-clear error pattern coincides with a strong peak in dust\nmixing ratio, reaching more than 2 10–7 kgkg–1 around 3.5 km altitude. The presence of dust seems to affect the quality of\n515\nRayleigh-clear measurements without influencing the normalized useful signal and thus leading to an underestimation of the\nEE. Reason could be linked to a cross-talk. while outliers with EEs of less than 5 ms–1 (Fig. 9b) can be spotted above 15km and below 5km. This error structure is\nsurprising, as both the normalized useful signal and error estimation curves are similar to the one of the cloud-free case study\nin Fig. 7b and 7c. 9e, the atmospheric conditions in the co-location area\nwere completely cloud free at 07:30, with some low level cloud further south of the island. The radiosonde profile shown in Fig. 9a indicates that Aeolus primarily measured in the Rayleigh channel along this orbital segment. Rayleigh-clear measurements\n510\nappear to be consistent with radiosonde wind measurements throughout the mid-troposphere between 5 km and 15 km altitude, 9a indicates that Aeolus primarily measured in the Rayleigh channel along this orbital segment. Rayleigh-clear measurements\n510\nappear to be consistent with radiosonde wind measurements throughout the mid-troposphere between 5 km and 15 km altitude, 23 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Figure 9. Same as Figs. 7 and 8 but for the case study with dust. Here, the radiosonde was launched from the Sal airport at 06:50 UTC on\nthe 21th September 2021, while Aeolus overpassed the co-location area, with radius of 60km, on a descending node between 07:28:44 UTC\nand 07:29:07 UTC. The SAFNWC CT shown in (e) corresponds to 07:30 UTC. Figure 9. Same as Figs. 7 and 8 but for the case study with dust. Here, the radiosonde was launched from the Sal airport at 06:50 UTC on\nthe 21th September 2021, while Aeolus overpassed the co-location area, with radius of 60km, on a descending node between 07:28:44 UTC\nand 07:29:07 UTC. The SAFNWC CT shown in (e) corresponds to 07:30 UTC. However, in panel 9c, we see that the Rayleigh-clear error pattern coincides with a strong peak in dust\nmixing ratio, reaching more than 2 10–7 kgkg–1 around 3.5 km altitude. The presence of dust seems to affect the quality of\n515\nRayleigh-clear measurements without influencing the normalized useful signal and thus leading to an underestimation of the\nEE. Reason could be linked to a cross-talk. 515 24 5\nConclusions In this study, we conducted a cross-Atlantic validation of Aeolus wind observations using radiosondes in the scope of the Joint\nAeolus Tropical Atlantic Campaign (JATAC). Of the total 20 radiosonde profiles included in this work, 11 were launched from\n520\nPuerto Rico and St. Croix in the Caribbean and 9 from Sal Airport on Cape Verde between August and September 2021. The\nadvantage of radiosondes is that they provide good vertical coverage, providing 384 Rayleigh-clear bin-to-bin comparisons\nfrom the surface to an altitude of 20 km and 59 Mie-cloudy comparisons, mainly restricted to the presence of clouds and\naerosols. After having applied several Quality Control (QC) and adaptation grid procedures, we quantified the quality of Aeolus Tropical Atlantic Campaign (JATAC). Of the total 20 radiosonde profiles included in this work, 11 were launched from\n520\nPuerto Rico and St. Croix in the Caribbean and 9 from Sal Airport on Cape Verde between August and September 2021. The\nadvantage of radiosondes is that they provide good vertical coverage, providing 384 Rayleigh-clear bin-to-bin comparisons\nfrom the surface to an altitude of 20 km and 59 Mie-cloudy comparisons, mainly restricted to the presence of clouds and\naerosols. After having applied several Quality Control (QC) and adaptation grid procedures, we quantified the quality of 25 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Rayleigh-clear, Mie-cloudy and to a lesser extent Rayleigh-cloudy observation types, with respect to co-location aspects as\n525\nwell as atmospheric conditions such as cloud cover and dust concentration. Rayleigh-clear, Mie-cloudy and to a lesser extent Rayleigh-cloudy observation types, with respect to co-location aspects as\n525\nwell as atmospheric conditions such as cloud cover and dust concentration. According to our statistical analysis, the total systematic error of Rayleigh-clear is –0.5 ± 0.2 ms–1, which is in agreement\nwith the ESA recommendation of 0.7 ms–1. The random error was calculated from the standard deviation of the difference\nbetween radiosonde and Aeolus measurements accounting for radiosonde observation errors estimated at 0 7±0 28 ms–1 and Rayleigh-clear, Mie-cloudy and to a lesser extent Rayleigh-cloudy observation types, with respect to co-location aspects as\n525\nwell as atmospheric conditions such as cloud cover and dust concentration. According to our statistical analysis, the total systematic error of Rayleigh-clear is –0.5 ± 0.2 ms–1, which is in agreement\nwith the ESA recommendation of 0.7 ms–1. The random error was calculated from the standard deviation of the difference\nbetween radiosonde and Aeolus measurements, accounting for radiosonde observation errors estimated at 0.7±0.28 ms–1 and between radiosonde and Aeolus measurements, accounting for radiosonde observation errors estimated at 0.7±0.28 ms–1 and\nrepresentativeness errors ranging from 1.5 to 2.5 ms–1. In the altitude range of 2–16 km and 16–20 km, the random error\n530\nis 3.8 – 4.3 ms–1 and 4.3 – 4.8 ms–1, respectively, which is above the ESA-specified values of 2.5 ms–1 and 3 ms–1, respec-\ntively. In general, Rayleigh-clear shows no error dependency with respect to co-location radius, even for distances reaching\n340 km, whilst being more sensitive to co-location time, especially if the radiosonde measurement is ahead of Aeolus’ over-\nflight time, which presumably corresponds to low altitude measurements. In addition, the systematic and random errors are representativeness errors ranging from 1.5 to 2.5 ms–1. In the altitude range of 2–16 km and 16–20 km, the random error\n530\nis 3.8 – 4.3 ms–1 and 4.3 – 4.8 ms–1, respectively, which is above the ESA-specified values of 2.5 ms–1 and 3 ms–1, respec-\ntively. In general, Rayleigh-clear shows no error dependency with respect to co-location radius, even for distances reaching\n340 km, whilst being more sensitive to co-location time, especially if the radiosonde measurement is ahead of Aeolus’ over-\nflight time, which presumably corresponds to low altitude measurements. In addition, the systematic and random errors are height-dependent, with larger errors occurring in the upper troposphere, mainly caused by the reduction in signal return from\n535\ndecreasing air density, and in lower levels, most likely caused by the signal attenuation by clouds and dust. The error estimate\nlikewise follows a similar form to the observed height error dependency, as it is inversely proportional to the squared root\nof the normalized useful signal. In cases where the normalized useful signal is strongly attenuated by clouds or dust, the er-\nror estimate is generally underestimated, with measurements exhibiting non physical features and departures from radiosonde height-dependent, with larger errors occurring in the upper troposphere, mainly caused by the reduction in signal return from\n535\ndecreasing air density, and in lower levels, most likely caused by the signal attenuation by clouds and dust. The error estimate\nlikewise follows a similar form to the observed height error dependency, as it is inversely proportional to the squared root\nof the normalized useful signal. In cases where the normalized useful signal is strongly attenuated by clouds or dust, the er-\nror estimate is generally underestimated, with measurements exhibiting non physical features and departures from radiosonde winds larger than the error estimate. A redefinition of the Rayleigh-clear error estimate could account for this underestimation\n540\nby including other sources of noise, such as detector noise or readout noise, which increase for reduced signal levels. Fur-\nthermore, a cross-talk, i.e. the leakage of the Mie signal into the Rayleigh receiver, could also explain this underestimation,\nespecially in the case of strong Mie returns. However, this supposition was not investigated in the context of this study. Outliers,\ndefined as measurements with small error estimate and large absolute differences, are found under all conditions, i.e. for all co-location radii, co-location times, altitudes as well as cloud and dust cover. Their origin does not appear to be correlated\n545\nwith low signal levels but seem to be inherent to the statistical nature of the error distribution. Taking other terms into account\nwhen defining the error estimate, such as the influence of temperature, pressure or scattering ratio on the Rayleigh response,\ncould certainly contribute to improving the error characterisation. The ECMWF model equivalents of Rayleigh-clear are found\nto have a significantly better agreement with the radiosonde wind measurements compared to the Rayleigh-clear observations. co-location radii, co-location times, altitudes as well as cloud and dust cover. Their origin does not appear to be correlated\n545\nwith low signal levels but seem to be inherent to the statistical nature of the error distribution. Taking other terms into account\nwhen defining the error estimate, such as the influence of temperature, pressure or scattering ratio on the Rayleigh response,\ncould certainly contribute to improving the error characterisation. The ECMWF model equivalents of Rayleigh-clear are found\nto have a significantly better agreement with the radiosonde wind measurements compared to the Rayleigh-clear observations. This is a further confirmation that the co-location parameters used for this validation study are appropriate and that the model\n550\nequivalents provide a suitable reference for validating Aeolus. In addition, we demonstrate the existence of an orbital- and\naltitude-dependent bias in the Rayleigh-clear channel, which is visible with respect to both radiosondes and ECMWF model\nequivalents. This bias has already been documented by Borne et al. (2023) in West Africa using model equivalents and is now\nconfirmed observationally. The underlying cause for this bias, however, remains unknown. In addition, we find that Rayleigh- For Mie-cloudy, the statistical analysis yielded a systematic negative deviation of –0.9±0.3 ms–1 within ESA specifications\nwhen uncertainty is taken into account, and it is consistent across all orbital nodes and Cal/Val sites. The random error between\n2–16 km is 1.1 – 2.3 ms–1, which falls within the ESA recommendations. The general quality of Mie-cloudy winds does not 26 Competing interests. The authors declare that they have no conflict of interest. Author contributions. MB, PK and MW conceptualized the study and developed the methodology. MB, PK, MW and BW carried out the\n580\ninvestigation and validation. PK, MW, PV and RR provided financial support for the project. MB, PV and RR were responsible for data\ncuration. MB conducted the formal analysis and wrote the original draft of the paper. MB, PK, MW, BW, CF, PV and RR reviewed and\nedited the paper. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. depend on the co-location radius, while it is more sensitive to temporal differences. The errors appear to be larger at 5 km and\n560\nabout 1 km altitude, typically at the upper and lower limits of the Saharan Air Layer, where clouds frequently occur. According\nto Lux et al. (2022a), the Mie fringe of the Fizeau interferometer can be distorted in the case of strong backscatter gradients, e.g. at cloud edges. Interestingly, Mie-cloudy does not seem to sample within dust layers, as most bins with high dust concentrations\nare rejected by the QC. Furthermore, the systematic and random Mie errors decrease with the percentage of cloud cover, while depend on the co-location radius, while it is more sensitive to temporal differences. The errors appear to be larger at 5 km and\n560\nabout 1 km altitude, typically at the upper and lower limits of the Saharan Air Layer, where clouds frequently occur. According\nto Lux et al. (2022a), the Mie fringe of the Fizeau interferometer can be distorted in the case of strong backscatter gradients, e.g. at cloud edges. Interestingly, Mie-cloudy does not seem to sample within dust layers, as most bins with high dust concentrations\nare rejected by the QC. Furthermore, the systematic and random Mie errors decrease with the percentage of cloud cover, while they increase in the presence of dust. This may be attributed to the generally weak backscatter of dust, increasing the error of\n565\nthe Mie-cloudy winds. Similar to Rayleigh-clear, outliers with small error estimate and large absolute differences can be found\nfor all co-location distance, co-location time, altitude, dust concentrations and cloud cover. An improvement of the Mie EE is\nexpected from an optimisation of the Mie core algorithm, such as the fitting function or the classification algorithm. The presented study determined the error dependencies of the different Aeolus observation types and error estimates with y\ny\nrespect to tropical clouds and dust. The acquired information are valuable to further improve the processing algorithms in order\n570\nto meet the requirements of the mission. respect to tropical clouds and dust. The acquired information are valuable to further improve the processing algorithms in order\n570\nto meet the requirements of the mission. Code availability. The analysis was conducted using the Python language and the code can be provided on request. Code availability. Acknowledgements. We would like to thank Thorsten Fehr for organising the Joint Aeolus Tropical Atlantic Campaign (JATAC), which\n585\nmade this cross-Atlantic study possible. We would also like to thank the German Aerospace Center (DLR), which assisted in the transport\nof radiosonde material and helium tanks from Germany to Sal in Cape Verde. Sincere gratitude also goes to the large teams involved in the\nradiosonde launches. For the Cape Verdean team, our special thanks go to Azusa Takeishi, Tanguy Jonville and Cedric Gacial for their hard\nwork and dedication. The analysis was conducted using the Python language and the code can be provided on request. Forecasts (ECMWF) model-equivalents are available in the ECMWF Meteorological Archival and Retrieval System (MARS) operational\n575\narchive. Dust mixing ratio data is openly available and can be downloaded in the Copernicus Atmosphere Monitoring Service (CAMS) Data\nStore. The NW SAF Cloud Type (CT) product can by obtained using the software package NWC/PPS available from http://www.nwcsaf.org/. The radiosonde data corresponding to launches from Puerto Rico are publicly available at http://dx.doi.org/10.5067/CPEXAW/DATA101. The other radiosonde data can be provided on request. Author contributions. MB, PK and MW conceptualized the study and developed the methodology. MB, PK, MW and BW carried out the\n580\ninvestigation and validation. PK, MW, PV and RR provided financial support for the project. MB, PV and RR were responsible for data\ncuration. MB conducted the formal analysis and wrote the original draft of the paper. MB, PK, MW, BW, CF, PV and RR reviewed and\nedited the paper. 27 References\n590 J., Langland, R.,\net al.: Lidar-measured wind profiles: The missing link in the global observing system, Bulletin of the American Meteorological Society, Baker, W. E., Atlas, R., Cardinali, C., Clement, A., Emmitt, G. D., Gentry, B. M., Hardesty, R. M., Källén, E., Kavaya, M. J., Langland, R.,\net al.: Lidar-measured wind profiles: The missing link in the global observing system, Bulletin of the American Meteorological Society,\n95, 543–564, 2014. 00 95, 543–564, 2014. 600\nBedka, K. M., Nehrir, A. R., Kavaya, M., Barton-Grimley, R., Beaubien, M., Carroll, B., Collins, J., Cooney, J., Emmitt, G. D., Greco, S.,\net al.: Airborne lidar observations of wind, water vapor, and aerosol profiles during the NASA Aeolus calibration and validation (Cal/Val)\ntest flight campaign, Atmospheric Measurement Techniques, 14, 4305–4334, 2021. Belanger, J., Jelinek, M., and Curry, J.: A climatology of easterly waves in the tropical Western Hemisphere, Geoscience Data Journal, 3,\n40–49, 2016. 605 Belanger, J., Jelinek, M., and Curry, J.: A climatology of easterly waves in the tropical Western Hemisphere, Geoscience Data Journal, 3,\n40–49, 2016. 605\nBelova, E., Kirkwood, S., Voelger, P., Chatterjee, S., Satheesan, K., Hagelin, S., Lindskog, M., and Körnich, H.: Validation of Aeolus winds 605 40–49, 2016. 605\nBelova, E., Kirkwood, S., Voelger, P., Chatterjee, S., Satheesan, K., Hagelin, S., Lindskog, M., and Körnich, H.: Validation of Aeolus winds\nusing ground-based radars in Antarctica and in northern Sweden, Atmospheric Measurement Techniques, 14, 5415–5428, 2021. Bormann, N., Saarinen, S., Kelly, G., and Thépaut, J.-N.: The spatial structure of observation errors in atmospheric motion vectors from\ngeostationary satellite data, Monthly Weather Review, 131, 706–718, 2003. Belova, E., Kirkwood, S., Voelger, P., Chatterjee, S., Satheesan, K., Hagelin, S., Lindskog, M., and Körnich, H.: Validation of Aeolus winds\nusing ground-based radars in Antarctica and in northern Sweden, Atmospheric Measurement Techniques, 14, 5415–5428, 2021. Bormann, N., Saarinen, S., Kelly, G., and Thépaut, J.-N.: The spatial structure of observation errors in atmospheric motion vectors from\ngeostationary satellite data, Monthly Weather Review, 131, 706–718, 2003. Borne, M., Knippertz, P., Weissmann, M., Martin, A., Rennie, M., and Cress, A.: Impact of Aeolus wind lidar observations on the rep-\n610\nresentation of the West African monsoon circulation in the ECMWF and DWD forecasting systems, Quarterly Journal of the Royal\nMeteorological Society, 2023. References\n590 Chen, S., Cao, R., Xie, Y., Zhang, Y., Tan, W., Chen, H., Guo, P., and Zhao, P.: Study of the seasonal variation in Aeolus wind product\nperformance over China using ERA5 and radiosonde data, Atmospheric Chemistry and Physics, 21, 11 489–11 504, 2021. Chen, S., Cao, R., Xie, Y., Zhang, Y., Tan, W., Chen, H., Guo, P., and Zhao, P.: Study of the seasonal variation in Aeolus wind product\nperformance over China using ERA5 and radiosonde data, Atmospheric Chemistry and Physics, 21, 11 489–11 504, 2021. Dabas, A., Denneulin, M., Flamant, P., Loth, C., Garnier, A., and Dolfi-Bouteyre, A.: Correcting winds measured with a Rayleigh Doppler\n615\nlidar from pressure and temperature effects, Tellus A: Dynamic Meteorology and Oceanography, 60, 206–215, 2008. Derrien, M. and Le Gléau, H.: MSG/SEVIRI cloud mask and type from SAFNWC, International Journal of Remote Sensing, 26, 4707–4732, Dabas, A., Denneulin, M., Flamant, P., Loth, C., Garnier, A., and Dolfi-Bouteyre, A.: Correcting winds measured with a Rayleigh Doppler\n615\nlidar from pressure and temperature effects, Tellus A: Dynamic Meteorology and Oceanography, 60, 206–215, 2008. Dabas, A., Denneulin, M., Flamant, P., Loth, C., Garnier, A., and Dolfi-Bouteyre, A.: Correcting winds measured with a Rayleigh Doppler\n615\nlidar from pressure and temperature effects, Tellus A: Dynamic Meteorology and Oceanography, 60, 206–215, 2008. Derrien, M. and Le Gléau, H.: MSG/SEVIRI cloud mask and type from SAFNWC, International Journal of Remote Sensing, 26, 4707–4732,\n2005. Derrien, M., Farki, B., Harang, L., LeGleau, H., Noyalet, A., Pochic, D., and Sairouni, A.: Automatic cloud detection applied to NOAA- p\np\n,\ny\ngy\ng p y,\n,\n,\nDerrien, M. and Le Gléau, H.: MSG/SEVIRI cloud mask and type from SAFNWC, International Journal of Remote Sensing, 26, 4707–4732,\n2005. 2005. Derrien, M., Farki, B., Harang, L., LeGleau, H., Noyalet, A., Pochic, D., and Sairouni, A.: Automatic cloud detection applied to NOAA- Derrien, M., Farki, B., Harang, L., LeGleau, H., Noyalet, A., Pochic, D., and Sairouni, A.: Automatic cloud detection applied to NOAA-\ni\ni\nf\ni\n6\n6\n6 Derrien, M., Farki, B., Harang, L., LeGleau, H., Noyalet, A., Pochic, D., and Sairouni, A.: Automatic cloud detection applied to NOAA-\n11/AVHRR imagery, Remote Sensing of Environment, 46, 246–267, 1993. 620 11/AVHRR imagery, Remote Sensing of Environment, 46, 246–267, 1993. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Acknowledgements. We would like to thank Thorsten Fehr for organising the Joint Aeolus Tropical Atlantic Campaign (JATAC), which\n585\nmade this cross-Atlantic study possible. We would also like to thank the German Aerospace Center (DLR), which assisted in the transport\nof radiosonde material and helium tanks from Germany to Sal in Cape Verde. Sincere gratitude also goes to the large teams involved in the\nradiosonde launches. For the Cape Verdean team, our special thanks go to Azusa Takeishi, Tanguy Jonville and Cedric Gacial for their hard\nwork and dedication. 28 References\n590 Abril-Gago, J., Ortiz-Amezcua, P., Bermejo-Pantaleón, D., Andújar-Maqueda, J., Bravo-Aranda, J. A., Granados-Muñoz, M. J., Navas-\nGuzmán, F., Alados-Arboledas, L., Foyo-Moreno, I., and Guerrero-Rascado, J. L.: Validation Activities of Aeolus Wind Products in the\nSoutheastern Iberian Peninsula, Atmospheric Chemistry and Physics Discussions, pp. 1–32, 2022. Alonso Lasheras, O., Sanz Diaz, A., and Lopez Cotin, L.: The Initial Operations Phase of the EUMETSAT’s SAF to Support Nowcasting Abril-Gago, J., Ortiz-Amezcua, P., Bermejo-Pantaleón, D., Andújar-Maqueda, J., Bravo-Aranda, J. A., Granados-Muñoz, M. J., Navas- Guzmán, F., Alados-Arboledas, L., Foyo-Moreno, I., and Guerrero-Rascado, J. L.: Validation Activities of Aeolus Wind Products in the\nSoutheastern Iberian Peninsula, Atmospheric Chemistry and Physics Discussions, pp. 1–32, 2022. Alonso Lasheras, O., Sanz Diaz, A., and Lopez Cotin, L.: The Initial Operations Phase of the EUMETSAT’s SAF to Support Nowcasting\n(NWC SAF) s, ESA Special Publication, 584, 4, 2005. 595 Alonso Lasheras, O., Sanz Diaz, A., and Lopez Cotin, L.: The Initial Operations Phase of the EUMETSAT o so\nas e as, O., Sa\na ,\n., a d\nope Cot ,\n.:\ne\nt a Ope at o s\nase o t e\nU\nS\ns S\nto Suppo t Nowcast g\n(NWC SAF) s, ESA Special Publication, 584, 4, 2005. (NWC SAF) s, ESA Special Publication, 584, 4, 2005. (NWC SAF) s, ESA Special Publication, 584, 4, 2005. 595\nBaars, H., Herzog, A., Heese, B., Ohneiser, K., Hanbuch, K., Hofer, J., Yin, Z., Engelmann, R., and Wandinger, U.: Validation of Aeolus\nwind products above the Atlantic Ocean, Atmospheric Measurement Techniques, 13, 6007–6024, 2020. Baker, W. E., Atlas, R., Cardinali, C., Clement, A., Emmitt, G. D., Gentry, B. M., Hardesty, R. M., Källén, E., Kavaya, M. J., Langland, R.,\net al.: Lidar-measured wind profiles: The missing link in the global observing system, Bulletin of the American Meteorological Society, Baars, H., Herzog, A., Heese, B., Ohneiser, K., Hanbuch, K., Hofer, J., Yin, Z., Engelmann, R., and Wandinger, U.: Validation of Aeolus\nwind products above the Atlantic Ocean, Atmospheric Measurement Techniques, 13, 6007–6024, 2020. ,\n,\ng,\n,\n,\n,\n,\n,\n,\n,\n,\n,\n,\n,\ng\n,\n,\ng ,\nwind products above the Atlantic Ocean, Atmospheric Measurement Techniques, 13, 6007–6024, 2020. Baker, W. E., Atlas, R., Cardinali, C., Clement, A., Emmitt, G. D., Gentry, B. M., Hardesty, R. M., Källén, E., Kavaya, M. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Folger, K. and Weissmann, M.: Height correction of Atmospheric Motion Vectors using satellite lidar observations from CALIPSO, Journal\nof Applied Meteorology and Climatology, 53, 1809–1819, https://doi.org/10.1175/JAMC-D-13-0337.1, 2014. Folger, K. and Weissmann, M.: Height correction of Atmospheric Motion Vectors using satellite lidar observations from CALIPSO, Journal\nof Applied Meteorology and Climatology, 53, 1809–1819, https://doi.org/10.1175/JAMC-D-13-0337.1, 2014. Garrett, K., Liu, H., Ide, K., Hoffman, R. N., and Lukens, K. E.: Optimization and impact assessment of Aeolus HLOS wind assimilation in of Applied Meteorology and Climatology, 53, 1809–1819, https://doi.org/10.1175/JAMC-D-13-0337.1, 2014. Garrett, K., Liu, H., Ide, K., Hoffman, R. N., and Lukens, K. E.: Optimization and impact assessment of Aeolus HLOS wind assimilation in Garrett, K., Liu, H., Ide, K., Hoffman, R. N., and Lukens, K. E.: Optimization and impact assessment of Aeo Garrett, K., Liu, H., Ide, K., Hoffman, R. N., and Lukens, K. E.: Optimization and impact assessment of Aeolus HLOS wind assimilation in\nNOAA’s global forecast system, Quarterly Journal of the Royal Meteorological Society, 148, 2703–2716, 2022. 630 g\ny\ny\ny\ng\ny\nGuo, J., Liu, B., Gong, W., Shi, L., Zhang, Y., Ma, Y., Zhang, J., Chen, T., Bai, K., Stoffelen, A., et al.: First comparison of wind observations\nfrom ESA’s satellite mission Aeolus and ground-based radar wind profiler network of China, Atmospheric Chemistry and Physics, 21,\n2945–2958, 2021. Horányi, A., Cardinali, C., Rennie, M., and Isaksen, L.: The assimilation of horizontal line-of-sight wind information into the ECMWF data\nassimilation and forecasting system. Part II: The impact of degraded wind observations, Quarterly Journal of the Royal Meteorological\n635\nSociety, 141, 1233–1243, 2015. Horányi, A., Cardinali, C., Rennie, M., and Isaksen, L.: The assimilation of horizontal line-of-sight wind information into the ECMWF data\ni il i\nd f\ni\nh i\nf d\nd d\ni d\nb\ni\nl\nl\nf h\nl\nl\ni\nl Horányi, A., Cardinali, C., Rennie, M., and Isaksen, L.: The assimilation of horizontal line-of-sight wind information into the ECMWF data\nassimilation and forecasting system. Part II: The impact of degraded wind observations, Quarterly Journal of the Royal Meteorological\n635\nSociety, 141, 1233–1243, 2015. References\n590 620\nDirksen, R., Sommer, M., Immler, F., Hurst, D., Kivi, R., and Vömel, H.: Reference quality upper-air measurements: GRUAN data processing\nfor the Vaisala RS92 radiosonde, Atmospheric Measurement Techniques, 7, 4463–4490, 2014. Flemming, J., Huijnen, V., Arteta, J., Bechtold, P., Beljaars, A., Blechschmidt, A.-M., Diamantakis, M., Engelen, R. J., Gaudel, A., Inness,\nA., et al.: Tropospheric chemistry in the Integrated Forecasting System of ECMWF, Geoscientific model development, 8, 975–1003, 2015. g y\ng\nDirksen, R., Sommer, M., Immler, F., Hurst, D., Kivi, R., and Vömel, H.: Reference quality upper-air measurements: GRUAN data processing\nfor the Vaisala RS92 radiosonde, Atmospheric Measurement Techniques, 7, 4463–4490, 2014. Flemming, J., Huijnen, V., Arteta, J., Bechtold, P., Beljaars, A., Blechschmidt, A.-M., Diamantakis, M., Engelen, R. J., Gaudel, A., Inness,i Dirksen, R., Sommer, M., Immler, F., Hurst, D., Kivi, R., and Vömel, H.: Reference quality upper-air measurements: GRUAN data processing\nfor the Vaisala RS92 radiosonde, Atmospheric Measurement Techniques, 7, 4463–4490, 2014. Flemming, J., Huijnen, V., Arteta, J., Bechtold, P., Beljaars, A., Blechschmidt, A. M., Diamantakis, M., Engelen, R. J., Gaudel, A., Inness,\nA., et al.: Tropospheric chemistry in the Integrated Forecasting System of ECMWF, Geoscientific model development, 8, 975–1003, 2015. Flesia, C. and Korb, C. L.: Theory of the double-edge molecular technique for Doppler lidar wind measurement, Applied Optics, 38, 432–440,\n625\n1999. Flesia, C. and Korb, C. L.: Theory of the double-edge molecular technique for Doppler lidar wind measurement, Applied Optics, 38, 432–440,\n625\n1999. 29 Huijnen, V., Pozzer, A., Arteta, J., Brasseur, G., Bouarar, I., Chabrillat, S., Christophe, Y., Doumbia, T., Flemming, J., Guth, J., et al.:\nQuantifying uncertainties due to chemistry modelling–evaluation of tropospheric composition simulations in the CAMS model (cycle\n43R1), Geoscientific Model Development, 12, 1725–1752, 2019. Huijnen, V., Pozzer, A., Arteta, J., Brasseur, G., Bouarar, I., Chabrillat, S., Christophe, Y., Doumbia, T., Flemming, J., Guth, J., et al.:\nQuantifying uncertainties due to chemistry modelling–evaluation of tropospheric composition simulations in the CAMS model (cycle\n43R1), Geoscientific Model Development, 12, 1725–1752, 2019. Iglewicz, B. and Hoaglin, D. C.: How to detect and handle outliers, vol. 16, Asq Press, 1993. 640\nIngmann, P. and Straume, A.: ADM-AEOLUS mission requirements document, Centre ESRaT, 2016. Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A.-M., Dominguez, J. J., Engelen, R., Eskes, H., Flem-\nming, J., et al.: The CAMS reanalysis of atmospheric composition, Atmospheric Chemistry and Physics, 19, 3515–3556, 2019. Iwai, H., Aoki, M., Oshiro, M., and Ishii, S.: Validation of Aeolus Level 2B wind products using wind profilers, ground-based Doppler wind Iglewicz, B. and Hoaglin, D. C.: How to detect and handle outliers, vol. 16, Asq Press, 1993. 640\nIngmann, P. and Straume, A.: ADM-AEOLUS mission requirements document, Centre ESRaT, 2016. Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A.-M., Dominguez, J. J., Engelen, R., Eskes, H., Flem\nming, J., et al.: The CAMS reanalysis of atmospheric composition, Atmospheric Chemistry and Physics, 19, 3515–3556, 2019. gmann, P. and Straume, A.: ADM-AEOLUS mission requirements document, Centre ESRaT, 2016. Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A.-M., Dominguez, J. J., Engelen, R., Eskes, H., Flem-\nming, J., et al.: The CAMS reanalysis of atmospheric composition, Atmospheric Chemistry and Physics, 19, 3515–3556, 2019. Iwai, H., Aoki, M., Oshiro, M., and Ishii, S.: Validation of Aeolus Level 2B wind products using wind profilers, ground-based Doppler wind\nlidars, and radiosondes in Japan, Atmospheric Measurement Techniques, 14, 7255–7275, 2021. 645 Iwai, H., Aoki, M., Oshiro, M., and Ishii, S.: Validation of Aeolus Level 2B wind products using wind profile Iwai, H., Aoki, M., Oshiro, M., and Ishii, S.: Validation of Aeolus Level 2B wind products using wind profilers, ground-based Doppler wind lidars, and radiosondes in Japan, Atmospheric Measurement Techniques, 14, 7255–7275, 2021. 645\nLux, O., Lemmerz, C., Weiler, F., Marksteiner, U., Witschas, B., Rahm, S., Geiß, A., and Reitebuch, O.: Intercomparison of wind observa-\ntions from the European Space Agency’s Aeolus satellite mission and the ALADIN Airborne Demonstrator, Atmospheric Measurement\nTechniques, 13, 2075–2097, 2020. L\nO L\nC W il\nF M k\ni\nU Wi\nh\nB R h\nS G iß A S h fl\nA\nd R i b\nh O R\ni\nl i Lux, O., Lemmerz, C., Weiler, F., Marksteiner, U., Witschas, B., Rahm, S., Geiß, A., and Reitebuch, O.: Intercomparison of wind observa-\ntions from the European Space Agency’s Aeolus satellite mission and the ALADIN Airborne Demonstrator, Atmospheric Measurement\nTechniques, 13, 2075–2097, 2020. Lux, O., Lemmerz, C., Weiler, F., Marksteiner, U., Witschas, B., Rahm, S., Geiß, A., Schäfler, A., and Reitebuch, O.: Retrieval improvements\nfor the ALADIN Airborne Demonstrator in support of the Aeolus wind product validation, Atmospheric Measurement Techniques, 15,\n650\n1303–1331, 2022a. Lux, O., Lemmerz, C., Weiler, F., Marksteiner, U., Witschas, B., Rahm, S., Geiß, A., Schäfler, A., and Reitebuch, O.: Retrieval improvements Lux, O., Lemmerz, C., Weiler, F., Marksteiner, U., Witschas, B., Rahm, S., Geiß, A., Schäfler, A., and Reitebuch, O.: Retrieval improvements\nfor the ALADIN Airborne Demonstrator in support of the Aeolus wind product validation, Atmospheric Measurement Techniques, 15,\n650\n1303–1331, 2022a. Lux, O., Witschas, B., Geiß, A., Lemmerz, C., Weiler, F., Marksteiner, U., Rahm, S., Schäfler, A., and Reitebuch, O.: Quality control and error 650 for the ALADIN Airborne Demonstrator in support of the Aeolus wind product validation, Atmospheric Measurement Techniques, 15,\n650\n1303–1331, 2022a. Lux, O., Witschas, B., Geiß, A., Lemmerz, C., Weiler, F., Marksteiner, U., Rahm, S., Schäfler, A., and Reitebuch, O.: Quality control and error\nassessment of the Aeolus L2B wind results from the Joint Aeolus Tropical Atlantic Campaign, Atmospheric Measurement Techniques,\n15, 6467–6488, 2022b. Lux, O., Witschas, B., Geiß, A., Lemmerz, C., Weiler, F., Marksteiner, U., Rahm, S., Schäfler, A., and Reitebuch, O.: Quality control and error\nassessment of the Aeolus L2B wind results from the Joint Aeolus Tropical Atlantic Campaign, Atmospheric Measurement Techniques,\n15, 6467–6488, 2022b. Martin, A., Weissmann, M., Reitebuch, O., Rennie, M., Geiß, A., and Cress, A.: Validation of Aeolus winds using radiosonde observations\n655\nand numerical weather prediction model equivalents, Atmospheric Measurement Techniques, 14, 2167–2183, 2021. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Portabella, M. and Stoffelen, A.: On scatterometer ocean stress, Journal of Atmospheric and Oceanic Technology, 26, 368–382, 2009. 665 Portabella, M. and Stoffelen, A.: On scatterometer ocean stress, Journal of Atmospheric and Oceanic Technology, 26, 368–382, 2009. 665\nRatynski, M., Khaykin, S., Hauchecorne, A., Wing, R., Cammas, J.-P., Hello, Y., and Keckhut, P.: Validation of Aeolus wind profiles using\nground-based lidar and radiosonde observations at La Réunion Island and the Observatoire de Haute Provence, EGUsphere, pp. 1–33,\n2022. Reitebuch, O.: The spaceborne wind lidar mission ADM-Aeolus, in: Atmospheric Physics, pp. 815–827, Springer, 2012. ground-based lidar and radiosonde observations at La Réunion Island and the Observatoire de Haute Provence, EGUsphere, pp. 1–33,\n2022. Reitebuch, O.: The spaceborne wind lidar mission ADM-Aeolus, in: Atmospheric Physics, pp. 815–827, Spr Reitebuch, O., Marksteiner, U., Rompel, M., Meringer, M., Schmidt, K., Huber, D., Nikolaus, I., Dabas, A., Marshall, J., de Bruin, F., et al.:\n670\nAeolus End-to-End Simulator and Wind Retrieval Algorithms up to Level 1B, in: EPJ Web of Conferences, vol. 176, p. 02010, EDP\nSciences, 2018. Rémy S Kipling Z Flemming J Boucher O Nabat P Michou M Bozzo A Ades M Huijnen V Benedetti A et al : Description Reitebuch, O., Marksteiner, U., Rompel, M., Meringer, M., Schmidt, K., Huber, D., Nikolaus, I., Dabas, A., Marshall, J., de Bruin, F., et al.:\n670\nAeolus End-to-End Simulator and Wind Retrieval Algorithms up to Level 1B, in: EPJ Web of Conferences, vol. 176, p. 02010, EDP\nSciences, 2018. Rémy, S., Kipling, Z., Flemming, J., Boucher, O., Nabat, P., Michou, M., Bozzo, A., Ades, M., Huijnen, V., Benedetti, A., et al.: Description\nand evaluation of the tropospheric aerosol scheme in the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Rémy, S., Kipling, Z., Flemming, J., Boucher, O., Nabat, P., Michou, M., Bozzo, A., Ades, M., Huijnen, V., Benedetti, A., et al.: Description\nand evaluation of the tropospheric aerosol scheme in the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated\nForecasting System (IFS-AER, cycle 45R1), Geoscientific Model Development, 12, 4627–4659, 2019. 75 Forecasting System (IFS-AER, cycle 45R1), Geoscientific Model Development, 12, 4627–4659, 2019. 675\nRennie, M., Isaksen, L., et al.: The NWP impact of Aeolus Level-2B winds at ECMWF, ECMWF Technical Memoranda, 864, 2020. Rennie, M. Martin, A., Weissmann, M., Reitebuch, O., Rennie, M., Geiß, A., and Cress, A.: Validation of Aeolus winds using radiosonde observations\n655\nand numerical weather prediction model equivalents, Atmospheric Measurement Techniques, 14, 2167–2183, 2021. Martin, A., Weissmann, M., and Cress, A.: Impact of assimilating Aeolus observations in the global model ICON: A global statistical\noverview, https://ucloud.univie.ac.at/index.php/s/DIP0XxSgKA6eXek, 2022a. Martin, A., Weissmann, M., and Cress, A.: Investigation of dynamical scenarios leading to particularly high impact of Aeolus on NWP\nforecasts EGUsphere pp 1 23 2022b\n660 Martin, A., Weissmann, M., Reitebuch, O., Rennie, M., Geiß, A., and Cress, A.: Validation of Aeolus winds using radiosonde observations\n655\nand numerical weather prediction model equivalents, Atmospheric Measurement Techniques, 14, 2167–2183, 2021. Martin, A., Weissmann, M., and Cress, A.: Impact of assimilating Aeolus observations in the global model ICON: A global statistical\noverview, https://ucloud.univie.ac.at/index.php/s/DIP0XxSgKA6eXek, 2022a. Martin, A., Weissmann, M., and Cress, A.: Impact of assimilating Aeolus observations in the global model ICON: A global statistical\noverview, https://ucloud.univie.ac.at/index.php/s/DIP0XxSgKA6eXek, 2022a. Martin, A., Weissmann, M., and Cress, A.: Investigation of dynamical scenarios leading to particularly high impact of Aeolus on NWP\nforecasts, EGUsphere, pp. 1–23, 2022b. 60 660 Martin, C. and Suhr, I.: NCAR/EOL Atmospheric Sounding Processing ENvironment (ASPEN) software, Version 3.4. 3, 2021. McKay, J. A.: Assessment of a multibeam Fizeau wedge interferometer for Doppler wind lidar, Applied opti ssessment of a multibeam Fizeau wedge interferometer for Doppler wind lidar, Applied optics, 41, 1760–1767 Naderi, F. M., Freilich, M. H., and Long, D.: Spaceborne radar measurement of wind velocity over the ocean-an overview of the NSCAT\nscatterometer system, Proceedings of the IEEE, 79, 850–866, 1991. 30 and Chaphalkar, N.: Impact of outlier detection on neural networks based property value prediction, in: Information systems E.: Observing the Diurnal Cycle of Coastal Rainfall over Western Puerto Rico in Collaboration with University of Puerto Rico Students,\n680\nBulletin of the American Meteorological Society, 104, E305–E324, 2023. Bulletin of the American Meteorological Society, 104, E305–E324, 2023. Sandbhor, S. and Chaphalkar, N.: Impact of outlier detection on neural networks based property value prediction, in: Information systems\ndesign and intelligent applications, pp. 481–495, Springer, 2019. Saunders, R. W. and Kriebel, K. T.: An improved method for detecting clear sky and cloudy radiances from AVHRR data, International\nJournal of Remote Sensing, 9, 123–150, 1988. 685 Journal of Remote Sensing, 9, 123–150, 1988. 685\nSchillinger, M., Morancais, D., Fabre, F., and Culoma, A. J.: ALADIN: the LIDAR instrument for the AEOLUS mission, in: Sensors,\nSystems, and Next-Generation Satellites VI, vol. 4881, pp. 40–51, SPIE, 2003. Stoffelen, A., Pailleux, J., Källén, E., Vaughan, J. M., Isaksen, L., Flamant, P., Wergen, W., Andersson, E., Schyberg, H., Culoma, A., et al.:\nThe atmospheric dynamics mission for global wind field measurement, Bulletin of the American Meteorological Society, 86, 73–88, 2005. Schillinger, M., Morancais, D., Fabre, F., and Culoma, A. J.: ALADIN: the LIDAR instrument for the AEOLUS mission, in: Sensors,\nSystems, and Next-Generation Satellites VI, vol. 4881, pp. 40–51, SPIE, 2003. Stoffelen A Pailleux J Källén E Vaughan J M Isaksen L Flamant P Wergen W Andersson E Schyberg H Culoma A et al : Systems, and Next-Generation Satellites VI, vol. 4881, pp. 40–51, SPIE, 2003. Stoffelen, A., Pailleux, J., Källén, E., Vaughan, J. M., Isaksen, L., Flamant, P., Wergen, W., Andersson, E., Schyberg, H., Culoma, A., et al.:\nThe atmospheric dynamics mission for global wind field measurement, Bulletin of the American Meteorological Society, 86, 73–88, 2005. Stoffelen, A., Pailleux, J., Källén, E., Vaughan, J. M., Isaksen, L., Flamant, P., Wergen, W., Andersson, E., Schyberg, H., Culoma, A., et al.:\nThe atmospheric dynamics mission for global wind field measurement, Bulletin of the American Meteorological Society, 86, 73–88, 2005. Stowe, L. L., Davis, P. A., and McClain, E. P.: Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification\n690\nalgorithm for the Advanced Very High Resolution Radiometer, Journal of atmospheric and oceanic technology, 16, 656–681, 1999. P., Isaksen, L., Weiler, F., de Kloe, J., Kanitz, T., and Reitebuch, O.: The impact of Aeolus wind retrievals on ECMWF global\nweather forecasts, Quarterly Journal of the Royal Meteorological Society, 147, 3555–3586, 2021. Rios-Berrios, R., Sakaeda, N., Jimenez-González, H. J., Nieves-Jimenez, A., Zayas, Y., Martin, E., Wu, S.-N., Homeyer, C. R., and Rodríguez, Forecasting System (IFS-AER, cycle 45R1), Geoscientific Model Development, 12, 4627–4659, 2019. 675\nRennie, M., Isaksen, L., et al.: The NWP impact of Aeolus Level-2B winds at ECMWF, ECMWF Technical Memoranda, 864, 2020. Rennie, M. P., Isaksen, L., Weiler, F., de Kloe, J., Kanitz, T., and Reitebuch, O.: The impact of Aeolus wind retrievals on ECMWF global\nweather forecasts Quarterly Journal of the Royal Meteorological Society 147 3555 3586 2021 Rennie, M., Isaksen, L., et al.: The NWP impact of Aeolus Level-2B winds at ECMWF, ECMWF Technical Memoranda, 864, 2020. Rennie, M. P., Isaksen, L., Weiler, F., de Kloe, J., Kanitz, T., and Reitebuch, O.: The impact of Aeolus wind retrievals on ECMWF global Rennie, M. P., Isaksen, L., Weiler, F., de Kloe, J., Kanitz, T., and Reitebuch, O.: The impact of Aeolus wind retrievals on ECMWF global\nweather forecasts, Quarterly Journal of the Royal Meteorological Society, 147, 3555–3586, 2021. ,\n,\n,\n,\n,\n,\n,\n,\n,\n,\n,\np\ng\nweather forecasts, Quarterly Journal of the Royal Meteorological Society, 147, 3555–3586, 2021. Rios-Berrios, R., Sakaeda, N., Jimenez-González, H. J., Nieves-Jimenez, A., Zayas, Y., Martin, E., Wu, S.-N., Homeyer, C. R., and Rodríguez, Rios-Berrios, R., Sakaeda, N., Jimenez-González, H. J., Nieves-Jimenez, A., Zayas, Y., Martin, E., Wu, S.-N., Homeyer, C. R., and Rodríguez, E.: Observing the Diurnal Cycle of Coastal Rainfall over Western Puerto Rico in Collaboration with University of Puerto Rico Students,\n680\nBulletin of the American Meteorological Society, 104, E305–E324, 2023. Sandbhor, S. and Chaphalkar, N.: Impact of outlier detection on neural networks based property value prediction, in: Information systems\ndesign and intelligent applications, pp. 481–495, Springer, 2019. Saunders, R. W. and Kriebel, K. T.: An improved method for detecting clear sky and cloudy radiances from AVHRR data, International E.: Observing the Diurnal Cycle of Coastal Rainfall over Western Puerto Rico in Collaboration with University of Puerto Rico Students,\n680\nBulletin of the American Meteorological Society, 104, E305–E324, 2023. Sandbhor, S. St\nA G R\ni\nM I\nk\nL d Kl\nJ M\nill\nG J St ff l\nA Fl\nt T Sti\nlit\nH D b\nA H b\nD\nt l Stowe, L. L., Davis, P. A., and McClain, E. P.: Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification\n690\nalgorithm for the Advanced Very High Resolution Radiometer, Journal of atmospheric and oceanic technology, 16, 656–681, 1999. Straume, A.-G., Rennie, M., Isaksen, L., de Kloe, J., Marseille, G.-J., Stoffelen, A., Flament, T., Stieglitz, H., Dabas, A., Huber, D., et al.:\nESA’s space-based Doppler wind lidar mission Aeolus–First wind and aerosol product assessment results, in: EPJ Web of Conferences,\nvol. 237, p. 01007, EDP Sciences, 2020. Straume, A.-G., Rennie, M., Isaksen, L., de Kloe, J., Marseille, G.-J., Stoffelen, A., Flament, T., Stieglitz, H., Dabas, A., Huber, D., et al.:\nESA’s space-based Doppler wind lidar mission Aeolus–First wind and aerosol product assessment results, in: EPJ Web of Conferences,\nvol. 237, p. 01007, EDP Sciences, 2020. Tan, D. G., Andersson, E., Kloe, J. D., Marseille, G.-J., Stoffelen, A., Poli, P., Denneulin, M.-L., Dabas, A., Huber, D., Reitebuch, O., et al.:\n695\nThe ADM-Aeolus wind retrieval algorithms, Tellus A: Dynamic Meteorology and Oceanography, 60, 191–205, 2008. Tan, D. G., Andersson, E., Kloe, J. D., Marseille, G.-J., Stoffelen, A., Poli, P., Denneulin, M.-L., Dabas, A., Huber, D., Reitebuch, O., et al.:\n695\nThe ADM-Aeolus wind retrieval algorithms, Tellus A: Dynamic Meteorology and Oceanography, 60, 191–205, 2008. Tripathy, S. S., Saxena, R. K., and Gupta, P. K.: Comparison of statistical methods for outlier detection in proficiency testing data on analysis\nof lead in aqueous solution, American Journal of Theoretical and Applied Statistics, 2, 233–242, 2013. Weiler, F., Kanitz, T., Wernham, D., Rennie, M., Huber, D., Schillinger, M., Saint-Pe, O., Bell, R., Parrinello, T., and Reitebuch, O.: Charac-\nterization of dark current signal measurements of the ACCDs used on board the Aeolus satellite, Atmospheric Measurement Techniques,\n700\n14, 5153–5177, 2021a. Tan, D. G., Andersson, E., Kloe, J. D., Marseille, G.-J., Stoffelen, A., Poli, P., Denneulin, M.-L., Dabas, A., Huber, D., Reitebuch, O., et al.:\n695\nThe ADM-Aeolus wind retrieval algorithms, Tellus A: Dynamic Meteorology and Oceanography, 60, 191–205, 2008. Tripathy, S. S., Saxena, R. K., and Gupta, P. K.: Comparison of statistical methods for outlier detection in proficiency testing data on analysis\nof lead in aqueous solution, American Journal of Theoretical and Applied Statistics, 2, 233–242, 2013. Tripathy, S. S., Saxena, R. K., and Gupta, P. K.: Comparison of statistical methods for outlier detection in proficiency testing data on analysis\nof lead in aqueous solution, American Journal of Theoretical and Applied Statistics, 2, 233–242, 2013. Weiler, F., Kanitz, T., Wernham, D., Rennie, M., Huber, D., Schillinger, M., Saint-Pe, O., Bell, R., Parrinello, T., and Reitebuch, O.: Charac-\nterization of dark current signal measurements of the ACCDs used on board the Aeolus satellite, Atmospheric Measurement Techniques,\n700\n14, 5153–5177, 2021a. 31 https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. https://doi.org/10.5194/egusphere-2023-742\nPreprint. Discussion started: 3 May 2023\nc⃝Author(s) 2023. CC BY 4.0 License. Weiler, F., Rennie, M., Kanitz, T., Isaksen, L., Checa, E., de Kloe, J., Okunde, N., and Reitebuch, O.: Correction of wind bias for the lidar on\nboard Aeolus using telescope temperatures, Atmospheric Measurement Techniques, 14, 7167–7185, 2021b. Weiler, F., Rennie, M., Kanitz, T., Isaksen, L., Checa, E., de Kloe, J., Okunde, N., and Reitebuch, O.: Correction of wind bias for the lidar on\nboard Aeolus using telescope temperatures, Atmospheric Measurement Techniques, 14, 7167–7185, 2021b. Weissmann, M., Busen, R., Dörnbrack, A., Rahm, S., and Reitebuch, O.: Targeted observations with an airborne wind lidar, Journal of Weissmann, M., Busen, R., Dörnbrack, A., Rahm, S., and Reitebuch, O.: Targeted observations with an airborne wind lidar, Journal of\nAtmospheric and Oceanic Technology, 22, 1706–1719, 2005. Atmospheric and Oceanic Technology, 22, 1706–1719, 2005. 705\nWitschas, B., Lemmerz, C., and Reitebuch, O.: Horizontal lidar measurements for the proof of spontaneous Rayleigh–Brillouin scattering in\nthe atmosphere, Applied Optics, 51, 6207–6219, 2012. Witschas, B., Lemmerz, C., Geiß, A., Lux, O., Marksteiner, U., Rahm, S., Reitebuch, O., and Weiler, F.: First validation of Aeolus wind\nobservations by airborne Doppler wind lidar measurements, Atmospheric Measurement Techniques, 13, 2381–2396, 2020. Witschas, B., Lemmerz, C., and Reitebuch, O.: Horizontal lidar measurements for the proof of spontaneous Rayleigh–Brillouin scattering in\nthe atmosphere, Applied Optics, 51, 6207–6219, 2012. Witschas, B., Lemmerz, C., Geiß, A., Lux, O., Marksteiner, U., Rahm, S., Reitebuch, O., and Weiler, F.: First validation of Aeolus wind Witschas, B., Lemmerz, C., and Reitebuch, O.: Horizontal lidar measurements for the proof of spontaneous Rayleigh–Brillouin scattering in\nthe atmosphere, Applied Optics, 51, 6207–6219, 2012. Witschas, B., Lemmerz, C., Geiß, A., Lux, O., Marksteiner, U., Rahm, S., Reitebuch, O., and Weiler, F.: First validation of Aeolus wind\nobservations by airborne Doppler wind lidar measurements, Atmospheric Measurement Techniques, 13, 2381–2396, 2020. Witschas, B., Lemmerz, C., Geiß, A., Lux, O., Marksteiner, U., Rahm, S., Reitebuch, O., Schäfler, A., and Weiler, F.: Validation of the Aeolus\n10\nL2B wind product with airborne wind lidar measurements in the polar North Atlantic region and in the tropics, Atmospheric Measurement\nTechniques, 15, 7049–7070, 2022. Witschas, B., Lemmerz, C., Geiß, A., Lux, O., Marksteiner, U., Rahm, S., Reitebuch, O., Schäfler, A., and Weiler, F.: Validation of the Aeolus\n710\nL2B wind product with airborne wind lidar measurements in the polar North Atlantic region and in the tropics, Atmospheric Measurement\nTechniques, 15, 7049–7070, 2022. Zuo, H., Hasager, C. B., Karagali, I., Stoffelen, A., Marseille, G.-J., and De Kloe, J.: Evaluation of Aeolus L2B wind product with wind\nprofiling radar measurements and numerical weather prediction model equivalents over Australia, Atmospheric Measurement Techniques, Techniques, 15, 7049–7070, 2022. Zuo, H., Hasager, C. B., Karagali, I., Stoffelen, A., Marseille, G.-J., and De Kloe, J.: Evaluation of Aeolus L2B wind product with wind\nprofiling radar measurements and numerical weather prediction model equivalents over Australia, Atmospheric Measurement Techniques, Zuo, H., Hasager, C. B., Karagali, I., Stoffelen, A., Marseille, G.-J., and De Kloe, J.: Evaluation of Aeolus L2B wind product with wind\nprofiling radar measurements and numerical weather prediction model equivalents over Australia, Atmospheric Measurement Techniques,\n15, 4107–4124, 2022. 715 15, 4107–4124, 2022. 715 15, 4107–4124, 2022. 715 32"