Final cleanup: remove duplicate section header, alphabetise refs, remove duplicate Eval 6 method paragraph

paper/paper.md

@@ -131,8 +131,6 @@ All downstream tasks use the trained VAE without modification — only the embed
 
 **Thermal comfort mapping (Eval 6):** Universal Thermal Climate Index (UTCI; Bröde et al., 2012) is computed at all 40 stations from measured air temperature, humidity, wind speed, and global solar radiation. Mean radiant temperature — required for UTCI but not directly measured — is estimated from global solar radiation using the solar gain method (Thorsson et al., 2007): solar position from pvlib (Holmgren et al., 2018), irradiance decomposition via Erbs et al. (1982), and MRT delta from the pythermalcomfort library (Tartarini and Schiavon, 2020). Wind speed is adjusted from 2 m measurement height to the 10 m reference height required by UTCI using a log-law wind profile. The same five stations held out in Eval 1 are held out here. A linear probe (Ridge regression) is trained to predict UTCI from the 6-d embedding on training stations, then applied to embeddings averaged from 5-nearest-neighbour training stations to interpolate UTCI at the held-out locations. A raw-variable baseline — averaging the raw weather observations from the same 5 neighbours, then computing UTCI from the averaged values — provides a direct comparison.
 
-**Thermal comfort mapping (Eval 6):** Universal Thermal Climate Index (UTCI; Bröde et al., 2012) is computed at all 40 stations from measured air temperature, humidity, wind speed, and global solar radiation. Mean radiant temperature — required for UTCI but not directly measured — is estimated from global solar radiation using the solar gain method (Thorsson et al., 2007): solar position from pvlib (Holmgren et al., 2018), irradiance decomposition via Erbs et al. (1982), and MRT delta from the pythermalcomfort library (Tartarini and Schiavon, 2020). Wind speed is adjusted from 2 m measurement height to the 10 m reference height required by UTCI using a log-law wind profile. The same five stations held out in Eval 1 are held out here. A linear probe (Ridge regression) is trained to predict UTCI from the 6-d embedding on training stations, then applied to embeddings averaged from 5-nearest-neighbour training stations to interpolate UTCI at the held-out locations. A raw-variable baseline — averaging the raw weather observations from the same 5 neighbours, then computing UTCI from the averaged values — provides a direct comparison.
-
 **Extreme scenario generation (§5.8):** To generate extreme weather scenarios, we first identify directions in the VAE's latent space that correspond to individual meteorological variables. A linear regression from the 6-d embedding to each observed variable yields a coefficient vector that defines the latent direction along which that variable changes most. This is a one-time direction-finding step — the actual weather generation is performed entirely by the VAE decoder. Scenarios are produced by shifting the full year of campus-mean embeddings along the identified directions (or combinations thereof), calibrating the shift magnitude via binary search to achieve a target perturbation, and decoding the shifted embeddings back to weather space through the VAE. The decoder enforces physically coherent cross-variable correlations learned from the training data, producing simultaneous changes in all six variables that are consistent with real observations. Decoded values are clipped to physical bounds and converted to EnergyPlus Weather (EPW) format with derived fields (dew point via the Magnus formula, radiation decomposition via the Erbs model, sky cover from clearness index, and infrared radiation via the Martin-Berdahl model). Three scenarios are generated: a baseline (observed campus-mean), a heatwave (+2°C), and an urban heat island intensification (+2°C with 23% wind reduction).
 
 ---
@@ -340,10 +338,6 @@ The temperature shift is uniform across the diurnal cycle (±0.3°C variation fr
 
 The generated EPW files are released alongside the dataset at https://huggingface.co/citysyntaxlab/campus-weather, enabling building energy researchers to evaluate cooling load sensitivity to heatwave and urbanisation scenarios using site-specific, multivariate-consistent weather data rather than simple uniform temperature offsets applied to airport TMY files.
 
-
-
-
-
 ---
 
 ## 6. Discussion
@@ -408,15 +402,13 @@ The NUS-40 dataset, trained model weights, embeddings, and all evaluation code a
 
 ## References
 
-Chandola, V., Banerjee, A., and Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys, 41(3), 1–58.
 Bröde, P., Fiala, D., Blazejczyk, K., Holmér, I., Jendritzky, G., Kampmann, B., Tinz, B., and Havenith, G. (2012). Deriving the operational procedure for the Universal Thermal Climate Index (UTCI). International Journal of Biometeorology, 56(3), 481–494.
 
-
+Chandola, V., Banerjee, A., and Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys, 41(3), 1–58.
 
 Crawley, D. B., Lawrie, L. K., Winkelmann, F. C., et al. (2001). EnergyPlus: creating a new-generation building energy simulation program. Energy and Buildings, 33(4), 319–331.
-Erbs, D. G., Klein, S. A., and Duffie, J. A. (1982). Estimation of the diffuse radiation fraction for hourly, daily and monthly-average global radiation. Solar Energy, 28(4), 293–302.
-
 
+Erbs, D. G., Klein, S. A., and Duffie, J. A. (1982). Estimation of the diffuse radiation fraction for hourly, daily and monthly-average global radiation. Solar Energy, 28(4), 293–302.
 
 Fortuin, V., Barber, D., and Rätsch, G. (2020). GP-VAE: Deep probabilistic time series imputation. Proceedings of the 23rd International Conference on Artificial Intelligence and Statistics (AISTATS).
 
@@ -425,9 +417,8 @@ Gao, H., Jiang, R., Dong, Z., Deng, J., Ma, Y., and Song, X. (2024). Spatial-tem
 Guo, R., Yang, B., Guo, Y., Li, H., Li, Z., Zhou, B., Hong, B., and Wang, F. (2024). Machine learning-based prediction of outdoor thermal comfort: Combining Bayesian optimization and the SHAP model. Building and Environment, 254, 111301.
 
 Hasan, A., Roozbehani, M., and Dahleh, M. (2024). WeatherFormer: A pretrained encoder model for learning robust weather representations from small datasets. arXiv:2405.17455.
-Holmgren, W. F., Hansen, C. W., and Mikofski, M. A. (2018). pvlib python: a python package for modeling solar energy systems. Journal of Open Source Software, 3(29), 884.
-
 
+Holmgren, W. F., Hansen, C. W., and Mikofski, M. A. (2018). pvlib python: a python package for modeling solar energy systems. Journal of Open Source Software, 3(29), 884.
 
 Hsu, W.-N., Zhang, Y., and Glass, J. (2017). Unsupervised learning of disentangled and interpretable representations from sequential data. Advances in Neural Information Processing Systems, 30.
 
@@ -442,12 +433,11 @@ Oke, T. R., Mills, G., Christen, A., and Voogt, J. A. (2017). Urban Climates. Ca
 Salman, A. G., Kanigoro, B., and Heryadi, Y. (2015). Weather forecasting using deep learning techniques. Proceedings of the International Conference on Advanced Computer Science and Information Systems (ICACSIS), 281–285.
 
 Shashua-Bar, L., Pearlmutter, D., and Erell, E. (2009). The cooling efficiency of urban landscape strategies in a hot dry climate. Landscape and Urban Planning, 92(3–4), 179–186.
+
 Tartarini, F. and Schiavon, S. (2020). pythermalcomfort: A Python package for thermal comfort research. SoftwareX, 12, 100578.
 
 Thorsson, S., Lindberg, F., Eliasson, I., and Holmer, B. (2007). Different methods for estimating the mean radiant temperature in an outdoor urban setting. International Journal of Climatology, 27(14), 1983–1993.
 
-
-
 Toparlar, Y., Blocken, B., Maiheu, B., and van Heijst, G. J. F. (2017). A review on the CFD analysis of urban microclimate. Renewable and Sustainable Energy Reviews, 80, 1613–1640.
 
 Wang, H., Yang, J., Chen, G., Ren, C., and Zhang, J. (2023). Machine learning applications on air temperature prediction in the urban canopy layer: A critical review of 2011–2022. Urban Climate, 49, 101499.