text
stringlengths
0
820
SMimaginar y
VODreal
VODimaginar y
2011 2012 2013 2014 2015 2016 201700.020.040.060.08–0.8–0.40.40.8
0
Corr (ENSO4, SM)
ENSO4 --> SM
(a) (b)
Authorized licensed use limited to: ASU Library. Downloaded on March 07,2024 at 22:07:36 UTC from IEEE Xplore. Restrictions apply.
100
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2021FIGURE 9. (continued ) (c) Convergent cross mapping. An example of the unbiased CCM application used in [88] to derive causal rela -
tionships among the variables accounting for photosynthesis [the gross primary productivity (GPP) from FLUXCOM initiative (www
.fluxcom.org)], temperature (T air from ERA Interim), and SM (from the European Space Agency’s climate change initiative [(v 2.0)]. Data
cubes at 0.5° and eight-day spatial and temporal resolutions, respectively, spanning 2001–2002 were used. Reasonable spatial-causal
patterns are observed for SM and T air on GPP; GPP drives T air mostly in cold ecosystems (probably due to changes in land surface albe -
do, such as snow/ice to vegetation changes); SM is mostly controlled by T air, which partially drives evaporation in water-limited regions;
and GPP dominates SM. (d) The additive noise model. Structural equation models in the form of an additive noise model using the
kernels in [21] for hypothesis testing. Assessing cause-and-effect relationships is also possible when time is not involved. Here we rely
on a look-up table generated by radiative transfer model (RTM), which gives the right direction of causation: state vectors (parameters)
cause radiances. The algorithms accurately detect this from pairs of data and can be used for retrieval model-data intercomparison and
RTM assessment.
1
0.8
0.6
0.4
0.2
0True Positive Rate
0 0.2 0.4 0.6 0.8 1
False-Positive-Rate
(d)C (0.7667)"
Cs(0.7953)"
Photosynthesis
Strength of Forcing Over SM
50
–50
–150 –100 –50 05 0100 150Latitude (°)0GPP
Temperature
Longitude (°)Strength of Forcing Over Air Temperature
50
–50
–150 –100–50 05 0100 150Latitude (°)0GPP
SM
Longitude (°)
Strength of Forcing Over GPP
50
–50
–150 –100 –50 05 0100 150Latitude (°)0Temperature
SM
Longitude (°)Air Temperature SM
(c)
Authorized licensed use limited to: ASU Library. Downloaded on March 07,2024 at 22:07:36 UTC from IEEE Xplore. Restrictions apply.
101
JUNE 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINEwidely used approach in Earth and climate sciences to
quantitatively identify relationships between time series. It
tests whether including past states of a variable, X, improves
the prediction of an output variable, Y, more than consider -
ing other covariates. GC is a linear test, but nonlinear (ker -
nel) versions have also been proposed [ 85]. In [ 86], a gen -
eralized kernel GC is presented, able to discover footprints
of El Niño-Southern Oscillation (ENSO) on soil moisture
(SM) and vegetation optical depth records [see Figure 9 (a)].
However, GC approaches have problems in nonstationary,
nonlinear, and deterministic relationships, especially in
dynamic systems with weak-to-moderate coupling.
The second family considers nonlinear state-space meth -
ods, such as convergent cross mapping (CCM) [ 87]. CCM
attempts to address GC problems by reconstructing the
variable’s state spaces, ( Mx, My), using time embeddings
and concludes, based on XY", whether points on Mx can
be predicted more accurately using nearest neighbors in My
as more points are used for prediction. However, CCM is
very sensitive to noise and time-series length. Recent works
have included bootstrap resampling to alleviate such prob -
lems and shown good results in identifying causal links
in long global records of carbon and water fluxes [ 88] [see
Figure 9 (b)].
The third family, collectively known as causal network
learning algorithms , relies heavily on conditional-inde -
pendence tests. Its methods iteratively remove the links
between pairs of variables, ( X, Y), if they are found to be
independently conditioned on any subset of the other vari -
ables. The PC algorithm (named after its inventors Peter
and Clark) allows us to identify parents and can be flexibly
implemented with different kinds of conditional-indepen -
dence tests, which can handle nonlinear dependencies and
variables that are discrete or continuous and are univariate
or multivariate. Finally, structural causal models (SCMs)
are used when time is not involved or the sampling frequen -
cy is too low. SCMs search for the causal direction within
Markov-equivalent classes by exploiting the asymmetries
between cause and effect. Additive noise models rely on the
principle of independence between the cause and the gen -
erating mechanism and have recently shown good results
in remote sensing and geosciences in cases where time is
not involved and only two variables are observed [ 21].
PERSPECTIVES
Although the field of machine and deep learning has tra -
ditionally progressed very rapidly, we observe that this is
not the case in tackling the challenge of learning causal
relationships from Earth observation data. The role that