Investigating the control of ocean-atmospheric
oscillations over global terrestrial evaporation using
a simple supervised learning method
Brecht Martens ∗ , Diego G. Miralles ∗ , Wouter A. Dorigo † , Willem Waegeman
‡
and Niko E.C. Verhoest
∗
∗ Laboratory
of Hydrology and Water Management, Ghent University, Ghent, Belgium
of Geodesy and Geo-Information, Vienna University of Technology, Vienna, Austria
‡ Department of Mathematical Modelling, Statistics and Bioinformatics, Ghent University, Ghent, Belgium
† Department
Abstract—Intra-annual and multi-decadal variations in the
Earth’s climate are to a large extent driven by periodic oscillations in the coupled state of atmosphere and ocean. Because
changes in climate impact terrestrial ecosystems, these oceanatmospheric oscillations are expected to influence land evaporation as well. In this study, we investigate the control of oceanatmospheric oscillations over global terrestrial evaporation using
a simple learning method and a large dataset of satellite-based
observations of climatic variables. Our results show that the
simple learning method is able to quantify this control and allows
for a better understanding of the link between ocean-atmosphere
dynamics and terrestrial bio-geochemical cycles. Finally, our
approach may help improve the prediction of changes in the
global water cycle.
II. M ETHODS AND DATA
A three-step approach (Figure 1) is used to investigate the
effect of ocean-atmospheric oscillations on terrestrial evaporation, and to identify the climatic drivers through which these
oscillations are acting upon land evaporation. First, the impact
of 16 leading ocean-atmospheric oscillations – diagnosed by
their corresponding climate oscillation indices (COIs) – on
land evaporation is analyzed using the LASSO regression
method [1]. The LASSO is chosen here because of its unique
feature to perform both variable selection and regularization
at the same time by finding the coefficients that minimize the
following objective function (OFLASSO ):
I. I NTRODUCTION
Intra-annual and multi-decadal variations in the Earth’s
climate are to a large extent driven by periodic oscillations
in the coupled state of atmosphere and ocean (e.g. El Niño
Southern Oscillation). Typically, these oscillations alter not
only the climate in nearby regions, but can also have an
important impact on the local climate in remote areas, a
phenomenon that is often referred to as ‘teleconnection’.
Because changes in local climate impact terrestrial ecosystems
through a series of complex processes and feedbacks, oceanatmospheric teleconnections are expected to influence land
evaporation, which is – primarily – driven by the available
energy at the land surface, near-surface air temperature and
water availibility.
Quantifying the impact of ocean-atmospheric oscillations
on terrestrial evaporation and understanding the mechanisms
through which these oscillations act on the latter are critical to
discern the effects of climate change on the global terrestrial
water cycle. Here, we investigate the effects of these intraannual and multi-decadal oscillations on global terrestrial
evaporation using satellite-based observations of different essential climate variables, and a simple supervised learning
method called the Least Absolute Shrinkage and Selection
Operator regression (LASSO, [1]). The objectives are thus to
(1) identify the dominant ocean-atmospheric oscillations over
terretrial evaporation for different regions across the globe, and
(2) find the climatic drivers of terrestrial evaporation through
which these oscillations are affecting the evaporative flux from
the land.
OFLASSO =
n
X
i=1

yi − β0 −
p
X
j=1
2
βj xij  + λ
p
X
|βj | (1)
j=1
where yi is the response at record i (i.e. terrestrial evaporation at i), n is the number of records in the dataset, β0
is the fitted intercept, βj is the fitted regression coefficient
corresponding to feature j, xij is the value of feature j at
record i (i.e. COI j at i), λ is the regularization parameter –
optimized using a 5-fold cross-validation – p is the number
of features in the model (i.e. 16), and βj is the regression
coefficient corresponding to feature j. As shown in Equation
1, the LASSO will shrink redundant regression coefficients –
related to uninformative features – towards zero, by applying
a penalty on the regression coefficients, resulting in a reduced
and simpler model. Note that before applying the LASSO
regression, all features and the response are standardized [1].
The 16 COIs used in this study are sourced from the National
Oceanic and Atmospheric Administration (NOAA) National
Weather Service (www.cpc.ncep.noaa.gov) and characterize
the major modes of ocean-atmospheric variability at interannual and intra-annual time scales. For terrestrial evaporation,
the Global Land Evaporation Amsterdam Model (GLEAM,
[2], [3]) v3.0a dataset is used, which spans the period 1980–
2014. GLEAM is a set of algorithms specifically designed
to estimate terrestrial evaporation and root-zone soil moisture
from satellite data. The dataset of terrestrial evaporation is
available at a spatial resolution of 0.25◦ and is re-processed
to obtain monthly anomalies of evaporation.
Fig. 1. Schematic overview of the three-step approach used here to investigate
the control of ocean-atmospheric oscillations over global terrestrial evaporation.
In Step 2, the control of the dominant climatic drivers on
terrestrial evaporation (i.e. net radiation, air temperature and
precipitation) across the globe is investigated using the same
LASSO approach as in Step 1. This analysis allows to detect
through which climatic drivers the principal COIs, as defined
in the first step, affect the land evaporation. The datasets of
net radiation and air temperature from the latest ERA-Interim
reanalysis are used here [4], together with the Multi-Source
Weighted-Ensemble Precipitation dataset (MSWEP, [5]). The
latter is a merger of satellite-based precipitation retrievals,
reanalysis data and in situ measurements.
To close the loop, in Step 3, we confirm whether the selected
COIs from Step 1 effectively impact the dominant climatic
drivers, by regressing the COIs against each of the climatic
drivers of terrestrial evaporation using the LASSO method.
III. R ESULTS
As an example, Figure 2 shows the coefficients of two
selected COIs from the first step in the analysis (see also
Figure 1): the Southern Oscillation Index (SOI, diagnosing
the El Niño Southern Oscilation) and the East Atlantic West
Russia index (EAWR, diagnosing the East Atlantic West
Russia pattern).
As shown in the top pannel of Figure 2, dynamics in
terrestrial evaporation over Australia, the South of Africa and
the Amazon basin seem to be strongly dominated by the El
Niño Southern Oscillation pattern. In addition, the dependency
seems to be positive in most of these regions, suggesting
that more evaporation occurs during positive anomalies in the
SOI (i.e. during La Niña), and less water is pumped into the
atmosphere during its negative phase (i.e. during El Niño).
An exception is the East and Central part of the Amazon,
where the opposite is occuring. Figure 3 – showing a trivariate
colormap [6] of the three regression coefficients resulting from
Step 2 in the analysis (see also Figure 1) – indicates that
evaporation in large parts of Australia, the South of Africa and
the East of the Amazon basin is strongly controlled by water
availibility (i.e. P ), suggesting an important impact of the El
Niño Southern Oscillation pattern on the moisture supply in
these regions. Exceptions are the North and Central part of
Australia and the East and Central part of the Amazon basin,
Fig. 2. LASSO regression coefficients for the Southern Oscillation Index
(SOI, top panel) and the East Atlantic West Russia index (EAWR, bottom
panel).
Fig. 3. Nemani map of the three LASSO regression coefficients resulting
from the regression of E against its principal climatic drivers; i.e. net radiation
(Rn ), air temperature (Ta ) and precipitation (P )
where the relative control of radiation on evaporation seems
to increase. This suggest an impact of the El Niño Southern
Oscillation on radiation as well, or decreases in interception
loss due to a reduced supply of rainfall as hypothesized by
Miralles et al. [7]. Future research should disentangle the
processes in these regions. Note that the results discussed
here are consistent with the ones obtained in Mirealles et al.
[7], which focussed on the impact of the El Niño Southern
Oscillation pattern on terrestrial evaporation.
The bottom pannel of Figure 2 shows a clear negative
relation between terrestrial evaporation and the EAWR in the
East of Europe and the West of Russia, suggesting negative
anomalies in E during the positive phase of the EAWR and
vice versa. According to Figure 3, the evaporation in these
regions is strongly radiation-limited, suggesting a negative
impact of the EAWR on the available energy.
IV. C ONCLUSION
Using a simple supervised learning method, the LASSO
regression, this study investigates the control of oceanatmospheric oscillations on global terrestrial evaporation. To
this end, we proposed a three-step approach to (1) detect the
dominant ocean-atmospheric oscillations affecting evaporation
for each region across the globe, (2) analyze the controlling climatic drivers (i.e. net radiation, air temperature and
precipitation) of evaporation in these same regions, and (3)
confirm the impact of the ocean-atmospheric oscillations on
the main climatic drivers through which terrestrial evaporation
is affected.
Preliminary results of this study show that the LASSO
regression is able to detect the control of these oscillations
over terrestrial evaporation and that the second step in the
analysis allows us to discern through which climatic variables
these remote ocean-atmospheric oscillations are acting upon
the regional dynamics in land evaporation. The results from
further studies will allow for a better understanding of the
link between ocean-atmosphere dynamics and terrestrial biogeochemical cycles, and may help improve the prediction of
future changes in the global water cycle.
R EFERENCES
[1] R. Tibshirani, “Regression shrinkage and selection via the lasso,” Journal
of the Royal Statistical Society. Series B (Methodological), pp. 267–288,
1996.
[2] D. G. Miralles, T. R. H. Holmes, R. A. M. De Jeu, J. H. Gash, A. G. C. A.
Meesters, and A. J. Dolman, “Global land-surface evaporation estimated
from satellite-based observations,” Hydrology and Earth System Sciences,
vol. 15, no. 2, pp. 453–469, 2011.
[3] B. Martens, D. Miralles, H. Lievens, R. van der Schalie, R. de Jeu,
D. Fernández-Prieto, H. Beck, W. Dorigo, and N. Verhoest, “Gleam v3:
satellite-based land evaporation and root-zone soil moisture,” Geoscientific Model Development Discussions, pp. 1–36, 2016.
[4] D. P. Dee, S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli,
S. Kobayashi, U. Andrae, M. A. Balmaseda, G. Balsamo, P. Bauer,
P. Bechtold, A. C. M. Beljaars, L. van de Berg, J. Bidlot, N. Bormann,
C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. Haimberger, S. B.
Healy, H. Hersbach, E. V. Hólm, L. Isaksen, P. Kallberg, M. Köhler,
M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J.-J. Morcrette, B.-K.
Park, C. Peubey, P. de Rosnay, C. Tavolato, J.-N. Thépaut, and F. Vitart,
“The ERA-Interim reanalysis: configuration and performance of the data
assimilation system,” Quarterly Journal of the Royal Meteorological
Society, vol. 137, no. 656, pp. 553–597, 2011.
[5] H. E. Beck, A. I. J. M. van Dijk, V. Levizzani, J. Schellekens, D. G.
Miralles, B. Martens, and A. de Roo, “Mswep: 3-hourly 0.25◦ global
gridded precipitation (1979-2015) by merging gauge, satellite, and reanalysis data,” Hydrology and Earth System Sciences Discussions, vol.
2016, pp. 1–38, 2016.
[6] R. Nemani, C. Keeling, H. Hashimoto, W. Jolly, S. Piper, C. Tucker,
R. Myneni, and S. Running, “Climate-driven increases in global terrestrial
net primary production from 1982 to 1999,” Science, vol. 300, pp. 1560–
1563, 2003.
[7] D. G. Miralles, M. J. van den Berg, J. H. Gash, R. M. Parinussa, R. A. M.
de Jeu, H. E. Beck, T. R. H. Holmes, C. Jiménez, N. E. C. Verhoest, W. A.
Dorigo, A. J. Teuling, and A. J. Dolman, “El Niño-La Niña cycle and
recent trends in continental evaporation,” Nature Climate Change, vol. 4,
no. 2, pp. 122–126, 2014.