PosterPDF Available

Monitoring the 2013-2014 drought in the São Paulo State in Brazil using GRACE equivalent groundwater measures April 2016 DOI: 10.13140/RG.2.1.3022.1207 Conference: DLR Conference on Climate Change, "Challenges for Atmospheric Research", in collaboration with the United Nations Office for Outer Space Affairs (UNOOSA)



in collaboration with the United Nations Office for Outer Space Affairs
Monitoring the 2013-2014 drought in the São Paulo State in Brazil using GRACE
equivalent groundwater measures
Ademir Xavier1,2 , Sergio Celaschi1
1Divisão de Tecnologia de Redes, CTI Renato Archer, CEP 13069-901, Campinas, SP, Brazil
2 Fundação de Apoio à Capacitação em TI, FACTI, 13069-901, Campinas, SP, Brazil
The period between 2013 and 2014 was characterized by a severe drought in the South American southeast, in particular, in the most populated
state of Brazil, São Paulo, with the influence of a high-pressure zone at 6000 m above sea level that blocked most of the humid fronts from the
Amazon forest (South Atlantic convergence zone, SACZ). The drought started by the end of the dry season in the southern hemisphere, October
2013, triggering one of the most severe water shortage crisis in the city of São Paulo. We present a study based on remote sensed GRACE
(Gravity Recovery and Climate Experiment) data for the state and the city of São Paulo showing a correlation between groundwater equivalents
and surface data obtained from the largest water supply system (WSS) of the city. Moreover, we provide advanced forecasts for the annual peak
recovery using both GRACE and WSS data confirming the existence of a time lag between surface and water ground equivalent measures.
by A Xavier, march 2016 CTI/FACTI -
ÕPlot of the time series volume variation for the main São Paulo city reservoir (serving 11 million people) , indicating a sharp
decline after August 2013 (the drought started at the end of the dry season!).
ØComparison between two extreme situations: "Isothickness" lines using GRACE data for EWT (in cm) for the São Paulo State in
October 2010 and October 2014. Map borders indicate the position of all São Paulo municipalities including the capital.
São Paulo city
Cantareira Water System
The investigation on the resurgence of the yearly peak in the water system accumulated
volume used a series of tools from statistical learning. Essentially, the aim of this work was to
project a statistical regressor capable of predicting the time and amplitude of the yearly peak
in both pluviometric (surface) and GRACE (satellite) time series – Equivalent Water Thickness
(EWT) – using as scenario the city of S. Paulo. Application of Least-Squares Support Vector
Machine (LS-SVM) to GRACE water equivalent data: LS-SVM is a statistical learning technique
that successfully traces a boundary frontier in an extended mathematical space allowing
quantitative prediction. In our case, GRACE time series was further complemented by surface
precipitation, which clearly shows a percolation dynamics (water thickness equivalents were 1-
2 month delayed in relation to surface precipitation data). Summary on LS-SVM: given the
dataset (N training instance number) ሼԦݔǡݕ௞ୀଵ
, minimize:
ܬݓǡ݁ ͳ
Subjected to ݕൌݓ߮ Ԧݔ൅ܾ൅݁
Such problem is best solved by applying a Lagrangian approach in which ߙare Lagrangean multipliers
ܮൌܬݓǡ݁ െ
ߙݓ߮ Ԧݔ൅ܾ൅݁
After a solution is found, the regressor is given by
ߙܭ Ԧݔǡ Ԧݔ൅ܾ
with ܭԦݔǡ Ԧݔൌ݁ݔ݌െ ԦݔെԦݔȀߪ.
Investigation started here
update (Dec-15)
References: Suykens JA, Van Gestel T, De Moor B and Vandewalle J (2002). Basic Methods of Least Squares Support Vector Machines. In Least Squares Support Vector Machines. World Scientific Publishing Co. Pte. Ltd; Famiglietti, J. S. (2014). The
global groundwater crisis. Nature Clim. Change, 4(11), 945–948.; Swenson, S. C. (2002). Techniques for recovering surface mass variability from satellite measurements of time-variable gravity, Ph.D. thesis, 133 pp, University of Colorado at Boulder.
Acknowledgments: We would like to thank financial support by the Brazilian Ministry of Science and Innovation (MCTI) and CEMADEN (Brazilian Center for Monitoring and Disaster Alerts). Special thanks go to NASA JPL teams for providing GRACE
land data as available at GRACE mission is supported by the NASA MEaSUREs Program and DLR.
ÕData processing based on
LS-SVM. Three sets of data are
extracted: training, validation
and “last”. Parameters are
optimized by minimizing the
Euclidian distance between
validation data and regressor
forecast (based on two
parameters: advanced
position (n) and regressor
dimension (p). Finally, optimal
parameters are applied to last
data available, providing the
ØValidation results for optimized
parameters and several advanced
periods (using GRACE SP-EWT).
ØDependence of the
error as a function of
information (dimension).
advanced the return of
peak for GRACE SP-
confirming the reduction
amplitude of the
maxima. For some
reason, minimum
generalization error
obtained using 22 months
previous data.
View publication statsView publication stats
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.