Self‐Validating Deep Learning for Recovering Terrestrial Water Storage From Gravity and Altimetry Measurements

Irrgang, Christopher ORCIDiD
Saynisch‐Wagner, Jan
Dill, Robert ORCIDiD
Boergens, Eva ORCIDiD
Thomas, Maik

DOI: https://doi.org/10.1029/2020GL089258
Persistent URL: http://resolver.sub.uni-goettingen.de/purl?gldocs-11858/8479
Irrgang, Christopher; Saynisch‐Wagner, Jan; Dill, Robert; Boergens, Eva; Thomas, Maik, 2020: Self‐Validating Deep Learning for Recovering Terrestrial Water Storage From Gravity and Altimetry Measurements. In: Geophysical Research Letters, 47, 17, DOI: https://doi.org/10.1029/2020GL089258. 
 
Saynisch‐Wagner, Jan; 1 Section 1.3: Earth System Modelling Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences Potsdam Germany
Dill, Robert; 1 Section 1.3: Earth System Modelling Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences Potsdam Germany
Boergens, Eva; 1 Section 1.3: Earth System Modelling Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences Potsdam Germany
Thomas, Maik; 1 Section 1.3: Earth System Modelling Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences Potsdam Germany

Abstract

Quantifying and monitoring terrestrial water storage (TWS) is an essential task for understanding the Earth's hydrosphere cycle, its susceptibility to climate change, and concurrent impacts for ecosystems, agriculture, and water management. Changes in TWS manifest as anomalies in the Earth's gravity field, which are routinely observed from space. However, the complex underlying distribution of water masses in rivers, lakes, or groundwater basins remains elusive. We combine machine learning, numerical modeling, and satellite altimetry to build a downscaling neural network that recovers simulated TWS from synthetic space‐borne gravity observations. A novel constrained training is introduced, allowing the neural network to validate its training progress with independent satellite altimetry records. We show that the neural network can accurately derive the TWS in 2019 after being trained over the years 2003 to 2018. Further, we demonstrate that the constrained neural network can outperform the numerical model in validated regions.


Plain Language Summary: Continuous monitoring of the distribution and movement of continental water masses is essential for understanding the Earth's global water cycle, its susceptibility to climate change, and for risk assessments of ecosystems, agriculture, and water management. Changes of continental water masses are encoded as coarse blob‐like patterns in satellite observations of the Earth's gravity field. Focusing on the South American continent, we introduce a self‐validating artificial neural network to recover detailed and accurate spatiotemporal information of continental water masses from such gravity field observations.


Key Points:

South American terrestrial water storage (TWS) is derived from satellite gravity observations with deep learning. A neural network accurately predicts multiscale monthly TWS anomalies in 2019 based on training data from 2003 to 2018. A data assimilation‐like training is introduced, allowing the neural network to validate itself with independent altimetry records.