GRAiCE: Terrestrial water storage anomalies reconstructions

Terrestrial Water Storage (TWS) is the total amount of freshwater stored on and below the Earth’s land surface, including surface water, groundwater, soil moisture, snow, and ice. As a result, TWS is a crucial variable of the global hydrologic cycle, representing an essential indicator of water availability. Since 2002, the Gravity Recovery and Climate Experiment (GRACE) mission and its follow-on (GRACE-FO) have been measuring temporal and spatial variations of TWS, namely the Terrrestrial Water Storage Anomalies (TWSA), enabling the monitoring of global hydrological changes over the last two decades. However, the lack of observations prior to 2002 along with the temporal gaps in GRACE/GRACE-FO time series limit our understanding of long-term variations of global freshwater availability. In this study, we use Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) neural networks and two sets of predictors to develop four global monthly reconstructions of TWSA from 1984 to 2021 at 0.5º spatial resolution (GRAiCE). The first set of predictors is given by a combination of five fundamental meteorological forcings and data on vegetation dynamics, whereas the second set of predictors includes the five meteorological forcings only. Specifically, the meteorological predictors are monthly averaged data of total precipitation, snow depth water equivalent, surface net solar radiation, surface air temperature, and surface air relative humidity. We derive data on vegetation dynamics from a long-term reconstruction of solar-induced fluorescence (SIF), which represents a proxy for photosynthesis. Each model is trained with monthly TWSA data from the GRACE JPL mascon dataset. The GRAiCE dataset accurately reproduces GRACE/GRACE-FO observations at the global scale and across different climatic regions. Moreover, we found that our models predict observed TWSA better than a previous reference reconstruction and produce reliable estimates of the water budget at the river basin scale. Beyond generating long-term continuous TWSA time series, our models allow us to detect and examine TWS changes due to climate variability/change. This repository contains the GRAiCE dataset and includes four files in netCDF format. The dataset provides monthly TWSA estimates from 1984 to 2021 at a 0.5º spatial resolution. TWSA values are expressed in terms of cm of equivalent water thickness. GRAiCE_LSTM.nc and GRAiCE_BiLSTM.nc files contain TWSA reconstructions obtained from LSTM and BiLSTM models fed with all predictors (i.e., including SIF data), respectively. GRAiCE_LSTMnoSIF.nc and GRAiCE_BiLSTMnoSIF.nc files contain TWSA reconstructions obtained from LSTM and BiLSTM models fed with meteorological forcings only (i.e., without SIF data).