An Indirect Approach Based on Long Short-Term Memory Networks to Estimate Groundwater Table Depth Anomalies Across Europe With an Application for Drought Analysis

This study introduced a number of hydrometeorological variables in addition to precipitation anomaly (pr_a) in the construction of Long Short-Term Memory (LSTM) networks to arrive at improved water table depth anomaly (wtd_a) at individual pixels over Europe in various experiments. The input variables in the experiments E1 and E2 are: E1.1: pr_a; E1.2: evapotranspiration anomaly (ET_a); E1.3: soil moisture anomaly (θ_a); E1.4: pr_a and ET_a; E1.5: pr_a and θ_a; E1.6: ET_a and θ_a; E1.7: pr_a, ET_a and θ_a; E2.1: pr_a, θ_a and scaled yearly averaged snow water equivalent (SWE_scaled); E2.2: pr_a, θ_a at the selected pixels and adjacent pixels; and E2.3: pr_a, θ_a at the selected pixels close to rivers and river stage anomaly (rs_a) at the adjacent pixels. The data files provide the TSMP-G2A ET_a, the TSMP-G2A θ_a, and the LSTM wtd_a data (obtained from E1.2 to E2.3) for the period 1996-2016, with a spatial resolution of 0.11 degrees. The TSMP-G2A yearly averaged snow water equivalent (SWE) data are also provided.