Using Long Short-Term Memory networks to connect water table depth anomalies to precipitation anomalies over Europe

Abstract. Many European countries rely on groundwater for public and industrial water supply. Due to a scarcity of near-real-time table depth (wtd) observations, establishing spatially consistent monitoring system at the continental scale is challenge. Hence, it necessary develop alternative methods estimating wtd anomalies (wtda) using other hydrometeorological observations routinely available near real time. In this work, we explore potential Long Short-Term Memory (LSTM) networks producing monthly wtda precipitation (pra) as input. LSTM are special category artificial neural that useful detecting long-term dependency within sequences, in our case time series, which expected relationship between pra wtda. proposed methodology, spatiotemporally continuous data were obtained from daily terrestrial simulations Terrestrial Systems Modeling Platform (TSMP) over Europe (hereafter termed TSMP-G2A set), with spatial resolution 0.11∘, ranging years 1996 2016. The separated into training set (1996–2012), validation (2013–2014), test (2015–2016) establish local selected pixels across Europe. modeled maps agreed well distributed dry wet events, 2003 2015 constituting drought Moreover, categorized performances based intervals yearly averaged wtd, evapotranspiration (ET), soil moisture (θ), snow equivalent (Sw), type (St), dominant plant functional (PFT). Superior performance was found < 3 m, ET > 200 mm, θ>0.15 m3 m−3, Sw<10 revealing significant impact factors ability process information. Furthermore, results cross-wavelet transform (XWT) showed change temporal pattern some pixels, can be reason undesired network behavior. Our demonstrate high-quality measured predicted large scales, such pra. This contribution may facilitate establishment an effective relevant management.