Assessing trends in irrigation water withdrawal using a data-driven temporal extrapolation framework

Abstract. Assessing long-term trends in irrigation water withdrawal (IWW) is essential for understanding the impacts of climate change on water resources and agricultural systems, yet long-term observations remain scarce. In France, annual IWW observations are only available from 2008 onward through the national water withdrawal database, limiting the analysis of long-term changes. To address this limitation, we develop a data-driven framework to reconstruct annual catchment-scale IWW over the 2000–2022 period. The approach relies on a generalized additive model (GAM), using structural (e.g., area equipped for irrigation) and hydro-meteorological (e.g., monthly mean temperature) predictors. The model explicitly identifies inter-catchment and inter-annual variability through a decomposition of total IWW into a base IWW term and a relative deviation term. The model is evaluated using a series of Leave-One-Period-Out cross-validation experiments. Results show robust predictive performance across all experiments (RMSE < 2 mm/yr) and reliable uncertainty estimates. The framework was then used to reconstruct annual IWW over the 2000–2007 period and to generate continuous time series of IWW at the catchment scale over 2000–2022. The reconstructed time series reveal widespread positive trends, with significant increases detected in 92 % of catchments. Trend analysis indicates that changes in total IWW are primarily driven by the relative deviation component. A regional synthesis further identifies the Mediterranean region as a major hotspot, combining high IWW and strong IWW positive trends.