Groundwater Level Data Imputation Using Machine Learning and Remote Earth Observations Using Inductive Bias

Sustainable groundwater management requires an accurate characterization of aquifer-storage change over time. This process begins with analysis historical water levels at observation wells. However, water-level records can be sparse, particularly in developing areas. To address this problem, we developed imputation method to approximate missing monthly averaged groundwater-level observations individual wells since 1948. impute wells, used two global data sources: Palmer Drought Severity Index (PDSI), and the Global Land Data Assimilation System (GLDAS) for regression. In addition meteorological datasets, engineered four additional features encoded temporal as 13 parameters that represent month year observation. extends previous similar work by using inductive bias inform our models on trends structure from existing observations, prior estimates behavior. We formed initial estimating long-term ground priors smoothing. These expected behavior long term allow regression approach perform well, even large gaps up 50 years. demonstrated Beryl-Enterprise aquifer Utah found imputed results follow observed hydrogeological principles, periods no data.