<p>A data-driven framework for establishing groundwater reference levels through hydrogeological process analysis is applied to the unconfined Chou-Shui Chi alluvial aquifer, central Taiwan. The model conceptualizes each monitoring well as an independent single-tank system, where groundwater levels fluctuate in response to rainfall recharge, pumping, and lateral flow. These processes are integrated into an ordinary differential equation, discretized using the Euler method, enabling the calibration of the model parameters through a least-squares curve-fitting approach. The model parameters, including the recharge coefficient (<i>α</i>), pumping coefficient (<i>β</i>), and lateral flow coefficient (<i>λ</i>), were calibrated using historical groundwater monitoring data from 33 wells. The model demonstrated strong predictive capability, with root-mean-square errors (RMSE) ranging from 0.07 to 0.24&#xa0;m. Model validation was conducted using a time-series expanding window cross-validation technique over a decade-long dataset, complemented by leave-one-out cross-validation (LOOCV) to assess the robustness of spatial parameter interpolation through inverse distance weighting (IDW). Thereafter, spatial analysis of the calibrated parameters reveals significant heterogeneity across the aquifer. <i>α</i> identified primary recharge zones, while <i>λ</i> helped in understanding the groundwater flow pattern, and elevated <i>β</i> values indicated areas of strong groundwater/surface-water interaction. <i>z</i>_b served as a baseline for evaluating groundwater sustainability. This approach provides a valuable tool for establishing reliable the groundwater reference level and offers insights into underlying hydrogeological processes, aiding sustainable groundwater management. The conceptual model can also be adapted for different aquifer configurations.</p>