Beyond the Metrics: Context‐Aware Calibration for Better Drain Flow Modelling

ABSTRACT Robust hydrological and water quality modelling is essential for advancing the understanding and management of engineered agricultural subsurface drainage systems. Achieving credible model output hinges on calibration strategies that are both rigorous and context sensitive. At the field scale, accurately capturing the dynamics of nutrient load curves, particularly under conditions of intermittent zero flow, is a critical challenge. These systems often exhibit heteroscedastic residuals, complicating traditional calibration approaches. While the heteroscedastic maximum likelihood estimator (HMLE) has shown promise in watershed‐scale rainfall–runoff modelling, its applicability to field‐scale, process‐based models such as DRAINMOD remains widely underexplored. This study systematically evaluates the performance of the HMLE over widely used metrics—NSE and Kling–Gupta efficiency (KGE)—across diverse error structures in subsurface (tile) drainage model residuals. Two key insights emerged: (1) KGE consistently outperformed the NSE across scenarios and (2) the HMLE's effectiveness was highly sensitive to the underlying error structure, ranging from superior to suboptimal. These findings underscore the necessity of adopting multiobjective calibration frameworks that integrate both statistical and graphical diagnostics. More critically, they challenge the notion of universal calibration metrics, emphasizing the imperative for modellers to tailor calibration strategies to the specific characteristics of their system and data.