Comparison of standalone and hybrid algorithms for SWH model parameterization: Reliable evapotranspiration simulation and partitioning via an ensemble selection mechanism

Reliable evapotranspiration (ET) simulation is critical for agricultural water management, yet parameter calibration of the Shuttleworth-Wallace-Hu (SWH) model often suffers from local optima due to its complex parameter space. This study coupled four global heuristic algorithms—Bayesian Optimization (BO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Sparrow Search Algorithm (SSA)—with the local optimizer Fmincon to develop hybrid algorithms (termed 'Algorithm-Fmincon'). To enhance robustness, we adopted an ensemble parameter averaging strategy to obtain the final parameters, using 2020 data for calibration and 2022 data for validation on summer maize in the North China Plain. The hybrid algorithms effectively resolved the accuracy bottlenecks of standalone models; notably, performance gains at the daily scale substantially surpassed those at the half-hourly scale. Among all combinations, BO-Fmincon exhibited the most balanced and robust performance, achieving test-set R² of 0.934 (half-hourly) and 0.904 (daily), with RMSE of 0.06 mm·h⁻¹ and 0.44 mm·d⁻¹ , respectively. It was therefore selected as the optimal calibration framework. However, it should be noted that the partitioned soil evaporation (E) and plant transpiration (T) are model-derived rather than independently observed, and the current validation is limited to a single site and two growing seasons. Future research integrating multi-site long-term observations and independent in sit component measurements is necessary to further verify the model's generalizability.