Multi-Model Intercomparison of Conceptual Hydrological Models Across Tropical Catchments in the Northern Pacific Region of Costa Rica

This study evaluates the performance of five conceptual hydrological models namely AWBM, BUDYKO, GR2M, IHACRES, and SIMHYD in simulating monthly streamflow across twelve tropical catchments in the Northern Pacific Region of Costa Rica. Building on existing regional hydrological modeling efforts, this study provides an intercomparison of lumped monthly conceptual models across steep, forest-dominated catchments characterized by bimodal precipitation patterns. The study aims to evaluate the relationship between model performance, complexity, structure, and calibration period duration. The Generalized Simulated Annealing (GSA) method was used for parameter optimization to assess its performance under these specific hydroclimatic conditions. Performance metrics, including KGE, PBIAS, D, and RSR, were computed to compare simulated and observed streamflows during calibration and validation periods. GSA proved effective in navigating complex, non-linear parameter spaces, handling diverse meteorological conditions. However, computational demands grew with model complexity. Simpler models converged faster, whereas higher-dimensional architectures increased costs without delivering substantial accuracy gains. All hydrological models exhibited satisfactory performance during calibration, with multiple independent metrics providing a more reliable assessment. These findings suggest that increased model complexity and overparameterization do not guarantee superior accuracy. Instead, parsimonious structures, such as GR2M and BUDYKO, often matched or outperformed heavier models, indicating that model performance is more influenced by structure and runoff generation representation than by complexity. Instead, for monthly streamflow simulation across the investigated tropical catchments, parsimonious structures, such as GR2M and BUDYKO, often matched or outperformed heavier models, indicating that model performance is more influenced by structure and runoff generation representation than by complexity. No single hydrological model universally outperformed the others, with model behavior remaining deeply context-dependent. Performance varied by catchment, underscoring that model strengths, weaknesses, and optimal choices depend on specific study objectives, catchment characteristics, and runoff generation representations. During validation, all models showed reduced simulation accuracy compared to calibration, particularly when calibrated on older data and validated on newer periods. This decline highlights challenges in temporal transferability caused by discrepancies in meteorological conditions and inter-annual climate variability. Given that no single model consistently outperforms the others, adopting multi-model intercomparisons frameworks offers a more reliable strategy. The methodology applied can be replicated in regions with similar conditions to assess water resources more accurately and support sustainable catchment management.