Bias-corrected precipitation products in a data-scarce Andean basin: performance and parameter transferability

Abstract

ABSTRACT Sparse monitoring networks and complex topography in mountainous regions reduce the reliability of precipitation records, limiting their use in hydrological and engineering analyses. In the tropical Andes, large-scale circulation patterns and strong orographic gradients amplify these challenges, making gridded precipitation products prone to systematic biases that require correction before operational use.This study evaluates the reliability and transferability of bias-corrected precipitation products in the upper Rímac River basin, Peru. Ground observations from six SENAMHI stations were combined with six satellite and reanalysis products (CHIRPS, ERA5-Land, MERRA-2, PERSIANN-CDR, PISCO, and RAIN4PE). Data quality was assessed through homogeneity tests, and missing values were reconstructed using three machine-learning methods (KNN, MissForest, and MICE-GBM). Five bias-correction techniques were ranked using a weighted index prioritizing hydrological efficiency metrics, and a Random Forest model was used to assess parameter transferability based on distributional similarity between stations. Locally calibrated products (PISCO and RAIN4PE) outperformed global datasets after correction, achieving KGE values up to 0.78 and RMSE reductions of 15-35%. Linear Scaling performed comparably to more complex quantile-mapping methods. Transferability was linked to statistical similarity between precipitation regimes rather than geographic proximity, while canyon stations remained weakly transferable. The proposed workflow improves precipitation inputs in data-scarce Andean basins.