Evaluation of NDVI-Based downscaled precipitation datasets in the Peruvian Andes

Abstract. Remotely sensed or model-based gridded precipitation products are increasingly used for hydrological assessments, especially in data scarce regions. However, their coarse spatial resolution often limits the accurate representation of daily rainfall variability and extremes, particularly in complex mountainous terrain. To address this, several methods have been developed to downscale these products and improve their spatiotemporal representation. In the Peruvian Andes, sparse and discontinuous rain gauge networks have led to a strong reliance on satellite-based precipitation products for hydrological assessments, despite persistent uncertainties in their spatio-temporal accuracy. In this study, we developed daily downscaled precipitation datasets covering the Andean region of Peru based on the Global Precipitation Measurement (GPM) IMERG product using parsimonious relationships, including exponential regression (EXP) and geographically weighted regression (GWR), incorporating environmental covariates such as NDVI, topography, and geographical location. These downscaled datasets were evaluated against in situ meteorological stations and benchmarked against regionally optimised products, such as PISCO and Rain4PE, using statistical performance indices, indicators of overall rainfall pattern representation, and extreme-value metrics. Results show that Rain4PE achieves the highest overall performance with a median Kling-Gupta Efficiency (KGE) of 0.67, whereas PISCO faces some limitations with overall performance (median KGE ≈ 0.25), but it excels at representing extreme precipitation. GPM-based datasets exhibit systematic limitations in both temporal coherence and detection of extremes. The GWR approach enhanced spatial detail while preserving the performance of the source GPM dataset (median KGE ≈ 0.28), outperforming the simpler EXP (median KGE ≈ 0.21) method without degrading temporal coherence and extreme-event representation. These findings highlight the potential of parsimonious downscaling strategies to improve precipitation datasets in complex, data-scarce mountainous regions, while underscoring the continued importance of regional gauge-informed products.