Satellite-Based Daily Precipitation Bias Correction in a Tropical Mountainous Region Using Functional Generalized Additive Mixed Models: A Case Study in Valle del Cauca, Colombia
Validation study demonstrates reduced rainfall estimation errors using functional models across tropical mountain stations, highlighting potential for water-balance monitoring.
Key Points
To develop and evaluate a functional generalized additive mixed model that corrects systematic biases in daily satellite precipitation estimates across complex tropical terrain.
Treated annual accumulated precipitation curves from 62 IDEAM ground stations in Valle del Cauca, Colombia (2012–2020) as functional responses and CHIRPS satellite accumulation curves as functional covariates.
Incorporated station-level random effects and the Southern Oscillation Index within a Functional Generalized Additive Mixed Model (FGAMM) utilizing penalized splines.
Compared cross-validation performance against linear regression, Support Vector Machines (SVM), and Random Forest benchmarks, followed by an evaluation on an independent national dataset.
The FGAMM achieved a mean cross-validation root mean square error (RMSE) of 0.68 mm/day (95% bootstrap CI: 0.61–0.75 mm/day).
The model produced statistically significant lower error rates than linear regression, SVM, and Random Forest on the primary regional dataset.
The performance advantage was not maintained when tested on an independent national-network dataset using a simplified concurrent approximation, reflecting differences in model smoothness constraints and validation design.
Cite This Study
Arango-Londoño et al. (2026) studied this question.