Jubones River Basin, a tropical mountainous basin in the Andes, Ecuador. Satellite precipitation products (SPPs) are essential for hydrological forecasting in data-scarce regions, yet their uncertainties increase at hourly timescales. This study evaluates the applicability of the Three-Cornered Hat (TCH) method for satellite-only precipitation fusion at hourly resolution and its hydrological value for machine learning–based runoff forecasting. TCH was applied to fuse IMERG, PERSIANN, and GSMaP precipitation estimates, and Random Forest runoff forecasts were developed for increasing lead times from 3 to 24 h. Results were benchmarked against a single-source SPP (IMERG-ER) and the multi-source MSWEP dataset, with particular emphasis on numerical issues arising during no-precipitation periods. (1) Frequent dry periods induce strong statistical dependence among SPPs, leading to singular difference covariance matrices that disable the classical TCH formulation. (2) Introducing Tikhonov regularization permits consistent application of the method without altering precipitation magnitudes or temporal variability, enabling continuous satellite-only fusion. (3) Runoff forecasting skill is comparable across precipitation scenarios; MSWEP slightly outperforms others in NSE, KGE, and RMSE, while the TCH-based product consistently reduces bias. Overall, although regularized TCH is technically feasible for hourly precipitation fusion, its added value for operational runoff forecasting is limited under dry-hour-dominated conditions. These findings highlight both the potential and constraints of satellite-only fusion for near-real-time hydrological forecasting in data-scarce regions. • First evaluation of the Three-Cornered Hat (TCH) method at hourly scale for runoff forecasting. • Identification of dry-hour-induced covariance singularity as the main limitation of hourly TCH. • Simple Tikhonov regularization enables robust satellite-only precipitation fusion. • Hourly TCH fusion is feasible but provides limited gains for operational runoff forecasting.
Abril et al. (Thu,) studied this question.