From Deterministic to Stochastic: A Hybrid Machine Learning Framework for Reliable Satellite Precipitation Merging over Greece

Accurate precipitation estimations are crucial for hydrological modelling and water resource management, especially in geographically complex regions like Greece. Satellite-based products are valuable as they encompass extensive spatial coverage with high data density, but their accuracy is limited compared to ground truth measurements. To address this bias, we leverage machine learning (ML) approaches. We present a hybrid machine learning framework that employs post-processing techniques to integrate satellite-derived precipitation data with ground-based gauge observations. The methodological framework upgrades a deterministic ML regressor (D-model) into a fully stochastic system (S-model) using Bluecat methodology. We use data for the period 2000-2021 over Greek territory, from gauge observations and Integrated Multi-Satellite Retrievals for GPM (IMERG). The S-Model significantly improves reliability and statistical consistency, effectively transforming the ML output into actionable, risk-aware intelligence.