A cloud-based geospatial machine learning framework for real-time 1 km hourly precipitation downscaling in complex terrains

Accurate high-resolution precipitation information is essential for flash-flood and landslide early warning in mountainous regions, yet reliable quantitative precipitation estimation (QPE) remains challenging because of complex terrain and sparse rain-gauge networks. This study presents a cloud-based geospatial machine learning framework implemented on Google Earth Engine (GEE) for real-time 1 km hourly precipitation forecasting over Hue City, Vietnam. The framework employs a Random Forest (RF) model to integrate six multi-source predictors, including wind speed and precipitable water from the Global Forecast System (GFS), enabling high-resolution precipitation prediction under complex orographic conditions while preserving the physical consistency of numerical weather predictions. Validation against rain-gauge observations demonstrates that the proposed framework reduces the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by up to 25% and decreases spatial bias by 20–30% compared with the original coarse-resolution precipitation products. The framework provides an efficient, scalable, and operational solution for high-resolution precipitation forecasting and supports real-time flash-flood and landslide early warning in data-scarce mountainous regions.