Disaggregating IMERG Satellite Precipitation Over Czech Republic: An Innovative Approach using Hybrid Extreme Gradient Boosting based on Fuzzy Spatial-Temporal Multivariate Clustering

Abstract

Achieving accurate precipitation estimation at high spatial and temporal resolution is critical in hydrology and meteorology, particularly in regions experiencing water resource degradation. Integrating original products of generally coarse-resolution hydrological parameters from multiple satellites is a promising technology for producing massive repositories of space-time varying datasets such as precipitation. A methodological modelling framework commonly known as downscaling or disaggregation is evolving as a viable approach for generating hydrological datasets with spatio-temporal scales suitable for operational usage. In this study, we propose a non-parametric method for generating a high-resolution dataset from a coarse-resolution integrated multiple satellite-based precipitation dataset. The disaggregation involves using a hybrid Extreme Gradient Boosting (XGBoost) approach combined with multivariate spatial-temporal Fuzzy clustering. This clustering relies on Integrated Multi-satellite Retrievals for GPM (IMERG) precipitation and Shuttle Radar Topography Mission (SRTM) Digital Elevation Data to establish eight distinct clusters.The proposed method is experimentally demonstrated implementing to downscale 255 months (June 2000 to September 2021) of IMERG satellite data from 11km to 1km spatial resolution over the Czech Republic. We utilized eight stations, one per cluster, for training and validation purposes, with the remaining 19 stations used solely for validation. Our findings demonstrate a strong agreement between the disaggregated monthly precipitation over the 20 years and ground-observed precipitation, suggesting that our proposed methodology substantially enhances the accuracy of IMERG precipitation data. This method holds promise for applications in other regions with remotely sensed data, especially where ground-measured station data is sparse, facilitating the generation of high-resolution, accurate precipitation data.