Hybrid Approach to Rainfall Disaggregation Using Artificial Intelligence and Multivariate Disaggregation Considering Spatial Correlation

High-resolution rainfall data are essential for various hydrological applications such as flood modeling, soil erosivity studies, and urban water management. However, the scarcity of subdaily data, particularly in remote regions, necessitates reliable disaggregation methods to derive finer temporal resolutions from daily rainfall records. This study introduces a novel hybrid approach combining the multivariate disaggregation of rainfall (MuDRain) model with artificial neural networks (ANNs) to enhance rainfall disaggregation. MuDRain disaggregates daily rainfall into hourly data while preserving spatial correlations across multiple locations. The methodology involved clustering of 222 grid points into different clusters using k-means, followed by disaggregation for both individual clusters and the entire data set (all stations approach) for Kerala, India. The results were evaluated using descriptive statistics. The “all stations approach” outperformed cluster-specific disaggregation. To further refine the results. A hybrid MuDRain-ANN model is developed further and demonstrates significant advancements in capturing high-intensity events compared to MuDRain alone. Additionally, intensity-duration-frequency (IDF) curves were generated for representative locations, revealing underestimations for shorter durations and better alignment with observed data for longer durations. This comprehensive approach not only enhances the temporal resolution of rainfall data but also provides critical insights for urban drainage design and agricultural planning.