Modeling Extreme Values in the IMERG Precipitation Dataset for Disaster Monitoring

Abstract:

Effective measurement and monitoring of extreme precipitation is an integral component of understanding the underlying nature of climate phenomena. The high resolution and near-real-time availability of the Integrated Multi-satellitE Retrievals for GPM (IMERG) product makes it ideal for monitoring global extreme precipitation. This project improves upon an existing extreme precipitation monitoring system that is based on the IMERG product. The proposed system uses Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) to partition a region of interest defined by IMERG measurements into nonoverlapping clusters. These clusters have significantly less variation in size compared with the recursive k-means scheme present in previous systems, increasing the confidence of fitted model parameters and decreasing overall noise. This method not only provides improved results but also comes at almost no additional runtime. The results are used to construct simple and intuitive average recurrence interval (ARI) maps that reveal how rare a precipitation event is. This information could be used by policy makers for disaster monitoring and prevention.