A Data-Driven Machine Learning Clustering of Rainfall Patterns: Is Reclassification of Tanzanian Rainfall Climate Zones Needed?

Abstract The wider accessibility of rainfall data and data-driven analytical approaches has boosted climate regionalization research worldwide, particularly in Tanzania. However, despite substantial research on rainfall trends and machine learning applications in Tanzania, we didn’t find a research study that has systematically analyzed the possibility for reclassification of the country's traditional rainfall climatic zones using data-driven approaches. As a result, this study fills that gap by investigating long-term rainfall variability and determining if the present climatic zoning is still valid under observed rainfall patterns. The study used a quantitative and comparative research design with 40 years of monthly rainfall data (1980-2020), mostly from the GPCC and CHIRPS satellite datasets. Machine learning techniques were used to reclassify zones: k-means, hierarchical clustering, and Partitioning Around Medoids (PAM) algorithms. The study also performed cluster validation and zone's agreement analysis (data-driven vs traditional zones) using silhouette, chi-square tests, Cramer's V, Cohen's Kappa, and the Adjusted Rand Index (ARI). The results show significant interannual rainfall variability among zones, with no statistically significant long-term trends. Agreement analysis, on the other hand, shows that traditional zoning is robust (it agrees with the data-driven clustering), despite small divergence within some transitional zones. The divergence indicates that some zones have internal heterogeneity and there is only a need for intra-zonal reclassification. The study offers an analytical framework for improving climate zoning in Tanzania, enhances geographic precision in agricultural planning, water resource management, and climate adaption measures.