Validating a Data-Driven Multi-Model Characterization Technique for Water Users: A Case of Pangani Basin in Tanzania

Through a literature review, it has been observed that water scarcity results from increased demand due to population growth, economic progress, and climate change, leading to disparities between required and available water resources. Addressing this challenge requires segmenting water users into homogeneous groups and thoroughly examining their characteristics regarding water utilization to develop efficient and effective water governance strategies. This study employed data-driven multi-model validation techniques to characterize water users in Pangani Basin in Tanzania. The Kmeans, Agglomerative Hierarchical, and Fuzzy C-means clustering algorithms were used to ascertain the efficacy of the characterization. Cluster validation showed that K-means outperformed Agglomerative hierarchy by owning a high Calinski–Harabasz Index and low Davies–Bouldin Index of 692.3 and 1.8, respectively, compared to Agglomerative hierarchy with values of 578.2 and 1.9, respectively. The clustered dataset was tested for prediction accuracy by fitting the logistic regression. K-means showed a prediction accuracy of 98.2% over 97.5% of the Agglomerative Hierarchical method. The four clusters identified were large-scale irrigation water users, moderate irrigation water users, community water supply entities, and domestic water users. We argue that understanding users’ characteristics could efficiently and effectively add value to water governance along the basins.