Clustering Hydrological Homogeneous Regions and Neural Network Based Index Flood Estimation for Ungauged Catchments: an Example of the Chindwin River in Myanmar

Abstract

A neural network-based regionalization approach using catchment descriptors was proposed for flood management of ungauged catchments in a developing country with low density of the hydrometric network. Through the example of the Chindwin River basin in Myanmar, the study presents the application of principal components and clustering techniques for detecting hydrological homogeneous regions, and the artificial neural network (ANN) approach for regional index flood estimation. Based on catchment physiographic and climatic attributes, the principal component analysis yields three component solutions with 79.2 % cumulative variance. The Ward’s method was used to search initial cluster numbers prior to k-means clustering, which then objectively classifies the entire catchment into four homogeneous groups. For each homogeneous region clustered by the leading principal components, the regional index flood models are developed via the ANN and regression methods based on the longest flow path, basin elevation, basin slope, soil conservation curve number and mean annual rainfall. The ANN approach captures the nonlinear relationships between the index floods and the catchment descriptors for each cluster, showing its superiority towards the conventional regression method. The results would contribute to national water resources planning and management in Myanmar as well as in other similar regions.