Importance of Regional Rainfall Data in Homogeneous Clustering of Data Sparse Areas: A Study in the Upper Brahmaputra Valley Region

Abstract

Delineation of homogeneous regions has found its way into many hydrological applications as it helps in addressing the challenges in understanding the behavior of rainfall distribution and its variability at a local scale. In the present study, rainfall data recoded by 83 tea gardens in the upper Brahmaputra valley region of Assam have been used to identify homogeneous rainfall regions by using fuzzy clustering analysis. Further, seven different cluster validity indices (CVs) were utilized to find out the optimum clustering in the fuzzy c-means (FCM) algorithm. The clusters thus formed were assessed for statistical homogeneity by performing homogeneity tests based on L-moment. Three different combinations of feature vectors were employed in FCM algorithm and the outputs were compared for attaining best solutions to regionalization. The results were further compared with previous regionalization studies. The analysis and comparison conclude that if regionalization needs to be done at a local scale, further sub-clustering of a larger clustered region to smaller regions may be required. Local rainfall data can be used for the purpose provided a good dataset with large number of station points are available within the region. Along with rainfall data, geographical location parameters (latitude, longitude and elevation) need to be taken into account for getting a definite conclusion.