Identification of robust catchment classification methods for Sahelian watersheds

The study was conducted in a Sahelian watershed located in Burkina Faso (West Africa). In this study, an inter-comparison procedure is proposed to investigate the effects of implementing various sets of explanatory variables and clustering algorithms on developing hydrologically homogenous regions. Six different sets of explanatory variables considered in this framework are generated using the combinations of topographic, land-use, climatic, and hydrological attributes. Also, seven different linear and nonlinear clustering techniques are implemented using the combinations of Principal Component Analysis (PCA), Non-linear Principal Component Analysis (NLPCA), Self-Organizing Maps (SOM), and K-means algorithm. The mean and maximum annual runoff are considered as two variables of interest for conducting a comparison and identifying the most robust classification methods. The study results indicate that the monthly Bagnouls-Gaussen index (BGI) is the most robust set of explanatory variables to be used for identifying the hydrologically homogenous regions considering both mean and maximum annual runoff. Additionally, compared with BGI, the combination of topographic and land-use attributes can provide competitive results while the land-use attributes alone cannot capture the hydrological heterogeneity of the catchments. Moreover, interestingly, the comparison results show that regardless of its simplicity, the K-means algorithm is superior to the other clustering techniques in terms of generating hydrologically homogenous regions based on the monthly BGI. • Implemented catchment classification intercomparison framework. • Thoroughly examined various sets of explanatory variables and clustering methods. • Consider the effects of the number of clusters in the inter-comparison process. • The Monthly Bagnouls-Gaussen Index excels in classifying Sahelian watershed regions. • K-means outperforms in finding homogeneous regions using Bagnouls-Gaussen Index.