Resumen
Climatic data archives, including grid-based remote-sensing and general circulation model (GCM) data, are used to identify future climate change trends. The performances of climate models vary in regions with spatio-temporal climatic heterogeneities because of uncertainties in model equations, anthropogenic forcing or climate variability. Hence, GCMs should be selected from climatically homogeneous zones. This study presents a framework for selecting GCMs and detecting future climate change trends after regionalizing the Indus river sub-basins in three basic steps: (1) regionalization of large river basins, based on spatial climate homogeneities, for four seasons using different machine learning algorithms and daily gridded precipitation data for 1975?2004; (2) selection of GCMs in each homogeneous climate region based on performance to simulate past climate and its temporal distribution pattern; (3) detecting future precipitation change trends using projected data (2006?2099) from the selected model for two future scenarios. The comprehensive framework, subject to some limitations and assumptions, provides divisional boundaries for the climatic zones in the study area, suitable GCMs for climate change impact projections for adaptation studies and spatially mapped precipitation change trend projections for four seasons. Thus, the importance of machine learning techniques for different types of analyses and managing long-term data is highlighted.