Application of Relevance Vector Machine to Downscale GCMs to Runoff in Hydrology

As relevance vector machine (RVM) can powerfully manage complexity to regression and classification basing on the concept of probabilistic Bayesian learning framework, it has been widely used in dealing with various recognition problems. In present study we applied RVM as a statistical downscaling method to climate change impact on hydrology and water resources. General circulation models (GCMs) are main tools for study global climate change, however, their simulate results cannot be used directly to evaluate the impact of climate change on hydrology in basin scale for their large and coarse scale. By applying downscaling approach based on RVM, the complex non-linear relationship between the climate factors of GCMs and runoff in basin scale was bridged. The impact of climate change on runoff was assessed by using the established relationship. Comparing with the other two downscaling approach, least support vector machine (LSSVM) and Back propagation neural network (BPNN), the results showed that RVM is suitable for assessing climate change impact on hydrology as rational modeling accuracy and fast modeling speed.