ABSTRACT
This chapter introduces artificial neural network (ANN) and presents with an example to show its implementation. One of the most common ways to downscale global climate models (GCM) future projections to a point of interest is to employ regression methods. There is a large set of candidate predictor variables for the multiple linear regression downscaling. The nonlinear regression with the ANN technique can be easily adopted to regression-based statistical downscaling as a simple linear regression. Functional relationships between GCM climate variables and local weather-station variables are employed to downscale the GCM output scenarios to local weather variables. Multiple linear regression models have commonly been used, accompanied with the predictor selection methods, such as stepwise regression (SWR) and Least Absolute Shrinkage and Selection Operator (LASSO). A possible alternative of the SWR was suggested by R. Tibshirani as LASSO.
