The works is discussing about Multivariate Adaptive Regression Splines (MARS). The MARS model composed through stepwise process which based on recursive partitioning with splines. This method is not very strict on assumptions as the classic methods are. This method is used to downscale the GCM to predict the monthly rainfall. The prediction result of MARS is compared to the principal component regression (PCR) and artifficial neural network (ANN) with correlation criteria and root mean square error. Based on these criteria, the result accurateness of the MARS, PCR, and ANN model is vary between locations. Empirically, the MARS method generates assumption with relatively stable and high accurateness compared to the PCR and ANN method.

Key word: regression, recursive partitioning, Multivariate Adaptive Regression Splines, statistical downscaling, GCM

Last Updated: 2006-04-26