Abstract

Rainfall data is usually of high variation that needs advanced statistical techniques to properly model the data while taking into account its variation. Statistical Downscaling methods are well known in climate modeling especially to analyze the relationship between the large-scale climate data with small-scale climate data. In this setting, the large-scale covariates are correlated and require techniques for shrinking the regression coefficients. Fused LASSO (Least Absolute Shrinkage and Selection Operator) is a very popular shrinking method. The fused LASSO is a generalization of the LASSO penalty in a sense that new penalty parameters are added to enforces sparsity in both the coefficients and their successive differences. This addition of new parameters is desirable in applications especially if the covariates can be ordered in some meaningful way. In this paper the Fused LASSO is employed to model the average monthly rainfall data at Indramayu Sub-District collected from January 1981 until April 2014. The rainfall data is treated as a response variable whereas the precipitation data is considered as large-scale covariates. These covariates are obtained from a combination of interpolate surface observations and satellite data based on GPCP (Global Surface Climatology Project) version 2.2. The results showed that based on the AIC and BIC loss function the Fused LASSO method can selects 28.57 % significant grids.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call