Statistical prediction of Sri Lankan rainfall during October to December

Abstract

Sri Lanka receives most rainfall during October to December (OND). Here we construct multiple linear regression models to forecast the OND Sri Lankan rainfall during 1979-2012 for lead times of 1 and 2 months. Correlation analysis was used to examine the relationship between Sri Lankan OND rainfall and global sea surface temperature (SST) anomalies. Three independent predictors were identified through partial least square regression method which includes the southern Atlantic SST tendency, southern Pacific SST tendency and western Pacific and Maritime Continent SST tendency at two different lead times. Three-year-out cross validation concludes that the multiple linear regression models can produce forecast the OND rainfall forecast at correlation coefficient skill of 0.69 and 0.68 for the 1 and 2 month lead times respectively. The physical processes associated with these three predictors show that they contribute to increase in OND rainfall of Sri Lanka.