Abstract:
The 56-years daily precipitation data (1950-2005) in DL2 (Dry zone, Low country) region in dry zone, Sri Lanka were used to calculate the starting dates of four critical dry spells namely Critical Dry Spell 1,2,3 and 4 (CDS1, CDS2, CDS3 and CDS4). The correlation analysis among onsets of dry spells showed that there is a significant correlation between starting dates of successive critical dry spells. As a result fitting regression models were explored to predict onsets of CDS2, CDS3 and CDS4 using onsets of previously occurred dry spells. The linear and non linear regression models were fitted using Proc REG and Proc MODEL in SAS software. Based on the R square, mean square error and model assumptions, model adequacy was assessed and the best models were selected to fit onsets of critical dry spells 2,3 and 4 in three locations Ampara, Trincomalee and Batticaloa in DL2 region. The type of the regression models selected were different among different critical dry spells as well as different locations. Onset of CDS2 in Ampara was modeled using onset of CDS1. Similarly the onset of CDS3 was modeled using the onset of CDS2 with first order autoregressive error model. When modeling onset of CDS4, natural log transformation was used for the dependent variable. Polynomial regression model of order three was fitted for onset of CDS3 in Batticaloa while both onsets of CDS2 and CDS3 were used to model the onset of CDS4 in the same location.