Journal of Climatology & Weather Forecasting

ISSN - 2332-2594


Predicting the Frequency and Intensity of Climate Extremes byRegression Models

Adnan M, Rehman N and Shahbir J

This study examines the relationship of climate extremes indices with the large-scale factors like Sea Level Pressure (SLP) and Sea Surface Temperature (SST). The prediction of extreme indices is carried out and is based on statistical downscaling using the extreme indices data, National Centers for Environmental Prediction (NCEP) monthly SLP and SST reanalysis data. For this purpose, five extreme indices (PRCPTOT, R95p, RX5day, TN90p and TX90p) are developed by using homogenized and high quality daily data of temperature and precipitation for the period 1961-2010 of 10 meteorological stations of monsoon-dominated region of Pakistan. These indices are then average to develop an average time series of each extreme index. To check the assumption of regression model, extreme indices data are tested for heteroscedasticity, auto-correlation and normality. All extreme indices are independent, normal and homogeneous. These indices data are then used as predictand and SLP & SST datasets are used as predictors in regression model. Data for period 1961-2000 and 2001-2010 is used for training and validation purpose respectively. Stepwise regression procedure is adopted to compute regression coefficients based on algorithm of Jennrich. Predictors having strong correlation with extreme indices are identified and a regression model is developed using these predictors and also apply cross-validation technique. Performance of regression model and cross validation models is tested by using statistical measures (Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and bias). The performance is seen reasonably high both in training and validation period. The actual and estimated values show a close agreement. It is seen that ensemble mean prediction obtain from crossvalidation models well estimated the extreme indices than the regression model. This study is useful because extremes have a large impact on human society & economy and causing huge losses of the country. The timely prediction of extremes is a major factor and will help the policy makers to take necessary measures for reducing the huge losses.