Prediction of summer rainfall in Pakistan from global sea‐surface temperature and sea‐level pressure

Abstract

Agriculture contributes around a quarter to Pakistan's economy and is closely linked with the variability of monsoon rainfall. The prediction of monsoon rains with sufficient lead time has immense importance for the planning and management of water resources and agriculture. In this study, Multiple Linear Regression (MLR) and Principal Component Regression (PCR) methods are employed to predict monsoon rainfall, and their performances are compared for June–September (JJAS) for the period 1961–2014 over the monsoon region of Pakistan. Rainfall data of Meteorological stations are used as the predictand. In the MLR method, predictors are carried out from sea level pressure (SLP) and sea surface temperature (SST) of the National Centers for Environmental Prediction (NCEP) reanalysis datasets. The PCR method first calculates principal components (PCs) from SLP and SST data, and these PCs are then combined with the regression technique and used as predictors. The performance of both models is tested using statistical measures such as root mean square error (RMSE), mean absolute error (MAE), bias and the correlation coefficient to evaluate the skill of the forecast. The agreement between actual and predicted rainfall data provides evidence for reasonably accurate predictions from both methods. The MLR and PCR models explained 84.6 and 92.2% of the variation of data, and the multiple correlation coefficients are 0.92 and 0.96 respectively. The correlation coefficient for the verification period (2005–2014) is 0.73 for MLR and 0.89 for PCR. The values of mean bias, MAE and RMSE are −5.5, 20.0 and 25.1mm for MLR, and −0.42, 16.2 and 16.6mm for PCR, respectively. The results indicate that the PCR model forecast is slightly better than that of the MLR model.