Principal components of sea surface temperatures as predictors of seasonal rainfall in rainfed wheat growing areas of Pakistan

Time-lagged relationships were explored between VARIMAX rotated principal components (RCs) of sea surface temperatures (SSTs) and rainfall periods that are important for rainfed wheat production in Pakistan. Seasonal forecasts were developed using Generalized Additive Models. The first 10 RCs explained 54% of the variance in the SST data. Individual RCs were strongly (r2 ≥ |0.5|) to moderately (r2 ≥ |0.3|) correlated with climatic indices of SST anomalies associated with the El-Nino Southern Oscillation, Pacific Decadal Oscillation, Indian Ocean Dipole, and the tropical Atlantic Ocean. Forecasts of monsoon (July to September), total growing season (November to April), early (November to January) and late season (February to April) rainfall (1961–2010) were developed for Chakwal, Talagang and Islamabad. Important, linear or non-linear, time-lagged relationships were found between the RCs of SSTs and rainfall. Cross-validated forecasts were compared with real-time forecasts to evaluate the ‘true’ forecasting ability of the models. Continuous and categorical probabilistic forecasts were tested with an array of skill scores. Skilful forecasts of pre-season, monsoon and late-season rainfall were produced for the drier sites Chakwal and Talagang and to a lesser extent for the wetter site Islamabad. These simple, statistical forecasts can be developed with minimal financial investment. However, consideration of the potential uses of such forecasts will require a reflective decision framework that engages stakeholders and addresses socio-economic and agro-ecological constraints not included here.