Supportive empirical modelling for the forecast of monsoon precipitation in Nepal



Seasonal prediction of the monsoon precipitation in Nepal has been a challenge. That is partly because Nepal's monsoon precipitation exhibits a distinct and strong quasi-decadal oscillation while not correlated with the El Nino-Southern Oscillation. The existing global and regional climate models are insufficient in deriving reliable precipitation prediction. This paper examines the prediction of Nepal's July to August (JA) mean precipitation using five different methods: three time series models in comparison with a persistence forecast (PF) and a climatology forecast (CF). The first model (P-AR) uses past precipitation data to forecast the future, based upon the recently uncovered quasi-periodic feature of the JA mean precipitation. The other two models (ARX-SST and ARX-GQ) add covariate sea surface temperature (SST) and global water vapour flux circulation (GQ) respectively. Based upon the evaluation of 1-year-ahead forecast, the three time series models performed better than PF and CF. Of those, the P-AR model has the least mean absolute error (MAE) of <1 mm day–1. Based upon the 2-year-ahead forecast results, the P-AR model performs slightly better than ARX-SST and ARX-GQ models. The forecast ability of the time series models appears better than that of operational numerical models such as the NCEP Climate Forecast System (CFS) and so, can be used as an effective alternative in predicting monsoon precipitation for Nepal.