Improving short to medium range GEFS precipitation forecast in India

Abstract This study aims to enhance daily precipitation forecasts over the Indian subcontinent through post-processing the Global Ensemble Forecast System (GEFS) outputs using Analog (AN) and Logistic Regression (LR) techniques. Both raw and post-processed GEFS precipitation forecasts were evaluated against the Indian Meteorological Department observed dataset using probabilistic and deterministic forecast evaluation metrics, namely Brier Skill Score (BSS) and Root Mean Square Error (RMSE), respectively. Results found that the LR and AN post-processing method considerably improved short to medium range (1–7 day) precipitation forecasts over India. A comparison of the techniques with GEFS version 12 (GEFSv12) forecasts across different basins suggests that both methods were able to provide skillful precipitation forecasts in all the river basins, except that the AN method underperformed for the western ghats. The seasonal analysis also showed that the raw and the post-processed forecasts under-performed during the monsoon season while performing comparatively well during the non-monsoon season. The comparison of LR and AN methods showed that LR outperformed the AN method. The forecasts enhanced using both the post-processing techniques were more reliable and skillful than the GEFSv12 precipitation forecast. Comparison of the recent GEFS version with the older version of GEFS showed that the performance of the earlier version was slightly better compared to the latest version, however none of the forecasts can be used for decision making purposes without post-processing.