The standardized precipitation index (SPI) is an important yet easy-to-calculate means to describe wet or dry conditions in very different climates. In this work, a new scheme for obtaining improved forecasts of this index is developed. The methodology is tested over Russia and West Africa, proving that it can be successfully applied to different forecasting models and world regions. For testing, we use two forecasting models: the semi-implicit semi-Lagrangian vorticity-divergence (SL–AV) model of the Hydrometeorological Centre of Russia and the Institute of Numerical Mathematics of the Russian Academy of Sciences for Russia and the Climate Forecast System Version 2 (CFSv2) of the National Center for Environmental Prediction (NCEP) for West Africa. Based on hindcast simulations of both models, we demonstrate relatively poor skills in obtaining direct zero to three month lead-time SPI forecasts in the regions of interest during summer season. In order to improve the accuracy of these forecasts, we utilize surface temperature, mean sea level pressure and 500 hPa geopotential height fields, obtained from the outputs of both models. The spatial patterns of crosscorrelations between previously obtained climatological fields and our target variable (SPI-1) are studied to identify informative co-variates, potentially affecting monthly scale precipitation variability. The cross-correlation structures between the different fields reveal relevant interdependencies between SPI-1, sea surface temperature, mean sea level pressure and 500 hPa geopotential height in different regions. Subsequently, we employ two different regression models based on statistical post-processing of regional climate model output. In the first model, we consider all combinations of pairs of the previously identified predictors in a set of linear regression equations, which generates an ensemble of individual SPI-1 forecasts. The second model is based on a multiple linear regression approach comprising the dependency between all predictor variables and the predictand (SPI-1) in a single equation. The resulting SPI-1 forecasts obtained from both regression models are subsequently analysed in both deterministic and probabilistic ways and checked by various verification metrics. We identify that the first proposed model provides a significant improvement in the SPI forecasting, pointing to the potential for its implementation in operational monthly precipitation forecasts.
