Seasonal Forecasts for Food Security Applications in the Upper Blue Nile River Basin

Abstract:

Current global seasonal prediction products are becoming useful tools in the decision process of various economic activities. Their application requires post-processing procedures that include systematic error correction and downscaling. In this study, the NOAA Climate Forecast System version 2 (CFSv2) dataset is used to support the planning decision in four agricultural communities in the upper Blue Nile River Basin (BNB). The dataset is the 6-hourly forecast outputs out to seven months with spatial resolution of 1°. Two types of bias correction methods are implemented and evaluated: a non-parametric quantile mapping and a parametric, distribution mapping method. We use the CFSv2 hindcast and the corresponding Global Data Assimilation System (GDAS) for the 2012-2018 (7 years) period to train the statistical models. We simultaneously bias-correct and downscale the raw forecast from 1° to 0.12° and thereby retrieve the spatial heterogeneity of the field. In addition, we stratify the 6-hourly dataset into four cycles (00Z, 06Z, 12Z and 18Z) and apply the same statistical models to assess the sensitivity of the performance to the diurnal variability of the fields. The evaluation is performed on post-processed forecasts initialized on April 1, 2019. The results show that diurnal stratification adds value to the post-processed products. This study will highlight the details of the results to indicate the method with the best performance.

Acknowledgment: This material is based upon work supported by the National Science Foundation under Grant No. 1545874.