UC Santa Barbara

A simple—yet powerful—indicator used to monitor weather-related food insecurity is the Water Requirement Satisfaction Index (WRSI). This water-balance model uses precipitation (PPT) and potential evapotranspiration (PET) data to estimate the water supply and demand a crop experiences over the course of a growing season. If the season is still in progress, climatological forecast data can be adjoined with to-date conditions to provide insight into potential end-of-season crop performance. However, if used incorrectly, these same indicators can become a detriment to early warning, resulting in a lack of, or misallocation of, precious humanitarian aid. While several prominent agencies and data centers use arithmetic average climatological data as proxies for “average conditions,” little published research has evaluated the effectiveness of this forecast method when used in crop-water balance models (i.e., WRSI). We use WRSI hindcasts of three African regions’ primary growing seasons from 1981-2019 to assess the adequacy of the arithmetic mean climatological forecast. We find that this method of forecasting end-of-season (EOS) WRSI results in crop-water condition predictions that are positively biased, i.e., they overestimate WRSI. This bias ranges from 2-23% positive bias throughout portions of east, west, and southern Africa. The proposed alternative is a scenario-based approach, which adjoins the to-date conditions with data from previous seasons to produce a series of historically realistic conclusions to the current season (one potential scenario generated from each year in the data record; 1981-2019). The mean of these scenarios is then used as the projected end-of-season WRSI, hereby referred to as the WRSI Outlook. We find this approach has a near-zero bias score, and correspondingly, it has an improved (lower) Root Mean Squared Error (RMSE) in comparison to the traditional arithmetic average climatological method. While an increase in accuracy is a welcome by-product, the slight decrease (or increase, as seen during wet years) in RMSE has less to do with forecast skill, and more related to the reduction of bias provided by the mean-scenario method. In total, the findings from this paper highlight the inadequacies of the existing arithmetic mean climatological forecast method when used to project EOS WRSI, and present a less-biased, and more accurate, mean scenario-based approach.