Early warning systems are essential tool for humanitarian preparedness and response. The diversity of inputs required, ranging from agricultural production estimates to market price variability and weather forecasts, means that interpreting food security signals is not an easy task. Each of these inputs is fraught with uncertainty which analysts need to assess when making projections about future food security. Understanding the accuracy rates of early warning systems is therefore of paramount importance to enable improvements to food security prediction. However, to date, limited analyses of early warning accuracy have been conducted. Here we analyze Famine Early Warning System Network (FEWS NET) early warning data for the Greater Horn of Africa and show that, despite accuracy in projections, there remain important challenges for food security projections. The two major sources of uncertainty are associated with complex weather phenomena and conflict – with uncertainty in weather forecasts being twice as important as conflict in overall FEWS NET accuracy. Indeed, the least accurate projections are recorded in seasons with particularly complex weather events such as the 2015/2016 El Niño Southern Oscillation as well as in zones that are affected by internal conflict (e.g. South Sudan). With respect to predicting crisis transitions, areas with more frequent transitions tend to be more accurate, possibly because predicting the drivers behind these transitions are better understood. Our novel analysis provides a framework to invest resources in specific aspects of early warning. We also hope that by measuring the reliability of these systems, we can increase the confidence of decision makers to act early to mitigate the growing risks posed by hunger and famine.

预警系统是人道主义备灾和响应的重要工具。从农业生产估算到市场价格变化和天气预报，所需的投入是多种多样的，这意味着解释粮食安全信号并非易事。这些输入中的每一个都充满不确定性，分析人员在对未来的粮食安全进行预测时需要评估这些不确定性。因此，了解预警系统的准确率对于提高粮食安全预测至关重要。但是，迄今为止，对预警准确性的分析有限。在这里，我们分析了非洲大角地区的​​饥荒预警系统网络（FEWS NET）预警数据，结果表明，尽管预测准确，粮食安全预测仍然面临重大挑战。不确定性的两个主要来源与复杂的天气现象和冲突有关-天气预报的不确定性是整体FEWS NET准确性冲突的两倍。确实，最不准确的预测记录在天气事件特别复杂的季节，例如2015/2016厄尔尼诺南方涛动以及受内部冲突影响的地区（例如南苏丹）。关于预测危机过渡，过渡频率较高的区域往往更准确，这可能是因为更好地理解了预测这些过渡背后的驱动因素的原因。我们的新颖分析提供了一个框架，可以在预警的特定方面投入资源。