Sampling Uncertainty is Critical for Setting Forecast-Based Anticipatory Action Triggers

As of 2022, nearly 5 million smallholder farmers across more than a dozen countries are covered by Anticipatory Action (AA) programs. AA programs are aid schemes triggered when seasonal forecasts of precipitation fall below a locally-determined drought threshold. These thresholds are generally calculated to meet a donor- or government-specified expected trigger frequency. However, the observational precipitation record in many regions with AA programs is limited, and interannual and interdecadal autocorrelation in the climate patterns driving seasonal precipitation further reduce the effective sample size used to calculate these thresholds. AA trigger thresholds are highly sensitive to this limited observational record, increasing the risk that action is not triggered when it is needed or that programs become financially unsustainable due to excessive triggers. In practice, forecasters are aware of this uncertainty and ad hoc decisions on whether to trigger action are often made if seasonal forecasts are close to the threshold. Seeking a more formalized, statistically sound alternative to these ad hoc approaches, governments and humanitarian agencies have asked for transparent and intuitive metrics of trigger uncertainty. To address this demand, we illustrate how a simple statistical method (bootstrap resampling) can quantify the practical uncertainty in drought thresholds due to a limited observational record. Examining United Nations-supported operational AA forecasts in sub-Saharan Africa, we demonstrate how this method quantifies critical underlying uncertainty, enabling these AA systems to meet the state-of-the-art demanded by donors and the WMO.