On the quantitative limits for triggering drought anticipatory actions in Mindanao, the Philippines

Abstract

Anticipatory Action (AA), which involves timely and informed actions ahead of a crisis/impact, is increasingly being used/promoted as a way to mitigate the impacts of extreme climatic hazards, including droughts. Actions are initiated in anticipation of the occurrence of the hazard and its impacts, and depend on the lead time, likelihood of impact, as well as the effectiveness of, and the capacity to undertake such actions. A decision to initiate actions is taken with a support of the so-called trigger model that forecasts likelihood or magnitude of expected impact. To build efficient and credible AA trigger models, quantitative assessments of relationships between observed climate and environmental conditions, weather/seasonal forecasts, and variables expressing current sectoral and societal vulnerability (collectively referred to as indicators) and expected impacts, are required at varied lead times. These quantitative assessments are needed to: (a) avoid over-weighting (placing excessive trust in) non-skillful indicators; (b) avoid using several co-varying and correlated indicators (over-emphasising their collective importance for the decision at hand); and (c) provide objective and defensible evidence for and consequently confidence in the AA trigger model. Motivated by the need to improve the current AA trigger model used for agricultural drought by FAO in Mindanao, a region of the Philippines which experiences periodic drought-related food insecurity, this study evaluates a range of climate and environmental indicators as a basis for developing a quantitative, objective trigger model. The analyses focus on: (i) an evaluation of efficacy of using a climate-only drought hazard index as an expression of impactful drought in the region, and (ii) an evaluation of the predictive utility of a set of indicators and formal statistical models combining these indicators, at various lead times. We show that the predictive utility of each indicator varies by season and lead time, highlight the varying skill of the trigger model and consequently advocate for transparent inclusion of model skill in the trigger mechanism.