Early warning meets optimization: reimagining proactive policy responses to agricultural risk

Thesis

Disasters like droughts inflict about $123 billion in agricultural losses globally each year. But targeted, preemptive aid programs—the focus of this thesis—can mitigate their impacts.

One approach to tackling these hazards is leveraging early warning systems, which combine satellite data, climate models, and algorithms to predict disasters before they escalate. Properly used, these systems can trigger anticipatory action, a new type of humanitarian aid that allocates aid resources before a hazard occurs: for instance, by distributing cash or drought-resistant seeds ahead of expected weather extremes. This proactive approach is distinct from traditional humanitarian aid efforts, which typically allocate aid after disasters occur.

Major global humanitarian organizations are accelerating efforts to develop early warning systems and anticipatory action initiatives. However, important challenges need to be addressed as these systems scale. First, early warning forecasts are often inaccurate, biased, or generic (e.g., “a drought is expected”) rather than fully impact-based (e.g., “an expected drought will cause crop losses of 30% in this region”). And anticipatory action itself often employs blanket approaches (e.g., providing the same amount of aid to everyone) rather than targeting resources to the most vulnerable. In many contexts, early warning systems are only loosely connected to anticipatory action frameworks, rather than functioning as a seamless, end-to-end approach.

A comprehensive review of the literature clarified the key challenges outlined above and informed the overarching research question that guides the thesis: How can early warning systems and anticipatory action be designed to maximize the impact of aid delivered to vulnerable agricultural communities? This question is addressed through four research papers: Paper I highlights how early warning systems and anticipatory action can be integrated; Paper II and Paper III advance scalable, impact-focused forecasts that predict actual crop losses, moving beyond generic alerts; and Paper IV integrates impact-based forecasts with mathematical optimization to target the most vulnerable, and evaluates the practical effectiveness of the framework using a simulation-based approach. This DPhil research aims to make anticipatory action more timely, equitable, and targeted. The research uses India as a case study, due to heightened climate vulnerability in its rice-growing areas. The proposed approaches are broadly applicable to other agricultural settings confronting climate risks.

This DPhil research aims to make anticipatory action more timely, equitable, and targeted. The research uses India as a case study, due to heightened climate vulnerability in its ricegrowing areas. The proposed approaches are broadly applicable to other agricultural settings confronting climate risks.

Authors: De Clercq, D

• DOI: 10.5287/ora-9evyanv6b
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: anticipatory action; agriculture; optimization; machine learning
• Subjects: Social data science