Combining spatio-temporal pest risk prediction and decision theory to improve pest management in smallholder agriculture

Summary

Pests and diseases are a major cause of crop loss, affecting food security and, in particular, the livelihoods of smallholder farmers. While some pest management practices are widely adopted, poorly informed decisions, such as over-application of pesticides, can severely impact human and environmental health as well as farm profits. Frequent crop monitoring is often recommended for making interventions more effective, but highly intensive monitoring is beyond the capacity of many farmers. By combining pest risk prediction and decision analysis, we developed a framework to support the decision of whether to apply pesticides preventively, monitor crops, or omit any crop protection measures for a given day and geographic location. We used a deep Gaussian process classification model for spatiotemporal pest risk prediction, incorporating new observations in near real-time. We then applied decision analysis to determine the best intervention for a given pest risk prediction. Monitoring is recommended when the Value of Information (VoI) exceeds the cost of monitoring. We applied this method to a case study of Tuta absoluta infestations in tomato production in Andhra Pradesh, India. Our model-based decision strategy would reduce average pest-related costs by 25.4±4.3% and pesticide use by 58.8±2.7% according to Monte Carlo simulations. When monitoring results are used to update the pest risk model and thus shared with other farmers, additional value can be generated for the community. We found that this community VoI exceeded the expected information value for the individual farmer (individual VoI). Our open-source Python model can easily be adapted to other crops and pathogens, and serve as a basis for pest risk-aware decision support systems.