Early detection and recommendation of higher precision in treatment of late blight in potato crops by using drone photos

Abstract: In this thesis an evaluation was performed for the potential use of drone photos in Decision Support System (DSS) to detect outbreak of late blight and provide recommend treatment with fungicides. Late blight is a common crop infection on potato in Sweden. A DSS with and without information from drones were developed in three steps. Firstly, risk classification model for late blight was specified based on weather data, output from a blight forecasting models (SIMCAST) and drone photos. Secondly, this risk model was calibrated and evaluated with data from a late blight field trial and drone images taken during the time of the trial. Thirdly, a decision model specified as a probabilistic (Bayesian) network was built integrating the risk model with management costs and the decisions to spray with fungicides and/or to collect more information by drones to improve the accuracy of the risk classification. The decision model was used to evaluate the cost-benefit of using drone photos in potato late blight management under scenarios of blight incidence and management costs. The findings showed that drone photos may reduce the expected cost under certain conditions such as low blight incidence and low costs. If costs for drones can be held back, there are several opportunities to combine the technology of unmanned vehicles with DSS in order to manage late blight in potato crops. However, a number of limitations exist and further research is needed to achieve a more accurate model and a decision analysis considering multiple decisions before the method can be applied in practice.