Economic Adoption Thresholds for Conventional and Prospective AI-Based Fungicide Decision Support Systems in Winter Wheat : Scenario-Based Analysis Using RISE Testbed Digitalised Agriculture Field Trials

Sammanfattning

Fungicide treatment in winter wheat is an economic decision made under uncertainty. Farmers must choose a treatment strategy before disease pressure is known. The consequences of a wrong decision are not equal. Missing a necessary treatment in a severe season can cause substantially larger losses than applying an unnecessary treatment in a mild one. This thesis evaluates whether information-based fungicide decision strategies can provide sufficient economic value to compete with standard practice under the conditions studied. Four strategies are compared: no fungicide, standard practice, a DSS-assisted strategy, and a prospective AI-based decision support system (AI-DSS). The analysis combines field-trial data from the RISE Testbed Digitalised Agriculture with literature-based parameters for DSS and AI-DSS performance. The no-fungicide and standard-practice strategies are based on observed field-trial outcomes, whereas the DSS-assisted and prospective AI-DSS strategies are modelled using published evidence. Strategies are evaluated using expected value and downside risk relative to standard practice, and a break-even annual licence fee is derived for prospective AI-DSS. The results reveal a strong asymmetry in the cost of missed treatment. The economic shortfall from not applying fungicide is approximately 1,468 SEK/ha in low-disease-pressure years but 7,687 SEK/ha in high-disease-pressure years. Under the primary specification, the DSS-assisted strategy generates a negative expected value of −1,610 SEK/ha relative to standard practice, whereas the prospective AI-DSS remains positive at 116 SEK/ha. This corresponds to a break-even annual licence fee of 116 SEK/ha. Sensitivity analysis identifies recommendation accuracy in high-disease-pressure years as the critical determinant of economic viability. The break-even fee reaches zero at 91.2 percent accuracy and becomes negative below this threshold. The study concludes that the economic value of fungicide decision support depends primarily on performance in high-disease-pressure years. For a prospective AI-DSS to justify a positive licence fee, it must maintain sufficiently high recommendation accuracy when the consequences of error are greatest.