The Bioeconomics of Perennial Crop Disease Management under Partial Observability: Evidence from Western X Disease in Cherry Orchards

Perennial crops affected by diseases with prolonged latent stages present a distinctive economic problem where infected trees can remain asymptomatic for years while continuing to transmit disease. As a result, farmers make irreversible capital decisions regarding tree removal and replanting without observing the true infection status of the orchard. We address three economic questions. (i) What is the economic infection threshold, i.e. the share of symptomatic trees in the orchard, beyond which treatment of individual infected trees result in negative return, (ii) what is the economic value of spatially targeted disease tests around symptomatic trees, relative to random surveillance, and (iii) when does disease management strategies that are adaptive to the information of disease tests outperform simpler static rules? We develop a spatially explicit Partially Observable Markov Decision Process (POMDP) bioeconomic framework to analyze optimal disease management in a perennial production system. The framework integrates three components: (1) a biological module governing farm-level infection dynamics, (2) an economic module capturing age-structured yields, production costs, and irreversible replanting decisions, and (3) a decision module in which growers update beliefs about each tree’s latent health state using visual scouting and imperfect diagnostic tests. Orchard disease dynamics evolve on a cellular automata grid where infection probabilities depend on the spatial proximity of symptomatic trees. Economic outcomes are evaluated over a 25-year horizon using discounted net present value (NPV). We implement a parallel rollout algorithm to approximate the optimal adaptive disease management policies. We consider Western X disease (WXD) affecting cherry orchards in the Pacific Northwest. We compare seven management scenarios spanning the range from no intervention to fully adaptive spatially targeted management. We account for several trade-offs in these scenarios. Only cutting infected trees suppresses disease but permanently lowers orchard density and overall yield. Cutting infected trees and replanting healthy ones maintains orchard density and yield at the costs of replanting expenditures and the risks that the newly planted trees become infected. The results show that unmanaged disease generates severe long-run losses through compounded infection over time. Relative to a disease-free benchmark, the no-management scenario reduces orchard NPVs by more than $330,000 per acre over 25 years. These losses arise from the disease yield penalties which increases as disease severity and prevalence rise over time. Each infected tree increases future infection pressure across neighboring trees and leaving infected trees in the orchard exponentially increases the disease damages. Consequently, we find that early disease management intervention that keeps the share of symptomatic trees below approximately 1% produces substantially higher economic returns than delayed action. The economic return becomes negative when the share of symptomatic trees is more than 7% of orchard and farmer choose to only remove the symptomatic trees. This threshold increases to 22% if the farmer chooses to remove the infected trees and replant the healthy ones instead. Additionally, we show that spatially targeted testing substantially improves economic gains compared to random testing by concentrating surveillance around symptomatic trees, where infection probabilities are highest. Importantly, the value of information is not fixed. The same spatially targeted test raises return when paired with a replant policy but lowers return when paired with removal alone, since replanting captures the benefit of earlier, more precisely targeted capital redeployment. The fully adaptive POMDP policy achieves the highest NPV of all scenarios, but only when the planning horizon is long enough for newly planted trees to recoup establishment costs. The optimal adaptive disease management strategy follows a three-phase structure. Early aggressive containment through removal without replanting, a middle recovery phase consisting of removal with replanting, and a terminal phase in which removal occurs without replanting because newly planted trees cannot recover establishment costs before the planning horizon ends. Also, we show that the optimal testing radius changes over time unlike previous studies who considered a fixed testing radius. Additionally, we identify a boundary condition for the well-supported findings of previous studies that adaptive, information-contingent management weakly dominates the best available static rule. This presumption fails once the managed system contains slow-maturing, lumpy capital whose returns compound over a horizon longer than the remaining decision-making window. This boundary condition is a general property of the interaction between capital-accumulation structure and planning horizon. This condition extends to other bioeconomic settings involving similarly structured, long-lived investments, such as timber rotation and aquaculture stocking.