Economical-Epidemiological Analysis of the Coffee Trees Rust Pandemic

Abstract

Coffee leaf rust is a prevalent botanical disease that causes a worldwide reduction in coffee supply and its quality, leading to immense economic losses. While several pandemic intervention policies (PIPs) for tackling this rust pandemic are commercially available, they seem to provide only partial epidemiological relief for farmers. In this work, we develop a high-resolution economical-epidemiological model that captures the rust pandemic's spread in coffee tree farms and its associated economic impact. Through extensive simulations for the case of Colombia, a country that consists mostly of small-size coffee farms and is the second-largest coffee producer in the world, our results show that it is economically impractical to sustain any profit without directly tackling the rust pandemic. Furthermore, even in the hypothetical case where farmers perfectly know their farm's epidemiological state and the weather in advance, any rust pandemic-related efforts can only amount to a limited profit of roughly 4% on investment. In the more realistic case, any rust pandemic-related efforts are expected to result in economic losses, indicating that major disturbances in the coffee market are anticipated.

Figures (5)

• Figure 1: A schematic illustration of the tree level CLR spread dynamics. Solid lines indicate changes in the epidemiological state of the coffee tree's branches. Dashed lines indicate berries production. Dotted lines indicate the production and interaction of urediniospores with the coffee tree's branches.
• Figure 2: A schematic view of the farm level CLR pandemic dynamics.
• Figure 3: Baseline dynamics of the epidemiological and economic processes in a single season (one year) without any CLR control measures. The results are shown as the mean $\pm$ standard deviation result of $n = 10000$ repetitions divided into 100 unique farms and 100 interactions for each farm.
• Figure 4: A comparison of the obtained NSP for several different PIP configurations implemented by the DRL agent. The results are shown as the mean $\pm$ standard deviation result of $n = 100$ repetitions. (a) Fully observable pandemic and Fully observable weather. (b) Fully observable pandemic and Model-based weather. (c) $\alpha$-Partial observable pandemic and Model-based weather, where $\alpha = 0.05$.
• Figure 5: Sensitivity analysis of the NSP with respect to the main epidemiological and economic properties of the system. The results are shown as mean $\pm$ standard deviation of $n = 100$ repetitions for the case of $\alpha$-Partial ($\alpha = 0.05$) observable pandemic and model-based weather DRL agent.