A spatially explicit negative‑prognosis framework for Cercospora leaf spot using remotely sensed leaf area index

Timely and spatially explicit forecasts of plant disease risk are essential for efficient fungicide use and sustainable crop protection, yet conventional epidemiological models commonly rely on weather data from single stations and assume spatially uniform canopies. Such simplifications overlook within-field variability in crop development that can be revealed by Earth observation data. Using the sugar beet–Cercospora beticola pathosystem, we tested whether integrating remotely sensed canopy metrics with epidemiological models can improve the spatial precision and timeliness of disease warnings.

We coupled a classic Cercospora leaf spot (CLS) negative-prognosis model with spatially explicit leaf area index (LAI) maps derived from multispectral satellite imagery (10 m), super-resolved imagery (1 m), and airborne campaigns (1 m). From these LAI time series, canopy closure dates were determined for each grid cell and used to initialise the CLS model. This workflow produced a distribution of earliest estimated epidemic onset dates (EEEOs) across the field, instead of a single area-averaged forecast. The method was benchmarked against ground observations of epidemic onset (EO) derived from repeated disease assessments in inoculated and non-inoculated plots.

EO occurred 156.9 days after sowing (DAS) in non-inoculated areas. Compared with the conventional, uniform-field prognosis, our spatially explicit approach predicted substantially earlier EEEOs—117.4 DAS for satellite, 114.0 DAS for super-resolved satellite, and 103.8 DAS for airborne imagery, with comparable trends in inoculated plots. These results confirm that accounting for within-field canopy heterogeneity allows earlier and more localised warnings, offering a pathway towards precision crop protection. The proposed workflow captures sub-field variability in canopy development missed by regional-scale disease models, thereby supporting more efficient scouting strategies and fungicide applications.

To promote transparency and reuse, all modelling and analysis steps are implemented in the open-source R package cercospoRa, which operationalises existing CLS rules and enables reproducible negative-prognosis modelling using FAIR (Findable, Accessible, Interoperable, Reusable) principles. The package provides a modular framework for integrating remote-sensing data, radiative transfer–based LAI retrieval, and epidemiological modelling. Beyond the case study presented here, cercospoRa can serve as an open hub for implementing new epidemiological components, such as inoculum distribution kernels or refined definitions of “epidemic onset”, and facilitate community-driven advance of spatial plant disease modelling.

Our results highlight that merging open data, Earth observation, and process-based modelling can bridge the current gap between plant epidemiology and agroecosystem monitoring. Future research aims to upscale these concepts towards landscape epidemiology, exploring how canopy heterogeneity, infection sources, and microclimatic variation combine across larger spatial scales to shape epidemic risk. By fostering open and reproducible workflows, we aim to advance data-driven, science-based decision support tools that contribute to sustainable, climate-resilient crop management.