Weather data sources influence predictive performance of disease forecasting models

Abstract

Abstract

Weather data from fine- and coarse-scale sources are increasingly used in disease forecasting models to support management decisions, yet comparisons between fine- and coarse-scale sources remain limited, largely due to the difficulty of acquiring fine-scale datasets across locations and seasons. Differences among weather sources can influence model outputs, resulting in variation in predictive performance and the timing of management decisions. This study evaluated four weather data sources, grouped as fine-scale (in-canopy and on-farm stations) and coarse-scale (regional gridded and point-based data from SILO, and Bureau of Meteorology, BOM, stations), across two production regions in Australia. Macadamia orchards were used as a case study to assess spatial variability in 10 weather variables associated with fungal infection over three seasons (2022–2024). Heterogeneous trends and associations were identified among the weather sources, primarily driven by differences in minimum temperature and minimum relative humidity. On-farm and SILO data showed high spatial variability, whereas in-canopy and BOM data were comparatively stable. Predictive performance of disease models based on each weather source was evaluated using airborne conidia dynamics for the fungal pathogens Botrytis , Cladosporium and Neopestalotiopsis / Pestalotiopsis , which cause flower blight in macadamia. In-canopy sources produced significantly lower predictive performance ( p < 0.05) than BOM sources, while on-farm, SILO and BOM sources showed comparable performance ( p > 0.05) across all pathogens. These findings indicate that the choice of weather data source, rather than spatial scale alone, is critical for reliable disease forecasting.