Machine learning vs. empirical models: Estimating leaf wetness patterns in a wildland landscape for plant disease management

Abstract

This study presents the development and application of models to estimate leaf wetness duration and their integration with drone-based imagery to analyze plant disease patterns across a coastal gradient. By comparing machine learning algorithms with empirical models, we identified that both approaches effectively predict leaf wetness, particularly in a temperate maritime ecosystem. The models were applied to study two manzanita species (Arctostaphylos tomentosa and A. pumila), revealing a strong correlation between leaf wetness and disease prevalence. This work highlights the role of microclimate conditions in shaping plant health and disease distribution in coastal shrublands. We compared nine popular machine learning algorithms and four empirical threshold models to characterize leaf wetness patterns in a spatially diverse temperate maritime wildland ecosystem. We suggest that simple empirical leaf wetness models based on dew point depression or relative humidity thresholds perform as well as machine learning techniques and should not be overlooked. The relationship between leaf wetness duration and the spatial distribution of plant disease along a coastal-to-inland climate gradient offers valuable insights into disease dynamics.