Decoupling Climate and Seed System Drivers of Wheat Yellow Rust Epidemics in Ethiopia: A 30‐Year Geospatial Machine‐Learning Approach

ABSTRACT

Efforts to predict outbreaks of wheat yellow rust ( Puccinia striiformis f. sp. tritici ), a major threat to food security, have been hindered by the failure to distinguish and quantify the respective roles of climate in shaping pathogen niches and agrisystemic factors in amplifying disease risk. The ‘conducive niche hypothesis’ was tested, proposing that epidemics arise from threshold‐based synergies within biophysical and socioeconomic niches. Using a 30‐year geospatial dataset (5478 field surveys, 1993–2023), high‐resolution climate data and machine learning, this study developed a framework to decouple these drivers. A biologically informed Random Forest model ( R ²  = 0.92) significantly outperformed linear models ( R ²  = 0.67), identifying a stable, elevational epidemic corridor (> 2000 m) in the Amhara and Oromia highlands. This climate‐defined niche has precise limits: permissive T _min (7.5°C–12.5°C), inhibitory T _max (< 25°C), and initiating precipitation (12–20 mm). Polygon trend analysis shows a significant structural shift ( p  < 0.001), with years conducive to outbreaks intensifying since 2008, indicating climate‐mediated reconfiguration of spatiotemporal risk. This niche is amplified by agrosystemic vulnerability due to extreme genetic concentration, with 78% of national seed production reliant on four or five varieties. This study provides the first integrated framework that sorts out climate‐driven niche stability from seed system‐driven outbreak magnitude. It delivers an outline for transitioning from reactive to pre‐emptive, precision disease management by targeting distinct drivers, offering a transferable model for building climate resilience in Ethiopia and across other similar agroecologies.