Abstract
Designing and identifying biologically meaningful weather-based predictors of plant disease is challenging due to the temporal variability of conducive conditions and interdependence of weather factors. Confounding effects of plant genotype further obscure true environmental signals within observed disease responses. To address these limitations, this study leveraged window-pane analysis with feature engineering and stability selection, to identify weather-based variables associated with latent environmental factors ( λ^ ) of a factor analytic model explaining genotype-by-environment (GEI) effects on disease severity in multi-environment trials. Using Stagonospora nodorum blotch of wheat as a case study and a two-stage feature engineering procedure, hourly weather data, i.e., air temperature (T), precipitation (R), and relative humidity (RH), were aggregated into 1,530 distinct time series, in the first stage feature engineering procedure. These series were correlated daily with λ^ throughout the second half of the wheat growing season. In the second stage procedure, significant daily weather variables were consolidated into optimal epidemiological periods relative to wheat anthesis, yielding 60, 19, and 28 second-level weather-based variables derived from the first ( λ^1 ), second ( λ^2 ), and third ( λ^3 ) environmental factor loadings, respectively. Among the weather-based predictors identified, fa1.41_18.TRH.13T16nRH.G80.daytime.sum_25 and fa1.11_5.R.S.dawn.sum_10, were positively associated with λ^1 (i.e., the dominant environmental gradient underlying variation in SNB severity across environments) pre-anthesis, during a period of 24 and 7 consecutive days, respectively. In contrast, fa1.22_16.TR.19T22nR.G0.2.dawn.sum_20 and fa1.2_-12.RH.L35.daytime.sum_15 were negatively associated with λ^1 at pre-anthesis and post-anthesis, respectively. Additional predictors derived from T, R, and RH, were identified up to 63 days pre-anthesis. However, no single predictor consistently maintained an association with λ^ during the entire study period. This framework advances the development of weather 'markers' for detailed environmental profiling of GEI drivers and improves upon prior approaches that limited window-pane analysis to disease outcomes from susceptible hosts to identify weather-based variables for predicting plant disease epidemics.