Disease progression and genotypes performance
RAB onset was observed during the pod formation stage in 2021 (Sept 6–12) and 2022 (Aug 30–Sept 5) (Supp. Fig 1). Earlier studies observed that the disease typically begins at flowering and worsens with crop maturity, progressing of severe blight in the reproductive phase due to canopy-induced humidity
(Amrate et al., 2021).
The disease progress was measured in terms of AUDPC. AUDPC values were significantly higher (Fcal=16733.41***, P<0.0001) in 2022 (1488.46) compared to 2021 (416.96), indicating more severe disease progression during the former year (Table 1). In 2021, rainfall was lower (266.40/ mm) and more evenly distributed (SD/ =/ 67.72; 2.63 rainy days). In contrast, 2022 had higher but more variable rainfall (360.22 mm; SD = 85.89; 3.43 rainy days), creating conditions favourable for disease progression. However, excessive rainfall during pod formation may also cause waterlogging, potentially suppressing
R.
solani by inducing anaerobic conditions or hindering sclerotia germination
(Kumar et al., 1999).
During the critical phase of RAB development, warm and humid conditions prevailed, with average Tmax of 29.39oC, Tmin of 22.54oC, RHM at 90.01% and RHE at 83.64%. This microclimate favoured RAB progression and sustained
R.
solani activity, aligning with earlier reports that temperatures of 25-30
oC and relative humidity of 85-95% promote mycelial growth and sclerotia formation (
Harikrishnan and Yang, 2004).
The analysis of AUDPC obtained through percentage disease incidence demonstrated a significant difference among genotypes (P<0.0001), year (P<0.0001) and genotypes x year (P<0.0001) included in the current study. This result indicated the influence of both genetic and environmental factors on RAB progression. It also reflected a complex relationship between host resistance and environmental conditions, consistent with earlier findings on genotype–environment effects in RAB development (
Rodriguez-Herrera et al., 2023).
Genotypes EC 325098 (959.24), EC 289099 (951.97) and EC 251541 (969.12) showed significantly higher AUDPC than JS 9560 (902.64) and EC 343312 (891.53). The higher AUDPC makes suitable these genotypes suitable candidates for studying disease progression and weather-based modelling. The variation among genotypes suggests differential resistance levels among susceptible genotypes (
Surbhi and Singh, 2020).
Weather variables correlation with RAB disease incidence
In 2021, no weather variables significantly correlated with disease incidence (p<0.05) in any genotype. However, in 2022, Tmax showed a strong positive correlation (0.76-0.80; p<0.01) in every genotypes, indicating its critical role in RAB progression. Elevated temperatures likely enhanced sclerotia germination, infection cushion formation and hyphal spread (
Bhamra et al., 2022;
Bashyal et al., 2022).
Pooled disease incidence across the tested genotypes showed a strong negative correlation with RF (–0.58, p<0.0001) and NRD (–0.62, p<0.0001), indicating increased rainfall and number of rainy days are strongly negatively associated with increased virulence of
R.
solani (
Amrate et al., 2021). The remaining
weather variables including Tmax (-0.28, p<0.01) and Tmin (0.31, p<0.01) were significantly correlated with the pooled disease incidence across the tested genotypes. These associations highlight the role of temperature in enhancing the virulence and epidemic potential of
R.
solani (Spurlock et al., 2016).
Model development
Two single-variable regression models were developed (Table 2). Model A, developed with RF and Model B, based on NRD, both exhibited strong negative associations with RAB progression. Previous studies also highlight rainfall and number of rainy days as a critical factor influencing the onset and progression of the disease, aligning with our findings
(Romero et al., 2021). Both models, showing relatively high R² and lower AIC, BIC, CVS, RMSE and PRESS values, were prioritized over others for their superior predictive performance (Table 2).
Among the both models, model B demonstrated greater explanatory power, accounting for 38 % (R2) of the variation in RAB incidence (Table 2). Model B showed superior predictive performance indicator than model A. RF and NRD influence canopy wetness, ambient humidity and leaf surface moisture, thereby creating microclimatic conditions highly favourable for
R.
solani infection and subsequent disease spread (
Rodriguez-Herrera et al., 2023). Earlier studies have demonstrated that epidemic onset and intensity are governed more by the duration and frequency of leaf wetness than by total rainfall, underscoring the pivotal role of rainfall distribution in driving RAB disease outbreaks
(Amrate et al., 2021; Surbhi and Singh, 2020;
Nainwal et al., 2024).
Model C was developed using stepwise and subset regression based on significantly correlated variables (Table 2). Model C incorporated two independent predictors, Tmin and NRD, which were identified as the key determinants of RAB incidence. It explained 13.63% to 25% more variation in RAB incidence than Models A and B. Additionally, Model C showed improved predictive performance, with lower AIC (225.21), BIC (233.85), RMSE (1.11) and PRESS (120.08) compared to the single-variable models.
These results indicate that Model C provided the best fit compared to single-variable models. The improved accuracy highlights the importance of integrating multiple climatic factors for reliable RAB forecasting
(Amrate et al., 2021). Similar observations were reported by earlier studies, where multiple regression models effectively captured the variability in RAB disease incidence in soybeans
(Amrate et al., 2021; Surbhi and Singh, 2020;
Nainwal et al., 2024).
Model validation and predictive power
All regression models were validated using cross-validation statistics calculated from LOOCV and cross-validated residual distributions (Fig 1 and Table 2). The predictive capacity of each model for future disease incidence was tested using ROC analysis (Fig 1). Lowest prediction error (CVS = 1.87) was observed for multiple-variable regression model C (Table 2), suggesting that integrating multiple environmental predictors substantially enhances model performance
(Amrate et al., 2021).
Cross-validated residuals from Model C showed a tighter clustering around the zero line compared to model A and B, indicating a better model fit with minimal de
viation between predicted and observed values (Fig 2) Similarly, cross-validated residual distributions have been effectively used to validate weather-based forecasting models for various other plant diseases
(Chaulagain et al., 2020; Ali et al., 2022).
Results from the ROC analysis indicated that all models were capable of accurately predicting disease severity at a threshold level of 0.1. Model C achieved the highest predictive accuracy with an AUC of 0.860, indicating excellent discriminatory ability in distinguishing diseased from non-diseased cases. Additionally, both model B and model C demonstrated high accuracy (85.90 %), specificity (84.70%), recall (100.00%) and F1-score (52.60%) (Supp Table 1), indicating strong predictive performance with a good balance between true positives and false positives
(Singh et al., 2021).
This study offers key insights for managing RAB in soybeans, especially in India’s central zone, which produces ~90% of the country’s soybeans and largely grows the susceptible cultivar JS 95-60
(Rajput et al., 2025; Amrate et al., 2018). These models can estimate RAB risk using weather data from the preceding 15 days and support region-specific disease advisories, optimized fungicide scheduling and identification of agroclimatic risk zones. The effectiveness of the proposed prediction models is contingent on the accessibility and precision of localized meteorological data.