Estimating ear germination and moldiness under continuous rainfall stress during wheat maturity

Wheat is one of the most crucial global staple crops for food security. However, the continuous rainy weather during its growth, particularly at maturation, can easily cause ear germination and moldiness, thus severely impacting the yield and quality. This study aims to accurately monitor and evaluate the germination and moldiness of wheat ears under continuous rainy weather stress during the maturity period. A case study was also conducted on the continuous rainy weather in the western part of the Huang-Huai region of China in late May 2023. The wheat ear germination and moldiness were tackled using meteorological and satellite remote sensing data, with emphasis on the disaster risk elements. Then, meteorological hazard factors were determined from the weather stress mechanisms. The resilience was also characterized using remote sensing parameters, according to the state and environment of the wheat. Thirdly, the modeling factors were selected for subsequent analysis. Spearman correlation and ReliefF method were also used for the feature selection in binary and severity classification tasks, while Pearson correlation was employed to predict the ear germination and moldiness index (EGMI). The optimal factors were then combined to form the SCF, PCF, and RFF factor groups, according to the meteorological and remote sensing types. Subsequently, five classification models (including Logistic regression, LGR) and five regression methods (including multiple linear regression, MLR) were applied for the binary classification and severity grading of wheat ear germination and moldiness, in order to predict and simulate the EGMI. The effectiveness of these models was then compared to identify and grade the wheat ear germination and moldiness. The results showed that the optimal factors were achieved in the identification and severity grading of germination and moldiness using different classifiers, from the perspective of the disaster-causing process of continuous rain and the three elements of disaster risk. The accuracy score (AC) ranged from 0.649 to 0.811 in the binary classification of wheat ear germination and moldiness identification, with the Kappa coefficients between 0.245 and 0.600. In the three-category classification of severity grading, the AC value ranged from 0.432 to 0.622, with the Kappa values between 0.099 and 0.414. The R² value of EGMI prediction ranged from 0.10 to 0.25, with an average mean absolute error (MAE) of 12.93 and an average root mean square error (RMSE) of 16.74. The PCF-XGBR model performed the best, with the R², RMSE, and MAE values of 0.25, 15.69, and 12.05, respectively, as well as the standard deviation (SDEV) and centered root-mean-square deviation (CRMSD) values of 13.10 and 15.55, respectively. Comparative analysis of the three models showed that the remote sensing model was superior to the meteorological model, in terms of the identification of germination and moldiness. While the meteorological model outperformed the remote sensing model, in terms of grading the severity of germination and moldiness. The meteorological-remote sensing model was integrated to balance their shortcomings for better performance and robustness. The estimation of continuous rainy weather disasters was achieved in the western Huang-Huai region, thus filling the technological gap in monitoring wheat ear germination and moldiness. The finding can provide the technical support to reduce the wheat disaster in post-disaster assessment.