Answers are grounded in this paper and cite their sources.
Late-spring frost during the flowering period is one of the most severe agro-meteorological disasters affecting apple production in China. Traditional low-temperature experiments and frost-risk index based on apple cultivars and did not fully consider the impacts of other important meteorological factors and the geographical features. Machine learning models could more comprehensively capture these impacts, although with relatively low interpretability. In this study, we developed an explainable frost risk prediction framework by integrating daily meteorological observations with geographic variables across the four major apple-growing regions of China (Bohai Bay, Loess Plateau, Southwest Cool Highland, and Xinjiang). Ten candidate predictors were screened using Pearson correlation and variance inflation factors to remove highly collinear variables. Two experimental datasets were constructed: a combination of only meteorological variables and a combination of meteorological and geographic variables. Five machine learning models (Random forest, RF; Support vector machine, SVM; Multilayer perceptron, MLP; Extreme gradient boosting, XGBoost; Long short-term memory neural network, LSTM) and an attention-enhanced deep learning architecture (LSTM-CAM) were then trained to predict late-spring frost risk during apple flowering. Model interpretability was achieved using Shapley Additive Explanations (SHAP). Results showed that RF achieved the best performance (overall accuracy = 0.90, AUC = 0.96). SHAP analysis identified minimum temperature as the dominant meteorological predictor and latitude as the most influential geographic factor. LSTM-CAM model effectively captured the interactions between meteorological and geographical variables by adaptively weighting feature importance through its attention mechanism. This study demonstrates that an explainable machine learning approach, which integrates multi-source meteorological and geospatial features, is effective in forecasting late-spring frost risks, thereby providing a robust decision-support tool for climate-resilient apple production and orchard management.
No takes yet. Share an insight, caveat, or question.
Zhou et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: