Quantifying the influencing factors and predictive analysis of cotton defoliation and maturation based on machine learning

• Key factors influencing cotton defoliation and ripening were identified across major Chinese cotton-producing regions. • Key influencing factors, such as defoliant dosage, temperature, and weather conditions, vary across different cotton regions. • Machine learning models (RF, SVM, GBM) were developed to predict defoliation percentage (DP), boll opening percentage (BOP), and increment of boll opening percentage (IBOP). Good defoliation and boll opening are essential for cotton mechanical harvesting. However, in actual production, many factors can affect the effect of defoliation and boll opening rate. Assessing the importance of factors affecting cotton defoliation and ripening, and predicting these factors, is critical to adjust the key influencing factors for achieving optimal defoliation and boll opening effect. This study, conducted from 2016 to 2022 across the three major Chinese cotton-producing regions – the Yellow River Valley, the Yangtze River Valley, and the Xinjiang Autonomous Region – aimed to evaluate the key factors influencing post-application defoliation and ripening processes and to develop models predicting defoliation and ripening progression using three machine learning methods: Random Forest, Support Vector Machines, and Gradient Boosting Machine. The results show that machine learning-selected variables accurately predict defoliation percentage (DP), boll opening percentage (BOP), and the increment of boll opening percentage (IBOP). Crucially for DP, considering the dosage of applied XSL, as well as other defoliants, is essential. Also important are the highest temperatures’ influence and pre-treatment boll opening percentage on defoliation in the Yellow River Valley and Xinjiang Autonomous Region. For BOP, enhancing the pre-treatment boll opening percentage is vital across all regions. Sunshine, humidity, and precipitation levels, especially in the Yellow River, Yangtze River, and Xinjiang Autonomous Regions, also require attention. Region-specific key factors affect IBOP; in the Yellow River Valley, these are minimum and average temperatures and pre-treatment boll opening percentage. In the Yangtze River Valley and Xinjiang Autonomous Region, the increment is mainly influenced by the pre-treatment boll opening percentage and the application of XSL. Having ranked the importance of influencing factors, these methods were applied to predict variables, with Random Forest demonstrating superior predictive performance (R 2 > 0.7, rRMSE < 15 %). In actual production, producers should pay special attention to the application of XSL to achieve effective defoliation and cultivate a favorable pre-treatment plant population structure to ensure a high boll opening rate post-treatment.