Prediction of short‑term drought variation in tea plantations using a LASSO-COX-NOMOGRAM approach

To identify early indicators of short‑term drought in tea plantations driven by climatic and environmental changes, this study develops a LASSO-Cox-nomogram predictive model to achieve accurate prediction of short term and localized drought variation in tea plantations. Corresponding variability quantification indices were designed for multisource climatic data collected by Internet of Things devices. Limma differential analysis was used to examine climatic variables under different drought severities. Combined with univariable Cox regression, this approach systematically screened key climatic factors significantly associated with drought severity and showing clear variation patterns across drought stages. A nomogram was then constructed using LASSO for variable selection and Cox regression for multivariate analysis to assess the impact of climatic changes on drought conditions. LASSO regression was used to screen modeling factors, and fivefold cross‑validation together with multivariate Cox analysis was applied to establish the model. A nomogram was then constructed, and a visual prediction system was developed using Shiny and DynNOM. The prediction model achieved AUC values of 0.776, 0.762, and 0.777 for soil moisture content changes exceeding - 5%, 0%, and 5%, respectively, in the training set. In the validation set, corresponding AUC values were 0.742, 0.799, and 0.710. The model demonstrates strong discriminative ability and effectively captures differences in soil moisture across distinct variation intervals. The calibration curves closely matched the ideal reference lines, and the temporal hold-out testing demonstrated an accuracy of 78.57%. The developed drought prediction system enables accurate forecasting of short-term, localized drought variations in tea plantations. It offers high precision with low computational demand, thereby providing a foundation for improving the yield and quality of Yunnan tea.