Environmental drivers and machine-learning analysis of grapevine sap flow on hilly slopes in Southwest China

Abstract

Global climate change has increased the frequency and intensity of high-temperature droughts, threatening fruit production in the fragile hilly regions of Southwest China. Grapevine ( Vitis labrusca×vinifera ‘Shine Muscat’ ) is widely grown there, but its sap flow dynamics under climate change remain insufficiently explored. This study monitored sap flow and environmental variables from 2023 to 2024 to characterize grapevine sap flow dynamics and their drivers, and evaluated Random Forest (RF) and XGBoost models for daily sap flow prediction. Results showed that: (1) Sap flow exhibited a pronounced seasonal single-peak pattern, with 68.33–88.29% of the annual sap-flow total occurring from April to September; (2) At the daily scale, sap flow generally increased after approximately 6:00 a.m. Under moderate atmospheric demand, a bimodal pattern was commonly observed, with 35.61–57.35% of the daily sap flow total, peaking between 10:00 a.m. and 3:00 p.m. During periods of high vapor pressure deficit (VPD > 2 kPa), the diel pattern shifted to an earlier single peak at approximately 10:00–11:00 a.m.; (3) Path analysis indicated that photosynthetically active radiation was the key factor affecting sap flow. Correlation analysis revealed that the meteorological factors exerted stronger direct effects on sap flow than soil factors (soil temperature, moisture, electrical conductivity, water potential); (4) With both meteorological and soil predictors, RF and XGBoost models exhibited high accuracy in predicting daily grapevine sap flow (test set R² : 0.888 and 0.891 respectively), and the XGBoost model had better generalization ability (RMSE: 0.086, MAE: 0.059). When soil predictors were excluded, the corresponding R² values decreased to 0.804 and 0.808. This study provides a scientific basis for the formulation of orchard irrigation schedules and water resource management of fruit trees under climate change on the hilly slopes of Southwest China.