Data Packet.

This study addresses the inefficiencies in ecological restoration on the Qinghai-Tibet Plateau (QTP), particularly concerning prolonged vegetation restoration cycles, slow soil quality improvement, and difficulties in quantifying manual intervention measures. An integrated Cubist regression tree model is developed using ecological environment data from the QTP and multi-source environmental monitoring data from 2019 to 2023. The model combines a lightweight self-attention mechanism (SA) with bidirectional gated recurrent units (BiGRU) to enhance the accuracy and adaptability of restoration efficiency prediction. The SA mechanism dynamically adjusts environmental factor weights to strengthen nonlinear relationship capture capabilities, while the BiGRU learner optimizes temporal feature representation to accommodate spatiotemporal variability in restoration processes. Input factors include fractional vegetation cover (FVC), temperature, precipitation, soil moisture, and manual intervention measures (irrigation volume, planting density), with outputs being vegetation restoration rate and soil quality improvement effects. Experimental results demonstrate that the model achieves less than 5% error in vegetation restoration rate prediction, with correlation coefficients exceeding 0.90, and 96% accuracy in soil improvement prediction. Temperature and precipitation show contribution rates of 32% and 25%, respectively, while soil moisture and NDVI jointly contribute 25%. Prediction accuracy remains above 90% across different altitude zones, indicating strong regional adaptability. Notably, in areas with annual precipitation below 200 millimeters, every 10% increase in irrigation volume leads to approximately 15% improvement in vegetation survival rate. This study provides quantitative and operational intervention guidelines for plateau ecological restoration, enhances the evaluation efficiency of manual intervention measures, and has significant practical application value.