Vegetation is established as an essential element of the global carbon cycle, playing a significant role in regulating the Earth’s carbon balance. Therefore, the accurate forecasting of vegetation changes is recognized as crucial for understanding global carbon balancing mechanisms. However, traditional point forecasting methods are limited by their failure to account for the uncertainty inherent in prediction outcomes. To address this issue, the Ordos section of the middle reaches of the Yellow River was selected as the study area. A trigonometric-function Harris hawks optimization (TTHHO) method was proposed to facilitate intelligent parameter-space exploration and adaptive optimization. Subsequently, the optimized parameters were integrated into a least squares support vector machine (LSSVM) to construct a nonlinear regression model, denoted as TLA. Additionally, probabilistic interval prediction of the vegetation drought index (VDI) was achieved through the employment of an adaptive bandwidth kernel density estimation (ABKDE) method. Experimental results demonstrate follows.Substantial performance improvements are achieved by the proposed TLA model, with prediction accuracies found to be 5.10%~16.87% higher than comparative models such as LSTM and Transformer (p < 0.05).An increase in the coverage width criterion (CWC) is observed as the confi-dence level decreases (from 95% to 5%).And high predictive accuracy (R² > 0.84) is maintained by the multi-objective TLA model, despite a moderate decrease in overall accuracy compared to the single-objective model. Consequently, a multi-objective TLA interval prediction framework is proposed, providing a rigorous scientific method for monitoring regional-scale vegetation dryness and contrib-uting to the development of predictive systems for global ecosystem management.