Monitoring and forecasting the spatiotemporal dynamics of vegetation across the Loess Plateau emerge as critical endeavors for environmental conservation, resource management, and strategic decision-making processes. Despite the swift advances in deep learning techniques for spatiotemporal prediction, their deployment for future vegetation forecasting remains underexplored. This investigation delves into vegetation alterations on the Loess Plateau from March 2000 to February 2023, employing fractional vegetation cover (FVC) as a metric, and scrutinizes its spatiotemporal interplay with precipitation and temperature. The introduction of a convolutional long short-term memory network enhanced by an attention mechanism (CBAM-ConvLSTM) aims to forecast vegetation dynamics on the Plateau over the ensuing 4 years, leveraging historical data on FVC, precipitation, and temperature. Findings revealed an ascending trajectory in the maximum annual FVC at a pace of 0.42% per annum, advancing from southeast to northwest, alongside a monthly average FVC increment at 0.02% per month. The principal driver behind FVC augmentation was identified as the growth season FVC surge in warm-temperate semi-arid and temperate semi-arid locales. Precipitation maintained a robust positive long-term association with FVC (Pearson coefficient > 0.7), whereas the temperature–FVC nexus displayed more variability, with periodic complementary trends. The CBAM-ConvLSTM framework, integrating FVC, precipitation, and temperature data, showcased commendable predictive accuracy. Future projections anticipate ongoing greening within the warm-temperate semi-arid region, contrasted by significant browning around the Loess Plateau’s peripheries. This research lays the groundwork for employing deep learning in the simulation of vegetation’s spatiotemporal dynamics.

监测和预测黄土高原植被时空动态已成为环境保护、资源管理和战略决策过程的关键工作。尽管用于时空预测的深度学习技术取得了迅速进展，但它们在未来植被预测中的部署仍未得到充分探索。本调查采用植被覆盖度（FVC）作为指标，深入研究了2000年3月至2023年2月黄土高原的植被变化，并仔细研究了其与降水和温度的时空相互作用。引入由注意力机制增强的卷积长短期记忆网络（CBAM-ConvLSTM）旨在利用 FVC、降水和温度的历史数据来预测未来 4 年高原上的植被动态。调查结果显示，最大年度 FVC 呈每年 0.42% 的上升趋势，从东南向西北推进，而月平均 FVC 增量为每月 0.02%。 FVC 增加的主要驱动因素被确定为暖温带半干旱和温带半干旱地区生长季节 FVC 激增。降水量与 FVC 保持着稳健的长期正相关关系（皮尔逊系数 > 0.7），而温度与 FVC 关系则显示出更大的变异性，并具有周期性互补趋势。 CBAM-ConvLSTM 框架集成了 FVC、降水和温度数据，展示了值得称赞的预测准确性。未来的预测预计，暖温带半干旱地区将持续绿化，而黄土高原周边地区将出现明显的褐变。这项研究为利用深度学习模拟植被时空动态奠定了基础。