Global assessment of climate-driven lag and cumulative effects on vegetation using Earth Observation data and multivariable machine learning models

Vegetation rarely responds instantaneously to meteorological variations; instead, its adjustment often reflects delayed and accumulated climatic influences. However, these temporal effects have not been comprehensively analyzed at the global scale, particularly with higher-resolution Earth Observation (EO) data and across multiple climate drivers. By leveraging EO datasets and modeling approaches, this study characterizes global trends in vegetation and climate, evaluates time effects, and develops climate–vegetation regression and prediction models. Our results revealed normalized difference vegetation index (NDVI) increase alongside shifts in key climatic variables during the study period. A large proportion of vegetated regions showed temporal responses, with heterogeneity in major climate zones. Incorporating time-effect variables into eXtreme Gradient Boosting (XGBoost) model influenced model performance across regions, while temperature (Tmp) and surface solar radiation downwards (SSRD), together with their corresponding temporal-effect predictors, were identified as the key influential climatic contributors. Among predictive approaches, Long Short-Term Memory (LSTM) showed the highest skill in capturing nonlinear vegetation–climate interactions. These findings underscore that temporal effects are fundamental to global vegetation dynamics and point to the need for EO-based, multivariable modeling frameworks to elucidate ecosystem response under ongoing climate change.