Abstract
Future variations of global vegetation are of paramount importance for the socio-ecological systems. However, up to now, it is still difficult to develop an approach to project the global vegetation considering the spatial heterogeneities from vegetation, climate factors, and models. Therefore, this study first proposes a novel model framework named GGMAOC (grid-by-grid; multi-algorithms; optimal combination) to construct an optimal model using six algorithms (i.e., LR: linear regression; SVR: support vector regression; RF: random forest; CNN: convolutional neural network; and LSTM: long short-term memory; transformer) based on five climatic factors (i.e., Tmp: temperature; Pre: precipitation; ET: evapotranspiration, SM: soil moisture, and CO(2)). The optimal model is employed to project the future changes in leaf area index (LAI) for the global and four sub-regions: the high-latitude northern hemisphere (NH), the mid-latitude NH, the tropics, and the mid-latitude southern hemisphere. Our results indicate that global LAI will continue to increase, with the greening rate expanding to 2.25 times in high-latitude NH by 2100 against the 1982-2014 period. Moreover, RF shows strong applicability in the global and NH models. In this study, we introduce an innovative model GGMAOC, which provides a new optimal model scheme for environmental and geoscientific research.