Deep learning projects future warming-induced vegetation growth changes under SSP scenarios

Abstract

Climate warming has been projected to enhance vegetation growth more strongly in higher latitudes than in lower latitudes, but different projections show distinct regional differences. By employing big data analysis (deep learning), we established gridded, global-scale, climate-driven vegetation growth models to project future changes in vegetation growth under SSP scenarios. We projected no substantial trends of vegetation growth change under the sustainable development scenario (SSP1-1.9) by the end of the 21st century. However, the increase of vegetation growth driven by climate warming shows distinct regional variability under the scenario representing high carbon emissions and severe warming (SSP5-8.5), especially in Northeast Asia where growth could increase by (6.00% ± 4.21%). This may be attributed to the high temperature sensitivities of the deciduous needleleaf forests and permanent wetlands in these regions. When the temperature sensitivity that is defined as permutation importance in deep learning is greater than 0.05, the increase in vegetation growth will be more prominent. In addition, an extreme temperature increase across grasslands, as well as changing land-use management in northern China may also influence the vegetation growth in the future. The results suggest that the sustainable development scenario can maintain stable vegetation growth, and it may be a reliable way to mitigate global warming due to potential climate feedbacks driven by vegetation changes in boreal regions. Deciduous needleleaf forests will be a centre of greening in the future, and it should become the focus of future vegetation dynamics modelling studies and projections.

Keywords

• Vegetation growth
• Climate change
• Deep learning
• Climate sensitivity
• Future projection