Towards a deeper understanding of long-term changes in global vegetation growth

Vegetation provides fundamental supports to human society and strongly affects the Earth system by mediating carbon, water, and energy exchanges. Thus, how vegetation responds to climate change has been a central scientific question of the research community. The prerequisite for answering this question is to have long-term temporal consistent data for monitoring the global vegetation. Despite significant advances in generating long-term vegetation indices data from the 1980s based on the Advanced Very High Resolution Radiometer series, recent evaluation studies have pointed out notable discrepancies between the existing data sets and suggested further improvements. The uncertainties in these monitoring data could be further propagated to the attribution of the changes in global vegetation growth. Here, we introduce our work aiming towards a deeper understanding of long-term changes in global vegetation growth. Based on a machine learning framework, we first evaluate the uncertainties in three widely used long-term leaf area index (LAI) products during the last four decades. This machine learning framework provides a valuable tool to fill the blank that these long-term LAI products before the 2000s have not been systematically evaluated. We then generated a new version of the GIMMS LAI data (i.e., the GIMMS LAI3.1g) based on the latest GIMMS NDVI3.1g data and an extensive multi-source remote sensing data collection. The GIMMS LAI3.1g data showed many improvements, especially in terms of temporal consistency. Combining the GIMMS LAI3.1g, the TRENDY ecosystem models, and machine learning models, we also discuss the opportunities and challenges of using machine learning to understand the underlying mechanisms of the long-term changes in global vegetation growth.