Estimation of Global Grassland Net Ecosystem Carbon Exchange Using a Model Tree Ensemble Approach

摘要

Understanding the net ecosystem CO_2 exchange (NEE) between terrestrial ecosystems and the atmosphere is crucial for accurate estimation of carbon budgets, which remains unclear for grassland ecosystems. Here we upscaled site-level NEE from 44 grassland flux towers (1,457 site-months) to the global scale by using a model tree ensemble approach that considers the management activities (grazing and cutting) (MTE-GM). Cross-validation showed that MTE-GM performs reasonably well in terms of among-site variability and seasonal variation, with a Nash-Sutcliffe efficiency of 0.90 and 0.86 and an R2 of 0.91 and 0.86, respectively. Radiation (shortwave and longwave), temperature, leaf area index, and fraction of absorbed photosynthetic active radiation had the highest relative explanatory power in predicting NEE. Based on MTE-GM, mean annual NEE of global grassland was 72 ± 4 g C m~(-2) year~(-1) (1.9 ± 0.11 Pg C year~(-1)) during 1982-2011, suggesting that the grassland ecosystems have been acting as a small carbon source during the past three decades. However, grasslands in temperate and continental regions had the largest carbon sink of -61.9 ± 5.7 and -51.8 ± 7.9 g C m~(-2) year~(-1), respectively. Moreover, we found that elimination of grassland management effect resulted in an extra emission of 1.7% CO_2 to the atmosphere (CO_2 sink from the management is 0.03 Pg C year~(-1)). From the 1980s to the 2000s, 38% (22%) and 17% (18%) of pixels showed an increased (decreased) carbon uptake and decreased (increased) carbon release, respectively. Uncertainty assessment suggested that there would be higher confidence in NEE estimates in most parts of middle- to high-latitude regions in the Northern Hemisphere.