Retrieval of grassland aboveground biomass across three ecoregions in China during the past two decades using satellite remote sensing technology and machine learning algorithms

Abstract

The aboveground biomass (AGB) is closely linked to the carbon cycle in grassland ecosystems worldwide. Accurately quantifying AGB variations is thus essential for assessing grassland carbon sequestration and its feedback on climate change. Although many studies have investigated grassland AGB, they are limited to local areas and few research efforts have been attempted to estimate AGB at large scales with the constraint of in situ quadrat harvested AGB. In this study, we used multi-source satellite remote sensing data from 2000 to 2021 and abundant harvest quadrats data to explore AGB estimation methods and then analyze the spatiotemporal patterns of AGB for various grassland types across China’s three ecoregions. The results indicate that: (1) The temporal resolution improvement of remote sensing data results in a higher correlation between satellite remotely sensed NDVI and in situ AGB. Therefore, the MODIS MCD43A4 dataset with higher temporal resolution has a better fit with the harvesting AGB data. (2) Compared to the statistical methods, the machine learning algorithms exhibit high accuracy in estimating grassland AGB. Among them, the random forest (RF) model performs the most robustly, with the highest R2 of 0.83 (explaining 83 % of the variation of the harvesting AGB), and the lowest RMSE of 43.84 gm−2. (3) The multi-year average annual maximum grassland AGB decreases from the southeast to the northwest, with the temperate steppe region having the highest, followed by the alpine vegetation region and the temperate desert region. (4) While approximately 61.94 % of grassland pixels show an increasing trend in average annual maximum AGB from 2000 to 2021, significant (P < 0.05) changes are mainly concentrated in the eastern areas of each ecoregion. Our study presents a valuable framework for estimating grassland aboveground biomass using satellite remote sensing and in situ quadrat datasets. Additionally, it provides a robust product of annual maximum AGB for China’s grasslands from 2000 to 2021, contributing to our understanding of long-term AGB changes in China’s grassland ecosystems.

Keywords

• Grassland aboveground biomass
• Statistical model
• Machine learning
• Linear trend
• Spatiotemporal patterns