A new AG-AGB estimation model based on MODIS and SRTM data in Qinghai Province, China. (December 2021)

• Record Type: Journal Article
• Authors: Zhao, Li; Zhou, Wu; Peng, Yiping; Hu, Yueming; Ma, Tao; Xie, Yingkai; Wang, Liya; Liu, Jiangchuan; Liu, Zhenhua
• Is Part Of: Ecological indicators. Volume 133(2021)
• Publication Date: 2021-12
• Subjects: AG-AGB -- XGBoost -- Estimation indicators -- New model; Environmental monitoring -- Periodicals; Environmental management -- Periodicals; Environmental impact analysis -- Periodicals; Environmental risk assessment -- Periodicals; Sustainable development -- Periodicals; 333.71405
• DOI: 10.1016/j.ecolind.2021.108378
• Languages: English
• ISSNs: 1470-160X

Abstract

Highlights: A new model for improving the estimation accuracy of AG-AGB. EVI, radiation, altitude, B5/B7, latitude, and precipitation were the key indicators. 1200 samples was the lowest sample size to construct the new model. The AG-AGB increased more than it decreased on Qinghai Province (2005–2018). Abstract: The ability to rapidly and effectively monitor the alpine grassland aboveground biomass (AG-AGB) is very important to grassland resource management and sustainable use. The present satellite-driven models for estimating AG-AGB still need be improved due to incomplete monitoring indicators and a lack of sufficient in situ measurements. Thus, this study proposes a new AG-AGB estimation model based on MODIS and SRTM data for improving the estimation accuracy of AG-AGB. In this new model, first six estimation indicators were obtained from 33 spectral and environmental indicators using Extreme Gradient Boosting (XGBoost) and correlation analysis. Second, multiple linear regression (MLR), a back propagation neural network (BPNN), a support vector machine (SVM), and the random forest (RF) algorithm were used to build the new model based on the long-term AG-AGB field measurements and the corresponding to six estimating indicators for 6128 samples. Third, the new model was used to map the spatial distribution of the AG-AGB in Qinghai Province, China. And, the trends of the AG-AGB changes from 2005 to 2018 were analyzed using Sen and Mann-Kendall trend analysis and Hurst exponent analysis. The results show that 1) the six estimation indicators (EVI, radiation, altitude, B5/B7, latitude, and precipitation) were confirmed to be important for estimating the AG-AGB. 2) Among the MLR, BPNN, SVM, and RF, the model based on MODIS and SRTM data constructed using the RF algorithm was determined to be the best new model, with an R 2 of 0.938 and a relative root mean square error (RRMSE) of 19.88%, for accurately estimating the AG-AGB based on 6128 samples. 3) Moreover, a set of 1200 samples was determined to be the lowest sample size for constructing the new model, with an R 2 of nearly 0.923 and an RRMSE of 25%. 4) The AG-AGB can be more accurately mapped using the new model ( R 2 of 0.800 and RRMSE of 23.91%), indicating that the new model based on MODIS and SRTM data has the potential to map the AG-AGB over a large area. And over the last 14 years, the AG-AGB overall trends were that the areas with increasing AG-AGB (67.65%) were much larger than those with decreasing AG-AGB (32.35%), and the area of the sustainable AG-AGB increase accounts for 39.75% of the total alpine grassland area in northeastern and southwestern parts of Qinghai Province. The proposed estimation model based on MODIS and SRTM data is more suitable, greatly improves the accuracy of AG-AGB modeling, and provides a scientific basis for the reasonable use and management of alpine grassland resources.