Abstract
The Three-River Headwaters Region (TRHR) is located in the interior of the Qinghai-Tibetan Plateau, which is a typical research area in East Asia and is of fragile environment. This paper studied the characteristics of grassland cover changes in the TRHR between 2000 and 2016 using methods of area division (AD) based on natural conditions and tabulate area (TA) dependent on Moderate-resolution Imaging Spectroradiometer (MODIS) 44B product. Further investigations were conducted on some of the typical areas to determine the characteristics of the changes and discuss the driving factors behind these changes. Classification and Regression Trees (CART), Random Forest (RF), Bayesian (BAYE), and Support Vector Machine (SVM) Machine Learning (ML) methods were employed to evaluate the correlation between grassland cover changes and corresponding variables. The overall trend for grassland cover in the TRHR towards recovery that rose 0.91% during the 17-year study period. The results showed that: (1) The change in grassland cover was more divisive in similar elevation and temperature conditions when the precipitation was stronger. The higher the temperature was, the more significant the rise of grassland cover was in comparable elevation and precipitation conditions. (2) There was a distinct decline and high change standard deviation of grassland cover in some divided areas, and strong correlations were found between grassland cover change and aspect, slope, or elevation in these areas. (3) The study methods of AD and TA achieved enhancing performance in interpretation of grassland cover changes in the broad and high elevation variation areas. (4) RF and CART methods showed higher stability and accuracy in application of grassland cover change study in TRHR among the four ML methods utilized in this study.
Highlights
The Three-River Headwaters Region (TRHR) is a typical study area with high ecology values [1,2,3]
Wang et al [12] examined the correspondence between Soil Moisture Content (SMC) and vegetation cover via MODIS13A3 Normalized Difference Vegetation Index (NDVI) product on Loess Plateau, while Liu et al [13] studied the vegetation variation through Global Inventory Modeling and Mapping Studies (GIMMS) NDVI3g product and Theil-Sen median analysis in Xinjiang
The tabulate area (TA) method applied in this study revealed the relationship between grassland cover changes and terrain factors, and the Machine Learning (ML) algorithms of Classification and Regression Trees (CART), Random Forest (RF), BAYE, and Support Vector Machine (SVM) provided another way to indicate the properties of the relationship
Summary
The Three-River Headwaters Region (TRHR) is a typical study area with high ecology values [1,2,3]. The main vegetation cover type in TRHR is grassland, which conserves water resource, preserves soil, and grazes livestock [4,5,6]. The grassland cover is traditionally studied by in-situ measurement, but it is inefficient in grassland cover change monitoring due to a large area of grassland [7,8]. The remote sensing images, product indicators, and methods were much more effective in researches of grassland cover change [9,10]. Si et al [16] validated the quantity and quality of grassland through Leaf Area Index (LAI) and Canopy Chlorophyll Content (CCC) indicators from Medium Resolution Imaging Spectrometer (MERIS) over Groningen and Friesland provinces in Netherland
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
