Knowledge Network Node

Estimating Above-ground Biomass of Grassland in Drylands of China at 30m Scale Based on Google Earth EngineCN

修晓敏

贵州师范大学

Abstract:Drylands in China is very vast,with 47%of the total land area and a large amount of population.The environment which people depend on living is extremely fragile due to its unique and complicated structure.In recent years,global climate has become warming,natural disasters have occurred frequently,ecological environment has become worse and worse,which has severe affected the quality of people’s lives.Grassland regarded as a renewable resource in drylands is of great significance for regulating climatic conditions,preventing wind and fixing sand,conserving water sources,maintaining ecological balance and increasing carbon sinks.The biomass of grassland is a good indicator for monitoring grassland growth status and productivity,and is an important index for evaluating grassland ecosystems.Therefore,accurate monitoring of the biomass of grassland in drylands of China plays a crucial role in ecological security,rational management and utilization.This article used China’s drylands as the study area,estabulished remote sensing estimated models of the above-ground biomass of grassland by applying CART(Classification and Regression Tree)and SVM(Support Vector Machine)two machine-learning methods based on Google Earth Engine cloud computing platform,with Landsat8 as the main datasets,SRTM(Shuttle Radar Topography Mission)and TRMM(Tropical Rainfall Measuring Mission)as auxiliary data.Firstly,the pre-processing work of the Landsat8,SRTM and TRMM data in the study area was completed on the Google Earth Engine platform.Secondly,according to the field samples’lantitude and longitude,extracted three categories 15 remote sensing feature information from the remote sensing image.Meanwhile,the correlation relationship between remote sensing feature information and the above-ground biomass of grassland was analyzed.Then,the study constructed remote sensing estimation models of above-ground biomass of grassland under free combination of three types of remote sensing feature information based on CART and SVM machine learning,and built one dimensional regression model based on vegetation index which is compared with the machine learning model.What’s more,through the verification analysis of the model’s modeling accuracy and predictive ability,the most suitable remote sensing estimation model of above-ground biomass of grassland is obtained.Finally,the distribution of above-ground biomass of grassland in this study area is inverted.The main conclusions of the study are as follows:(1)Research on the Google Earth Engine platform could extract the band spectrum(B2-B7),vegetation index(NDVI,RVI,DVI and ARVI)and DEM and meteorology(Elevation,Slope,Aspect,Hillshade,Precipitation)remote sensing feature information from Landsat8,SRTM and TRMM.After correlation analysis between remote sensing feature information and the above-ground measured biomass of grassland,the results showed that 15 remote sensing feature information and biomass measured data did not have strong correlation relationship,with correlation coefficient of-0.4~0.4.Due to the wide distribution of research areas and the different ecological conditions of field collection sites,the study freely combined three types of remote sensing feature information into band spectrum,vegetation index and DEM and meteorological,band spectrum and vegetation index,and vegetation index and DEM and meteorology,which can explore the biomass of grassland remote sensing estimation model with different information sets.(2)The study combined three types of remote sensing feature information with the above-ground biomass of grassland,and used cross-validation method to construct the remote sensing estimation model based on CART and SVM.The results showed that the accuracy of model based on CART and SVM machine learning method was higher than that only used one vegetation index to establish regression model.The judgement coefficient R~2 of CART,SVM and one demonsion regression model were 0.78~0.83,0.48~0.51 and 0.6,respectively.The predictive ability of models was verified by 10%of the reserved sample data.The analysis showed that the predictive ability based on the CART and SVM machine learning method model was better than regression model.Among these,the CART model and the SVM model had an average increase of 0.22 and 0.15 with respect to the ARVI index model,and the root mean square error RMSE(Root Mean Square Error)had increased by 27.20%and 5.83%,respectively.(3)Comprehensively analyzed modeling accuracy and predictive ability of the seven models,model established with all three types of remote sensing feature information had the best effect,which was better than the other models.Although the correlation relationship between 15 remote sensing feature information and above-ground biomass of grassland was not obvious,there had inherent links between them.The role of any information can not be ignored.(4)The three types of remote sensing feature information were used to invert above-ground biomass of grassland in drylands of China based on CART machine learning.The above-ground biomass of grassland in study areas located in 6 grades:<100g/m~2,100~500g/m~2,500~1000g/m~2,1000~1500g/m~2,1500~2000g/m~2 and>2000g/m~2.(5)The Google Earth Engine platform demonstrates powerful capabilities in remote sensing image processing,analyzing and computing.Remote sensing data is available to users on the GEE platform.Users can access large-area remote sensing image data quickly and complete their cloud computing tasks efficiently.The efficiency of work improved obviously.The GEE platform has formidable function which can provide a new approach to exploring remote sensing estimation of grassland biomass.
  • Series:

    (D) Agriculture

  • Subject:

    Animal Husbandry and Veterinary

  • DOI:

    10.27048/d.cnki.ggzsu.2019.000288

  • Classification Code:

    S812

Tutor:

杨广斌; 李晓松;

Download the mobile appuse the app to scan this coderead the article.

Download:223 Page:68 Size:2397K

Related Literature
  • Similar Article
  • Reader Recommendationr
  • Related Funding Articles
  • Citation Network
  • Study Results