Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Chine → 2019 → The Estimation of Shrub Coverage Based on Google Earth Engine in Four Mega-Sandy Lands in Northern China

贵州师范大学 (2019)

The Estimation of Shrub Coverage Based on Google Earth Engine in Four Mega-Sandy Lands in Northern China


Titre : The Estimation of Shrub Coverage Based on Google Earth Engine in Four Mega-Sandy Lands in Northern China

Auteur : 陈黔;

Grade : Master’s Theses 2019

Université : 贵州师范大学

The northern of China was a region with severe sand damage and serious soil erosion in the past.In the 1980 s,with the several large-scale forestry ecological restoration projects such as the Three-North Shelterbelt Project and the Beijing-Tianjin Sand Source Region carried out,the vegetation cover in the arid and semi-arid regions of northern China has undergone tremendous changes.The four-mega sandy lands in the northern farming-pastoral ecotone are extremely fragile and respond severely to climate change.Shrub is the dominant vegetation type in this region,plays an important role in sand control,food/timber product provision,and on.Mastering the shrub coverage in this area is of great significance for vegetation growth assessment and desertification monitoring.The lack of remote sensing products of shrub coverage in large-scale,medium-and high-resolution arid regions,a method for estimating shrub coverage in sandy areas is proposed.That is imminent.In view of this,this study selected the four mega-sandy lands as research areas,used Landsat-8 surface reflectance products as data sources,combined with Collect Earth sample collectors and field survey samples,estimated the coverage of shrub vegetation in sandy areas in arid areas based on the Google Earth Engine(GEE)remote sensing cloud platform.Firstly,the principle and sampling method of the Collect Earth sample collector are explained in depth,and the data processing flow and related remote sensing feature parameters are extracted through the GEE platform.Then,combined with the study,the three systems learning algorithms introduced systematically which is classification and regression tree(CART),random forest(RF)and support vector machine(SVM).In the same time compared the advantages and disadvantages about them.Finally,constructed three machine-learning models and verified them estimation accuracy.Research indications are followings :(1)Collect Earth is a free and open source software for land monitoring.It combines with multiple high-altitude resolution map archives,and proposes "augmented visual interpretation" to improve sample collection accuracy.This method can help researchers to capture samples quickly under Very High-Resolution(VHR)images.In the arid area with strong surface heterogeneity and sparse vegetation,Collect Earth can effectively obtain the ground-shrub coverage sample data,which can effectively distinguish shrubs from tall trees and herbaceous vegetation for shrub cover.That lays the foundation for estimation of shrub coverage.(2)Based on the GEE platform,three machine learning models was constructed respectively.All three models show the ability to estimate the coverage of shrubs in four mega-sandy lands.Among them,the SVM model has the best estimation result,the Estimated Accuracy(EA)is 66.8%,the Root Mean Square Error(RMSE)is 7.04%,and the model determination coefficient(R2)is as high as 0.93.However,the shortcomings of the SVM model are also obvious.In the arid area of the vegetation with medium and low coverage,the super-plane of the model structure has an insufficient prediction ability for high-coverage areas,and there is a serious underestimation.The other two models have relatively poor prediction ability,and the estimation accuracy is about 54%.The estimated root mean square error of CART model and RF model are 9.86% and 9.75%,respectively,the determination coefficient is 0.72.(3)The GEE platform stores nearly 40 years of remote sensing data on the global scale,including Landsat,Sentinel,MODIS and other major international satellite remote sensing platform data and other remote sensing products.The entire data storage capacity reaches BP level.Through the Application Programming Interface(API)provided by the platform,various complex map operations can completed,and the instantaneous distributed parallel computing is efficient.The data calculation and analysis which takes several days or weeks to complete on the supercomputer can be completed in a few hours on the GEE platform.GEE is good in reducing the data calculation cost and the data redundancy.It is an indispensable platform foundation for remote sensing research at the national,intercontinental and global scales.Overall,this study proposes a large-scale,medium-high resolution,rapid estimation remote sensing method for shrub coverage,which is sensitive to sandy land in arid areas and is distinct for tall trees and low-lying herbs.The status quo of shrub coverage remote sensing production provides effective complement in medium-resolution arid regions

Mots clés : four mega-sandy lands; shrub coverage; Collect Earth; Google Earth Engine; machine learning;

Présentation (CNKI)

Page publiée le 25 mars 2020