| Desertification is a global ecological problem,which seriously reduces the utilization efficiency of land resources.Fractional Vegetation Coverage(FVC)can reflect the growth status of plants on the soil surface and indirectly reveal the degree of desertification.Therefore,it is effective to measure the regional desertification degree by estimating the fractional vegetation coverage of desertification area.At present,the estimation of vegetation coverage in desertification areas is mainly realized by remote sensing inversion.However,due to the fragmentation of vegetation and the strong soil background in the desertification area,there are generally mixed pixels when using medium and low-scale images.In addition,there are technical bottlenecks such as high cost and large labor,so it is necessary to explore more accurate remote sensing inversion methods.In this paper,Beijing Tianjin sandstorm source ecological control project area is taken as the research area.The purpose of this paper is to construct a vegetation coverage inversion system based on MODIS data with low spatial resolution,supplemented by FVC data measured on the ground and meteorological data,and with the comparison of various models.The main purpose of this study is to use MODIS data,combined with the measured vegetation coverage of 160 sample plots of 500 m × 500 m in size of stratified random sampling survey,supplemented by meteorological data to extract characteristic variables,and use linear stepwise regression method to screen characteristic variables.In order to obtain the optimal model of vegetation coverage inversion,pixel dichotomy model,multivariable linear regression(MLR),random forest model(RF),k-nearest neighbors model(kNN),support vector machine(SVM)and gradient boosting regression tree model(GBRT)were constructed.The accuracy of six models was compared and analyzed,and the method of vegetation coverage inversion suitable for desertification area was discussed.On this basis,the temporal variation trend and spatial distribution characteristics of vegetation coverage in Beijing Tianjin sandstorm source area were analyzed qualitatively and quantitatively.After research,the following conclusions are obtained:(1)Among the six inversion methods,the inversion effect of random forest model is significantly better than other models.Among all the inversion models,the R2 of random forest model is 0.74(RMSE=0.11,rRMSE=18.36%).The accuracy of pixel dichotomy model is the lowest,R2 is only 0.53(RMSE=0.19,rRMSE=25.70%).The coefficients of determination of MLR,KNN and SVM are all about 0.70,but they all produce unreasonable estimates.The spatial distribution of vegetation coverage of random forest model is basically consistent with the actual situation,and the mapping effect is the best among all models.In conclusion,the random forest model can be used to retrieve the vegetation coverage in Beijing Tianjin sandstorm source area.(2)From 2000 to 2016,the vegetation coverage of Beijing Tianjin sandstorm source area showed an increasing trend.Based on the random forest method,the vegetation coverage of each year in the study area was obtained.After the statistics of the average vegetation coverage of each year,it was found that the vegetation coverage of the study area increased in general from 2000 to 2016.From the overall distribution characteristics,the spatial distribution of vegetation coverage presents a decreasing trend from east to west and from south to north.In recent 17 years,the vegetation coverage area of grade ?(FVC?0.7)and grade ?(0.5?FVC<0.7)increased significantly,while the vegetation coverage area of grade ?(0.3?FVC<0.5),grade ?(0.15?FVC<0.3)and grade V(FVC<0.15)decreased gradually.(3)From 2000 to 2016,the fractional vegetation cover in Beijing Tianjin sandstorm source area kept a good momentum of steady growth.By analyzing the change trend of annual average vegetation coverage in Beijing Tianjin sandstorm source area,it is concluded that 70.04%of the regional vegetation coverage value is basically unchanged from 2000 to 2016,the improved area accounts for 26.19%,the degraded area accounts for 3.77%,and the significantly improved and significantly degraded area accounts for 12.58%and 0.69%respectively.The area of vegetation improvement is obviously more than that of degradation,and the overall situation is good.(4)The vegetation coverage in Beijing Tianjin sandstorm source area is highly concentrated and shows strong regional characteristics.The results of spatial autocorrelation analysis showed that the vegetation cover change characteristics in Beijing Tianjin sandstorm source area showed significant spatial positive correlation,and the aggregation state was "high-high gathered" and "low-low gathered"... |
PDF Full Text Request