Desertification Information Extraction Based On Google Earth Engine And UAV Images

At present,desertification is one of the most serious environmental,social and economic problems that have perplexed human society all around the globe.Threatened by the desertification for a long time,governments at all levels in the north of China have carried out a series of projects for ecological protection and desertification prevention and control in recent years,keeping track of the implementation effect of the project in time,adjusting the relevant policy direction,and grasping the development and change of desertification quickly and accurately.Remote sensing technology plays an irreplaceable role in desertification monitoring,but now,there are still many problems,such as heavy workload of field data collection,limited accuracy of monitoring methods,chaotic index system and too much information cross redundancy between indicators,and difficulty in localization.Taking Zhenglan Banner,which is located in the hinterland of Hunshandak Sandy Land,as the research object and based on the UAV image obtained in the field,this paper studies how to quickly obtain the area ratio of mobile sandy land on the scale of satellite image.The Landsat5 and Landsat8 data were invoked in Google Earth Engine,and the desertification land in year of 2000,2004,2010,2015 and 2019 was extracted by mixed pixel decomposition method.The desertification comprehensive index was constructed to evaluate the dynamic change of desertification land in Zhenglan Banner.Finally,the driving force of desertification was analyzed.The main contents and conclusions are as follows:(1)Three classifiers,random forest,extreme gradient elevation and multi-layer perceptron,are selected to classify the desertified land into mobile and immobile desertified land.The results show that the accuracy of the three classifiers is similar,and that the overall accuracy of random forest is the highest,reaching 97.6%.Using the results of random forest classification to resample in Gee,the area ratio of mobile desertified land in the corresponding Landsat pixel is obtained,which is used for the accuracy verification in the following paper.(2)In order to eliminate the influence of bare soil on the extraction of desertified land in the study area,the linear spectral mixture model is used to extract the desertified land in gee.The model accuracy is tested based on the area ratio sample of mobile sand acquired by UAV,and the results show that R2 is 0.7648,RMSE is 0.1383,which indicates that the model accuracy meets the demand.According to the sample points obtained in the field,it is found that when the area ratio of quicksand is more than 18% and the vegetation coverage is less than 80%,the desertification land can be accurately extracted.From 2000 to 2019,the area of desertified land showed a downward trend,and only picked up from 2004 to 2010.In addition,based on Gee,an app for monitoring desertification land in Zhenglan Banner is designed,where users can observe the spatial change of desertification land in this area since 1984.(3)Towards calculating 14 desertification related indexes in Gee,NDVI,Albedo,TGDI,and wetness were selected through feature selection and correlation analysis,and the comprehensive desertification index(DI)was established by using Analytic Hierarchy Process(AHP).Based on DI and desertification land area,this paper has analyzed the overall situation of desertification from 2000 to 2019,the results show that the desertification in Zhenglan Banner has a significant reversal trend during the past 20 years.It is found that 2000 is the most serious year of desertification,2019 the least one,but the desertification has developed between 2004 and 2010.In addition,using DI to classify the desertified land,the results show that from 2000 to 2019,the area of non desertified land has been increasing at the rate of 2.2% per year,the area of light desertified land has been decreasing at the rate of 0.4% per year,the area of moderate desertified land decreasing at the rate of 2.7% per year,and the area of severe desertified land decreasing at the rate of 3.4% per year.In terms of spatial dynamic change characteristics,the total proportion of reversals and obvious reversals has reached about 37% from 2000 to 2019,and the total proportion of development and serious development is only about 5.4%,which shows that the desertification degree of Zhenglan Banner has been reduced and the ecological situation has been improved significantly in the past 20 years.(4)According to the residual trend analysis of the spatial effects of human activities and climate change in Zhenglan Banner from 2000 to 2017,the results indicates that 59.1% of the desertification areas are not significantly affected by human activities,20.0% are significantly positively affected by human activities,and 21.0% are significantly negatively affected by human activities.The results of factor analysis show that the process of desertification in the study area during the past 20 years has been affected both by meteorological and social factors,on the one hand,the total population and the stock of livestock at the end of the year have a greater impact in social factors,and on the other,the average temperature and annual precipitation in meteorological factors have a significant impact.