Remote Sensing Monitoring Of Cropland Change For The Recently 10 Years In Yunnan Based On GEE Platform

Cropland extraction and monitoring is the most important part of ensuring food security.At present,time-sensitive remote sensing images become an important means for cropland extraction and monitoring.However,due to space-time limit,the remote sensing classification based on local host faces great challenges in cropland extraction with large region and long time series.Yunnan province,particularly,has more difficulty in cropland extraction and monitoring because of the cloudy and rainy weather,complicated terrain,fragmented and diverse cropland.These problems make Yunnan province lack of of long time series cropland data seriously.With the development of remote sensing cloud service technology,it is possible to process and analyze remote sensing data in large region through cloud platforms.Google Earth Engine(GEE)is a cloud computing platform for remote sensing images on a global scale.With huge storage space and advanced cloud computing capabilities,it has advantages in extraction and monitoring of remote sensing information in large areas and long time series.Thus kindly,based on Landsat and Sentinel-2 time series data,this paper use GEE cloud platform to carry out remote sensing monitoring research on cropland change in Yunnan province.Above all,multi-season images data were used to obtain four cloud-free composite images of 2009,2012,2015 and 2018.On this basis,the time series data were constructed to obtain phenological feature,and the texture feature,tasseled cap trasformation feature,terrain feature and spectral feature are also calculated for feature selection by J-M distance.Once more,three classification methods were used to extract cropland,namely Random Forest,Support Vector Machine and K-Means algorithm.Then,the hybrid dynamic detection method based on texture feature was used to detect cropland cover change information of 2009,2012,2015 and 2018.Finally,a visualization platform of Yunnan cropland data was realized on GEE cloud platform.This study contains the following contributions:(i)semi-automatic sampling method is proposed.This method based on multi-period land cover products,solves the time consuming and energy consuming difficulty of sample data collection in large areas.(ii)The NDVI,EVI and MSAVI time series were constructed to analyze the cropland phenological feature,and the cropland extraction accuracy is improved effectively by combining the texture feature,tasseled cap trasformation feature,terrain feature and spectral feature.(iii)In addition,a hybrid dynamic detection method based on texture feature is proposed,which uses texture feature pixel ratio in different phases to make up for the short-comings of Post Classification Comparison method.The results show that :(1)The time series of ternary vegetation index can obtain phenological feature more accurate than that of single vegetation index.(2)Multiple optimized features combinations have higher accuracy than single feature classification.(3)Random Forest classification has high accuracy.Based on the method,the average overall accuracy(OA),average user accuracy(UA),average producer accuracy(PA)and average Kappa coefficient reaches 0.974,0.940,0.945,0.926 respectively.(4)In 2009,2012,2015 and 2018,the total area of cropland in Yunnan province remain stable basically,reaching 5.94387 million ha,5.80882 million ha,6.19480 million ha and 6.30400 million ha respectively.The cropland is mainly distributed in Qujing city,Honghe hani yi autonomous prefecture,Wenshan zhuang and miao au-tonomous prefecture,Zhaotong city in eastern and Kunming city in central Yunnan.(5)From 2009 to 2018,cropland in Yunnan province shows a trend of small growth.Cropland remained stable basically from 2009 to 2012,but increased by 148.77 million hectares from 2012 to 2015 and 114,700 hectares from 2015 to 2018 respectively.