A Research Of Land Cover Monitoring In The Northeast Of China With FY-3 MERSI Data

The northeastern region of China is populated sparsely,and the land is so fertile that it is the major grain production area of corn,rice,etc.In recent years,deforestation,farmland increasing and water resources reducing lead to a large change on land cover in part of the Northeast.It is significant to obtain the changes of land cover for policy formulation in fields of agricultural production,urban planning,environmental protection.Taking the Sanjiang Plain in Heilongjiang Province as an example,adopting the FY3 MERSI data combining with crop phenology information and field sampling data to study and compare different land cover classification algorithms.Then utilizing random forest classification algorithm which has better classification accuracy to make land cover classification of the whole northeast region of China,and get northeast area land cover classification figures of different years from 2009 to 2016.In order to realize the monitoring of land cover changes in the northeast of China,the main research work is as follows:(1)Selecting land cover classification algorithm in the Sanjiang Plain.Firstly,selecting high quality FY3 MERSI data in growing season and harvest time for different crops on the basis of crop phenology information,and then going to the Sanjiang plain for field data sampling work.Secondly,making classification for the data of the 2016-year growing season in the Sanjiang Plain by using random forest,decision tree,neural network and support vector machine(SVM).The result shows that the classification accuracy on the basis of random forest classification algorithm is between 88% and 93%,which has an obvious advantage on classification accuracy over other classification algorithms.At last,employing majority voting method to fuse land cover classification results of 2016.The overall accuracy is 89.87%,and the Kappa coefficient is 0.85.(2)Researching on land cover monitoring in the northeast of China.Establishing FY3 satellite 250 meters' resolution MERSI data set in the northeast of years from 2009 to 2016 and preprocessing the data set in batches.The random forest classification algorithm is used to make land cover classification of the northeastern region.The land cover classification maps of the northeast are obtained in years of 2016,2015,2014,2013,2012,2011,2010 and 2009.The overall classification accuracy of land cover classification maps is between 80% and 88%.Using the annual land cover classification map to monitor the land cover changes in the northeastern region.Since the year 2010,the agricultural land has increased and the cover type has changed from tree to corn in the western part of Heilongjiang province and northern part Jilin province.Since the year 2013,building land has increased obviously,especially in Liaoning Province.The land cover in the rice planting areas of the Sanjiang Plain and Songnen Plain had a steady growth,the vegetation coverage of the original forest in the Xing'anling and Changbai Mountain did not change obviously.(3)Designing and implementing land cover monitoring system.Through the analysis of the whole classification process and principle,on the basis of the above research results,the land cover monitoring system is designed and developed by VS platform for the quick and easy FY data classification in batches.The classification system covers data preprocessing model,data classification and fusion model,classification result verification model and result mapping model,etc.