Study On Improving The Accuracy Of Globeland30 Data By Coupling Geo-statistics And Eco-geographical Regions

Land cover is the most commonly used indicator to represent human or natural processes,and remote sensing is the only effective means to obtain land cover data in a wide range.Due to spectral confusion,image resolution limitation and the complexity of the features themselves,the LULC products obtained from remote sensing image classification must contain a large number of misclassified or uncertain pixels.Classification accuracy is generally not high,secondary classification is more difficult to obtain reliable results and local accuracy evaluation system is missing.Therefore,quantitative assessment of the distribution of precision and correction of classification errors are of great significance to the study of global change,the census of geographical conditions,social and environmental planning,and the management of ecological resources.At present,the commonly used methods to improve the classification accuracy are mainly divided into two categories: one is the use of geological knowledge rules,the other is the application of geo-statistics.The commonly used confusion matrix for evaluating the classification accuracy can only give the overall accuracy,and cannot reflect the spatial variation of the classification accuracy.The information provided to users of land cover products is incomplete and uncertain,which may lead to poor application effect.Therefore,it is urgent and necessary to quantify the uncertainty of classification on the basis of the ground cover classification obtained by the conventional classification algorithm,and how to effectively evaluate the variation of LULC product precision with space and improve LULC accuracy.In this paper,a method to improve the precision of land cover classification products by coupling ecological geographical zoning and Markov chain geosciences statistical simulation is proposed.Firstly,verification points from various channels are collected from the network,and some verification points are interpreted manually to form a sample data set.The surface cover classification products that need to be improved in accuracy are used as auxiliary data(Globeland30 data is selected in this study),and they are used as auxiliary data together with the expert knowledge of ecological geography division for collaborative simulation.Secondly,the sample data set and auxiliary data set generate the transfer probability graph model for Markov chain co-simulation.Finally,the accuracy of the simulation results is verified.The results show that the accuracy of Globeland30 data can be improved by nearly 10% by using the method of coupling ecological geographical partition and Markov chain geo-statistical simulation.