Research On Information Extraction And Spatial Sampling Methods Of Accuracy Assessment For Woodlands

Woodlands are a key component of natural resources and play an important role in global climate change mitigation,the maintenance of ecological balance,and improvements in the ecological environment.Information detection,target recognition,investigation and evaluation of woodland resources at different scales are achieved based on remote sensing data because of its spatial and temporal features.Frequent human activities exacerbate the transformations between woodlands and other land use and land cover types.Therefore,it is the most fundamental and key to acquire reliable remote sensing classification map.According to the new demands for remote sensing mapping of woodland resources,this study takes Beijing as an example,and carries out the researches on information extraction and spatial sampling methods of accuracy assessment for woodlands based on the comprehensive utilizations of MODIS,Landsat TM and all kinds of remote sensing data products.Breakthroughs of extraction method for remote sensing information of woodlands and spatial sampling layout optimization method for accuracy assessment of remote sensing classification are adopted to solve the information acquisition and the reliability problem of accuracy assessment for regional remote sensing mapping of woodland resources,respectively.The main research results in this study are as follows:(1)Information extraction of woodlands based on fractal features.This study presents an extraction method for woodland areas based on the fractal features of MODIS NDVI time series data.The research results show that fractals can reveal clear separations of different targets at different scales.The overall accuracy,kappa coefficient,and error coefficient of the extracted results of woodlands for Beijing were 90.54%,0.74,and 8.17%,respectively.Compared with the extracted results for woodlands based only on the MODIS NDVI time series,the average error coefficient decreased from 30.2%to 7.38%based on the fractal features of the MODIS NDVI time series.The information extraction method developed in this study can accurately and effectively extract the information of woodlands.(2)Multidimensional sampling space construction and spatial layout optimization of sampling sites.This study proposes a construction method for multidimensional sampling space of the fusion of feature space with auxiliary variable and geographical space.The spatial stratification and sample types division were adopted for sampling optimization in feature criterion space.The spatial simulated annealing and minimization of the mean of shortest distances as an optimization objective function were used for sampling optimization in geographical space.Finally,the optimizated results were integrated into a spatial sampling layout optimization method for accuracy assessment of remote sensing classification.The application case in Beijing showed that the sampling optimization method in this study performed well in the representative and balancing in feature and geographical space and ensured the accuracy and reliability for accuracy assessment of remote sensing classification.(3)The comparison of different spatial sampling layout optimization methods for accuracy assessment of remote sensing classification.This study set up comparative experiments of different spatial sampling layout optimization methods.The results show that the overall accuracy,root mean square error and the standard deviation of stratified even sampling method developed in this study were 71.36%～73.91%,13.46%and 0.96%,and this method was much better than the spatial even sampling,stratified random sampling and simple random sampling methods.Accuracy assessment of the fractal extraction results of woodland information in Beijing based on stratified even sampling method was achieved,and the overall accuracy,relative accuracy,root mean square error and standard deviation were 78.67%～81.76%,86.89%～90.30%,10.66%and 1.16%,respectively.Comparison experiments and application case results also confirmed that stratified even sampling method developed in this study was very accurate and effective for accuracy assessment.The optimized design scheme has better representativeness and balance in sampling feature space and geographical space.(4)Information extraction method for woodlands and spatial sampling layout optimization method for accuracy assessment developed in this study are being applied to the land use and land cover change detection of Beijing-Tianjin-Hebei region,and the evaluation of major forestry and ecological government projects in China.The acquisition and reliable accuracy of regional land use and land cover remote sensing classification maps are achieved and adopted for supporting the government decision-making and macro management for the monitoring and supervision of natural resources.Good application results have been achieved in this study.