Research Of Land Use Classification Based On Multi-source And Multi-temporal Remote Sensing Data In Guanzhong Area

To obtain the land use information accurately in time could provide important basis for the management of land resources and the protection activities of the ecological environment,and could be used to calculate the water requirement of agricultural irrigation,which is of great significance to the management of water resources and the request of agriculture informatization.In order to obtain the land use information in Guanzhong area,explore the factor influencing the accuracy of land use classification,based on MOD13Q1 data,the combined method of Iterative Self-organizing Data Analysis Techniques Algorithm(ISODATA)with Classification and Regression Decision Algorithm(CART)was adopted to classify the land use in four vegetation indexes cases.The classification results were evaluated in the aspects of spatial precision and quantitative accuracy.The differences were compared between the results of land use classification based on different vegetation indexes time series to obtain the best vegetation index case with highest accuracy.The Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model(ESTARFM)was used to realize the fusion of blue,red and NIR band of Landsat 8 data and MODIS data to obtain fusion images with high temporal and spatial resolution.The quality of fusion data was assessed from both spectrum and classification results.In order to improve the classification accuracy of the fusion remote sensing data further,DEM and slope data were considered as the basis for land use classification as well.And the results of different classification cases were compared to explore the factors that affect the classification accuracy of land use in Guanzhong area.The main results of this study were shown as follows:(1)Land use classification results with the combined method of ISODATA,an unsupervised classification with CART algorithm,a supervised classification,were in line with the reality of Guanzhong.The overall accuracy and Kappa coefficient of the classification results derived were over 96% and 0.94 respectively,and the extraction accuracy of cultivated land and orchard could reach 89.36% and 65.47% respectively in four vegetation indexes cases.And this method was suitable for the land use classification in Guanzhong area,which was more accurate for classification than single CART algorithm.(2)The land use classification results varied with the different vegetation indexes.The results based on combined vegetation indexes were more accurate than that based on single vegetation indexes.The precision was higher when the EVI time series was prior.And EVI+NDVI time series was the best vegetation indexes case for classification.(3)The ESTARFM algorithm could realize the fusion of MODIS data and Landsat 8 data.The fusion images had temporal resolution of MODIS data and spatial resolution of Landsat 8 data.The correlation coefficients were all above 0.8 between the predicted reflectance of fusion data and the reference one of Landsat 8 data for blue,red,and NIR bands.The the fused images were of high quality.Compared with the original MODIS data,the spatial precision of land use classification has been improved to some extent based on the fusion images.The extraction accuracy of cultivated land was improved by 5.19%.Due to the “pepper and salt phenomenon” probably,the extraction accuracy of orchard did not increase yet.(4)The land use information based on multi-source data of fusion data from MODIS and Landsat 8,DEM,and slope information alleviated the “Salt and pepper phenomenon” of fusion remote sensing data effectively.And the overall accuracy and Kappa coefficient increased up to 97.60% and 0.9700 respectively,and the extraction accuracy of cultivated land and orchard reached 99.53% and 72.17% respectively.(5)The factors,influencing the accuracy of land use classification,mainly included: the system error and quality of remote sensing data,temporal and spatial resolution of remote sensing data,spatial and temporal fusion algorithm,and parameters used for the classification,DEM,slope,distribution and other spatial features.The land use classification results with higher accuracy are obtained based on multi-source data of MODIS,Landsat8 time series images,DEM and slope,and the indicators were explored influencing land use classification precision.Nevertheless,there are some deficiencies restricted by time and personal ability in this study: the work of data processing is heavy because a few of images could cover the study area limited by the breadth of Landsat 8,and the influencing factors were analyzed qualitatively only.Two issues should be addressed in future:(1)the batch software is suggested to be developed to simplify the processing work and reduce the workload;(2)the factors affecting the accuracy of land use classification need to be studied further by calculating the landscape metrics.