Study On Wetland Information Extraction From Multi-sourced Remote Sensing Imagery And Its Scale Effects Using OBIA Method

Zhongwei City, Ningxia Hui Autonomous Region is located in the northwest area of arid and semi-arid climate, Which is an important wetland distribution and play an important role on the maintenance of ecological balance and human aspects of production and life. With the social development, urban construction speed up the process, human strengthen the transformation and utilization of wetlands. This results in a substantial shrinking wetland, biodiversity loss, ecological functions of wetlands declining, scientifically rational use and protection of wetlands have become increasingly prominent. Because of wetland complex geographical environment and accessible poor, remote sensing data are able to obtain a large area of wetland observation information and detect dynamic changes of wetland timely, which is widely used in the wetland investigation work.In this paper, based on the experimental data of five kinds of spatial resolution images which are 0.5m Worldview-2 converged data, 2m Worldview-2 multispectral data, 8m GF-1 multispectral data, 15 m Landsat OLI converged data and 30 m Landsat OLI multispectral data, the author integrated data mining the PART classification algorithms in Object Based Image Analysis(OBIA) to build the wetland information extraction classification rules set for wetland classification, and discusses the spatial scale effect problem of wetland information of remote sensing images in the study area. Main conclusions are as follows:The classification rule set models built by the integrated method of OBIA and PART classification algorithms have good accuracy and interpretation, screening out more features for image classification, achieving a set of classification rules establish automated, avoiding manual classification of subjectivity, simulating and completing the expert knowledge base, improving the efficiency and accuracy of classification.With the increase of remote sensing image spatial scale, the complexity of classification rules set models built by the integrated method of OBIA and PART classification algorithms show a decreasing trend. Because of more prominent objects features in high-resolution images, there are more texture and shape features in the classification rule set models for wetland extraction.With the increase of remote sensing image spatial scale, the smallest class polygon area can be identified in images becomes larger, the ability to judge category weakened, wetlands extraction accuracy reduced. In all the image classification results, classification accuracy of 30 m Landsat OLI image is lowest, whose overall accuracy is 84.36% and Kappa coefficient is 0.77.The spatial pattern of different wetland types are consistent in the macro in five kinds of spatial resolution images in study area, but different spatial pattern index have a big difference, which mainly correlate with its own plaque shape, size and spatial distribution. With the decease of remote sensing image spatial resolution, average patch area, minimum and maximum patch area and the average length of patch boundary of four kinds of wetlands including lakes, ponds, reed and paddy land presented an increased trend. on the contrary, the patch numbers, average fractal dimension, average patch fragmentation and shape index showed a decreasing trend. The average fractal dimension, average patch fragmentation and shape index had a large range of values change, indicating four wetlands types in the study area are more sensitive to the spatial resolution of images.