Dynamics And Driving Forces Of Landscape Pattern Changes Based On Remotely Sensed Landsat Images Classification

In the process of urbanization development, rational use of land resources is an urgent problem to be resolved. Analyzing landscape pattern dynamics and corresponding driving forces is of great significance to local ecological protection and sustainable development.Taking Jinsha township of Nantong city as the case study, using Landsat 5 TM and Landsat 8 OLI remotely sensed imagery as the major data sources, determining landscape classification systems according to land uses was first implemented in the analysis followed by comparing the pixel-based classification and object-oriented elements of classification.The preprocessing procedures were implemented in ENVI5.1 environment. The supervised classification algorithms, including the maximum likelihood and neural networks classifiers were implemented, followed by a validation to assess classification performance. The multi-scale image segmentation was conducted by using eCognition8.7 package to specify the parameters and the optimal segmentation scales for different landscape types.Carrying the single-level object-oriented decision trees classifier and the multi-level object oriented decision trees classifier,followed by a validation to assess classification performance.Choose higher precision method to classify the 2009 remotely sensed images.Carrying the dynamic analyses of landscape patterns through Fragstats4.2.The forest landscape transformation was conducted by using ArcGis10.2,and computed the spatial interactions of forest and urban landscape, forest and agricultural landscape.In combination with Jinsha's yearbooks, driving forces contributing to landscape patterns change were qualitatively analyzedThe principal findings from the current work were summarized as follows:(1)According to the pixel-based classification,the maximum likelihood classifier had an overall accuracy of 85.25%, with a kappa coefficient of 0.7985, neural networks classifier obtained an overall accuracy of 88.67%, with a kappa coefficient of 0.8848.(2)The weights for seven multi-spectral bands in eCognition8.7 package were specified as 1.06:1:1.10:1.09:1.87:1.45:1.21 according to their corresponding DN values standard deviation, the shape factor as 0.2, the compactness factor as 0.5. In addition, the optimal segmentation scales for different landscape types were specified as follows: 70 for forest, 80 for water bodies, 100 for urban land uses and 70 for agricultural land uses. The single-level object-oriented decision trees classifier used a segmentation scale of 70,the validation showed its overall accuracy of 89.19%, with a kappa coefficient of 0.8493 while the multi-level object oriented decision trees classifier gained an overall accuracy of 93.59% and a kappa coefficient of 0.9107. In this study area, the object-oriented classification has the higher accuracy than pixel-based classification.The multi-level object oriented decision trees classifier has the highest accuracy,which using the optimal segmentation scales for different landscape types.Thus, the multi-level object oriented decision trees classifier was accordingly applied to classify the 2009 remotely sensed images.(3)Dynamic change analysis of landscape pattern. In class type level,From 2009 to 2014,the patch number and patch area of Jinsha's forest landscape showed growing trend,which is inextricably linked to the current ecological forest project in Tongzhou.Urban landscape contiguous in space and has the clearly higher dominance.The agricultural landscape is getting an accelerating fragmentation. The connectivity of forest landscape and urban landscape are increasing, suggesting the expanding of patches;in landscape level,the overall landscape fragmentation degree increased,the shape complexity of landscape patches was medium,diversity index decreased, mainly due to the accumulation of urban landscape,the spatial distribution of landscape types tended to be complicated.Agricultural land and water bodies landscapes became more decentralized so that the number of small patches increased, evenness decreased.(4)Analysis of forest landscape transformation.Based on the need of urban development,32.97% of forest landscape retained during the period 2009 to 2014,31.91% of the forest landscape transferred to the urban landscape,30.79% transferred to the agricultural landscape.(5)Spatial interactions of forest and urban landscape,forest and agricultural landscape. During 2009 to 2014, the spatial interaction of forest and urban landscape rose, suggesting their spatial connectivity got more closely, while the spatial interaction of forest and agricultural landscape reduced, indicating their separation degree increased.(6)Driving forces contributing to landscape patterns change were qualitatively analyzed in terms of demographic growth, technical development, economic boom,urbanization level,forestry policies and land price.The urban landscape area increased significantly due to the increasing population of study area,which means more land used for residential and consumer; with the new technology,Jinsha develops the modern agriculture, promoting agricultural landscape pattern change;the service industry accounted for increased significantly, promoting respective urban landscape developing; in the process of urbanization development,the township population concentrated,promoting changes in the surrounding landscape;forestry policies promote the development of forest; the pursuit of the land price makes more construction landscape.