Research For Informations To Be Extracted Form Dongting Lake Wetland Based On Decision Tree

The Dongting wetland is located in the Yangtze River. It's a great significance of theecological balance and the regional economic sustainable development in this area. However,because of the change of natural environment,long-term sediment siltation and humanactivities increasing,the Dongting Lake wetland has been destroyed seriously. The wetlandarea decreased sharply in Dongting Lake area. Human activities and the river water sandregime is restricting the lake evolution,reclamation from lakes and lake sediment depositionaccelerated decline of Dongting Lake. In open water area reduces,bottomland area graduallyincreased,the Dongting Lake wetland structure and function is undergoing tremendouschanges.In this study based on different resolution remote sensing image,using the remote sensingdata of spectrum and texture features by combination with other auxiliary data we extracteDongting Lake wetland information. At the same time according to the multiple image dataclassification results,it reveales the evolution of wetland.The result showes that:(i)By using flood control dam on Dongting Lake wetland zoning,combined with theknowledge of GIS spatial analysis function,studying of Dongting Lake wetland distributioncharacteristics in winter and summer,considering the phase different,sufficient mining data,a decision tree is constructed to obtained achieve the study area wetland types. Than wecompared classification results to the traditional maximum likelihood supervised classification.Which shows: The method using of knowledge of decision tree classification to classify typesof wetlands, increased by12.05%to the traditional maximum likelihood supervisedclassification overall accuracy; overall kappa coefficient was increased by0.1407; Such aswoodland, reed beaches, mudflats, water coverage types' producer accuracy and user accuracywere greatly improved outside the Lake area. (ii) SPOT-5high resolution images is used to classification land coverage type.inDongting lake wetland. We selected panchromatic as texture features to calculate data source.Through the various J-M distance for the selected sample of wetland types we determined thebest texture scale.,than we used QUEST algorithm for remote sensing image spectral,textureinformation form data sets ofr data mining to constructe the decision tree model for highresolution image classification.The results show that by selecting the optimal texture scale combination,using a decisiontree on the spectral data and multi scale texture data for high resolution remote sensing imagehaving a high classification accuracy of78.57%. The spectroscopic data classification andcombining single scale texture data classification accuracy were71.98%and76.76%. It showsthat texture information can effectively improved the wetland object recognition level,at thesame time multiscale texture can better described the features of texture features and moreeffective to solved the classification results of the foreign body in the same spectrumphenomenon. It helped to improve the accuracy and efficiency of high resolution remotesensing image classification.(iii) Water,mudflat,carex,reed,forest land and paddy field are the most important typesof land coverage in the study area. Water,carex area at the selected time span decreases first,then tends to be stable; Forest land to explosive increase,then tends to be stable.From1987to1996,in the study area the land coverage types have a acute change. Themain transformation is happened between mudflat,carex,reed and water to forest land. From1987to2004forest land area continues to increase,at the same time the mudflat continued todecline. In the years2004to2009interval, the forest land outside the Dongting lake dam areais decline,but not obvious. It has become a major ground cover types in the Dongting Lakearea especially in Western Dongting Lake area.(iv) The study area's overall landscape diversity and heterogeneity changed little. Thesuperior class proportion appears first decrease and then increase. Landscape types ofagglomeration degree increases,the degree of dispersion is reduced. Land coverage types in the study area, plaques in overall plaque tended to thefragmentation and miniaturization trend from1987to1996. Form1996to2009it shows asplaque tends to polymerization tendency. The study area to class plaque density changesoverall appears first increased and then decreased. Patch fragmentation peaked in1996to2004,which showed it has a great degree improvement and landscape tends to be integrity.Patch boundary density in1996to2009has a overall decline. The study area shape tends to beregular which has a more intense human disturbance. The class of the study area presents aaggregation degree increasing trend.