Application Of Data Mining In Wetland Remote Sensing Mapping

Remote sensing has great advantages of obtaining information of ground objects conveniently compared with conventional methods.With the rapid development of remote sensing data sources regarding quantity and quality,higher expectation occurs.However,there are still issues which limit the application of remote sensing in wetlands.For example,satellite data with well-balanced spatial and temporal resolution can be hardly obtained.Large-scale updating of thematic map remains highly labor-intensive.Due to the fact that wetland is hard to be defined clearly and it possesses non-uniform values of texture or reflectivity,it is more difficult to acquire thorough information of wetland in comparison with other land-cover types.In this paper,we mainly focus on two issues.1).Slender rivers and the boundary of open water are all categorized as wetland for their shallow water.Nevertheless,given that they are too small in size,we prefer analysis on a large scale.Therefore,how can we extract such detailed information of water by keeping spatial and temporal resolution in balance? 2).Thematic map demands intensive manual labor and is therefore generally time-consuming.Can we take advantages of historical maps when the map updating is frequent and urgent? We aim to find solutions for these two questions.1).How to extract detailed water body from images with coarse spatial resolution? 2).How to make full use of thematic map in automatic updating.The main contents and conclusions are listed as follows.1.For the regions with water boundary as well as slender water,root causes of leaking classified points and sub-phenomenon mistakes is explored.Only “mixed pixels” will miss some information,which should be taken into consideration together with “adjacent effect”.2.We utilized a spatial template to locate the mixed pixels,which is also a terrestrial-aquatic transverse zone,performing adaptively and robustly.Without supervised information,this function performed better than the threshold method.3.A system is introduced for wetland updating with the assistance of historical thematic map,which has reduced subjective judgments and is more automatic.The method,IIR(Iterative interpretation and Reorganization),can obtain a more accurate classification result than that of the supervised method in various four study areas.Furthermore,IIR can detect wetland types which didn’t exist in historical classification systems.The main innovations are as follows: 1.We proposed a method can consider two main effects comprehensively.It has a more accurate location of mixed pixels as well as quantity of water body than that of supervised methods.2.IIR,which can utilize existing thematic map,has an overall accuracy from 70% to 90%,which is better than that of supervised method in the same study area.