| Remote sensing techniques offer timely, up-to-date, efficient and relatively accurateinformation for sustainable and effective management of wetland vegetation over a largearea. Taking the Sanjiang Plain in northeast China as an experimental area, quantitativeresearch and studies were made for wetlands classification, with multi-season LandsatTM/ETM+images as data sources, support of3S technology and three times field surveydata in2000,2006and2012. Multi-scale segmentation algorithm is applied to extractobjects with different segmentation scales. Characteristics such as variety of spatialspectral bands, phenological and seasonal aspect information and the object-orientedclassification method were used as well. And the use of remote sensing data, the paperanalyzes the research areas cover the space characteristics of the wetland resources of2000-2012, and analyzes the characteristics of type of wetlands. The results showed that:1. The producers accuracy of extraction, user accuracy and overall accuracies ofwetland classification are above85%in2000,2006and2012. The object-orientedclassification method could efficiently use the image information so that higherclassification accuracy could be obtained. Due to the complexity of wetland distributionand broken, and the moderate resolution imaging features fuzzy boundaries, imagesegmentation, the scale is too large is difficult to accurately reflect the wetland distributionand boundary, object segmentation scale is too small are not conducive to the extractionand the use of spatial information. The cooperated usages of different segmentation scalesare good for wetland information collection in moderate resolution images.2. Season in the use of the image data, to improve the classification effect andprecision of the seasonal changes in the wetland, wetland to get higher classificationprecision, the rainfall amount in the selection and phonological rules is the key factors.Selection of wettest image for a particular time period and understanding of phenologicallaws are crucial for brief classification of wetland pattern information. The usage ofmulti-seasonal image data makes the classification of seasonal wetland clearer, especiallyin delineation of paddy fields’ boundaries and classification of forested wetlands, andimproves the accuracy of wetland classification. It can be concluded that: multi-seasonremote sensing images incorporated with object-oriented processing provide a low-cost andefficient method for meticulous and high-accuracy classification of wetlands in a larger region.3. The study area wetland area, type pattern in the same year in different seasons havedifferent amplitude change, revealed a general trend of increase slowly reduced. The studyarea of wetland have each stage conversion, mainly in wetland and other land cover, mostof the transformation between the marsh and the grass. Wetland distribution andtransformation of wetland area mainly concentrated in the low altitude area and low gradearea. Wetland transformation and precipitation have close relationship with temperatureand wetland vegetation phenology.4. The wetlands were mainly distributed on low-lying areas in the Sanjiang Plain.Paddy fields are the most primary wetland of the study region and the primary landscapetype. Herbaceous, marsh follows as the second primary, and other types of wetlands area ofsmaller proportion. Natural wetlands mostly locate inside the nature reserve, along theriversides and national boundaries. The rest scatters at other places. Wetlands in differentstages of the2000,2006and2012in the study area cover changes significantly, mainly inwetlands and wetland, wetland internal transformation. The area of wetland cover is stablein the nearly decade, and the process of loss of marsh in Sanjiang plain is get slowly. |
PDF Full Text Request