ZHANG Shiqiang, LU Jian, LIU Shiyin. Deriving Glacier Border Information on Qinghai Tibet by TM High Spectrum Image[J]. Geomatics and Information Science of Wuhan University, 2001, 26(5): 435-440.
Citation: ZHANG Shiqiang, LU Jian, LIU Shiyin. Deriving Glacier Border Information on Qinghai Tibet by TM High Spectrum Image[J]. Geomatics and Information Science of Wuhan University, 2001, 26(5): 435-440.

Deriving Glacier Border Information on Qinghai Tibet by TM High Spectrum Image

  • Qinghai Tibet is one of the most sensitive areas in the world among the research of global climatic changes.The fluctuation of modern glaciations recorded the climatic changes of decades.The extent of the change was achieved by means of classifying remote sensing images.FK(W33?40ZQThe main problem of classifying lies in the tongue part of the ice which includes various grayscale values,so it is very important to get a method which can identify the ice areas from the image.Four methods,threshold values,supervised classification by training sites,non-supervised classification,relationship between bands,are used in experiments on a Landsat TM image including 7 bands.The result was compared on the basis of the interpreting by means of eyes and experience.Threshold values,which had been widely used in early single band image for identifying snow or ice,was first tested.There is a great gap between the fixed ice extent and uncertain ice areas by using the experience thread 200 and 60 in band 2,so it didn't performed well.The threshold values of NDSI,which can be expressed as (band2-band5)/(band2+band5),but it is difficult to achieve the precise value.Supervised classification is the most widely used method based on statistic analysis in classing multi-band image.Five categories of main land cover types based on the field observation and previous report on the land cover distributions,namely the ice and snow,dirty snow,ice tongue,bare hill,shallow area,were fixed to generate the unsupervised and supervised land cover classification images maps for 255 clusters.Based on the knowledge of the study area and glacier,five training polygons were carefully defined and their signature was made.Two methods,minimum distance classifier and maximum likelihood classifier,were used respectively.In unsupervised classification of the composition image of band 1,2 and 5,the image was labeled into the same five classes,it was worse than the supervised classification poorly in discrimination of ice tongue and shadow area.The main principle of relationship between bands is from knowledge of the spectrum of the well known land cover categories to discriminate all kinds of land cover in the remote sensing image.By finding out the relationship between each band of multi-band image,or extended ration images,or other different images,and evaluating the relations,the different land covers can be discriminated.In this experiment,by observing the signature of the five classes,it can be found that there lies a great difference between band 2 and band 5 for identifying ice areas (including ice or snow,dirty ice and ice tongue) and non-ice areas,so the thread value of 5 was applied in the ratio image of band2/band5.Then supervised classifier was applied in ice areas to discriminate the three classes.The result is better than the others.It can be concluded that the relationship between bands is the most efficient method in identifying interesting areas.The main causes maybe lies in the wide range spectrum of the ice tongue.The remaining problem was discussed in the end.
  • loading

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return