GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Land Cover/Land Use

Integration of multi source data for Land Cover Monitoring

Toshiya Gotoh, koji Kajiwara,Ryitaro Tatelishi
Remote Sensing and image Research Center, Chiba University
1-33 Yayoicho ,Chiba-si Japan.


Abstract
Asian region was classified into 15 to 17 land cover types by unsupervised classification method using NOAA / GVI and AVHRR channel -4 data in 1987 Classification was carried out using 1) GVI data only 2) channel -4 data only or 3) Combination if GVI channel -4 data .The relation of resultant classes elevation were also examined,

Introduction.
There have been rising a need for global scale land cover monitoring for this purpose it is necessary to build Geographic information system includes integrated data from various source such as satellite data and geographic data.

In this study part of these data NOAA Global Vegetation Index (GVI) data NOAA AVHRR Channel -4 data elevation data latitude data were used The GVI data were already to classify land cover is some studies in the view point of vegetation penology since its availability coarse resolution and relatively short interval are suitable for classification of large area from sub continental to global scale AVHRR channel 4 data were used for land cover monitoring since channel 4 data have the information of surface temperature and snow cover elevation and latitude data were also used to investigate to the relationships between them and classified land cover types.

Data.
  1. GVI data.

    GVI data is products of NOAA / NESDIS (National Environment Satellite, Data and information service) it is produced from AVHRR channel -1 and -2 scaled Normalized Vegetation index (SNVI) is calculated as follows.


    The format of data used in this study was plate career with the resolution of 16 km at the equator size of 2500 by 904 latitude and longitude grid an the region between 75° north and 55° south it is available in the form of weekly composite data.

  2. AVHRR channel-4 data

    GOES count (Intended originally for use with geostationary satellites) derived from AVHRR channel -4 data isavialable.

    The format and observation interval of GOES count were same as GVI data. GOES count is scaled brightness temperature calculated as follows,

    GOES = C x TA + D

    Where TA is absolute temperature in Kelvin C and D are constants.

  3. Elevation data.

    The data called ETOPO5 is a product of NOAA / National Geophysical data center (NGDC) it contains alter metric and bathymetric data with the unit of one meter the resolution is 5' and 5' latitude longitude in order to process with other data it was converted in to GVI format as mentioned above it was also used as the preliminary mask to extracting land area.
Data preprocessing.
  1. Production of monthly composite GVI data and correspond channel -4 data.

    Though original weekly GVI data were generally cloud for there still remain un removable clouds in some parts. For reduction of such clouds and convenience of the following processing monthly GVI data were produced from four or five weekly GVI data included a month Corresponding SNVI values of weekly data were compares pixel by pixel if a weekly SNVI value was exceptionally lower than the others the value was considered as noise and omitted lowest value of the reminder was picked up as monthly GVI data. Monthly channel -4 data were also produced from weekly channel -4 data in order that the selected week of a pixel is same as the picked up week in the production of the corresponding monthly GVI data. Note that this operation still could not eliminate clouds completely but produced data are acceptable for employ temporal analysis.

  2. Extraction of land data.

    In order extract land area from all image data in use land data were produced mainly from ETOP05 elevation date. Land area are binary data in which 1 means land and 0 means sea and lakes as a pixel and the area consists of these pixel were considered as and area this was modified and registered by comparing GVI data and maps.

  3. Extraction of Asian region.

    Almost whole Asia was extracted from image data. The size was 1000 pixels by 600 lines, which is equivalent to the area between the 75° north, and 11.2° south latitude and between 36-east longitude and 180° -east west longitude.
Cluster analysis
Cluster analysis of Asian in 1987 were made in the following cases using 12 monthly GVI data 2) Using 12 monthly channel 4 data 3 ) using both 12 monthly GVI and channel -4 data simultaneously).
  1. Classification results -------- GVI data only.

    Fig 1 shows geographical distribution of seventeen classes as the result of cluster analysis using annual GVI data. Since NDVI mainly reflects amount of vegetation the area of deserts and tall mountains almost regard as same area since both area gas slight vegetation the Middle East and Tibet Plateau located in same class.

    Fig 3(a) shows multi temporal plots NDVI monthly mean value in each class this shows the seasonal vegetation dynamics. Each curve is able to be discriminated from others generally at peak point of NDVI curve its value and inflection points if any however low NDVI in winter in high latitude region were affected try by of low solar zenith angle and snow cover.

  2. Classification results channel 4 only.

    Channel 4 indicates surface temperature if not affected by clouds generally the higher the latitude the lower the temperature in mid latitude zone there are high elevation area of relatively low temperature such as Tibet plateau low latitude zone divided in to two classes by the boundary of longitude about 75° east ester part is part is roughly forest and western part approximately desert.


    Figure 1. Geographical distribution of 17 classes as the results of cluster analysis using GVI data.


    Figure 2. Geographical distribution of 15 classes as the results of cluster analysis using GVI and channel-4 data.

  3. Classification results ---- combination of GVI data and channels 4 data.

    Fig 2 shows the same as fig 1 except using annual GVI data channel -4 data simultaneously .One apparent difference with these figures is discrimination between the middle east and Tibet Plateau both of the these areas has same seasonal pattern of NDVI that is flat and low values .On the other hand brightness temperature of the middle east is high flat as compared with that of Tibet Plateau. The other difference is that geographical distribution of classes in fig 2. Is closer than fig 1 especially for latitude higher than 45°.

    Fig 4 show multi temporal Plot of NDVI and brightness temperature of these classes .The horizontal axis in mouth left vertical axis is brightness temperature in Kelvin and right vertical axis is NDVI solid line displays NDVI broken line displays brightness temperature. There are some classes that brightness temperature curves differ each other but corresponding NDVI classes 4,5,6, and 17 opposite case is also found e.g classes 1 and 2.

    Fig 5 shows elevation range and latitude range of each class. A point marked with o is a mean point of a class. Horizontal bar means range of elevation between 25 percentile and 75 percentile vertical bar means same range of latitude as elevation. Concerning latitude there is a gap at 30 as for the elevation there is no obvious thres hold mean point become broader class 17 has range for latitude and second broadcast for elevation. This class distributes high latitude area such as Novaya Zemlya or highest range such as Kara Coram range.

Figure. 3 (a) Multi-temporal plots of mean NDVI for 17 classes in the case of using GVI data.
(b)Multi-temporal plots of mean brightness temperature for 15 class in the case of using channel-4 data.


Figure 4. Multi-temporal plots of NDVI and brightness temperature in the case of using GVI and channel-4data. Solid line displays mean NDVI curves and broken line displays mean brightness


Figure 5. Mean points and ranges for elevation and latitude for 17 classes in the case of using GVI and channel-4 data. Horizontal axis displays range of elevation between 25 percentile. Vertical axis displays same range of latitude.

Conclusion
Asian region in 1987 was classified by cluster analysis .It was divided in to fifteen to seventeen classes according to the data used though clustering GVI data is less discriminating than channel 4 data clustering of the combination of GVI data and channel -4 data is most discriminating for example low elevation deserts can be distinguished from mountains by this method The distribution range of each class elevation and latitude were clarified.

Acknowledgement
The authors wish to thank Mr. K. Okumura for making clustering program and Mr. H. kuki for assistance in parts of data analysis.

References
  1. SDSD, 1986 "NOAA POLAR ORBITER DATA USER'S GUIDE" Satellite Data Service Division (SDSD) National Climatic Data Center NESDIS, NOAA U.S. Department of commerce Washington D.C. USA.

  2. SDSD 1990 "Global VEGETATION INDEX USER'S GUIDE Satellite Data Service Division (SDSD) National Climatic Data center NESDIS NOAA U.S. Department of commerce Washington D.C. USA.

  3. Justice C.O. Townshend J.R.G and Choudhary B.J. 1989 Comparison of AVHRR and SMMR data for monitoring vegetation phenology on a continental scale int J. of Remote Sensing 10 1607 1632,

  4. Kajiwara K. Tateishi R, 1990 Integration of satellite data and geographic data for global land cover monitoring "Proceedings of ISPRS Com IV Sym Tsukuba Japan.