GISdevelopment.net ---> AARS ---> ACRS 1997 ---> Poster Session 1

Land cover Mapping by Combining Satellite Estimated NDVI and Surface Temperature

V.Saravanapavan, T.Saravanapavan and R.Shibasaki
Institute of Industrial Science, University of Tokyo
7-22-1 Roppongi, Minato-Ku, Tokyo 106, Japan
Tel: (81)-3-3402-6231 Fax: (81)-3-3479-2762
E-mail:http://www.gisdevelopment.net/aars/acrs/1997/ps1/vasu@shunji.iis.u-tokyo.ac.jp

Abstract
Accurate land cover mapping, which provides information on the distribution of vegetation on earth surface , is essential for monitoring and maintaining the natural resources of the earth. Usage of satellite remote sensing is becoming a potential tool as it describes the spatial and temporal variation prominently. The conventional method o0f classification by employing Vegetation Indices has some uncertainties, especially in discriminating the active agricultural land form forest in the southeast Asia. Since satellite observed surface temperature measurements are the additive compositions of TIR from background soils and overlying vegetation canopy and the NDVI measurements is and estimate of the vegetation cover, we employ the combination of surface temperature and NDVI in land cover mapping. Our results result reveal appreciable improvement in classifying land cover by Surface temperature and NDVI combination and we produce classified map of southeast Asia.

Introduction
Fast development occurring in Southeast Asia causes higher demand for agricultural development and urbanization and food. Therefore, monitoring which can be achieved by land cover mapping, and maintaining of natural resources are becoming an essential issue. Time series of land cover map for southeast Asia are essential for land cover change and its change analysis especially to determine the change in natural resources. Land cover mapping can be carried out by using ground measured information. Since the ground measured information are time, money and labor consuming and suitable for small, areas remote sensing has the potential to map large area of land cover .

Vegetation index data derived from the National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer is the main feature for land cover change in broad scale (Lambin 1994 a). Eventhough the NDVI is used for identifying and separating vegetation type there are some limitation to the NDVI s a discriminant of vegetation type, particularly for high biomass. Its is likely that if physiognimically different but phenologicaly same vegetation type are considered together. Therefore, by employing NDVI only it sis difficult to differentiate some active agricultural vegetation form forest. Since surface temperature provides the information on back ground soil and vegetation and NDVI measures the information of vegetation, the combination provides more information on land cover. Lambin (1996) used the ratio of surface temperature and NDVI in order to separate the weed land form forest in Africa.

Background
A good understanding of surface temperature and NDVI is essential when land cover is classified by using these two parameters. A negative relation is observed between vegetation indices and surface temperature ( Goward et al, 1985, Price 1990). Te basis for this relationship lies in the unique spectral reflectance- emittance properties of plant relative to soil. Thus vegetated surface show low surface temperature than soil . surface temperature is the function of evapotranspiration. Green biomass increment is often associated with the reduction in surface resistance to evapotranspiration and greater transpiration and a larger latent heat transfer. The presence of transpiring vegetation increases the share of latent heat flux with regard to total available energy.

The loses due to evapotranspiration by vegetation depends on stomatal resistance, canopy height and structure, rooting depth of plants. Moreover surface resistance is influenced by amount and type of vegetation an soil characteristics 9 Rosenberg et al, 1983). Therefore different types of vegetation differ in evapotranspiration. Lambin (1996) used surface temperature and NDVI ratio for land cover classification in Africa since NDVI and surface temperature are having correlation regarding biome.

Data and methodology
The National Oceanic and Atmospheric Administration (NOAA) Advanced Very High resolution Radiometer (AVHRR) data was used in this study. NDVI ( Normalized Difference Vegetation Index ) data an channel 4 and channel 5 data were collected form EROS Data Center for the study area covering southeast Asia form April 1992 to March 1993. surface temperature was obtained. NDVI, surface temperature was calculated based on Price formula 9198 3, 1984) and monthly composite data were obtained . NDVI, surface temperature and the ration of surface temperature were used separately as a measure for the classification . unsupervised classification was performed since there was insufficient information on the study area. Land use map of Thailand, 1990 was used for the validation.

Results and Discussion
The classification result ( fig. 1) based on NDVI shows some misidentification. Some agricultural area and orchard area were identified as forest with I Thailand. For example, in southern part of Thailand is partly cultivated with orchard according to land use map., but this intensive cropping is practiced but is misclassified as forest. Therefore, by using NDVI alone it is difficult to separate some active agricultural area from forest.


Fig 1. Classification results by NDVI

The classification result ( fig.2) obtained by using the ration of surface temperature to NDVI, which was used for African biome calcification ( Lambin , 19960, shows prominent separation of forest fro active agricultural area. Orchard area was separated from forest and some active agricultural area near Bangkok also separated from forest. But in Mekong delta , intensive agricultural area still shows some mis-classification. But the area misclassified as forests is smaller than the area misclassified as forest earlier by using NDVI alone.


Fig 2. Classification results by Ts/NDVI ratio

In order to reduce the misclassification we computed the squared value of surface temperature and the ration of squared surface temperature and NDVI was used as a feature for classification. The classification result ( Fig 3) obtained by using squared surface temperature and NDVI ratio shows improvement in separating forest form active agricultural are. In southern part of Thailand , forest is separated well than the other method . like wise most part of the Mekong delta agricultural area also separated from forest. But some misidentification is still observed in the this area. The misclassified area of agricultural area as forest in Mekong delta is small when compared to other methods. Thus this method shows a distinct improvement in differentiating forest from intensive agricultural area then the other methods employed in this study.


Fig 3. Classification results by Ts2/NDVI ratio

Reference:
  • Goward, S.N., Cruickshanks, G.D., and Hope, A.S>, 1985. Observed relation between thermal emission and reflected spectal radiance of a complex vegetated landscape. Remote sensing of environment =, 18, 137-147.
  • Lambin, E.F., and STRAHLER, A.H., 1994a. Change-vector analysis; a tool to detect and categorize land-cover change processes using high temporal-resolution satellite data. Remote sensing Environment, 48, 231-244.
  • Lambin, E.F., and Rhrlich, D., 1996. the surface temperature vegetation index space for land cover and land cover change analysis. Int .j.Remote sensing, 17, 463-487.
  • Price, j.C., 1990. using spatial context in satellite data to infer regional scale evapotranspiration. I.E.E.E Transaction in Geoscience and Remote sensing 28, 940-948.
  • Price, J.C., 1984. land surface temperature measurements from split window channels of NOAA-7 advance very high resolution radiometer. J. of Geophysical Res. 89, D5:7231-7237.
  • Price, J.C.1983. Estimating surface temperature form satellite thermal infrared data-a simple formulation for atmospheric effect. Remote sensing of Environment 13:353 -361
  • Rosenberg, N.J., Blad, B.L., and Verma, S.B., 1983. Microclimate ; The biological environment ( New York ; John Wiley & sons. ).