Use Of Satellite Data To Estimate Areal Evapotranspiration From A Tropical Watershed
The Data sets
Daily input data are rainfall, air temperature, humidity and insulation. The other required parameters reflected the physical characteristics of the watershed, including vegetation and soils description of the parameters and meterorological input data sets follow a brief explanation the watershed subdivided into GRUs and LAI estimation from the satellite data.
Land cover classification
The Landsat Tm data of February 26, 1993 were use together with aerial photographs, topographic, land use and soil maps for land cover classification. ERDAS software was used for the image processing and generating data overlays and vector data were stored and processed in ARC/INFO. The TM data were classified using an unsupervised clustering algorithm to identify the dominant categories of land cover characteristics of watershed. The image was classified into 6 spectral classes. These spectral classes were easily grouped into three broad informational classes namely forest, oil palm and Hevea. If two vegetation types were equally predominant in a GRU, a mixture of two was considered. Classified ground cover classes of Trolak are shen in figure 1, since the spatial extent of the study are is tool large to consider finer details like species variation, this scheme should be fairly reasonable. A vector image containing the watershed and subwatershed boundaries was overlaid on the classified image for the demarcation of tentative boundaries of GRUs based on the classified ground cover classes.
Figure 1: Classified ground cover classes of Trolak watershed
A GIS approach based on elevation, slope and aspect was used to subdivide the tentative GRUs with large elevation differences into smaller GRUs with relatively smaller elevation differences differenes into smaller GRUs with relatively smaller elevation difference. It was assumed that generally a GRU with relatively smaller elevation differences. It was assumed that generally a GRU will not have an elevation difference of greater than 300m. these GRUs were then overlaid on the raster containing vegetation type classes so as the verify the partitioned GRUs and to ensure that no modification is needed. However, some sample field data will be needed to ascertain as to what each ground cover class represent on the ground. The masked of these GRUs were used ot prepare different data layer for estimating model parameters and input variables of each GRU. The detailed procedure of raster/data preparation and watershed subdivision into GRUs is available in Amin et al. (1996).
Leaf area index (LAI)
In order to estimate LAI from TM data, a simple form of LAI as a function of TM data was established. A vegetation-indices 9VI) based model was used to relate 46 field measured LAI of 10 sites from forest oil palm and rubber stands to the vegetation index form the Landsat TM data. In this model, the LAI of three vegetation stands was defined as the dependent variable. Vegetation-Indices based mode is obtained by rearranging he Baret and Guyot (1991) model in which the vegetation index expressed s a function of LAI, using modified Beer's law. The detailed procedure of field assessment of LAI and relationship between the LAI measurements and the vegetation indices form Landsat TM data are available in Amjad and Amin (1997).
The meteorological data
The maximum and minimum temperature data were taken from Trolak Utara, the nearest weather station located about 6.5 km north of Trolak bridge and monthly lapse rates were determined using the data form Cameron Highlands. Daily rainfall was taken from Trolak, about 1 km north of Trolak bridge. Daily shortwave radiation was estimated on the slope-aspect combination of each GRU. The required inputs for the calculation are the latitude, the slope inclination and aspect, and the day of the year.
