INGENIEUR
classes which were vegetation , sparse vegetation ( high ), sparse vegetation ( low ). The details of the classes are shown in Table 2 .
Class Name Class Number NDVI values Vegetation 1 > 0.60 Sparse Vegetation ( High ) Sparse Vegetation ( Low )
2 > 0.45
3 < 0.45
Table 2 : Details of classes involved in unsupervised classification .
( iv ) Terrain and Digital Elevation Model ( DEM ) This analysis used the GeoTIFF metadata with a resolution of 30 metres . The basic terrain analysis was carried out using the SAGA-GIS software which by default produces 16 terrain and hydrological metric outputs as shown in Figure 2 . Once the output had been generated , a three dimensional view of the terrain was generated along with its contour lines . The Topographic Wetness Index was also generated to show the local topographic impact on hydrology and simulation of spatial moisture distribution and saturation of the soil in the area of study .
( v ) River Detection The river located inside the area of interest is considered to be a narrow waterbody since its average width is around two to three metres during the dry seasons . To detect this specific river or any other small water bodies , a sub-pixel analysis was carried out to extract the endmember abundance in pixels using Linear Spectral Unmixing tool ( Sall et al ., 2020 ) . This function can be found in remote sensing software such as ENVI or Earth Resources Data Analysis System ( ERDAS IMAGINE ). From this analysis , we also tried to generate the Tasseled Cap Wetness Index to detect the contrast between the northern part of the river and the southern part . These analyses were done using Landsat 4 and 5 TM image dated December 10 , 1990 and SPOT-5 image from November 30 , 2012 .
POST-CLASSIFICATION
This process is a method to detect changes in two sets of images with different dates . The method used in this study was Confusion Matrix and is
Figure 2 : Outputs expected after generating the terrain analysis .
usually applied in Supervised and Unsupervised Classifications . This method is commonly used for detecting any changes ( increment , decrement or unchanged ) in an area . For this study , two different satellite images with different dates were acquired . The criteria for the selection of images were the oldest and the most recent Landsat TM data available over the area of study . This process generated the confusion matrix charts and tables for the change detection analysis .
RESULTS AND DISCUSSIONS
In this section we show the output from the Google Earth images for the NDVI and the river detection analysis . The results of the NDVI and LST analysis of the area of interest in a time span of 32 years along with the results from the unsupervised classification of the area of interest from the year February 20 , 1988 until July 5 , 2020 is discussed . This section also focuses on the terrain analysis and how it affects the oil palm land scheme , and provides the method and steps used to check the presence of river inside the area of interest as well as the larger river , Sungai Kerian inside the same image .
( i ) Google Earth Image Observation Due to its high resolution of 0.31 metres , images from Google Earth , which was acquired from DigitalGlobe were analysed to give an overall
54 VOL 84 OCTOBER-DECEMBER 2020