Separation (Recognition) of tree species or species composition in an old growth forest
Plantation in Peninsular Malaysis using the Vegetation Index approach
Nor Azman Hussein1 and Mazlan Hashim2
1Forest Research Institute Malaysia (FRIM)
Tel: 03-6302117, Fax: 03-6367753,
E-mail: norazman@frim.gov.my
2Faculty of Engineering and Geoinformation Science
University Teknoligi Malaysia (UTM)
Locked Bag 791, 80990 Johor Bahru
Tel: 07-5502940, Fax: 07-5566163,
E-mail: mazlan@fksg.utm.my
Abstract
Most of the standard image classification procedures extract information form remotely sensed data using spectral characteristics alone. Although these procedures are quite effective for general mapping purposes, purely spectral classifiers have limitation when used to map detailed information such as tree species or composition, especially when there is an abundance of mixed pixels (mixels) or if there is substantial overlap in the spectral definition of the classes. These factors have significantly contributed to the failure to apply remote sensing technology to forest management, especially at operational level. This limitation is especially felt in the tropical forest environment when classification by tree species or species composition is required. Quantitative analysis using Vegetation Index (VI), simple regression, and a modeling approach have been proven effective to estimate vegetation properties independent of vegetation types. Of the three, VI is the most commonly used; that is by linearly combining or rationing reflectance in the red and in the near infrared (NIR) spectral region. The use of these spectral regions is due to their "characteristic" responses to vegetation properties and as such, VI is highly correlated to biophysical factors such as biomass density, leaf index (LAI), vigorousness, canopy and leaf structure, etc. In this paper we examine the capability of 6 different types of VI derived from Landsat TM data to differentiate or recognize tree species or species composition of a well-documented old growth forest plantation test site at Bukit Lagong in Peninsular Malaysia. Result of the study show that there was a positive correlation of VI to tree species or species composition with an overall accuracy of 60%, an encouraging factor for pursuing the study
Introduction
There has been rapid development in multispectral remote sensing since its introduction in the 70's. However, the development of remote sensing technology for application in tropical forestry has been rather slow. This is mainly due to the complex characteristics of the tropical forest and the lack of quantitative analysis of multispectral remote sensing data to actual ground information. With better understanding of the remote determined and their full potential can be realized.
The typical spectral reflectance curve for vegetation is primarily caused by variations in biophysical aspects of the plant and its stand structure. This is because these factors absorb, reflect and emit EMR different part of the spectrum. These differences are attributed to the intensity of the pigments (maily chlorophyll), internal cellular structure of plant leaves, moisture content and leaf thickness. Chlorophyll, the green pigment in plant leaves, strongly absorbs EMR at about 0.45mm (blue band) and 0.67 mm (blue band), and
Reflects green energy (0.52-0.6 mm) which causes healthy vegetation to appear green. Variation in intensity of chlorophyll in the plant, either naturally or caused by stress, abnormal growth and productivity, result in reduction of absorption of the blue and red bands. Theoretically, characteristics of reflectance in the chlorophyll-absorptance region can be used to separate different vegetation and/or vegetation classes in remote sensing data.
Plant internal structure also varies considerably either under natural conditions or abnormal growth and stress. Inherent characteristics of the internal structure of healthy plant leaves cause 40 to 50 percent of the NIR portion of the EMR spectrum (0.7-1.3 mm) incident upon it to be reflected. Distinct characteristics of spectral reflectance in the NIR region is another aspect that would enable separation of vegetation that otherwise looks the same in visible wavelengths. The use of spectral reflectance in the NIR in combination with other optical bands, particularly red, could further enhance the recognition of vegetative types.
The method of using 'infra-red' bands in combination with optical bands to study vegetation is termed Vegetation Indices (VI). It is defined as "mathematical transformations designed to assess the spectral contribution of vegetation to multispectral observation" (Elvidge and Chen, 1995). The most basic assumption in VI is that algebraic combination of remotely sensed spectral bands correlate to the presence of green vegetation in the pixels scene. This is based on the characteristic of the green vegetation spectru; there is intense chlorophyll pigment absorption in the red (R) against the high reflectivity of plant materials in spectrum are widely used. There is a vast number of publications that discuss R and NIR use of VI to estimate vegetation variables such as percent green cover, leaf area index (LAI), absorbed photosynthetically active radiation (APAR) and others either for general vegetation studies or related to forestry (Fisher, 1994; Huete et al., 1994; Myneni & Williams, 1994; Spanner et al., 1994) The R and NIR combination is normally in the form of a ratio, a slope, or other formulation that can generally be separated into three categories, namely intrinsic indices, soil-line related indices, and atmospheric-corrected indices (Rondeaux, Steven and Baret; 1996). Nonetheless, there are other less commonly used algebraic combinations which utilize multiple spectral bands.
