Establishment of landsat MSS spectral classes of fairly homogeneous tropical peat swamp forests by transect analysis on the Orser batch-oriented image analysis system

Establishment of landsat MSS spectral classes of fairly homogeneous tropical peat swamp forests by transect analysis on the Orser batch-oriented image analysis system

A.B. Ismail
Research Officer,
Malaysian Agricultural Research and Development Institute (MARDI).
GPO Box 12301. 50774 Kuala Lumpur

B. J. Turner
Reader. Department of Forestry,
Australian National University, Canberra.

Introduction
The hydromorphic nature of peat areas in Malaysia render the conventional methods of survey based on extensive filed work ineffective methods of survey based on extensive field work ineffective due to difficult accessibility. Remote sensing with its wide synoptic format and variable spectral discrimination, offers an attractive alternative for this purpose. Past study using Landsat MSS data (MAHMOOD) et. al., 1983), however, indicated that spectral reflectance's within peat swamp forest in Peninsular Malaysia are fairly homogeneous.

Peat swamps in Malaysia are formed in water saturated saucer-shaped basins. Therefore, peat thickness along the perimeter is thineer as compared to the centre of the basins. The differences in ecology is expected to influence the forest composition occurring in the areas. especially from the point of view of root anchorage and nutrient availability. Natural zoning of forest compositions around the centre of the basins if therefore anticipated. For peat swamps in Sarawak, there are indications of dome-shaped surface structure ( ANDERSON, 1964: LIONG and SIONG, 1979) and zoning of forest compositions around the domes ( ANON, 1957: BRUNIG, 1970).

Variations in forest compositions are demonstrated by the differences in dominant species, which among others, have a different pigmentation and leaf structure. Variations in pigmentation are detectable in visible spectrum, and structural differences in the spongy mesophy11 layer of leaves are indirectly observable in the near infrared region ( RHODE. 1971; RHODE and OLSON. 1971). It is therefore expected that the variations in forest compositon can be Detected by spectral analysis of Landsat MSS data. Furthermore, these areas are fairly flat and hence the possibility of spectral variation due to topographic effect is minimal. As such, a study was undertaking to assess the effectiveness of Landsat MSS data in detecting spectral variations within fairly homogeneous peat swamp forests in Malaysia.

Study areas
Two peat swap basins, one in Pahang and the other in Sarawak, were selected for the study. They are expected to represent different peat environments occurring in Malaysia. Standard Computer Compatible Tapes (CCTs) for these areas were recorded by Landsat 3 (August 8, 1978) and Landsat 4 (June 27, 1985), respectively. Majority of the study areas were free of cloud cover.

Data processing and analysis
The MSS data were processed and analysed using the ORSER image analysis software system (TURNER et al., 1978) run on a VAX 11/780 computer at the Australian National University. The system, which urns on batch oriented mode, was developed at the Pennsylvania State University. USA. In the following discussion, the ORSER routines employed in the study are indicated in parenthesis. Visual display and hardcopy printouts were done using the UNIRAS software on Tektronix 4113 and 4695, respectively.

Difficult accessibility of the study areas render the stud areas render the normal method of selecting training areas ineffective. Spectral analysis along several transects across the basins, however, is expected to give maximum possibility for detecting spectral variations. Hence, spectral classes can be manual established and then used as training areas for supervised classification routine.

After subsetting (SUBSET) the study areas from the CCTs, bringhtness classification mapping (NMAP) was carried out to visually verify the respective peat swamp areas. This was followed by homogeneity classification (UMAP) in order of high contrast and their boundaries. Based on these information, and coupled with various collateral data such as the peat basins were identified. For maximum possible variations, those transects were aligned across the widest parts of the basins and extended from the inland hilly areas to the sea. Three transects for each study areas were chosen for further analysis.

Along the transects, a succession of blocks of 10X10 pixels were statistically analysed (STATS). Mean spectral reflectance values for all bands were then plotted along the transects. Vegetation index values, (VI-(5-7)/(5+7)). Were also plotted. Subsequently, covariance matrices from the statistical analysis were used in canonical analysis (CANAL) to enhance block separability.

Coupled with information from cluster analysis (CLUS) and various collateral data, the information obtained from various spectral or signature classes found along the transects. The spectral classes for all the transects within each study area were then compared, and the similar ones were discarded. As such, a set of spectral classes for each study area was identified.

The established spectral classes and their specified limits were coded in various characters and then used to classify (CLASS) the study areas. Minor modification of the mean spectral reflectance values for various classes and their limits, however, might be required. The study areas were then geometrically corrected, rotated and rubber-sheet stretched (SUBGM and DISPGM). This was followed by cleaning of stray symbols and production of compressed maps (DISPLAY). Finally, the compressed maps were used for visual display and colour hard copy print out after scaling.

Spectral variations within the peat swamp forests
The transect analysis indicated that there were significant variations in the spectral reflectances of the Landsat MSS data over these peat swamp forests. Examples of the mean brightness and vegetation index values of all the blocks plotted along selected transects in Pahang and Sarawak, as well as the block separability, are presented in Figure 1 and 2, respectively. Generally, the spectral responses, especially those in the near infrared regions, changed toward the center of the peat areas. The vegetation indicates also indicated the decreasing biomass in the same direction.

Figure 1. Landsat MSS brightness values and vegetation values and
vegetation index of 10 X 10 pixels blocks, and their separability
based on canonical analysis, along a selected transect over
Pahang peat swamp forest.

Figure 2. Landsat MSS brightness values and vegetation values and
vegetation index of 10 X 10 pixels blocks, and their separability
based on canonical analysis, along a selected transect over
Sarawak peat swamp forest.

Spectrally, the Sarawak peat swamp forest is more variable, as indicated by the many more blocks that were statistically separabl, as compared to the Pahang area. This may be due to the variable forest compositions of more areas were affected by logging activities in the former. The spectral veriations in the latter were mainly observed near the parameter, and this may be transitional to the non-peat areas.

Supervised classification, using the established spectral training areas from the transect analysis, confirmed the expected zoning nature of these peat swamp forests. Concentric zonation of spectral classes around the centre of subbasins was obvious (Figure 3 and 4). Moreover, reported nature and forest compositions, especially for the Sarawak area.

Figure 3.

Figure 4.

Conclusion
The study shows that, with proper approach and methodology, the ORSER batch-oriented image analysis systemable to detect, classify and map Landsat MSS spectral variations over fairly homogeneous peat swamp forests in Malaysia. In addition, the study also indicates possibility of using these data for characterization of the forests and soils occurring in these swamps.

References

ANDERSON, J.A.R. ( 1964). The structure and development of peat swamps of Sarawak and Brunei; Trop. Geography 18:7-15
ANON ( 1957). Land use map of Sarawak. Land and Survey Department, Kuching, Sarawak; Series NC. 11.
BRUNIG, E.F. ( 1970). Stand structure, physiognomy and environmental factors in some lowland forest in Sarawak. Trop/. Ecology 11: 26-43.
LIONG, T.Y and K. H. SIONG, ( 1979). A review of lowland organic soils of Sarawak. Technical Paper No. 4. Dept. of Agric., Kuching, Sarawak.
MAHMOOD, N.N.: M. BRUNEAU: H. LE Men (1983). A pilot study on the use of satellite remote sensing data for agroecology mapping of Peninsular Malaysia: MARDI (ASAS 02-83), Kuala Lumpur.
RHODE, W. G. ( 1971). Multispectral enhancement of disease in forest stand - In Color Aerial Photography in Plant Science; Third Biennial workshop, American Society of Photogrammetry; pp 131-143.
RHODE, W. G.; C.E. OLSON, Jr. ( 1971). Estimating foliar moisture content from infrared reflectance data - In Color Aerial Photography in Plant Science; Third Biennial Workshop, American Society of Photogrammetry; pp 114-146.
TURNER, B.J.: G.M. BAUMER: W. L. MYERS ( 1975). The ORSER remote sensing analysis system. A users' manual. Res. Publ 109/)OR. ORSER, The Pennsylvania State University.