GISdevelopment.net ---> AARS ---> ACRS 1989 ---> Land Use

Application of Landsat MSS Data in Land Use Mapping: A Malaysian Experience

Darsu Ahand
Agricultural officer, Land Use Survey Section
Department of Agriculture, Malaysia

Introduction
One of the functions of the Malaysian Department of Agriculture (DOA) is to provide land use information for various levels and lad development planning in the country. Such information not only covers agricultural land utilization, but also other non-agricultural land use activities namely urban, minings, forested areas etc.

The application of remote sensing technology in DOA upto the present stage of development has been confined to the visual interpretation of black and white panchromatic aerial photographs, expensive as compared to satellite remote sensing. Furthermore, due to the longer period of time needed for aerial photographs to be taken and the large number of photographs that have to be interpretated visually, some of the information generated might have lot their value by the time the survey has been completed.

In view of these experiences and favourable reports made elsewhere, it is felt that the digital image processing and classification of satellite data could play quite a significant role in meeting the need and the increasing demands for more upto data land use information as the country is presently experiencing rapid development. This techniques offers certain unique advantages over conventional aerial photo interpretation. Not only does it furnishes synoptic repetitive coverage but also the availability of satellite data in digital format makes them rapidly suitable for computer assisted analysis; thereby facilitating information extraction. Studies in other countries have shown that the synoptic view from satellite platforms together with modern sensors and computer technologies have much to offer to planners who need accurate and timely land use information for various land development planning and crop yield forecasting.

Objective of the Study
The objective of the study is to assess the suitability and the capability of Landsat MSS data and land use mapping and updating in Malaysia. This exercise was the first in a series of studies on the applications of satellite data, which would also include Landsat TM and SPOT data, carried out by the department of agriculture in their efforts to utilize satellite remote sensing techniques for land use surveys in the country. The study emphasized on the development of a suitable survey procedure for use under local conditions, and on the applicability of data, particularly the degree do details, for the production of reliable land use information to assist planners in various land development plannings.

The study was conducted as the training project at the Ontario Center for Remote sensing (OCRS) under the Malaysia-Canada Joint-Project on Agricultural information system funded by the Canadian International Development Agency.

Material and Method
The study was carried out based on three bands (4,5 and 7) of a Landsat 4 MSS scene, path-row 128-56 dated March 23, 1985 with less than 10% cloud cover. The frame encloses the north-west region of Peninsular Malaysia covering the states of perils, Kedah, Penang and part of perak.

Digital data analysis was performed using DIPIX systems at the Ontario center for remote sensing. The data was first corrected to its real world position so that it could be correlated with the available references on the existing land use maps which were prepared based on aerial photo interpretation. Existing land use maps and other related ground information as well as piece-wise contrast stretching enhancement technique were used to assist in the selection and delineation of training sites. The image was then classified using supervised technique of a maximum likelihood classification algorithm.

The resultant maps were plotted at the scales of 1:10,000 and 1:50,000 for further evaluation and subsequent incorporation into a geographic database.

Results and Discussion
The present land use classification legend of Malaysia based on1:250,000 aerial photo interpretation contains 34 land use classes which were grouped into 9 main categories (appendix 1). The use of aerial photographs at this scale permit mapping of details upto a unit size of 0.6 hectares (Wong 1980). Major crop types such as rubber, oil palm and coconut could also be subdivided further into three different growth stages i.e the young or immature state, the mature stage and sensile or moribund stage. Further reduction in the scale of photography would proportionally increase the number of mapping units would have to be enlarged to 1.6 hectares and as a result, the number of mappable land uses were reduced to 27 classes.

However, the use of 80m resolution Landsat MSS data only allows for a board classification of the land covers in the study area. As a result, only 13 different classes of land covers were able to be classified and mapped. These can be grouped into three land cover categories as shown in Appendix II.

Major features such as areas under agricultural land us and the primary forests which include swamps and other wetland forests were easily identifiable on the Landsat MSS data. Conversely, a more detailed identification among and within individual land use types were found to be not possible at this stage of technology development. This is due to the nature of agricultural land utilization in the area which is predominantly characterized by small farm sizes and irregular cropping pattern. The farms can be as small as 0.5 hectares and in most cases, they are with mixed cropping as well as with intercropping. As such, the classification can only be done on the basis of broad crop groups or mixed roping. However, there are cases where certain major crop types such as sugarcane, irrigated paddy and mature rubber occurring in large contiguous blocks were found to be mappable.

Some difficulties were encountered in the identification of individual oil palm plantation found within mature rubber growing areas. The spectral reflectance’s from these two crops seem to be almost similar during that particular time of data captured eventhough they are botanically different from each other. This may be attributed to the absence of certain spectral bans in the Landsat MSS system which restrict finer discrimination among certain croptype. This observation is in line with the finding made by Bauer, et. al (1977) where they have recognized the importance of middle and thermal infrared portions of the spectrum, which are absent in Landsat MSS system, for more accurate crop identification. However, this difficulty is expected to be resolved with the use of MSS data captured during wintering season when rubber shed their leaves.

There is also the tendency to have other mixed cover classes comprising small units of crops or land uses which occur in close proximity. An example of such cover class is the mapping units representing small areas of mature/sensile rubber, secondary forest and some swampy forest when these covers are found close to each other. Such complication will always be encountered in mapping the land use of small holdings due to the small farm sizes and irregular cropping pattern.

Despite the shortcomings, the synoptic view of Landsat data provides the opportunity to obtained reliable information over very large area. Such example is on the irrigated paddy land in the study area. The data which was recorded in March 23, 1985 showed the distribution of fallow paddy fields for the entire MADA area for which for different stages of moisture. Flooding conditions could be easily classified and mapped (Appendix II). Based on this findings, it is envisaged that the different phonological growth stages of paddy crop cultivation in a particular area which then can be the basis for forecasting of paddy production.

The results of the study show that the satellite data can be one of the important sources for land use mapping and updating in this country. There is a need, however, to have a suitable procedure for use under local condition. The development of the survey procedure has to consider the four following factors : -
  1. The limited availability of low or cloud free data in a humid tropical country like Malaysia.
  2. The characteristics of small farm sizes, irregular and mixed pattern of the agricultural land utilization in Malaysia.
  3. The different degree of information details of the various satellite systems, and
  4. Aerial photography remains the most important sources of data for mapping and monitoring of land resources. It is an essential element when multi-stage procedure is applied to the interpretation of satellite imageries.
The proposed land use survey procedure for use in Malaysia, as shown in Appendix III, is basically a modified version of the procedure developed at the laboratory for applications of Remote Sensing, Purdue University. Slight modifications to the original survey procedure were made to suit the local conditions. It is felt appropriate at this juncture that the discussion should revolve more on the aspects pertaining to the selection of training sites since it could be considered as the most critical stage in any classification exercise. Therefore, a good understanding on the above factors would be of a great assistance in the selection of the proper training sites to arrive at successful classification results.

The selection of training sites is normally carried out based on the variation in spectral reflectance, both in terms of spectral values and visual appearances of the images. However, consideration should also be given to the positively identified land cover information should also be given to the positively identified land cover information that are transferable from the existing maps to the image on the screen via a digitizer. A software with the capability of performing accurate registration (or geometric correction) according to eh local map projection system is therefore required.

Training sites have to satisfy one important criterion that they must adequately represent the variations of spectral reflectance’s present in the cover types throughout the area to be classified. These variations are mainly due to the different conditions of cover types which among others, include differences in phonological growth stages and agronomic practices. The use of multi-temporal data at the beginning stage is undoubtedly necessary for the analysts to acquaint themselves with the variation in spectral reflectance’s particularly for annual crops. Together with appropriate enhancement techniques to highlight certain features, they can be of a great assistance in the delineation of proper training sites and amp legend development. In addition, this multi-temporal techniques could also be one of the possible solutions to the problem of limited availability of low or cloud free satellite data in a humid tropical country like Malaysia. Spatial information such as texture should also be used to improve on the level of classification accuracy attainable using the spectral data alone (Swain 1977).

Statistics of every spectral classes had to be calculated for the purpose of assessing the separability of these classes. An evaluation on the classification results is also needed to make sure that the classification accuracy and area estimates are within the acceptable limits as determined by the requirement of the project.

With consideration refinement of the registration techniques, it is possible for the classified imagery to be used or incorporated directly as a map at the standard 1:50,000 topographic base. It is preferable to have a software that allows the incorporation of the classification results automatically into a geographic information system for subsequent data processing and analyses; thereby avoiding manual digitization of the map data – a task which is quire labourious and time consuming.

Land use surveys require a good understanding on several aspects which in one way of another contributes to successfully classification results. These includes : -
  1. Nature and trend of land utilization in the country.
  2. Cropping pattern, crop calendars and agronomic practices in the agricultural sector.
  3. Certain botanical characteristics and physiological behaviour the some crops.
With respect to the land use survey based on satellite data the above are among the key factors which explain the variation in spectral reflectance patterns resulting from the interaction of solar radiation with various features on the earth surface.

Concluding Remarks
Experience from this preliminary study has shown that remote sensing data especially those captured by space borne platforms based on computer assisted analysis could serve as the basis for further improvement of existing procedures for land use data collection in the department of agriculture. However, one should realize the fact that there would be a trade-off between quick delivery f an up-to-date information with the lower degree of details as well as with a slightly higher initial cost of operation when this new technology is going to be implemented. Further more considerable emphasis also to be given to the types of satellite data to be used since they contribute to the degree of information detail required by the planners. Landsat MSS data for example, could only provide information at level 1 of the present land use classification legend of Malaysia. Low spatial resolution of 80m was observed to be less suitable for a more detailed mapping of the overall Malaysian agricultural land utilization due to the fact that it s predominantly characterized by small farm sizes and irregular cropping pattern. In addition, limited number of spectral bands, to a certain extent, dos not allow for a more detail discrimination among and within crop types. Despite these short comings, it was observed that the Landsat MSS could still act as an importance source of data for crops cultivated on large contiguous blocks.

With some refinements to the survey techniques and the availability of better satellite data together with advanced computer technologies, it is envisaged that the technology is capable of providing reliable and timely landuse information to assist planners in their decision makings.

Acknowledgement
The author wish to tahnk Mr. Siew Kam Yew and Dr. Zulkifli Kamaruzzman, Senior agricultural Officers, for their valuable comments during the preparation of this paper. The permission of the direction general of agriculture, Malaysia, to publish this paper is gratefully acknowledged.

References
  • Bauer, M.E. Hixson, M.M, Davis, B.J. and J.B Etheridge 1977), ‘crop identification and Area Estimation by Computer –Aided Analysis of Landsat Data, Proc. Of the symposium on Machine processing of Remotely Sensed Data, IEEE 77CH1218-7MPRSD.
  • Schowengerdt, R.A. (1976) Techniques for Image Processing and Classification in Remote Sensing. Academic Press, New York.
  • Swain, P.H (1977) Advancements in Machine-Assisted Analysis of Multispectral Data of Land Use Applications, Proc. Of the Symposium on Machine Processing of Remotely sensed Data, IEEE 77CH1218-7MPRASD.
  • Wong, IFT (1980) Aerial Photography Survey, Its Usefulness, Cost of Operation and Personnel and Time Span Involved – A Malaysian Experience. Department of Agriculture, Peninsular Malaysia.
Appendix I
Land Use Classification Legend for Malaysia Based on Aerial Photo Interpretation

  1. Settlements and Associated Non-Agricultural Areas
      Urban and Associates Areas.
      Estate Buildings and associated Areas
      Tin Mining Areas
      Power Line Right of Way
  2. Horticultural Lands
      Mixed Horticulture
      Market Gardening
      Agricultural Stations
  3. Tree, Palm and other permanent Crops
      Yong Rubber
      Mature Rubber
      Sensile Rubber
      Young Oil Palm
      Mature Oil Palm
      Sensile Oil Palm
      Young Coconut
      Mature Coconut
      Sensile Coconut
      Pineapple
      Tea
      Coffee
      Cocoa
      Pepper
      Sugarcane
      Orchards (Rembutan, Druain, Nutmeg, Citrus etc)
      Sago
      Banana
      Fish and Hyacinth Ponds
      Arecanut
  4. Crop Land
      Paddy
      Diversified Crops
      Shifting Cultivation
  5. Improved Permanent Pasture.
  6. Grasslands
      Lalang, Unimproved coarse Pasture and/or Scrub – Grassland Grass covered Erosion Scars and Landslides
  7. Forest Land
      Forest
      Scrub
      Newly Cleared Land
  8. Swamps, Marshland and Wetlands Forest
  9. Unused Land
Appendix II
Land Use Classification Legend based on Landsat MSS Data

  1. Agricultural Areas
    • Irrigated paddy areas
        Flooded paddy fields (deep)
        Flooded Paddy fields (shallow)
        Flooded paddy fields (very shallow)
        Paddy fields (very moist)
    • Rural settlement areas with coconut and mixed horticulture (mainly in MADA area)
    • Rainfed paddy (fallow fields, grass lands, some mixed horticulture and cleared lands.
    • Mainly matured rubber with some oil palm and scrub or secondary forests
    • Mainly sugarcane with scattered rainfed paddy fields and cleared lands.
    • Mainly matured oil palm with some rubber and scrub or secondary forests.
  2. Forest
    • Primary forest
    • Swamps, Marshlands and wetland forests and coconut (on the coastal plain)
    • Mainly scrub or secondary forests with some orchards, mature and sensile rubber
  3. Non-Agricultural Areas
    • Hihghways and main roads
    • Urban, mining & quarries, exposed soils and other built-up areas.

Proposed Flowchart of Land Use Survey Procedure for Malaysia Based on Satellite Data (Modified from Bauer et.al (1977)