GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Agriculture and Forestry

The use of Landsat TM data to estimate rubber growing area of Thailand

Dr. Suvit Vibulsresth
Dr. Darasri Downreang
Dr. Surachai Ratanas ermpong
Miss. Supapis Polngam
Mrs. Thanomsri Rangsikanbham
Miss Uthaiwan Pornprasertchai

Remote sensing Division, National Research Council

Dr. Likit Nualsri
Dr. Veth Thainukul
Mr. Somyot Sinthurahat
Mr.Udorn Charoensang
Mr.Sutus Danskulphol
Mr. Somjert Phatummin
Rubber Research Institute, Department of Agriculture Bangkok 10900, Thailand


Abstract
A certain part of Nakhon Si Thammarat province, Southern part of Thailand was selected as a test site for the application of simple image processing techniques to identify rubber plantation stages by the use of Landsat Thematic Mapper data. These images were taken during the shedding period of rubber trees.

To produce optimal contrast and variation for color composition, contrast enhancement and histogram equalization techniques were applied to the images with various band combinations. It was found that a combination of band 4, band 5 and band 2 provided the best identification of rubber growing area. This combination was most promising for the separation of rubber plantation from other vegetated area as well as differentiation of rubber of different stages of growth such as young rubber area including nearly cleared area that will be planted with rubber area including nearly cleared nearly cleared area feature and mature rubber area. The results reveal that within the total area of 1,090 sq. km. Under study young rubber plantations with rubber trees 1-4 years of age account for 162. 139 sq. km. While 241.986sq.km. are mature rubber with rubber trees 5-9 years of age, and 162.500 sq. km. are mature rubber with rubber trees more than 10 years of age. The accuracy of classification is more than 80% for rubber plantation area.

Introduction The study described in this paper is a part of an integrated approach to develop procedure for rubber plantation area estimation particularly for Southern Thailand, Where rubber are mainly planted. From past experiences, multispectral characteristics of rubber plantation and its separability from other cover types, has largely been studied with respect to MSS data. The information in the four MSS bands can essentially represent only two classes, one of which, termed "young rubber area", and the other is mature rubber area (Somyot, 1986)X. In addition, rubber plantation area is subject to confusion with oil palm plantation area.

With the development of higher spatial and spectral resolution satellite data such as Landsat TM the possibility to differentiate various land cover types becomes greater. Therefore, in this paper an attempt is made to explore the potential of TM data as applied to the identification of rubber plantation at different at ages via digital image processing means
  1. The study area

    The study area covers approximately 1, 090 sq.km and is located in Nakhon Si Thammarat province in South Tailand (Figure) It is generally a gently undulation landscape form with limestone feature. The major cultivation are rubber plantation which it is typical for the South, and oil palm plantation located throughout the area.




Data acquisition
  1. Satellite Data
    The Landsat TM Digital image (in Computer Compatible Tape (CCT) format) that is cloud free and coincides with the shedding season of rubber trees was selected from the existing archives at TRSC (Thailand Remote Sensing Center). The CCT product is of geocoded format for ease in geographical orientation, and correlation with topographic maps. The Geocoded scence was acquired on March 1990.

  2. Ground data
    Information on ground cover and stages of rubber growth were obtained for most part of the study area. The information was collected in may through July with assistance from the Rubber Research Institute at 12 training sites. Each training site represented different stage of rubber plantation (see Figure 2). This information was used for regional rubber plantation signature identification in preliminary step of mapping. Additional ground data collection was also made to check the accuracy of mapping.


Data analysis
Analysis of landsat Tm image was performed on the meridian Mapping system located at the TRSC. The first step was to assist in visualization and discrimination of spectral and textural features on the image. False color composite of bands 4, 5 and 2 (in Red, Green and Blue) was generated with histogram equalization and linear contrast enhancement. Using this combination, the distribution of rubber plantation at different stages could be clearly identified from the coloring of orange, pink and light blue (see Figure 2). Table 1 describes the correlation between color and surface pattern.

Table 1 The correlation between color and surface pattern.
Color Description
Light blue Orange Pink young rubber plantation; 1-4 years of age Productive rubber plantation; 5-9 years age mature rubber plantation, more than

Then classification was made from analysis of digital values for selected Classless in the study area. Representative data sets of training sites were chosen to correspond with area in which ground information was available. The spectral signatural files were constructed for all classes and were used to classify the whole study area by using nearest Neighbour method.

Results
  1. The results of classifications

    Classification of various rubber plantation stages are given in Table 2 and Figure 3, The results for rubber area estimation within the study area appear to be consistent with the reported rubber areas. However, for this area, there were some difficulties in separating young rubber plantation at 1-2 years old from young coffee plantation or other young vegetation due to similar color and texture.

    Table 2 Area estimation of the various stages of rubber with in the study area
    Class No. Description Area in Sq. Km. Percent
    1 Young rubber plantation with rubber trees 1-4 years of age 162.139 14.88
    2. Productive rubber plantation with rubber trees 5-9 years of age 241.986 22.20
    3. Mature rubber plantation with rubber trees more than 10 years 162.500 14.91
    4 Unidentified or non rubber area 523.375 48.01
      Total study area 1,090.000 100.00

    But, young rubber of more than 2 years could be clearly classified. Post classification filtering technique, which was used to assimilate the smaller groups of pixels into larger mapping units could help remove much of the speckle effect often found in the unfiltered map.

  2. Classification Accuracy The results of the digital classification were found to be comparable with ground data, the existing rubber plantation map and also report from the rubber experimental station. Classification accuracies for rubber plantation areas using landsat TM data which are shown in table 3 indicate the high degree of spectral separation of rubber plantation areas.

    Table 3 Classification accuracies (%) using Landsat TM data, based on bands 4,5and 2
    No. Young rubber % Productive Rubber % Mature Rubber %
    Young Rubber productive Rubber Mature Rubber Unclassified pixel 90.2 0.0 0.0
      0.0 90.5 0.0
      0.0 0.0 89.5
      9.8 9.5 10.5
Conclusion
This paper describes how computer aided techniques can be used for analysis of rubber plantation at different stages. It is essential to select proper spectral bands and dates of satellite data. These was good evidence that rubber plantation at various stages could be distinguished from all other vegetation on March image, using bands 4, 5 and 2 combination. Such the period has an advantage that the reflectance of rubber leaves are more or less predictable. During February through April, rubber leaves change color to red orange before shedding. With in this period the reflectance of rubber leaves are expected to be relatively high. In addition, the nearest neighbor classification seems to work well for rubber identification providing reasonably high accuracies.

Acknowledgements
The authors wish to thank the Rubber Research Institute, Department of Agriculture and the National Research Council for supports provided for the project.

References
  1. REEVES, R.G. ed. "Manual of remote Sensing " , vol.2, American Society of Photogramentry , 1975

  2. ROBERT A. SCHOWENGERAT, "Techniques for Image Processing and Classification in Remote Sensing", University of Arizona, Tucson, Arizona, 1983.

  3. SAWIN, PHILIP H.: DAVIS, SHIRLEY M. ed. "Remote Sensing: The Quantitative Aproach ", Mc Graw -Hill, 1987.

  4. SOMYOT and staff , " Survey of Rubber Growing Areas in Tailand Using Landsat MSS Data , 1987