GISdevelopment.net ---> AARS ---> ACRS 1997 ---> Forestry

Detecting Tropical Deforestation Using Satellite Radar Data: A Case Study From Central Sumatra, Indonesia

Belinda Arunarwati Yousif Ali Hussin Michael Weir
The International Institute for Aerospace Survey and Earth Science (ITC)
7500 AA, Enschede, The Netheralnds
Fax (31)53-4874-399
E-mail : HUSSIN@ITC.NL


Abstract
Indonesia is the most richly tropical forested national in southeast Asia. Over most of the country, forest land use is regulated by a system of land assessment known as TGHK or "consensus of forest land use". The key to wisely managing forest land and its resources in information. In the case of Indonesian deforestation, information is required not only about the rate and the extent of deforestation, but also about some other related information such as the presence of deforestation in relation to THK classes, its location within each class of TGHK, and also its type. Data derived from remote sensors are increasingly being utilized as a data source in GIS. Conventional methods of remote sensing using optical system have failed in some part of Indonesia due to cloud cover. Radar, being free from the time and weather restrictions may be a useful alternative sourced of the remote sensing data in Indonesia. The main objectives of this research was to investigate the potential of satellite radar data to detect, differentiate, and classify deforestation.

Introduction
It is generally recognized that forest resources should be sustainably managed to support the needs of present and future generations. The capability of the land to produce on a sustainable basis is, however, loggers want to fell trees for timber, and environmentalists want conservation and protection. The problems of deforestation and environmental degradation as a result of the activities of the former actors (farmers and loggers) are become alarming.

Deforestation is expressed by the rate of change of forest cover area caused by changing the use of forest land from forest to non forest. Several factors are directly responsible for the forest change, such as commercial logging, pasture, colonization programs or spontaneous migration, slash-and -burn agriculture, expansion of agricultural activities and other land use practices such the logging activity. It has been suggested that more than 80 percent of deforestation can be attributed to agricultural expansion (De gier, 1995). Logging, indirectly contributes to this major cause of deforestation by providing to farmer through the construction of timber extraction roads.

Indonesia, a part of Malaysian botanical region, is the richest tropical rainforest in the world. Indonesia's forest represent 10 percent of the world remaining tropical rainforest. Of its total land area of 193 million hectares, about 144 million hectares or nearly 75 percent are classified as "forest land". More than 95 percent of those forest land are outside java island (80 percent in Sumatra, Kalimantan, and Irian Jaya). A result of growing population pressure issues of land use have becoming increasingly important in Indonesia. On Java, the encroachment of landless farmer into upland forests have led to increasing soil erosion. Similar issues arise in the outer Java, where the conversion of forest land to agriculture use, is compounded by the commercial exploitation of forest resources.

In order to mange the forest land in sustainable way, the Government of Indonesia instigated classification the consensus of Forest Land Use of TGHK (Tata Guna Hutan Kesepakatan). Essentially, this consensus classifies the forest land into 5 categories according to their usage, namely: protection forest, convservation forest including national parks and reservation forests, limited production forests, permanent production forests, and convertible production (conversion) forest. A number of important issue related to spatial information such the location of deforestation within TGHK classes, and the deforestation pattern distribution, are worth investigation to support the decision maker's objectives in order to realize the sustainable forest management in Indonesia.

Remote sensing is an important source of spatial data. Remote sensing systems can be divided into active sensors (using their own energy), and passive sensors (using natural (sun) energy). Radar is an active sensor which transmits and receives a microwave signal. Some advantages of the use of radar in remote sensing are that it can be used during both day and nith, has all weather capability (cloud cover, rain, fog, atmospheric dust, etc.), and its energy has th ability to penetrate through some surficial features. Radar is expected to be used for a wide range of applications in forestry, including activities in forest management, monitoring of deforestation, shifting cultivation, colonization settlements, and land use change. The need for reliable information about deforestation within different forest land use (TGHK) classes, such as location, distribution, and probably also its types, is obviously important to monitor the deforestation occurring in relation the TGHK. Most tropical countries use satellite imagery as a source of data about their forest resource . Landsat TM and SPOT satellite image are the most important source, but problems due to lack of image quality as result of cloud cover occur in the tropical areas like Sumatra island in Indonesia (Sumatra is notorious area for cloud cover.) This problem can be solved with the use of radar images as data source. Radar can be used as an effective tool for detection the particular pattern of human activity such as logging, agriculture expansion, (shifting cultivation ) and illegal settlement in the tropical forest which objectives of this research was to investigate the ability of satellite radar data to detect, differentiate, and classify deforestation.

Study Area
The study area is located approximately in 01o25'00" to 01o45'00" Latitude South and 102o15'00" to 102o45'00" Longitude East, lies in jambi Province, Bungo Tebo Subdistrict (Kabupaten Bungo Tebo), between Kota Jambi and Gunung Kerinci, in the central of Sumatra island, Indonesia. The study area covers about 50 km x 35 km(1750 km2), and comprises a mountainous nature reserve area in the west and timber concession areas in the east. Apart from timber concession, the study area also contains the orginal Jambi villages along Batang Tebo, Batang hari, and Batang Tabir rivers, transmigration villages (Kuamang Kuning), and oil palm plantatin/oil palm estate.

The forest concession (HPH) has been selectively logged by PT Sylva Gama (30.000 ha). Part of the concession will be clear felled to establish an Industrial Forest Plantation (HTI). Part of the forest has been allocated for research on forest regeneration and management, and for education purposes for Faculty of Forestry Gadijah mada University.

According to the forest land use planning (TGHH) map and forest vegetation and Land cover maps of Jambi (at scale 1:25.000), the TGHK classes of study area are only two: Production Forest (HP), and conversion Forest (HPK), also one Other Land Use (APL). While land cover of the study area are Unproductive dryland (lktp), Agriculture (Ptn), and Lowland forest (Hr). All the classification above are only on the map,, the reality in the field might be not exactly same as those above classification. Based on the ground verification had been done, there are seven broad classes (main classes) of land cover types and land use combination, in the study area. Each class also has several sub classes. These are: forest classes including: logged over forest, old secondary forest, young secondary forest; rubber classes including: rubber plantation, untreated rubber plantation, jungle rubber, rubber agro forestry; oil palm plantation classes; rice classes including: wet rice field, dry rice field; clear felled classes including: buth fallow, lading and fallow; water classes; other classes including: temporary agriculture land, dry agriculture land, built up area, home graden, alang-alang grassland, grazing area.

Materials and Methods
The following data were used for this research project: Landsat-5 TM data of September 15, 1993, Spot XS data of March 21, 1993, ERS-1 images of October 17, 1993, June6, 1994, and July 7, 1994, and JERS-1 of August 16, 1993. Research methodology is illustrated in Figure 1.


Figure 1. Detailed Research Methodology

Results and Discussions
The wavelength of the radar system involved has a singnificant affect on the depth of penetration of the radar signal into a forest canopy, and therefore on the resulting backscatter. Different wavelengths( different backscatter) will give different interpretation results. The capability of radar to penetrate the forest canopy or surface layers in increased with the longer wavelengths, as can be seen clearly when both singly L-and C- band are interpreted visually. Actually, both of them can be used for detecting the clear cut as one of the present deforestation type in the study area. Howevery, in the dase of other type of deforestation (rubber) ERS-1 radar image with the shorter wavelength(C-bands) can not distinguish those rubber from the forest, while JERS can. Table 1 shows that JERS-1 image the L-bands can recognize more classes than the ERS-1 with C-bands. It should be noted that ERS-1 with shorter wavelength is better for detecting the settlements and oil palm plantation. The settlements are relatively clearly interpretably by shorter wavelength, because of more backscatter from the roof of houses or other buildings (corner reflection). For the JERS, the settlements are not visible, because the backscatters not noly comes from the surface (roof) of the buildings, but also come from the soil. Thus, there is not enough appearance of corner reflection. Oil palm can also be detected more clearly by the shorter wavelength. In the forest, the C-band (5.7 cm) radar signals will be reflected by the small and medium size branches in the canopy, and the incoming L-band (23 cm) radar energy will be reflected by the large branches and poles (tree trunks) of the trees. When interpreting radar surface or foliage of the canopy. Similarly with the oil palm plantation, where the plants are planted in rows, the backscatter from the canopies is affected by those rows (line pattern).

Table 1. Differences of JES-1 and ERS-1 radar images visual interpretation
JERS (Lband) ERS (C band)
Recognize 11 classes Recognize 8 classes
Can separate old secondary forest and young secondary forest Can separate old secondary forest and young secondary forest
Can distinguish rubber from forest Can distinguish rubber from forest
Not good to distinguish the settlement Good to distinguish settlement
Can distinguish annual crop from agriculture land Can distinguish annual crop from agriculture land
Can detect clear cut Can detect clear cut
Plantation pattern of oil palm is not so clear Plantation pattern of oil palm is not so clear

JERS can separate the forest into three classes (log over forest, and even old secondary and young secondary forests). Texture, tone, location and association are the most important image elements used for those detection. Similarly with the agriculture area, using the same image elements, JERS can distinguish the agriculture area into agriculture with other vegetation, with houses, and agriculture with the annual crops.

The angle between the radar beam and a line perpendicular to the surface (Hoffer, et al., 1995), defines the relationship between the incoming radar singnal and the actual slope of the ground. Therefore, the slope of the ground plays and important role. However, because the slope of the study area is relatively flat (horizontal surface), the effect of incidence angle is considerably uniform over the whole image.

The effect of incidence angle may only be recognized from the clear cut deforestation. Clear cut are difficult to distinguish from forested areas in incidence angle less than 30o, but nore clear in incidence angle more than 30o (Hoffer, et al., 1995). Clear cut can be seen in both, but in ERS (with incidence angle 23o) clear cut is more difficult to distinguish from the forest, while clear cut in JERS (with incidence angle 35o) can be seen more clearly. This is based on an assumption that the attenuation of microwave energy will increase with the increasing incidence angle. At higher incidence angle, the microwave will have to travel longer distances to reach the surface target, and consequently will lose the energy. Therefore, for the same object (clear cut), the smaller incidence angle (ERS) will give more bright appearance compared with the higher incidence angle (JERS).

Because both JERS and ERS have "Like Polarized" (VV for ERS) and (HH for JERS), it is believed that polarization makes no difference between them. Even during the interaction between the radar energy and the surface, the polarization will be modified based on the property of that surface.

Radar interpretation can be separated into two approaches: (1) visual and (2) digital. Visual interpretation is still the simplest and the most powerful approach to classify the radar imagery in which always difficult to be interpreted by the presents of the speckle noise. However, classification needs the knowledge about the study area obtained by field data checking. The image elements used for visual interpretation are tone, texture, pattern, location and association. These last element was the most important element. Some tone or texture differences are detectable from the backscatter differences. However, without the knowledge about the area, those detectable and delineable areas are almost impossible to the known.

An alternate process of classification is digital image processing by supervised classification. This classification mainly involve the computer and again also the prior knowledge of the area. Actually the visual interpretation of radar imagery is not pure manual step of image classification, because, before the radar images are interpreted visually, some processing by computer was also done, (image subsetting, georeferencing, and filtering), in the same way as for the digital image classification approach by supervised classification.

Table 2. Comparison between two data sets (DS1 and DS2)
First Data Sets (DS1)(ers1:red, ers2: green, ers3:blue) Second Data Sets (DS2)(jers:red, ers2:green, res3:blue)
Recognize 6 classes Recognize 7 classes
Can separate old secondary forest and young secondary forest Can not separate the forest into log over forest and secondary forest
Can not recognize oil palm, because mixed with the forest itself Can see oil palm separate from forest, but can not distinguish it from rubber
Can recognize the wet area Can recognize the wet area
Clear cut can be seen but bit difficult to detect Clear cut easy to detect
Rice is more easy to detect Rice can be seen, but not so clear because mixed with water and wet area
Agriculture mixed with the tree (rubber and forest itself) Agriculture mixed with the tree (rubber)
Can not separate the forest stand from another perennial tree like rubber Can not separate the forest stand from another perennial tree like rubber

Table 3. The comparison of the result of both classification approaches
Some remarks Visual Interpretation Digital Image processing
  Single JERS Single ERS Data sets with JERS and ERS Data sets with ERS only
Number of classes 11 8 7 6
Separate log over and secondary forest yes yes no no
Separate old and secondary forest yes no yes no
Separate rubber from forest yes no yes no
Detect clear cut yes yes yes Yes but not so clear
Separate oil palm from forest Yes, but not so clear yes yes No

Reference
  • De Gier, A., 1995, Your 7,000 Square Meters. Inaugural address, International Institute for Aerospace Survey and Earth Sciences (ITC), Enschede, The Netherlands.
  • Hoekman, D.H., 1990, Radar Remote Sensing Data For Application in Forestry, Ph d. Thesis, Wageningen Agriculture University, The Netherlands.
  • Hoffer, R.M., Maxwell, S., Ochis, H., 1995, Use of Radar For Forestry Applications, Colorado State University, fort Collins, Colorado.
  • Hussin, Y.A. and Shaker, S.R., 1995, Tropical Rain Forest Land Use Detection Analysis Using Remotely Sensed Data and GIS: A Case Study From Sumatra, Indonesia, Annual conference and Exposition Proceedings, Volume, 1, 1995, Nashville, Tennessee.