Land Use Dynamics from Multi-temporal Remotely Sensed Data: A Case Study Northern Thailand

Suthinee Dontree
Department of Geography, Faculty of Social Sciences, Chiang Mai University
239, Huai Kaew Road, Amphoe Mueang, Chiang Mai 50200, Thailand
[email protected]

1. Introduction
The land conflicts in northern Thailand has increased since the past. The causes of the conflicts came from the government policies to give forest concessions to private and governmental companies, promulgation of national reserve forests, watershed classification and classification of national reserve forests. These policies have provoked unequal access to natural resources and illegal land occupation despite the fact that several households have already occupied upland and highland areas for field crop cultivation before the quoted enactment. During the last 20 years and especially after Thailand’s economic collapse, there is significant increasing rate that local middlemen, landless farmers have encroached more and more into forestlands.

The study area is focused at the agricultural areas of Tho Saman village in Phrae province, North of Thailand covering an area of 292.3 km2. The village was established over 100 years ago at the vicinity of the Song national reserve forest, enacted in 1973. The majority of the villagers’ agricultural plots are situated in the national reserve forest, considered to be illegal predicament.

The topography comprises of three landform types. The first, the lowland area or the Song lower flood plain, with an elevation less than 240 MSL. It is situated in the south and southwestern parts of the Song reservoir. There are two agricultural types in the lowland: irrigated and rainfed. The second, the uplands or undulating hills are defined as the north-eastern and south-western parts of the reservoir, going from the flat flood plain until the nearest mountain ridge before the Nam Sai Song valley, these areas lie at elevations between 240-500 MSL. The major land use types are mixed deciduous forest and upland field. The last landform type is the highland area, defined as the area east of the upland area, located on the north-eastern parts with an elevation of 500- 1,200 MSL. The highland area is divided in several parts by narrow valleys of many small streams. Forestlands are major land use composing of mixed deciduous and dry dipterocarp.

The objective of the study is to apply remotely sensed data to investigate land use dynamics concerning the land utilization in the study area in each period from forest concession to modern market-oriented agriculture effected by some government strategies on forestlands and the construction of Song reservoir. The reservoir construction has severely altered the conditions for agriculture in the Song watershed. Since the dam construction obscured the old road to the highland agricultural fields and increased the possibility for intensive farming due to a larger irrigated area downstream from the reservoir, large changes in land use have occurred. The following analyses will quantify the changes that can be observed primarily with the use of available aerial photos and satellite images. Finally, the observed changes will be analysed in combination with other sources of data in an attempt at offering causal explanations for the changed land use.

2. Data and sources of data
The data are composed of:

1. Satellite imagery: Landsat MSS (01-02-1977), Landsat TM (11-01-1989) and Landsat ETM+ (14-03-2000);
2. Aerial photos in 1989, 1991 and 1996, scale 1:50,000;
3. Field survey and GPS coordinate registration of land use samplings in April and May 2001;
4. Semi-structured interview in May 2001 for land use history;
5. Mae Song watershed GIS database established in 2000 by the author at the Geo-informatics and Space Technology Center (Northern region), Faculty of Social Sciences, Chiang Mai University.

3. Methodology

1) Preprocessing
Three Landsat satellite images were rectified to UTM coordinates using a topographic map of the area. The 1989 image was rectified an UTM topographic map, the remaining two were rectified to this image on an image-to-image basis. All image manipulation was made using the CHIPS (Copenhagen Image Processing System) software package (figure 1). Channel 1 in both the 1977 and 2000 image have some noises on the image. The 1977 image was polluted by repeated scan lines, whereas the signal in the 2000 channel one was obscured by haze, possibly smoke from vegetation fires during summer time of the North (Dontree et al., 2002). False color composite are

Then, all three dates of the aerial photos are converted to digital format using a scanner. The 1996 aerial photos are rectified to UTM coordinates referred from a topographic map. The 1989 and 1991 aerial photos are geocoded to the 1996 rectified photos by image to image method. All geometric rectifications are processed by IDRISI 32 Release 2.

Figure 1 Landsat satellite imageries of the study area 2) Visual interpretation of the rectified aerial photos
The main disadvantage of relying on aerial photography in Thailand is the very low time resolution as sometimes there may be a gap of over five years between photos. For northern Thailand the most recent data set is from 1996, and the previous data set was from 1989. Thus rapid changes in land use are very difficult to study using this method. However, 8 similar land use types: paddy, orchard, field crop, disturbed forest, dense forest, village, construction area of Song reservoir and water body have been categorised for the photographs from the three dates. The aerial photographs were interpreted using three main criteria, grey scale, textural and proximity factors, grey scale relates to colour variations. In the upland areas dark grey areas with a smooth texture that tended to be located far from cultivated areas were classified as dense forest. Lighter grey areas with a rough texture that tended to be located close to cultivated areas were classified as disturbed forest. Light grey areas with a smooth texture that tended to be situated along the valleys were classified as field crops. All three classified data are converted to digital data by a scanner and re-rectified to UTM coordinates. The results were verified during the second ground truth campaign in May 2001 (figure 2)

Figure 2 Aerial photo interpretation in 1989, 1991 and 1996

3) Satellite image classification
The maximum likelihood classification is selected to distinguish land cover/land use. As the satellite data were acquired within different seasons and from different platforms and having different resolution. The Landsat MSS, 80 meter resolution, was acquired in on February 1, 1977, the Landsat TM, 30 meter resolution, on January 11, 1989 and the Landsat ETM+, 30 meter resolution, on March 14, 2000. These three months have different vegetation conditions according to the climate. In Northern Thailand, January is the month of cool season where moisture in soil and forest still kept after rainy season. February is an intermediate month between the cool season and summer where moisture becomes to decrease. And March is the beginning of summer where moisture is lesser than the two previous months. In addition, according to farmers’ practice, since the end of February, they start to prepare their fields by burning residual vegetation, and some collect wild vegetation that give new leaves after being burnt. As consequence, there is a lot of haze in the 2000 Landsat ETM+ from field burning and forest fires. It is difficult to distinguish burnt agricultural fields from burnt forestlands as they become similarly burnt bare soil. These factors give effects on vegetation status causing reflectance values of land use quite difficult to compare. However possible similar nomenclatures are set up based on physical characteristics of land use despite different number of land use types. There are 13 classes for MSS, 15 classes for TM-5 and 17 classes for ETM. The classified images were regrouped into 6 major classes for Landsat MSS and Landsat TM-5 and into 7 major classes for Landsat ETM+. (figure 3)

Despite the different vegetation status among the three satellite images and a lot of confusion among several land use types from one image to the others, we can identify land use dynamics of the study area well comparable with the results from the interpretation of aerial photos

Figure 3 The results of image classification 3. Change detection and discussion

The land use pattern along several periods has had the same characteristics. Dense forest is the major land use, usually along steep slopes in inner highlands in the eastern parts of the watershed. Agriculture is the second largest land use type after dense forest. The major crops identified by area were paddy, field crop and orchard, respectively. Paddy is located in rainfed and irrigated lowland at altitude less than 220 MSL. Orchards or mixed fruit trees are normally situated near to the lowland villages. Field crop comprises of two major cash crops: maize and cotton. These crops are scattered along the upland and highland valleys at altitude more than 220 meters. Disturbed forest is the third major land use, located near to the cultivated fields in uplands and highlands, along stream valleys where logging concessions were allowed (1973-1989). After the construction of Song reservoir, water body increases respectively since 1995.

It is difficult to use the results of image classification to investigate land use changes because of the class confusion and uncertainty of some land use types among the three dates. As the aerial photographs have more advantage on resolution, they are selected to compare land use changes. During the two periods (1989-1991 and 1991-1996), the land use dynamics has an identical trend (figure 4 and table 1). Dense forest was changed to disturbed forest and field crops and, some disturbed forests were altered to field crops but at a decreasing rate. It is remarkable that the increase was most profound in the first two years of the seven-year period. The borderzone in-between the uplands and lowland were the most affected. These land use types have common physical and geographical inter-connections, as an area increase in one type of land use category will be associated with a area decrease in another land use category. The changes happened in the areas with soil fertile enough to grow crops (normally along highland valleys and low and high terraces), less steep slopes, and close to former field crop areas. The diminishing rate of decreased dense forest and the increased disturbed forest and field crop could possibly indicate that land occupation of arable areas has reached its limit in the watershed.

Figure 4 Land use changes from 1989-1996

Table 1 Land use types and changes from 1989-1996 (area unit: km2)

Source: Visual interpretation of aerial photographs in 1989, 1991 and 1996.

The satellite data covers larger temporal space than the aerial photographs (1977-2000). During the period of logging concession from 1973-1989, there are traces of logging areas in upland and highland from Landsat MSS 1977, appearing in bluish color at the western part of the study area. From 1980 – 1990, land became more shortage for the poor in lowland, they have to expand their cultivated areas towards forestlands along small valleys following logging trails after tree cutting. Landsat TM 1989 depicts well this situation. Expanded fields appear as pale cyan narrow strips along stream valleys image in the eastern part. The lost of forest areas has continued until 2000. Despite the haze problem, Landsat ETM+ indicates more encroachment trend into former forestlands in inner upland and highland. At the same time, the advantage of Song reservoir, finished since 1993, also increases more irrigated areas in lowland. From our case study, The satellite images give more advantage in time scale to evaluate the land use dynamics since 1977 to present time. The problem of change detection technique should be considered to select data with similar environmental conditions as much as possible during each data capture, in order to facilitate comparable image classification. The resolution of Landsat images is still large comparing with the size of swidden fields, increasing confusion with surrounding environment. For the aerial photographs, they can provide more detailed information of land use but with limited time repetition. It is also time and budget consuming to rectify and mosaic many photos of several dates.

5. Acknowledgements
The author thanks Danced for supporting the research through SLUSE programs in Denmark, Thailand and Malaysia, the headmen of Tho Saman Mu 9 and 11 for logistic assistance, and the inhabitants of Tho Saman village for their kind co-operation and patience. A special thanks goes the Department of Geography, Faculty of Social Sciences, Chiang Mai University for the data processing.

6. Reference

• Dontree, Suthinee. 2000. GIS Database and Data Dictionary for Mae Song Sub Watershed Database, SLUSE Joint Field Course 2000. October 15 – November 2000, Phrae Province, Northern Thailand, Chiang Mai: Geo-Informatics and Space Technology Center (northern region), Faculty of Social Sciences, Chiang Mai University.
• Dontree, Suthinee et al. 2002. “Land Use Dynamics in Conservation and Economic Forest Zones: A Case Study of Tho Saman Village, Phrae Province, Thailand” in Problems of Sustainable Land Use and Natural Resource Management in a Community at Song Watershed, Phrae Province, edited by Catherine Helen Tranor, Sithinat Prabudhanitisarn, Peter Oksen, Suthinee Dontree and Christopher Saarnak, pp. 21-71. Chiang Mai: TUCED-SLUSE, DUCED-SLUSE and SLUSE-M.
• Ganjanapan, Anan. 2000. Local Control of Land and Forest: Cultural Dimensions of Resource Management in Northern Thailand. Bangkok: Amarin Printing and Publishing.
• Jensen, John R. 2000. Remote Sensing of the Environment, An Earth Resource Perspective. Upper Saddle River: Prentice Hall.
• Meyer, William B. and Turner II, B.L. (eds.). 1994. Changes in Land Use and Land Cover: A Global Perspective. Cambridge: Cambridge University Press.