Land Use-Cover Change Detection Using Knowledge based approaches: Remote Sensing and GIS
Mariamni Halid
Kalaysian Centre for Remote Sensing (MACRES)
CD 100, Tth Floor, City Square Centre
50300 Jalan Tun Razak, Kuala Lumpur Malaysia
Tel: (603)-2645640 Fax: (603)-2645560
E-mail: ipa@macres.gov.my
Abstract
Knowledge-based approaches were used to link the field data, local knowledge and the spectral land cover classes to generate the land use change. Field knowledge and spectral land cover was formalized into a "belief" factor. In this method, the maximum values of "belief" factors from land cover and land use change were calculated to detect land use changes. Land use in years 1 and 2 was automatically detected using this approach. If land cover classes were grouped overall accuracy of 78 per cent was reached for 1995. The knowledge-based classification and maximum likelihood classification of 1995 gave lower mean overall accuracy of 44 per cent. However, knowledge-based classification has advantage of quick classification and les field work than maximum likelihood classification.
Introduction
At the urban fringe, complex areas of land cover change are often found, including transformations from rural land uses to residential, commercial, industrial and recreational uses. These change can be monitored using remotely sensed data (in combination with ground survey), either by photo interpretation, enhanced false-colour composite imagery from different dates or by digital analysis of the imagery using change detection techniques (Quarmby 1989). For large areas, however, to the ground surveys and aerial photo-interpretation are impractical due to large amounts of data to be collected (in space and time). On the other hand, aerial photographs in malaysia generally have a 'restricted' status and are classified under military control, and access to them is difficult. Against this background, the potential of other forms of remote sensing needs to be explored. In particular, data from Landsat Thematic Mapper with its spatial resolution and repeated coverage, appear to offer possibilities and potential for land cover map production and monitoring land cover-land use change.
Research has shown that mapping of land cover often is significant by improved using of combination of different techniques (Manier et al. 1984). One example is to use knowledge-based approaches which utilize additional geographical data beside satellite images often with great success (Hutchinson 1982, Peddle and Franklin 1992, Bronsveld et al 1994). Janssen and Middeloop (1992) show used a knowledge-based approach for crop classification of a Landsat-TM image, reported classification improvements of 6 to 20 per cent compared to maximum likelihood classification alone. With the knowledge-based approach interrelationships are
Formalized into a set of rules, and individual pixels are classified by determining the chance/probability of a certain land cover type. Probabilities can be assigned using artificial intelligence or simply by applying a set of logistical decision rules (Shafer and Logan 1987, Srinivasan and Richards 1990).
The main objective of this research is to assess automatic knowledge-based approaches in digital land use change by integrating remote sensing and a geographic information systems. This objective could be achieved through:
-
spectral classification of multitemporal images in order to produce land cover maps;
- derivation of land use information from land cover classes by using the out dated land use maps supported by the field knowledge expressed in terms of rules and,
- combination of knowledge on land cover and land use change in order to produce land use change map.
The approach used in this study to detect the land use-land cover changes involved digital change detection and geographic information systems. The study will assess the capability of knowledge based change detection approach which less dependent on field observations and secondary data.
Study Area and Materials
The study area selected eas Rawang sub-district and surrounding areas of Ulu Selangor, Selangor, This area is located within corrdinates of latitude 3o 16' N and 101o 28' O" E longitude and 3o 27' 45" N and 101o 39' 30" e. The total study area is 441 km2. Annual precipitation in this area approximately 240 cm.
Precipitation occurs in two seasons, mainly in April to May and September to November. The relative humidity is 80 per cent due to high temperature and high rate of evaporation. Mean annual temperature ranges from 22.8oC to 33.8oC. The areas is about 87,000. The ethnic group consist of 34 per cent malay, 34 per cent Chinese, 27 per cent Indian and 5 per cent others. In this areas land use-land cover is composed of forest, rubber plantation, oil palm plantation, grass land, closed mining, build up area, homestead garden and village.
De Bie et al., (1995) defined land use as a series of operations on land, carried out by humans, with the intention to obtain products and or benefits through using land resources. Land use is captured by describing the land use purpose (s) and the operations sequence. In this study land use refers only to the purpose (in term of crop or cover types) but not include the operations. The classes are mixture of land cover and land use, they will be referred to as land use or land cover-use.
The materials used were Landsat TM dated 17 April 1988 and 14 October 1995 (path/raw - 127/58), Topographic Map 1994 scale 1:50 000 sheets no 3658 and 3758, Land cover map 1991 scale 1:50 000, Land use Map 1988 scale 1:125 000, ILWIS 1.4.
Methodology
In this study geometric correction was done for the image 1988 based on topographic map sheets no. 3658 and 3758. Image to image registration was done in order to register the image 1995 with geocoded image 1988 (master image) Keeping the root mean square error (RMSE ) less than 0.5 pixels, an image to image transformation model (third order)
polynomial) and nearest neighbour resampling was calculated and the slave images were resampled for its pixel by pixel correspondence with the master image. Image enhancement was applied to improve the appearance of images for human visual analysis. For that purpose band selection using Optimum Index Factor (OIF) were use for images 1988 and 1995. All the selected bands were used to produce false colour composite.
