Evaluation of RADARSAT-1 for monitoring and mapping land use, land cover in Thailand
Date
2003
Authors
Filion, Rébecca
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Shrimp farming is an important part of the world aquaculture market. Thailand has, for the last 10 years, been the world's leader in the intensive shrimp culture industry. However, intensive shrimp farming has major social, economic and environmental impacts. It is therefore paramount to manage this activity in order to make it sustainable. In response to the Thai government's search for a cost effective approach to the regular monitoring of large areas, a project was started in 1998 using optical LANDSAT data to identify shrimp farms, as well as other land cover and land use. The high frequency of cloud cover, however, made regular data collection impossible. RADARSAT-1 imagery, on the other hand, has great potential for this type of application as the images are firstly highly influenced by the presence of water and, secondly, not influenced by clouds and other typical effects of tropical weather such as dust and haze.
In this thesis, RADARSAT-1 imagery was evaluated as a tool to monitor and map shrimp farms as well as different land use and land cover in the area of Changwat Chachoengsao, southeastern Thailand. Multi-temporal and multi-angle image combinations were used in order to augment the quantity of information usually available when only one image channel is used. Five fine beam modes (Fl, F4 and F5) RADARSAT-1 scenes were taken between May and September 2001.
Image pre-processing techniques such as radiometric correction, geometric correction, principal components analysis, speckle filtering and texture analysis were performed on all the images. Subsequent pixel-based classifications, unsupervised and supervised, as well as segmentation-based classifications, were performed on the pre-processed images. A comparison was carried out between possible combinations of multi-temporal and multi-angle classified images.
The best results were obtained using a supervised maximum likelihood classification of four (rice paddies, human settlements, orchard plantations, and all water bodies) and five (rice paddies, human settlements, orchard plantations water bodies, and shrimp farms) land use / land cover classes using multi-temporal mean texture components produced from the two images having steeper incidence angles : May 24 2001 (Fl) and July 11 2001 (Fl).