Decision of Threshold Values for Extraction of Mangrove Area with TM Data of Landsat 5
Kazuhiro Sato*, Minoru Nakajima**, and Takashi Hoshi***
* : College of Agriculture, University of the Ryukus,
Senbaru 1, Nishihihara, Okinawa 903-0129, Japan
Tel. 81-98-895-8792, Fax. 81-98-895-8734,
E-mail : sato4408@agr.u-ryukyu.ac.jp
** : Omodaka Electronics Co. Ltd., 101 Column Otsuka,
Ostuka 4-43-9, Bunkyou-ku, Tokyo 112-0012, Japan
Tel. 81-3-3944-8661, Fax. 81-3-3944-8972,
E-mail : omodec@mub.biglobe.ne.jp
***: Faculty of Engineering, Ibaraki University,
Nakanaarisawa 4-12-1, Hitachi, Ibaraki 316-8511, Japan
Tel. 81-294-38-5133, Fax. 81-294-37-2223,
E-mail : hoshi@cis.ibaraki.ac.jp
Abstract
Mangrove area is an important coastal ecosystem in the tropical and sub-tropical coastal regions. One of the characteristics is to have both almost similar crown layer with land common forests and quite different forest floor with them. The floor is flooded above every high tide times and water content of the floor is almost saturated even low tide times. One mask procedure had been suggested for effective extraction of mangrove area using TM data of Landset 5 to pay attention to the matter.
In this paper, it was examined how to decide the threshold values without drawing of profiles of CCT count. One empirical and practical way was suggested because statistical characteristic is not still clear on the CCT count values of pixels distributed along the boundary between mangrove area and land area or water. A systematic decision of the threshold values and preparation of masks become possible by addition of some devices to existing software.
Introduction
We have examined qualitative and quantitative simultaneous expression of information on mangrove forest using remote sensing technics4.5). It is assumed that qualitative information are forest type, operation type, species, mixture ratio and so on, and quantitative information are tree height, diameter at breast height (DBH) and stand stock or biomass. Some classifiers are useful for qualitative classification and regression analysis is useful for quantitative estimation. One of the authors and other core searchers reported relationships among TM data of Landsat 5 and stand parameters on mangrove forest1.2.3). As it is deserved there is not so large difference among CCT count of each pixels next to each other is relatively smaller. For band 4 and 5, such value becomes large for the pixels distributed both sides of the boundary between mangrove area and land area or waters. This matte should be the key to decide threshold values between mangrove area and land area or water. But the statistical examination on the matter is just started. It will take more time to clear this problem.
In this paper, we examined and suggested illogical but practical way to search and decide fit threshold values. It depends on fast processing quickness of a personal computer system and our judgment.
Method
We used a set of TM data of Landsat 5 without band 6 acquired on 25 April 1990. It covered Mandah area distributed mangrove forest in this mouth of the Inderagiri River, Sumatera.
In the system for image or remote sensing processing, frequency distribution curve of CCT count is drawn frequently. Two curves for band 4 and 5 were shown in fig.-1. There are two clear large peaks on the curves of band 4 and three peaks for band 5. For band 4, tapering left peak is a group from water and right peak is a group from mangrove area and land area. In this band there is no clear partition between mangrove area and land vegetation. For band 5, the highest and narrow left peaks comes from water, higher central peak dose from mangrove area and small and narrow left peak dose from land area. A small notch is observed between mangrove area and land area.
Fig.-1 Frequency distribution curves of CCT count of band 4(left) and band 5 (right), Mandah area, Sumatera, Indonesia
