FOREST MAPPING BY MEANS OF DIGITAL LANDSAT-TM DATA
Ali A. Darvishsefat
University of Tehran, Faculty of Natural Resources, Karaj, IRAN
Phone: +41 261 222045 / Fax: +41 21 9007988
E-Mail adarvish@chamran.ut.ac.ir
Commission VII, Working Group 3
KEY WORLD: Landsat-TM, Quantitative Analysis, Forest mapping, Plane forested land, Ground-
truth, Accuracy
ABSTRACT
In order to determine the potential of information from Landsat-TM data for generating forest map in
flat forested land of the northern Iranian provinces, a thematic mapper scene of Nour-Park
(Mazandaran) area dated Sept. 1988 is analysed. Qualitative and quantitative analysis on geometric and
radiometric errors of the TM data reveals existing error of banding but is insignificant (£1DN) in TM
band 4.
Because of plenty undesired pixels existing in TM band 7, it has been discarded in this study.
Geometric registration of the image data is implemented using a ground control point method. The
RMS error (19 meter) is calculated to be much less than a single pixel.
A forest extension map in the region, which has been prepared by ground surveying, is digitised, up-
date and used as ground truth. For the forest vs. non-forest classification a supervised method using
maximum likelihood (ML) and parallelepiped using TM bands and ratios is proposed. The best band-
set is organised employing Jeffrey Matuzita (Leiss, 1993) method. Highest overall accuracy is obtained
from the band-set TM 3 and 4 (95.14%). This high accuracy indicates that Landsat TM data have a
high potential in forest mapping in flat forested areas in the Caspian region.
1. INTRODUCTION
An essential condition for a correct planning
and decision making in natural resources
management, including forests, is to have true
information on relevant ground truth.
Generating and updating information on
location and extend of forests in large scales is
- not possible trough Fieldwork only since 1t 1s
demanding high costs and time. In addition it is
possible that the area of forests might change
during the compilation and publishing process
of the map.
In contrast, satellite data with their own
characteristics such as being able to cover large
areas, their high revisit frequency, their
constant spatial and spectral resolution, and
finally their possibility of automatic
interpretation has created a new potential in
forest mapping. A large number of researchers
profit from these data sources to make forest
maps under different forest conditions and
growth stadiums (e.g. Borry et al, (1990) and
Itten et al. (1993)). In Irans forests it is
convenient to study the ability of this data in the
form of an integrated plan in plain and alpine
forested areas in the northern and western parts
of the country, so that it will be possible to
derive automated methods for updating forest
maps.
2. DATA
The Landsat Thematic Mapper data available
for this investigation is a subsection of the
scene 168/39, imaged in August 1988. Because
of it's thermal characteristic and higher spatial
resolution, the TM band 6 is not used in this
study. The TM scene 1s system corrected to
level 5 data by the receiving station and has a
spatial resolution of 28.5 m. To geometrically
correct the satellite data, a topographic map of
the study area (Nour) in the scale of 1:25'000
is used to located the ground control points.
Another map of the forested areas of the Nour
Forest Park is also used as ground truth, after
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 223
