Forest change detection in a montaineous terrain using multitemporal 
Landsat TM images introducing a relative atmospheric correction. 
A case study : Forest of M'sila (Oran-Algeria-). 
Khatir Benhanifia & Mustapha Bouziane 
Laboratory of Remote Sensing 
National Centre of Spatial Techniques 
BP 13, Arzew 31200, Algeria 
ABSTRACT. 
Cartographic and statistical updating of forest maps highly affected by anthropozoic phenomenon are periodically undertaken by 
forests. 
In the past, the updating was done using aerial photographs but it seemed that this operation was a moneyconsumming because of 
the number of photos to be taken. 
At present, the spatial images are used as a powerful tool to undertake a multitemporal study to evaluate change leading to a forest 
map change. 
This paper presents a study done on the area called forest of M'sila situated in the west of Oran (Northwest of Algeria), using. 
multitemporal remotely sensed data such as Landsat TM images acquired on April 7, 1984 and on March 15, 1993. 
A preliminary geometric correction needed to the superposition of images of different acquisitions with a DEM was done; 
Furthermore, a relative atmospheric correction based on DEM derived data was performed; 
Finally, a forest change map was obtained integrating the results together with complementary information into a GIS. 
The results are satisfactory however, it is necessary to check them with ground truth. 
Since the study area is a mountaineous region, it is, however, very difficult to localise the control points necessary to the geometric 
correction. 
Hence, the results are not very reliable especially when the size of the point is less or equal to 900 m? ( case of Landsat TM). 
Consequently, it is necessary to take care of the geometric correction since the atmospheric correction needs the slope and aspect of 
the study area. 
1. INTRODUCTION 
This paper presents a method for detecting forest cover change 
using relative atmospheric correction of multitemporal satellite 
images. Forest change detection is one of the major 
applications of remotely sensed data and is useful in the 
analysis of environmental change such as land use change, 
deforestation and damage assessment. 
This approach is based on the estimation. of relative 
atmospheric effects between satellite images taken at different 
times, such as the radiance change arising from the difference 
in atmospheric conditions and sun angles, by taking one'image 
as reference; the comparison is a pixel to pixel. 
It is noted that the relation between radiance values of 
remotely sensed data acquired at two different dates has a 
linear form. The coefficients in this linear equation only 
include atmospheric parameters at different dates, and are 
estimated by using radiance values of ground objects which are 
assumed to have almost constant reflectances with time. 
Considering the topographic features of pixels, a DEM and its 
derived slope and aspect maps were introduced. : 
The spectral surface reflectance ratio at two different times is 
obtained from digital numbers of the corresponding pixel in 
two different images. 
Hence, the change detection using the surface reflectance ratio 
is performed by introducing a threshold of change. 
The result is a significant radiance evolution map. But this 
would never mean that radiance change is automatically a 
ground change; therefore, methods relating significant change 
to ground truth are necessary. 
2. PRACTICAL PROBLEM 
2.1. Data used 
The present method was applied to Landsat TM images taken 
on April 7, 1984 and March 15, 1993. Further data such as 
topographic and land use map of 1984 were used. 
A DEM and the derived slope and aspect data describing pixel 
features were also used. 
2.2. Study area 
The study area covers the forest of M'sila in the west of Oran 
city in the Northwest of Algeria. The region is covered by the 
scene 198/35 and is estimated to 1100 ha, 
(35°36N <  < 35°42N and 0°47W < À < 0°54W). 
The forest is composed by different stands. Pine (pinus 
halepensis) and oak constitute 80% of the forest area while 
bush constitute the remaining area. 
3. METHODOLOGY 
3.1. Atmospheric correction 
Target and scattered radiances of the atmosphere are measured 
by the sensor embarked on the satellite. The atmospheric 
corrections are, hence, applied. When the surface is 
676 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
La: 
me 
Te 
su 
tin