Barbosa, Paulo 
  
DETECTION OF CHANGES IN FOREST LANDCOVER TYPE AFTER FIRES IN PORTUGAL 
Paulo M. Barbosa, Märio R. Caetano, and Teresa G. Santos 
Centro Nacional de Informacäo Geogräfica (CNIG), Portugal 
barp@cnig.pt, mario@cnig.pt, tags@cnig.pt 
KEY WORDS: Landsat TM, forest fires, vegetation indices, land cover change, multitemporal analysis 
ABSTRACT 
We present a methodology to detect terrain mobilization in the years following a fire using multitemporal Landsat TM 
imagery. Furthermore, we show the spectral evolution of the land cover since fire occurrence, and analyze the 
feasibility of detecting the type of land cover a number of years after fire occurrence. 
The methodology developed here is a spectral change identification technique based on vegetation index differencing. 
Spectral change identification has the advantage that it is based on the detection of physical changes between image 
dates, which avoids the propagation of accuracy errors. The vegetation index used was the Atmosphere Resistant 
Vegetation Index (ARVI). Although ARVI differencing was obtained through a simple subtraction between two 
different dates, the critical element was to decide the thresholds between change and no-change. 
The spectral evolution of the land cover was done through a multitemporal analysis from 1990 until 1998, and can be 
seen as a first step to help identifying the post-fire vegetation species. 
To assess the efficiency of this methodology in detecting the different land cover changes after a fire, a change 
detection error matrix was used. The reference data used for the validation was a land use map from 1990 in order to 
detect the land cover type before the fire, ortho-rectified aerial photographs from 1995 in order to detect the land cover 
type after the fire, and also burned area maps for each year. This study demonstrated the feasibility of using earth 
observation data to assess land cover changes in burnt forest areas. 
1 INTRODUCTION 
The problem of forest fires in Portugal is intimately linked to land cover and land use change purposes. Forests are 
periodically burned with higher or lower intensity, resulting in an immediate change of the land cover towards a burned 
surface. This can result in the recover of the natural vegetation or of the forest species that was present before the fire. 
However, it is not uncommon that human intervention leads to a land cover type different from the one that was present 
before the fire. The major land cover changes that might occur in Portuguese forests after fires are growth of shrub-land, 
re-growth of the forest stands, and terrain mobilization followed by forest plantation. Less common are the changes of 
the burned surface into urban or agricultural areas. 
We are currently developing methodologies to characterize post-fire land cover maps using satellite data, under an 
European Commission funded project - APERTURE (ENV4-CT97-0437). This project aims at promoting the use of 
Earth Observation (EO) data in judicial and extra-judicial proceedings to enforce the application of environmental law. 
In Portugal, we are addressing the implementation of burned forested area legislation, which requires the 
characterization of land cover, before and after fire occurrence, with EO data. Our most important goal, within 
APERTURE, is the development of methodologies that use EO data to identify areas where illegal land cover 
transitions occurred. Illegal actions entail, for example, the plantation of eucalyptus in areas occupied with pine forest 
before fire occurrence. This explains our interest in separating different types of land cover after a fire. 
The study reported here aims to characterize post fire land cover change with a particular emphasis on terrain 
mobilization. Our main goal is to detect terrain mobilizations in burned forest surfaces, and analyze the possibilities of 
identifying the forest species used in the plantation. 
Most of the reforestation actions, which are due to economic interests, take place in the first few years after fire 
occurrence. For this reason, the algorithm developed for post-fire detection of terrain mobilisation was tested only on 
the first two years after fire occurrence. However, the same technique can be used for a longer time period. Three fire . 
seasons were considered in the post-fire change analysis: 1991, 1992, and 1993. 
  
126 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.