|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
Using
Remote Sensing and GIS to Assess Wildfire Damage Throughout the Mediterranean Introduction The dynamics of fire is a very important aspect of ecosystems throughout the Mediterranean region, and its study is an essential task for forest management entities. Following a forest fire, forest managers require a quick, complete assessment of the environmental and financial damages that were wrought. As a result, a territorial description of the fire impact becomes necessary. This description of the level of damage, together with resources that might include a map of the previous vegetation, can be very useful when making decisions for restoring vegetation in the affected areas. This information can be derived from remote satellite imagery and ancillary data, but it is essential to develop a method that is operational, timesaving and cost-effective.
Remote sensing is a valuable tool in the study of forest fires not only for prevention and detection, but also for cartography. Satellite imagery offers the wide spatial and temporal covering provided by satellite sensors, as well as providing information collected in both visible and non-visible areas of the spectrum. Wildfires lead to a number of physical changes (reduction or disappearance of the vegetation cover, changes in surface color, etc.), which are reflected in the spectral response of them affected vegetation cover. Based on this, there are numerous works in which remote sensing, and more specifically LANDSAT scenes (TM/ETM), are used for mapping the burned areas. Geographic Information System (GIS) techniques provide the tools to create, transform and combine geo-referenced information. While remote sensing and GIS are often used together to develop environmental applications, there is little experience in combining fire-severity maps (derived from satellite data) and previous-vegetation information to obtain cartography of management recommendations for restoring affected ecosystems. Converting these techniques into operating solutions demands a great deal of research and development, as well as the ability of users to adapt to new technologies, something that is only achieved through narrow collaboration between users of that information and the people and organizations who generate it. The Forestry Department of Córdoba University (Andalucía, Spain) and the Environmental Information and Evaluation Service of the Andalucían government have applied a methodology, based upon remote sensing and GIS techniques, to discriminate burned areas and create an elaborate cartographic representation of pre-fire vegetation levels and damage-intensity levels, plus a restoration management plan. Eighteen wildfires that occurred in the Andalucía region of southern Spain between 1995 and 2001 have been analyzed. Data and Pre-Processing In order to generate pre-fire vegetation maps, aerial photographs (1:60,000 color) and geometrically corrected satellite IRS-PAN images were employed, each with a spatial resolution of five meters. In order to assign wildfire damage intensity levels, a single LANDSAT (TM/ETM) image per fire was used, acquired about a month after the forest fire. Every scene was geometrically corrected and resampled to a pixel size of 30 meters. Furthermore, a Digital Elevation Model (DEM) with a resolution of 20 meters was developed by the Andalucían government and resampled to 30 meters. Burned Area Discrimination The perimeter provided by the government of Andalucía (as determined by GPS coordinates) was corrected over an RGB color composite of bands TM4-TM3-TM7. A visual analysis of the post-fire LANDSAT image (TM4-TM3-TM7) for the burned area offered better results than did traditional methods used by forest managers. Although there are several automatic algorithms that discriminate burned areas via satellite data–undoubtedly useful when one works at a national or regional scale–determining the perimeter of a burned area through visual analysis can be both quick and accurate, provided that information about the location of the fire is readily available. On Figure 1, it is possible to see how the fire perimeters have been corrected. From these new perimeters, the burned area is computed for each separate fire. Previous Vegetation Different vegetation-cover polygons were determined by using pre-fire aerial photographs and IRS-PAN images (Figure 2). Afterwards, detailed information on each polygon was collected during the fieldwork campaign. With this information, polygons have been labeled as "previous vegetation maps" (Figure 3). These maps were more detailed than currently available land-use and vegetation maps of Andalucía. Damage Intensity Levels Three damage intensity levels were distinguished (Figure 4). These include the following designations:
In order to produce the cartography for damage levels of this intensity, supervised classification techniques have been employed. It became necessary to know the ground truth in a number of field sampling sites, at training sites where the classification was carried out, and at many of the validating sites in order to certify the produced cartography. Following each fire, a field campaign was carried out that used topographic cartography at 1:10,000 to locate several sites for each damage level (both moderate and extreme). An area corresponding to 80 pixels of a LANDSAT scene per damage level was located for the training phase, and an area corresponding to 40 pixels per damage level was located for the validating phase. Most often, plots of 3x3 pixels per damage level were chosen. Wherever possible, the same number of plots for the three non-affected classes were located and considered as the training phase (with lush vegetation, with scarce or null vegetation, with water, and shaded areas). The next step in the process was the digitalization of the field training sites. During visits to the area of study, an insufficient number of non-affected plots were located. Therefore, plots were selected visually on the LANDSAT scene to complete the number of pixels recommended for the training and validation phases. Afterwards, the LANDSAT post-fire scene (six reflectance bands) was classified using the maximum likelihood algorithm concept. The three non-affected classes were grouped into a single class. A majority filter of 3x3 was applied to the then-generated thematic scene. The damage-intensity-level assessment showed good results, with high separability values between whichever damage intensity levels were proposed. The global accuracy of damage-level cartography, computed for four wildfires, was greater than 89 percent. The separabilty between classes increased if different normalized band ratios were included in the classification process (band 4-band 3 / band 4+band 3 or band 4-band 7 / band 4+band 7), but global accuracy did not show any significant differences. Something similar occurred if an illumination band was included (derived from solar angles and DEM), where separability increased but global accuracy of the cartography remained the same. Global accuracy was higher for those wildfires that showed homogeneous vegetation prior to the fire. Restoration Plan The ultimate aim of this project was to produce information for making a number of restoration proposals in a burned ecosystem by using GIS and remote sensing techniques. By means of overlaying the previous-vegetation map and the damage-intensity-level map, a third map was created that offered various restoration proposals in which polygons smaller than 0.5 hectares were removed. These proposals were based upon the fire response of the various Mediterranean ecosystems. The cartography of restoration-management practices makes recommendations about the theoretic evolution of the vegetation and optimal management actions. The restoration-proposals map can be integrated with other thematic maps, such as geologic, soils, aspects, and slopes into a GIS, offering detailed information to forest managers. All these processes were carried out with software programs ERDAS IMAGINE (v. 8.4.), ARC/INFO (v.7.1.2) and ARCVIEW (v. 3.2). The analysis of each wildfire took a qualified individual about two weeks to complete. Conclusions The results showed that information about previous vegetation generated from photo-interpretation, and damage-intensity levels generated from the supervised classification of post-fire LANDSAT scene, is entirely accurate. It was also shown that the methodology herein employed is much more operative (quicker and lower-priced) than are other, more traditional methods such as in-field inventory or standard photo-interpretation. As a result, this process can form the basis of a management restoration plan for any burned area. This analysis is but a first step for integrating new technologies with the science of forest management. At the moment, the Department of Forestry at the University of Córdoba is working to improve their methodology, especially those aspects that will allow them to make a more detailed assessment of fire-damage-intensity levels. The authors are faculty members of the Department of Forestry Engineering at the University of Córdoba (Andalucía, Spain). They have been involved in several collaborations with the Environmental Information and Evaluation Service of the Andalucían Government in order to evaluate fire-intensity-damage levels and post-fire regeneration using remote sensing. They may be reached via e-mail at the following respective addresses:
|
Figure
1. Burned area limits over a TM4-TM3-TM7 color composition (Quesada wildfire).