Correcting the Fire Scar Perimeter of a 1983 Wildfire Using USGS-Archived Landsat Satellite Data

600

GIScience & Remote Sensing, 2011, 48, No. 4, p. 600–613. http://dx.doi.org/10.2747/1548-1603.48.4.600Copyright © 2011 by Bellwether Publishing, Ltd. All rights reserved.

Correcting the Fire Scar Perimeter of a 1983 Wildfire Using USGS-Archived Landsat Satellite Data

Foula Nioti, Panayotis Dimopoulos, and Nikos Koutsias1

Department of Environmental and Natural Resources Management, University of Ioannina, G. Seferi 2, GR-30100 Agrinio, Greece

Abstract: In July 1983, a large wildfire occurred on the island of Karpathos in Greece. However, only a general sketch of the burn perimeter was available and this lacked detailed spatial information, particularly for unburned patches within the fire scar perimeter. A study was undertaken to correctly map the area burned using USGS-archived Landsat data by applying several digital image processing techniques. This paper summarizes and discusses the main findings of that study and provides some general recommendations on the use of remote sensing and archived Landsat data for reconstructing fire history. Remote sensing along with geographic information systems can provide an excellent framework for fast, reliable data capture, measure-ment, and synthesis, all of which are essential for thorough eco-environmental analy-sis. Satellite data of multiple types offer an unlimited source of information due to their rich spectral and spatial information content. Satellite mapping of burned areas is considered a standard technique in creating maps of fire scars at multiple scales as a function of the satellite sensor’s geometric resolution.

INTRODUCTION

In order to study the ecological patterns of post-fire vegetation regrowth and suc-cession, and understand the mechanisms of the underlying processes, it is necessary to accurately map the area burned in detail. Other important information about potential environmental factors such as moisture, temperature, terrain, soils, and human activi-ties is also essential (Kalabokidis et al., 2007). In Greece, local forest authorities are responsible for mapping burned areas. However, maps of these areas lack a detailed mapping scheme and usually only indicate general fire perimeters. This information, unless updated, does not provide the critical data needed to study landscape-wildfire dynamics, understand the recovery processes of burned ecosystems, or explain the observed restoration patterns. Unburned patches within the fire scar perimeter are not indicated, and these are important in succession processes, especially for vegetation types whose regeneration pattern depends on the existence of unburned patches.

Additionally, lack of detailed spatially explicit descriptions of fire occurrence reduces, from the management point of view, the ability to protect and restore fire-affected natural ecosystems. Appropriate spatio-temporal data of fire occurrence help fire scientists and managers to understand the reasons for fire ignition and spread

1Corresponding author; email: [email protected]

fire scar perimeter correction 601

(Koutsias et al., 2010) and the role of climate change on fire regimes (Goldammer and Price, 1998), investigate and understand post-fire vegetation regeneration (Mitchell and Yuan, 2010), estimate post-fire soil erosion risk (Mallinis et al., 2009), and avoid mistakes when applying environmental policy and fire management (Kalabokidis et al., 2007). The temporal extent of such information varies considerably. Fire-history reconstruction may refer to a few decades for which spatially explicit information may be acquired from remote sensing observations, or to centennial or millennial scales where other tools (e.g., tree rings, charcoal analysis) are used to acquire the required data (Conedera et al., 2009). At short-term temporal scales (e.g., previous 30 years), numerous remote sensing studies can be found in the literature devoted to detection of fires, mapping of burned areas and burn severity, and monitoring vegetation recov-ery (López-Garcia and Caselles, 1991; Kasischke et al., 1992; Chuvieco and Martin, 1994; Viedma et al., 1997; Barbosa et al., 1998; Koutsias and Karteris, 1998; Rogan and Yool, 2001; Stroppiana et al., 2002; Roy et al., 2005; San-Miguel-Ayanz et al., 2005; Weber et al., 2008; Huang et al., 2009; Kontoes et al., 2009; Weber et al., 2009). In these studies, satellite data of multiple spatial, spectral, radiometric, and temporal resolutions constitute the prime source of information. Their choice depends mainly on the specifications arising from the scale, purpose, and objectives of the study. For instance, fire detection and monitoring requires satellite data of high temporal resolu-tion (i.e., NOAA AVHRR). On the contrary, the satellite data used for burned-land mapping and monitoring may originate from different sources (i.e. NOAA AVHRR, Landsat) and may have various characteristics, usually determined by the study’s observational and operational scales (Koutsias et al., 1999).

In July 1983, a large wildfire occurred on the island of Karpathos in Greece. According to local forest authorities, this wildfire burned 4475 ha of forests, com-posed mainly of Pinus halepensis subsp. brutia. For this wildfire event, we organized a study that sought to map the restoration patterns and status of vegetation 27 years after the fire. This study uses terrestrial and satellite data to identify and understand potential environmental factors that influenced the recovery process. However, no accurate, detailed map of the area burned was available for this wildfire. Instead, only a general sketch of the burn perimeter was available and this lacked detailed spatial information, particularly for unburned patches within the fire scar perimeter (Fig. 1). Therefore, almost three decades later, we attempted to correct the fire scar perimeter of this wildfire using USGS-archived Landsat multitemporal satellite data. The aim of our study was to apply and evaluate some methods already used successfully in other burned land mapping studies and provide improved information on the fire scar perimeters for this particular wildfire.

STUDY AREA AND SATELLITE DATA

The study area is located on the island of Karpathos in the East Aegean Sea (Fig. 1) where in 1983, as noted above, a large wildfire burned. Karpathos Island has a Mediterranean climate and significant ecological value due to its high diversity of plant taxa, endemic species, and habitat types. The island’s vegetation types result from the complex interactions of bioclimatic factors, relief, low water availability, and human interference. Two post-fire Landsat TM satellite images, acquired on June 12, 1984 and August 8, 1990, were downloaded from the U.S. Geological Survey (USGS)

602 nioti et al.

Earth Resources Observation and Science (EROS) Center (http://glovis.usgs.gov/). The spectral signal of the burned surface was still evident in the 1984 satellite image even though the image was captured one year after the fire. Therefore it seems that the recovery process was minimal or even non-existent during the first year. Unfortunately, satellite images taken before the fire event were not available. However, in the 1990 satellite image, the spectral signal of the fire scar has disappeared mainly as a result of vegetation regeneration. Therefore, the 1990 image was not used to map the recovery process in this study, but rather as a “pre-fire” image in a multitemporal approach used to improve the classification accuracy. Although we recognize that the treatment of the 1990 image as a “pre-fire” image is an unconventional approach, it is adopted here to increase the amount of available useable information. Multitemporal approaches can be superior to single-date ones (Koutsias et al., 1999), because the former mini-mize confusion with certain permanent land cover types that possess a similar spectral behavior (Pereira et al., 1997), as in our study. Peterson et al. (2002) listed some stud-ies in which classification approaches using multitemporal imagery discriminated and classified agricultural and grassland areas better than an approach using single date Landsat TM images. However, Henry (2008) found single-date mapping to be slightly more effective than multi-date for burned-land mapping using Landsat data. Using one image (rather than multitemporal analysis) eliminates the problem of relative or

Fig. 1. Geographical location of the study area together with the two post-fire Landsat images used to map the burned surface. The 1990 Landsat image served as a “pre-fire” image, given that the spectral signal of the fire scar had disappeared as a result of vegetation regeneration.

fire scar perimeter correction 603

absolute radiometric correction between dates, spatial registration errors, frequent cloud cover, and phenological differences (Henry, 2008).

METHODOLOGY

Overall Methodology Outline

The objective of our study was to apply and evaluate some methods already used successfully in other burned-land mapping studies and provide improved fire scar perimeters for one particular wildfire. Before applying these methods, a pre- processing step including radiometric and geometric correction was initiated. For the classifica-tion of satellite images in order to identify the burned area, we applied the maximum likelihood method to three classification schemes: (1) single-date satellite image (post-fire); (2) multi-date satellite image (pre- and post-fire); and (3) multispectral enhanced post-fire satellite image (forward/backward principal components analysis [PCA]). Following the multispectral classification of satellite data, we applied a 3 × 3 moving window with a majority function to the classified images of the three classification schemes after recoding them into two classes, burned and unburned. Finally, the clas-sified images were converted from raster to vector, followed by manual editing.

Satellite Data Pre-processing

USGS-archived Landsat data are already processed to Standard Terrain Correction (Level 1T), when possible, which provides systematic radiometric and geometric cor-rections by incorporating ground control points while employing a digital elevation model (DEM) (http://landsat.usgs.gov). Therefore, correction to eliminate geometric errors induced by different sources including the irregular terrain was not necessary; the images were only re-projected to the Greek National projection system. Despite the geometrically corrected product, a further control (image overlay) was implemented to assure the geometric consistency of the multitemporal images; however, no incom-patibilities were observed. Additionally, 14 control points established on both images resulted in an RMSE of 0.15 pixels, indicating the very good geometric matching of both satellite images.

USGS-archived Landsat data are provided in Digital Numbers (DN). Similar to Rogan and Franklin (2001), we corrected for atmospheric path radiance and converted to reflectance units using the dark object subtraction approach. The image-based approach followed is relatively simple to apply, and does not require any in situ mea-surements or atmospheric parameters (Chavez, 1996). This approach assumes 1% sur-face reflectance for dark objects. Additionally, no atmospheric transmittance loss and no diffuse downward radiation at the surface has been assumed (Rogan and Franklin, 2001). The recorded DNs were converted to radiance at sensor and then radiance at sensor was converted to surface reflectance for each spectral channel.

Multispectral and Multitemporal Transformations

In remote sensing studies, especially those in which multispectral and/or multi-temporal satellite data are used (i.e., Landsat series), various multivariate statistical

604 nioti et al.

methods are often applied for image enhancement and feature extraction (Ribed and López, 1995; Siljeström and Moreno-López, 1995; Patterson and Yool, 1998; García-Haro et al., 2001). Principal component analysis, a dimensionality reduction method, has been applied successfully in burned-land mapping (Richards, 1984). In the pres-ent study we followed the forward/backward PCA (F/B PCA) method developed by Koutsias et al. (2009) to enhance the spectral signal of burned surfaces, because the original signal was not very strong as the satellite image was taken one year after the fire event. This method creates a new spectral space that preserves the original spectral characteristics while enhancing particular structures of the original satellite data. The burnt surfaces constitute a spectrally enhanced feature after the selective removal of the spectral information from the original Landsat TM data. This spectrally enhanced dataset was then classified using the classical maximum likelihood classifier.

Classification Algorithm

The maximum likelihood classifier is based on the Gaussian estimate of the prob-ability density function for each class, assuming that the data in each spectral channel follow a normal distribution (Jensen, 2005). Training areas corresponding to six land cover classes (Fire Scar, Bare Land, Forests, Shrublands, Sea, Clouds) were estab-lished in the satellite images and served to estimate the parameters required by the algorithm. Additionally, a random sampling procedure was used to select a total of 684 points that were then used to estimate the accuracy of the classification results. Because no detailed information exists on the actual burned area and the fire occurred nearly 30 years ago, making it impossible to collect data in the field, these reference points were labelled through photo-interpretation of the post-fire satellite image. Although this approach seems problematic, it does provide some indications of the accuracy of the classified images, especially when various classification approaches are applied to the same case study. Overall and individual per class accuracy (user’s and producer’s) and the Kappa coefficient of agreement were estimated. A pairwise test statistic z was also applied to the Kappa coefficient of agreement to statistically compare the results of the different classification schemes (Congalton et al., 1983).

Post-classification Processing and Extraction of the Area Burned

Multispectral classification approaches (e.g., maximum likelihood) that rely only on information extracted from single pixels (known as per-pixel spectral classifiers), allocate each pixel to an output classification class on the basis of a relative similar-ity (distance) of the pixel’s vector x to the mean vector of each class derived after user-selected training data (Mallinis and Koutsias, 2008). However, these methods do not consider the spatial information of the surrounding region of each pixel, defined as the tendency of neighboring pixels to present similar characteristics. The spatial information inherent in satellite data can be incorporated into the classification process either during post-processing or prior to pixel labelling. The incorporation of spatial information is based on the principle that the distribution of any spatial phenome-non is explained not only by structural factors (e.g., spectral data in remote sensing) but also by spatial factors (e.g., neighbouring effects and associations) (Chou et al., 1990). Although other more advanced and sophisticated approaches are available to

fire scar perimeter correction 605

incorporate the spatial information into the digital classification, in the present study we applied a 3 × 3 moving window with a majority function to the classified images after recoding them to two classes, burned and unburned. The central pixel was assigned to the class corresponding to the majority value. Accuracy estimation allowed us to evaluate these products compared to the original classified images. Finally, they were converted from raster to vector format by applying typical techniques in a GIS envi-ronment supported by ArcGIS 9.3.

RESULTS

Tables 1 and 2 present the classification results and accuracy of: (1) the maximum likelihood classification of the three classification schemes; and (2) the post-processing 3 × 3 majority filter. Additionally, the entire process is presented graphically in Figure 2. Based on these results and taking into consideration the overall and individual per class accuracy, the best accuracy observed for the maximum likelihood classification corresponds to the multitemporal data set consisting of the pre- and post-fire satellite image (overall accuracy is 93.57%), the classification scheme with the fewest omis-sion errors in the burned class (11.95%). The multispectral transformation made in the original satellite data increased the overall classification accuracy (93.42%) compared to that of the single-date Landsat image ( 92.84%).

The 3 × 3 majority filter applied to the classified images increased the overall accuracy of all three classification schemes. However, we decided to keep the post-processing results of the multi-date Landsat scheme, because the difference between the omission and commission errors (3.71%) is the smallest of the three (Table 2).

DISCUSSION

In this study we investigated the performance of three classification schemes to map a burned area using Landsat TM satellite images. Classification accuracies appear to be in accordance with other published research findings and current knowledge on remote sensing of burned areas. Below we summarize and discuss the main findings

Table 1. Classification Results of the Three Classification Schemes Arising From the Two Processing Concepts Used in this Study

ProcessingClassification scheme (area in hectares)

1984–1990 1984 1984 f/b PCA

ML classificationa

Burned 3,122 2,757 2,913Non-burned 6,660 7,025 6,869

3 × 3 moving windowBurned 3,126 2,750 2,914Non-burned 6,655 7,032 6,868

aML = maximum likelihood.

606 nioti et al.

of our work concerning (1) the use of multitemporal versus single-date satellite data, (2) the use of multispectral transformations with emphasis on principal component analysis, and (3) the application of post-classification processing by a 3 × 3 majority filter. Finally we attempt to give some general recommendations on the role of remote sensing and the use of past archived Landsat satellite data for reconstructing short-term (e.g., 30 years) fire history, given that in most cases no reference ground truth data exist to evaluate mapping accuracy.

It has been demonstrated that methods using a multitemporal data set are more effective than those using only a single post-fire image, because the former minimize the confusion of land cover types with similar spectral behavior as the burned areas but remain unchanged in the multitemporal data set (Koutsias et al., 1999). However, single post-fire methods present superior performance over multitemporal ones in terms of cost and time requirements needed for the acquisition and processing of the multitemporal data set. One of the most critical issues in the multitemporal approach concerns the radiometric and geometric adjustments needed to ensure the spatial and

Table 2. Accuracy Statistics of the Three Classification Schemes Evaluated by the Two Processing Concepts, in percenta

Processing Accuracy statisticsClassification scheme

1984–1990 1984 1984 f/b PCA

ML classificationb

Burned Omission 11.95 17.70 14.60Commission 7.87 4.62 5.85Difference 4.08 13.08 8.75

Non-burned Omission 3.71 1.97 2.62Commission 5.77 8.18 6.89Difference –2.06 –6.21 –4.27Overall accuracy 93.57 92.84 93.42Kappa 0.853* 0.832* 0.848*

3 × 3 majority filterBurned Omission 10.62 18.58 12.83

Commission 6.91 2.65 4.37Difference 3.71 15.93 8.46

Non-burned Omission 3.28 1.09 1.97Commission 5.14 8.48 6.07Difference –1.86 –7.39 –4.10Overall accuracy 94.30 93.13 94.44Kappa 0.870* 0.838* 0.872*

aAll figures in percent except for Kappa, which is dimensionless.bML = maximum likelihood.* = no statistically significant differences found.

fire scar perimeter correction 607

Fig. 2. Three classification schemes were considered to map the burned area using the maximum likelihood classifier and a 3 × 3 majority filter. The final vectors of the fire scar perimeter are overlaid over the original satellite image.

608 nioti et al.

spectral matching of the images used. Radiometric and geometric mis-registration may produce unpredictable errors, which in turn may result in either under- or over-estimation of the burned areas. We observed that the majority of the errors are dis-tributed in the borders between burned and unburned vegetation where mixed pixels are found. In that sense such problems can be enhanced when the issue of geometric mis-registration is concerned. In our study the accuracy achieved by the multitemporal data set is higher than the accuracy of the single-date classification; however these dif-ferences were not statistically significant. The multitemporal data set helped to reduce the omission error from 17.70% to 11.95%, and therefore the burned area becomes less under-estimated (Table 2).

Multivariate statistical methods are widely applied in remote sensing applications, especially those that utilize a multidimensional data set. These methods, mainly those dealing with dimensionality reduction, such as principal components analysis (Hudak and Brockett, 2004) or vegetation indices (Chuvieco et al., 2002), aim to separate the spectral information distributed in the original spectral channels into those few new components that are more interpretable (Richards, 1984). If the dimensionality reduc-tion and the information separation are accomplished successfully, then the desired information can be achieved easily by applying simple further processing such as thresholding (Koutsias et al., 2000). The forward/backward PCA increased the overall classification accuracy from 92.84% to 93.42%, while the omission errors of burned areas decreased from 17.70% to 14.60%; however no statistically significant differ-ences were found (Table 2).

It has been demonstrated that methods which consider the spatial information of the satellite data can be superior to those that do not consider it, for example in burned-land mapping (Koutsias, 2003). Purely pixel-based multispectral clas-sification approaches only consider the spectral information at pixel basis and do not account for the spatial component by taking the spectral information of the sur-rounding pixel neighbors into consideration (Mallinis and Koutsias, 2008). Although other more advanced and sophisticated approaches are available to incorporate spa-tial information into digital classification either during post-processing or prior to pixel labelling, in our study we applied a simple 3 × 3 majority filter at the post- processing level. The overall accuracy of the three classification schemes increased after the application of the 3 × 3 majority filter by reducing the “salt and pepper” effect usually observed in satellite data. Again, however, these differences were not statistically significant.

Different tools exist to map burned areas. Mapping can be achieved by: (1) field surveys using on site human-made observations; (2) the use of black-and-white, color, or infrared aerial photography; and (3) remotely sensed data acquired by various satel-lite systems. A field survey is considered to be a time-consuming and expensive, but highly accurate method, although it poses serious limitations for burned land mapping because it only provides general statistics due to time and cost limitations (Koutsias et al., 1999). High accuracy, combined with the high cost and time requirements, make this method suitable only in very specific situations (local scale) in which the measure-ment scale (resolution, detail) is more significant than the geographic scale (extent of the study area). Aerial photography covers larger geographical areas than field sur-veys and processes the data with less cost. However its usage is minimal in burned-land mapping, mainly due to cost constraints. Satellite remote sensing appears to be a

fire scar perimeter correction 609

suitable approach to map and monitor burned areas compared to other more traditional methods, but this method is not error free. Various land cover classes present simi-lar spectral responses to the burned surfaces, thus resulting in spectral confusion and errors, such as with water bodies and cloud shadows as observed in NOAA AVHRR data (Chuvieco et al., 2002). Given the large geographic or time extent of burned-land mapping, it is clear that remote sensing provides an ideal alternative for gathering and processing the required information. Mallinis et al. (2011) state that remote sensing along with geographic information systems can provide an excellent framework for fast, reliable data capture, measurement, and synthesis, all of which are essential for thorough eco-environmental analysis. Satellite data of multiple resolutions offer an unlimited source of information guided by their rich spectral and spatial information content. Burned-area mapping from satellite sources is presently very common and is considered a standard and effective technique to create maps of fire scars at multiple scales as a function of the geometric resolution of the satellite sensor used (Koutsias et al., 1999).

Remote sensing has been found to be useful not only to wildfire management but also to other natural hazards, as it provides early warnings, monitors the hazards in real time, and makes it possible to assess damage. Even if the hazard cannot be prevented, in many cases disastrous situations can be avoided or mitigated (San-Miguel-Ayanz et al., 2000). Current global fire products, such as those based on MODIS (Justice et al., 2002), offer a unique dataset, although these are mostly devoted to continental-scale studies such as characterizing global fire regimes (Chuvieco et al., 2008) or esti-mating global biomass burning emissions (Korontzi et al., 2004). Annually resolved fire perimeters are also provided by the European Forest Fire Information System (EFFIS) of the European Commission Joint Research Centre and these are also based on MODIS satellite data in an effort to provide consistent fire statistics over Europe (Camia et al., 2009). However, at the local scale such systematic fire products are not common, mainly due to cost constraints on gathering and processing medium- or high-resolution satellite data series.

USGS-archived Landsat scenes processed to Standard Terrain Correction (Level 1T) where possible, are available to the public at no charge from U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center (http://glo-vis.usgs.gov/). These historical archives cover large spatial and temporal extents at continental scale and provide a unique opportunity to overcome cost constraints when reconstructing fire history at local scales. The Landsat satellite series has a long his-tory of data capture that commenced with the launch of Landsat 1 on July 23, 1972, with Landsat Multispectral Scanner (MSS) on board (originally known as the Earth Resources Technology Satellite [ERTS]). Since then Landsat satellites have been tak-ing repetitive images of the Earth’s surface at continental scale, thus creating a huge historical archive that can be used to reconstruct the past. To understand dynamic pro-cesses such as wildfires and assess their effects on landscape dynamics by understand-ing the recovery processes of the affected ecosystems and explaining the observed restoration patterns, we require observations in temporal and spatial extents that allow these processes to be captured. The Landsat archives are a valuable data source that allows reconstruction of short-term fire occurrences by mapping past fire events in cases where no data or only rough estimates are available.

610 nioti et al.

CONCLUSIONS

Lack of detailed spatial explicit descriptions of fire occurrence reduces the abil-ity to protect and restore fire-affected natural ecosystems. Additionally, appropriate spatio-temporal information on fire occurrence helps fire scientists and managers to better understand the reasons for fire ignition and spread, and avoid mistakes when applying environmental policy and undertaking fire management. Fire occurrence data should be collected and analyzed by cost-effective, quick, and accurate methods. Remote sensing with geographic information systems can provide an excellent frame-work for fast and reliable data capture, measurement, synthesis, all of which are essen-tial for thorough eco-environmental analysis. Satellite data of multiple types offer an unlimited source of information afforded by their rich spectral and spatial information content. Burned-area mapping from satellite sources is presently quite common and is considered to be a standard and effective technique for creating maps of fire scars at multiple scales as a function of the geometric resolution of the satellite sensor used.

ACKNOWLEDGMENTS

The research leading to these results has received funding from the European Union’s Seventh Framework Programme (FP7/2007- 2013) under grant agreement No. 243888. Landsat data available from the U.S. Geological Survey (http://www.usgs .gov).

REFERENCES

Barbosa, P. M., Pereira, J. M. C. and J.-M. Gregoire, 1998, “Compositing Criteria for Burned Area Assessment Using Multitemporal Low Resolution Satellite Data,” Remote Sensing of Environment, 65:38–49.

Camia, A., San-Miguel-Ayanz, J., Oehler, F., Santos de Oliveira, S., Durrant, S. Kucera, J., Boca, R. et al., 2009, Forest Fires in Europe 2008, Luxemburg: Office for Official Publications of the European Communities, Eur 23971 EN-2009, 83 pp.

Chavez, P. S., Jr., 1996, “Image-Based Atmospheric Corrections—Revisited and Improved,” Photogrammetric Engineering and Remote Sensing, 62(9):1025–1036.

Chou, Y. H., Minnich, R. A. and L. A. Salazar, 1990, “Spatial Autocorrelation of Wild-fire Distribution in the Idyllwild Quadrangle, San Jacinto Mountain, California,” Photogrammetric Engineering & Remote Sensing, 56(11):1507–1513.

Chuvieco, E., Giglio, L., and C. Justice, 2008, “Global Characterization of Fire Activi-ty: Toward Defining Fire Regimes from Earth Observation Data,” Global Change Biology, 14(7):1488–1502.

Chuvieco, E. and M. P. Martin, 1994, “A Simple Method for Fire Growth Mapping Using AVHRR Channel 3 Data,” International Journal of Remote Sensing, 15(16):3141–3146.

Chuvieco, E., Martín, M. P. and A. Palacios, 2002, “Assessment of Different Spectral Indices in the Red–Near-Infrared Spectral Domain for Burned Land Discrimina-tion,” International Journal of Remote Sensing, 23(23):5103–5110.

fire scar perimeter correction 611

Conedera, M., Tinner, W., Neff, C., Meurer, M., Dickens, A. F., and P. Krebs, 2009, “Reconstructing Past Fire Regimes: Methods, Applications, and Relevance to Fire Management and Conservation,” Quaternary Science Reviews, 28(5–6):555–576.

Congalton, R. G., Oderwald, R. G. and R. A. Mead, 1983, “Assessing Landsat Clas-sification Accuracy Using Discrete Multivariate Analysis Statistical Techniques,” Photogrammetric Engineering & Remote Sensing, 49(12):1671–1678.

García-Haro, F. J., Gilabert, M. A., J. and Meliá, 2001, “Monitoring Fire-Affected Areas Using Thematic Mapper Data,” International Journal of Remote Sensing, 22(4):533–549.

Goldammer, J. G. and C. Price, 1998, “Potential Impacts of Climate Change on Fire Regimes in the Tropics Based on MAGICC and a GISS GCM–Derived Lightning Model,” Climatic Change, 39(2–3):273–296.

Henry, M. C., 2008, “Comparison of Single- and Multi-date Landsat Data for Mapping Wildfire Scars in Ocala National Forest, Florida,” Photogrammetric Engineering and Remote Sensing, 74(7):881–891.

Huang, S., Siegert, F., Goldammer, J. G. and A. I. Sukhinin, 2009, “Satellite-Derived 2003 Wildfires in Southern Siberia and Their Potential Influence on Carbon Sequestration,” International Journal of Remote Sensing, 30(6):1479–1492.

Hudak, A. T. and B. H. Brockett, 2004, “Mapping Fire Scars in a Soutern African Savannah Using Landsat Imagery,” International Journal of Remote Sensing, 25(16):3231–3243.

Jensen, J. R., 2005, Introductory Digital Image Processing a Remote Sensing Perspec-tive, Upper Saddle River, NJ: Prentice Hall.

Justice, C. O. et al., 2002, “The MODIS Fire Products,” Remote Sensing of Environ-ment, 83:244–262.

Kalabokidis, K. D., Koutsias, N., Konstantinidis, P., and Vasilakos, C., 2007, “Multi-variate Analysis of Landscape Wildfire Dynamics in a Mediterranean Ecosystem of Greece,” Area, 39(3):392–402.

Kasischke, E. S., Bourgeau-Chavez, L. L., French, N. H. F., Harrell, P. and N. L. Christensen, Jr., 1992, “Initial Observations on Using SAR to Monitor Wildfires Scars in Boreal Forests,” International Journal of Remote Sensing, 13(18):3495–3501.

Kontoes, C. C., Poilvé, H., Florsch, G., Keramitsoglou, I. and S. Paralikidis, 2009, “A Comparative Analysis of a Fixed Thresholding vs. a Classification Tree Approach for Operational Burn Scar Detection and Mapping,” International Journal of Applied Earth Observation and Geoinformation, 11(5):299–316.

Korontzi, S., Roy, D. P., Justice, C. O., and D. E. Ward, 2004, “Modeling and Sensitiv-ity Analysis of Fire Emissions in Southern Africa during Safari 2000,” Remote Sensing of Environment, 92(2):255–275.

Koutsias, N., 2003, “An Autologistic Regression Model for Increasing the Accuracy of Burned Surface Mapping Using Landsat Thematic Mapper Data,” International Journal of Remote Sensing, 24(10):2199–2204.

Koutsias, N. and Karteris, M., 1998, “Logistic Regression Modelling of Multitempo-ral Thematic Mapper Data for Burned Area Mapping,” International Journal of Remote Sensing, 19(18):3499–3514.

612 nioti et al.

Koutsias, N., Karteris, M., and Chuvieco, E., 2000, “The Use of Intensity-Hue-Satura-tion Transformation of Landsat-5 Thematic Mapper Data for Burned Land Map-ping,” Photogrammetric Engineering & Remote Sensing, 66(7):829–839.

Koutsias, N., Karteris, M., Fernandez, A., Navarro, C. Jurado, J., Navarro, R., and A. Lobo, 1999, “Burned Land Mapping at Local Scale,” in Remote Sensing of Large Wildfires in the European Mediterranean Basin, Chuvieco, E. (Ed.), Berlin-Heidelberg, Germany: Springer-Verlag, 157–187.

Koutsias, N., Mallinis, G., and M. Karteris, 2009, “A Forward/Backward Principal Component Analysis of Landsat-7 ETM+Β Data to Enhance the Spectral Sig-nal of Burnt Surfaces,” ISPRS Journal of Photogrammetry and Remote Sensing, 64(1):37–46.

Koutsias, N., Martínez-Fernández, J., and B. Allgöwer, 2010, “Do Factors Causing Wildfires Vary in Space? Evidence from Geographically Weighted Regression,” GIScience & Remote Sensing, 47(2):221–240.

López-Garcia, M. J. and V. Caselles, 1991, “Mapping Burns and Natural Reforestation Using Thematic Mapper Data,” Geocarto International, 1:31–37.

Mallinis, G., Emmanoloudis, D., Giannakopoulos, V., Maris, F., and N. Koutsias, 2011, “Mapping and Interpreting Historical Land Cover Changes in a Natura 2000 Site Using Earth Observational Data: The Case of Nestos Delta, Greece,” Applied Geography, 31:312–320.

Mallinis, G. and N. Koutsias, 2008, “Spectral and Spatial-Based Classification for Broad-Scale Land Cover Mapping Based on Logistic Regression,” Sensors, 8(12):8067–8085.

Mallinis, G., Maris, F., Kalinderis, I., and N. Koutsias, 2009, “Assessment of Post-Fire Soil Erosion Risk in Fire-Affected Watersheds Using Remote Sensing and GIS,” GIScience & Remote Sensing, 46(4):388–410.

Mitchell, M. and F. Yuan, 2010, “Assessing Forest Fire and Vegetation Recovery in the Black Hills, South Dakota,” GIScience and Remote Sensing, 47(2):276–299.

Patterson, M. W. and S. R. Yool, 1998, “Mapping Fire-Induced Vegetation Mortality Using Landsat Thematic Mapper Data: A Comparison of Linear Transformation Techniques,” Remote Sensing of Environment, 65:132–142.

Pereira, J. M. C., Chuvieco, E., Beaudoin, A., and N. Desbois, 1997, “Remote Sensing of Burned Areas: A Review,” A Review of Remote Sensing Methods for the Study of Large Wildland Fires, Chuvieco, E. (Ed.), Alcala de Henares, Spain: Universi-dad de Alcala, 127–183.

Peterson, D. L., Price, K. P. and E. A. Martinko, 2002, “Discriminating between Cool Season and Warm Season Grassland Cover Types in Northeastern Kansas,” Inter-national Journal of Remote Sensing, 23(23):5015–5030.

Ribed, P. S. and A. M. López, 1995, “Monitoring Burnt Areas by Principal Com-ponents Analysis of Multitemporal TM Data,” International Journal of Remote Sensing, 16(9):1577–1587.

Richards, J. A., 1984, “Thematic Mapping from Multitemporal Image Data Using the Principal Component Transformation,” Remote Sensing of Environment, 16:35–46.

Rogan, J. and J. Franklin, 2001, “Mapping Wildfire Burn Severity in Southern California Forests and Shrublands Using Enhanced Thematic Mapper Imagery,” Geocarto International, 16(4):89–99.

fire scar perimeter correction 613

Rogan, J. and S. R. Yool, 2001, “Mapping Fire-Induced Vegetation Depletion in the Peloncillo Mountains, Arizona and New Mexico,” International Journal of Remote Sensing, 22(16):3101–3121.

Roy, D. P., Jin, Y., Lewis, P. E., and C. O. Justice, 2005, “Prototyping a Global Algorithm for Systematic Fire-Affected Area Mapping Using MODIS Time Series Data,” Remote Sensing of Environment, 97(2):137–162.

San-Miguel-Ayanz, J., Ravail, N., Kelha, V., and A. Ollero, 2005, “Active Fire Detec-tion for Fire Emergency Management: Potential and Limitations for the Opera-tional Use of Remote Sensing,” Natural Hazards, 35(3):361–376.

San Miguel-Ayanz, J., Voft, J., De Roo, A., and G. Schmuck, 2000, “Natural Hazards Monitoring: Forest Fires, Droughts, and Floods—the Example of European Pilot Projects,” Surveys in Geophysics, 21(2–3):291–305.

Siljeström, R. and A. Moreno-López, 1995, “Monitoring Burned Areas by Principal Component Analysis,” International Journal of Remote Sensing, 16(9):1577–1587.

Stroppiana, D., Pinnock, S., Pereira, J. M. C., and J.-M. Gregoire, 2002, “Radiomet-ric Analysis of SPOT Vegetation Images for Burnt Area Detection in Northern Australia,” Remote Sensing of Environment, 82:21–37.

Viedma, O., Melia, J., Segarra, D., and J. Garcia-Haro, 1997, “Modeling Rates of Ecosystem Recovery after Fires by Using Landsat TM Data,” Remote Sensing of Environment, 61:383–398.

Weber, K. T., Seefeldt, S., and C. Moffet, 2009, “Fire Severity Model Accuracy Using Short-Term, Rapid Assessment Versus Long-Term, Anniversary Date Assess-ment,” GIScience and Remote Sensing, 46(1):24–38.

Weber, K. T., Seefeldt, S., Moffet, C., and J. Norton, 2008, “Comparing Fire Severity Models from Post-Fire and Pre/Post-Fire Differenced Imagery,” GIScience and Remote Sensing, 45(4):392–405.