Modeling and sensitivity analysis of fire emissions in southern Africa during SAFARI 2000

This paper uses three recently generated southern African satellite burned area products for the month of September 2000 in a sensitivity study of regional biomass burning emissions for a number of trace gases and particulates. Differences in the extent and location of areas burned among products generated from Moderate Resolution Imaging Spectroradiometer (MODIS), Systeme Pour l'Observation de la Terre (SPOT-VEGETATION), and Along Track Scanning Radiometer (ATSR-2) data are significant and result in different emissions estimates for woodland and grassland land cover types. Due to the different emission profiles in woodlands and grasslands, favoring relatively more products of incomplete combustion in woodlands compared with products of complete combustion in grasslands in the late dry season, these changes are not proportional to the differences in the burned area amounts. The importance of accurate burned area information not just in terms of the total area but also in terms of its spatial distribution becomes apparent from our modeling results. This paper highlights the urgent need for satellite data producers to provide accuracy assessments associated with satellite-derived products. Preferably, these accuracy data will be spatially explicit, or defined in a way that can be applied in a spatially explicit modeling context, to enable emissions uncertainties to be defined with respect to different landscape units in support of greenhouse gas emissions reporting.     

Reconstructing long time series of burned areas in arid grasslands of southern Russia by satellite remote sensing

Fire is an important natural disturbance process in many ecosystems, but humans can irrevocably change natural fire regimes. Quantifying long-term change in fire regimes is important to understand the driving forces of changes in fire dynamics, and the implications of fire regime changes for ecosystem ecology. However, assessing fire regime changes is challenging, especially in grasslands because of high intra- and inter-annual variation of the vegetation and temporally sparse satellite data in many regions of the world. The breakdown of the Soviet Union in 1991 caused substantial socioeconomic changes and a decrease in grazing pressure in Russia's arid grasslands, but how this affected grassland fires is unknown. Our research goal was to assess annual burned area in the grasslands of southern Russia before and after the breakdown. Our study area covers 19,000 km² in the Republic of Kalmykia in southern Russia in the arid grasslands of the Caspian plains. We estimated annual burned area from 1985 to 2007 by classifying AVHRR data using decision tree algorithm, and validated the results with RESURS, Landsat and MODIS data. Our results showed a substantial increase in burned area, from almost none in the 1980s to more than 20% of the total study area burned in both 2006 and 2007. Burned area started to increase around 1998 and has continued to increase, albeit with high fluctuations among years. We suggest that it took several years after livestock numbers decreased in the beginning of the 1990s for vegetation to recover, to build up enough fuel, and to reach a threshold of connectivity that could sustain large fires. Our burned area detection algorithm was effective, and captured burned areas even with incomplete annual AVHRR data. Validation results showed 68% producer's and 56% user's accuracy. Lack of frequent AVHRR data is a common problem and our burned area detection approach may also be suitable in other parts of the world with comparable ecosystems and similar AVHRR data limitations. In our case, AVHRR data were the only satellite imagery available far enough back in time to reveal marked increases in fire regimes in southern Russia before and after the breakdown of the Soviet Union.     

A simulation analysis of the detectability of understory burns in miombo woodlands

The miombo woodlands of southern Africa are one of the most extensively burned biomes in the tropics. The detectability of understory burns in these woodlands was assessed with a sensitivity analysis approach, based on a hybrid geometrical-optical radiative transfer model. Model input data were obtained from a variety of sources, including field biometry and spectroradiometry, and satellite data. The effects of variable tree percent cover, leaf area index, stand density, burn scar age, illumination and observation geometry, and spectral region, were taken into account. Detectability of understory burns was defined as the spectral separability of burned and unburned understory, measured with the Jeffries-Matusita distance, for all possible combinations of the green, red, and near-infrared channels of the Moderate Resolution Imaging Spectrometer (MODIS) sensor. Single channels, or pairwise combinations of channels perform poorly at detecting understory burns, but a large improvement in detectability is obtained for the combination of the three spectral domains. The detectability of understory burns is largely insensitive to the effects of stand structure and illumination/observation geometry, and depends primarily on burn scar age. Our results agree with those of previous satellite-based studies of burns scar detectability in African savanna woodlands.     

Assessing the feasibility of a global model for multi-temporal burned area mapping using SPOT-VEGETATION data

The possibility of using the Syst@me Probatoire de l'Observation de la Terre (SPOT)-VEGETATION (VGT) data for global burned area mapping with a single algorithm was investigated. Using VGT images from south-eastern Africa, the Iberian Peninsula and south-eastern Siberia/north-eastern China, we analysed the variability of the spectral signature of burned areas and its relationship with land cover, and performed the selection of the best variables for burned area mapping. The results show that in grasslands and croplands, near-infrared (NIR) and short-wave infrared (SWIR) reflectance always decreases as a result of fire. In forests and woodlands, there may occur a simultaneous decrease of SWIR and NIR or an increase of SWIR and a decrease of NIR. Burning of green vegetation (high values of the Normalized Difference Vegetation Index (NDVI)) tends to result in an increase of the SWIR. The best variables for burned area mapping are different in each region. Only the NIR allows a good discrimination of burned areas in all study areas. We derived a logistic regression model for multi-temporal burned area mapping in tropical, temperate and boreal regions, which handles the spectral variability of burned areas dependent on the type of vegetation. The results underline the feasibility of a single model for global burned area mapping.     

Analysis of the relationship between spatial pattern and spectral detectability of areas burned in southern Africa using satellite data

Fires in Africa affect atmospheric emissions and carbon sequestration, landscape patterns, and regional and global climatic conditions. Studies of these effects require accurate estimation of the extent of measurable fire events. The goal of this study was to assess the influence of burned area spatial patterns on the spectral detectability of burned areas. Six Landsat‐7 ETM+ images from the southern Africa were used for burned area mapping and spatial pattern analysis, while contemporaneous MODIS 500 m spatial resolution images were used to measure the spectral detectability of burned areas. Using a 15 by 15 km sample quadrats analysis, we showed that above a burned area proportion threshold of approximately 0.5 the spectral detectability of burned areas increase due to the decrease in the number of mixed pixels. This was spatially related to the coalescence of burned patches and the decrease in the total burned area perimeter. Simple burned area shapes were found at the Botswana site, where the absence of tree cover and the presence of bright surfaces (soil and dry grass) enhanced the spectral contrast of the burned surfaces, thus enabling better estimates of burned area extent. At the Zambia and Congo sites, landscape fragmentation due to human activity and the presence of a tree vegetation layer, respectively, contribute to the presence of small burned area patches, which may remain undetectable using moderate spatial resolution satellite imagery, leading to less accurate burned area extent estimates.     

Comparison of burned area estimates derived from SPOT-VEGETATION and Landsat ETM+ data in Africa: Influence of spatial pattern and vegetation type

An algorithm for burned area mapping in Africa based on classification trees was developed using SPOT-VEGETATION (VGT) imagery. The derived 1 km spatial resolution burned area maps were compared with 30 m spatial resolution maps obtained with 13 Landsat ETM+ scenes, through linear regression analysis. The procedure quantifies the bias in burned area estimation present in the low spatial resolution burned area map. Good correspondence was observed for seven sites, with values of the coefficient of determination (R²) ranging from 0.787 to 0.983. Poorer agreement was observed in four sites (R² values between 0.257 and 0.417), and intermediate values of R² (0.670 and 0.613) were obtained for two sites. The observed variation in the level of agreement between the Landsat and VGT estimates of area burned results from differences in the spatial pattern and size distribution of burns in the different fire regimes encompassed by our analysis. Small and fragmented burned areas result in large underestimation at 1 km spatial resolution. When large and compact burned areas dominate the landscape, VGT estimates of burned area are accurate, although in certain situations there is some overestimation. Accuracy of VGT burned area estimates also depends on vegetation type. Results showed that in forest ecosystems VGT maps underestimate substantially the amount of burned area. The most accurate estimates were obtained for woodlands and grasslands. An overall linear regression fitted with the data from the 13 comparison sites revealed that there is a strong relationship between VGT and Landsat estimates of burned area, with a value of R² of 0.754 and a slope of 0.803. Our findings indicate that burned area mapping based on 1 km spatial resolution VGT data provides adequate regional information.     

Regionally adaptable dNBR-based algorithm for burned area mapping from MODIS data

Recent advances in instrument design have led to considerable improvements in wildfire mapping at regional and global scales. Global and regional active fire and burned area products are currently available from various satellite sensors. While only global products can provide consistent assessments of fire activity at the global, hemispherical or continental scales, the efficiency of their performance differs in various ecosystems. The available regional products are hard-coded to the specifics of a given ecosystem (e.g. boreal forest) and their mapping accuracy drops dramatically outside the intended area. We present a regionally adaptable semi-automated approach to mapping burned area using Moderate Resolution Imaging Spectroradiometer (MODIS) data. This is a flexible remote sensing/GIS-based algorithm which allows for easy modification of algorithm parameterization to adapt it to the regional specifics of fire occurrence in the biome or region of interest. The algorithm is based on Normalized Burned Ratio differencing (dNBR) and therefore retains the variability of spectral response of the area affected by fire and has the potential to be used beyond binary burned/unburned mapping for the first-order characterization of fire impacts from remotely sensed data. The algorithm inputs the MODIS Surface Reflectance 8-Day Composite product (MOD09A1) and the MODIS Active Fire product (MOD14) and outputs yearly maps of burned area with dNBR values and beginning and ending dates of mapping as the attributive information. Comparison of this product with high resolution burn scar information from Landsat ETM+ imagery and fire perimeter data shows high levels of accuracy in reporting burned area across different ecosystems. We evaluated algorithm performance in boreal forests of Central Siberia, Mediterranean-type ecosystems of California, and sagebrush steppe of the Great Basin region of the US. In each ecosystem the MODIS burned area estimates were within 15% of the estimates produced by the high resolution base with the R² between 0.87 and 0.99. In addition, the spatial accuracy of large burn scars in the boreal forests of Central Siberia was also high with Kappa values ranging between 0.76 and 0.79.     

Mapping burned areas in Mediterranean countries using spectral mixture analysis from a uni‐temporal perspective

The main aim of this study was to evaluate the usefulness of spectral mixture analysis (SMA) for mapping forest areas burned by fires in the Mediterranean area using low and medium spatial resolution satellite sensor data. A methodology requiring only one single post‐fire image was used to carry out the study (uni‐temporal techniques). This methodology is based on the contextual classification of the fraction images obtained after applying SMA to the original post‐fire image. The results showed that the proposed method, using only one image acquired post‐fire, could accurately identify the burned surface area (Kappa coefficient>0.8). The spatial resolution of the satellite images had practically no influence on the accuracy of the burned area estimate but did affect the possibility of detecting areas inside the perimeter of the burned area which were only slightly damaged.     

The role of topographic correction in mapping recently burned Mediterranean forest areas from LANDSAT TM images

Operational use of remote sensing as a tool for post‐fire, Mediterranean forest management has been limited by problems of classification accuracy arising from confusion of burned and non‐burned areas. Frequently, this occurs as a result of slope illumination and shadowing effects caused by the complex topography encountered in many forested areas. Cloud shadows can also be a problem. The aim of this work was to investigate how image classification results could be improved by removing the illumination effects of topography from satellite images. This was achieved by applying supervised classification to both uncorrected and topographically corrected LANDSAT TM data for a site on the Greek island of Thasos. The classification methodology included atmospheric and geometric correction, field‐based training, seperability/contingency analysis and maximum likelihood processing. The classification scheme was determined on the basis of consultation with the Greek Forest Service. Overlay of the resulting class maps enabled comparison of the total burned area and its spatial extent using the two different approaches to processing. The results of each approach were compared with the forest perimeter map generated by the Forest Service using traditional survey methods. Accuracy assessment and error analysis clearly indicated that the removal of the topographic effect from the satellite image before its classification resulted in more accurate mapping of the burned area. It is concluded that operational use of satellite remote sensing for forest fire management depends on accurate, robust, widely available and proven techniques. Topographic correction should now be regarded as an essential element of any classification methodology which will be used for operational, post‐fire management of forests in complex Mediterranean landscapes.     

Assessing the feasibility of sub-pixel burned area mapping in miombo woodlands of northern Mozambique using MODIS imagery

The goal of this study was to evaluate the feasibility of sub-pixel burned area detection in the miombo woodlands of northern Mozambique, using imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS). Multitemporal Landsat-7 ETM+ data were acquired to produce a high spatial resolution map of areas burned between mid-August and late September 2000, and a field campaign was conducted in early November 2000 to gather ground truth data. Mapping of burned areas was performed with an ensemble of classification trees and yielded a kappa value of 0.896. This map was subsequently degraded to a spatial resolution of 500 m, to produce an estimate of burned area fraction, at the MODIS pixel size. Correlation analysis between the sub-pixel burned area fraction map and the MODIS reflective channels 1-7 yielded low but statistically significant correlations for all channels. The better correlations were obtained for MODIS channels 2 (0.86 µm), 5 (1.24 µm) and 6 (1.64 µm). A regression tree was constructed to predict sub-pixel burned area fraction as a function of those MODIS channels. The resulting tree has nine terminal nodes and an overall root mean square error of 0.252. The regression tree analysis confirmed that MODIS channels 2, 5, and 6 are the best predictors of burned area fraction. It may be possible to improve these results considering, as an alternative to individual channels, some appropriate spectral indices used to enhance the burnt scar signal, and by including MODIS thermal data in the analysis. It may also be possible to improve the accuracy of sub-pixel burned area fraction using MODIS imagery by allowing the regression tree to automatically create linear combinations of individual channels, and by using ensembles of trees.     

Spatiotemporal problems with detecting and mapping mosaic fire regimes with coarse-resolution satellite data in savanna environments

Fire is a prominent disturbance factor and a major force of environmental change especially in the African savannas. The development of an accurate system to map and monitor fires on the African continent is a priority of numerous international research centers and programs. This effort has produced a flurry of research projects in recent years to detect and map areas affected by fires at the continental scale using coarse-resolution satellite imagery. The end product of these projects consists of weekly or monthly maps of burned area, several of which are available to the user community on the internet. It is argued here that the algorithms used to generate these products are designed to capture relatively large and contiguously burned areas and that the heterogeneous patterns of burn scars created by mosaic burning regimes pose a problem for current detection methodologies. Fine-scale burned area maps are generated using a series of Landsat ETM+imagery covering the 2002-2003 fire season for the study area in the savanna of southern Mali. These maps document a seasonal-mosaic pattern of burning in which burning begins early in the dry season and continues for several months ultimately affecting over 50% of the landscape. The majority of these fires burn relatively small areas producing a highly fragmented landscape pattern. A comparison of the fine scale maps with those from two widely available coarse-resolution products finds that the latter fail to detect approximately 90% of the burned area. A general argument is developed which suggests that the documented bias in the coarse resolution products is a function of low-resolution bias which derives from the fine-scale spatiotemporal pattern of burning not uncommon to savanna and other frequently burned environments. The study demonstrates how low-resolution bias can result in a significant underestimation of burned areas and/or a shift in the seasonal burned area profile in areas where mosaic burning occurs. The findings have implications for the development of broad-scale burned area detection algorithms as well as their applications to natural resource management and global environmental change research.     

Temporal and spatial variability in biomass burned areas across the USA derived from the GOES fire product

Burned area is a critical input to the algorithms of biomass burning emissions and understanding variability in fire activity due to climate change but it is difficult to estimate. This study presents a robust algorithm to reconstruct the patterns in burned areas across Contiguous United States (CONUS) in diurnal, seasonal, and interannual scales from 2000-2006. Specifically, burned areas in individual fire pixels are empirically calculated using diurnal variations in instantaneous fire sizes from the Geostationary Operational Environmental Satellites (GOES) WF_ABBA (Wildfire Automated Biomass Burning Algorithm) fire product. GOES burned areas exhibit diurnal variability with a temporal scale of half hours. The cumulative burned area during 9:00-16:00 local solar time accounts for 65%-81% of the total daily burned area. The diurnal variability is strongest in croplands compared to shrublands, grasslands, savannas, and forests. Analysis on a seasonal scale indicates that over 56% of burning occurs during summer (June-August). On average, the total annual burned area during the last seven years is 2.12 × 10⁴ ± 0.41 × 10⁴ km². The algorithm developed in this study can be applied to obtain burned area from the detections of GOES active fires at near real time, which can greatly improve the estimates of biomass burning emissions needed for predicting air quality.     

A MODIS direct broadcast algorithm for mapping wildfire burned area in the western United States

Improved wildland fire emission inventory methods are needed to support air quality forecasting and guide the development of air shed management strategies. Air quality forecasting requires dynamic fire emission estimates that are generated in a timely manner to support real-time operations. In the regulatory and planning realm, emission inventories are essential for quantitatively assessing the contribution of wildfire to air pollution. The development of wildland fire emission inventories depends on burned area as a critical input. This study presents a Moderate Resolution Imaging Spectroradiometer (MODIS) - direct broadcast (DB) burned area mapping algorithm designed to support air quality forecasting and emission inventory development. The algorithm combines active fire locations and single satellite scene burn scar detections to provide a rapid yet robust mapping of burned area. Using the U.S. Forest Service Fire Sciences Laboratory (FiSL) MODIS-DB receiving station in Missoula, Montana, the algorithm provided daily measurements of burned area for wildfire events in the western U.S. in 2006 and 2007. We evaluated the algorithm's fire detection rate and burned area mapping using fire perimeter data and burn scar information derived from high resolution satellite imagery. The FiSL MODIS-DB system detected 87% of all reference fires > 4 km², and 93% of all reference fires > 10 km². The burned area was highly correlated (R² = 0.93) with a high resolution imagery reference burn scar dataset, but exhibited a large over estimation of burned area (56%). The reference burn scar dataset was used to calibrate the algorithm response and quantify the uncertainty in the burned area measurement at the fire incident level. An objective, empirical error based approach was employed to quantify the uncertainty of our burned area measurement and provide a metric that is meaningful in context of remotely sensed burned area and emission inventories. The algorithm uncertainty is ± 36% for fires 50 km² in size, improving to ± 31% at a fire size of 100 km². Fires in this size range account for a substantial portion of burned area in the western U.S. (77% of burned area is due to fires > 50 km², and 66% results from fires > 100 km²). The dominance of these large wildfires in burned area, duration, and emissions makes these events a significant concern of air quality forecasters and regulators. With daily coverage at 1-km² spatial resolution, and a quantified measurement uncertainty, the burned area mapping algorithm presented in this paper is well suited for the development of wildfire emission inventories. Furthermore, the algorithm's DB implementation enables time sensitive burned area mapping to support operational air quality forecasting.     

Retrieving vegetation height of forests and woodlands over mountainous areas in the Pacific Coast region using satellite laser altimetry

The challenge to retrieve canopy height from large-footprint satellite lidar waveforms over mountainous areas is formidable given the complex interaction of terrain and vegetation. This study explores the potential of GLAS (Geoscience Laser Altimeter System) for retrieving maximum canopy height over mountainous areas in the Pacific Coast region, including two conifers sites of tall and closed canopy and one broadleaf woodland site of shorter and sparse canopy. Both direct methods and statistical models are developed and tested using spatially extensive coincident airborne lidar data. The major findings include: 1) the direct methods tend to overestimate the canopy height and are complicated by the identification of waveform signal start and terrain ground elevation, 2) the exploratory data analysis indicates that the edge-extent linear regression models have better generalizability than the edge-extent nonlinear models at the inter-site level, 3) the inter-site level test with mixed-effects models reveals that the edge-extent linear models have statistically-justified generalizability between the two conifer sites but not between the conifer and woodland sites, 4) the intra-site level test indicates that the edge-extent linear models have statistically-justified generalizability across different vegetation community types within any given site; this, combined with 3), unveils that the statistical modeling of maximum canopy height over large areas with edge-extent linear models only need to consider broad vegetation differences (such as woodlands versus conifer forests instead of different vegetation communities within woodlands or conifer forests), and 5) the simulations indicate that the errors and uncertainty in canopy height estimation can be significantly reduced by decreasing the footprint size. It is recommended that the footprint size of the next-generation satellite lidar systems be at least 10 m or so if we want to achieve meter-level accuracy of maximum canopy height estimation using direct and statistical methods.     

Pattern validation for MODIS image mining of burned area objects

An image mining method was applied to Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data to estimate the area burned by forest fires occurring in Galicia (Spain) between 4 August and 15 August 2006. Five different inputs were considered: post-fire near-infrared reflectance (NIR) band, post-fire Normalized Difference Vegetation Index (NDVI) image, pre-fire and post-fire NDVI difference image and 4-μm and 11-μm thermal bands. The proposed image mining method consists of three steps: a pre-classification step, applying kernel smoothing, based on the fast Fourier transform (FFT), a modelling step applying Gaussian distributions on individual grid cells with deviating values, and a thresholding step classifying the model into burned and unburned classes. Polygons collected in the field with Global Positioning System (GPS) measurements from a helicopter permitted validation of the burned area estimation. A Z-test based on the κ statistic compared the accuracy of this estimation with the accuracies achieved by traditional methods based both on spectral changes in reflectance after the fire and active fire detection. The results showed a significant improvement (7.5%) in the accuracy of the burned area estimation after kernel smoothing. Burned area estimation based on the smoothed difference image between pre-fire and post-fire NDVI image had the highest accuracy (κ = 0.72). We conclude that the image mining algorithm successfully extracted burned area objects and that extraction was best when smoothing was applied prior to classification. Image mining methods based on using the κ statistic thus provide a valuable validation procedure when selecting the optimal MODIS input image for estimating burned area objects.     

Downscaling real-time vegetation dynamics by fusing multi-temporal MODIS and Landsat NDVI in topographically complex terrain

Canopy phenology is an important factor driving seasonal patterns of water and carbon exchange between land surface and atmosphere. Recent developments of real-time global satellite products (e.g., MODIS) provide the potential to assimilate dynamic canopy measurements with spatially distributed process-based ecohydrological models. However, global satellite products usually are provided with relatively coarse spatial resolutions, averaging out important spatial heterogeneity of both terrain and vegetation. Therefore, bias can result from lumped representation of ecological and hydrological processes especially in topographically complex terrain. Successful downscaling of canopy phenology to high spatial resolution would be indispensable for catchment-scale distributed ecohydrological modeling, aiming at understanding complex patterns of water, carbon and nutrient cycling in mountainous watersheds. Two downscaling approaches are developed in this study to overcome this issue by fusing multi-temporal MODIS and Landsat TM data in conjunction with topographic information to estimate high spatio-temporal resolution biophysical parameters over complex terrain. MODIS FPAR (fraction of absorbed photosynthetically active radiation) is used to provide medium spatial resolution phenology, while the variability of vegetation within a MODIS pixel is characterized by Landsat NDVI. The algorithms depend on the scale-invariant linear relationship between FPAR and NDVI, which is verified in this study. Downscaled vegetation dynamics are successfully validated both temporally and spatially with ground-based continuous FPAR and leaf area index measurements. Topographic correction during the downscaling process has a limited effect on downscaled FPAR products except for the period around the winter solstice in the study area.     

Mapping forest degradation in the Eastern Amazon from SPOT 4 through spectral mixture models

In this paper, we present a methodology to map classes of degraded forest in the Eastern Amazon. Forest degradation field data, available in the literature, and 1-m resolution IKONOS image were linked with fraction images (vegetation, nonphotosynthetic vegetation (NPV), soil and shade) derived from spectral mixture models applied to a Satellite Pour L'observation de la Terre (SPOT) 4 multispectral image. The forest degradation map was produced in two steps. First, we investigated the relationship between ground (i.e., field and IKONOS data) and satellite scales by analyzing statistics and performing visual analyses of the field classes in terms of fraction values. This procedure allowed us to define four classes of forest at the SPOT 4 image scale, which included: intact forest; logged forest (recent and older logged forests in the field); degraded forest (heavily burned, heavily logged and burned forests in the field); and regeneration (old heavily logged and old heavily burned forest in the field). Next, we used a decision tree classifier (DTC) to define a set of rules to separate the forest classes using the fraction images. We classified 35% of the forest area (2097.3 km²) as intact forest. Logged forest accounted for 56% of the forest area and 9% of the forest area was classified as degraded forest. The resultant forest degradation map showed good agreement (86% overall accuracy) with areas of degraded forest visually interpreted from two IKONOS images. In addition, high correlation (R²=0.97) was observed between the total live aboveground biomass of degraded forest classes (defined at the field scale) and the NPV fraction image. The NPV fraction also improved our ability to mapping of old selectively logged forests.     

A performance evaluation of a burned area object-based classification model when applied to topographically and non-topographically corrected TM imagery

Operational use of remote sensing as a tool for post-fire Mediterranean forest management has been limited by problems of classification accuracy arising from confusion between burned and non-burned land, especially within shaded areas. Object-oriented image analysis has been developed to overcome the limitations and weaknesses of traditional image processing methods for feature extraction from high spatial resolution images. The aim of this work was to evaluate the performance of an object-based classification model developed for burned area mapping, when applied to topographically and non-topographically corrected Landsat Thematic Mapper (TM) imagery for a site on the Greek island of Thasos. The image was atmospherically and geometrically corrected before object-based classification. The results were compared with the forest perimeter map generated by the Forest Service. The accuracy assessment using an error matrix indicated that the removal of topographic effects from the image before applying the object-based classification model resulted in only slightly more accurate mapping of the burned area (1.16% increase in accuracy). It was concluded that topographic correction is not essential prior to object-based classification of a burned Mediterranean landscape using TM data.     

Development and analysis of a 12-year daily 1-km forest fire dataset across North America from NOAA/AVHRR data

Fires in boreal and temperate forests play a significant role in the global carbon cycle. While forest fires in North America (NA) have been surveyed extensively by U.S. and Canadian forest services, most fire records are limited to seasonal statistics without information on temporal evolution and spatial expansion. Such dynamic information is crucial for modeling fire emissions. Using the daily Advanced Very High Resolution Radiometer (AVHRR) data archived from 1989 to 2000, an extensive and consistent fire product was developed across the entire NA forest regions on a daily basis at 1-km resolution. The product was generated following data calibration, geo-referencing, and the application of an active fire detection algorithm and a burned area mapping algorithm. The spatial-temporal variation of forest fire in NA is analyzed in terms of (1) annual and monthly patterns of fire occurrences in different eco-domains, (2) the influence of topographic factors (elevation zones, aspect classes, and slope classes), and (3) major forest types and eco-regions in NA. It was found that 1) among the 12 years analyzed, 1989 and 1995 were the most severe fire years in NA; 2) the majority of burning occurred during June-July and in low elevation zones (< 500 m) with gentle slopes (< 10°), except in the dry eco-domain where more fires occurred in higher elevation zones (> 2000 m); 3) most fires occurred in the polar eco-domain, sub-arctic eco-division, and in the taiga ( boreal forests), forest-tundras and open woodlands eco-provinces in the boreal forests of Canada. The tendency for multiple burns to occur increases with elevation and slope until about 2500 m elevation and 24° slope, and decreases therefore. In comparison with ground observations, the omission and commission errors are on the order of 20%.     

Comparing ten classification methods for burned area mapping in a Mediterranean environment using Landsat TM satellite data

Various methods have been developed during the past three decades to improve the classification accuracy in burned area mapping using satellite data captured by different sensors. In this article, we compare ten such classification approaches using Landsat Thematic Mapper (TM) imagery on three Mediterranean test sites by evaluating the classification accuracy using (i) a traditional pixel-based approach, (ii) the concept of the Pareto boundary of efficient solution and (iii) linear regression analysis. Additionally, we make a discrimination of errors depending on their distribution and causal factor. The classification approaches compared resulted in not statistically significant differences in the accuracy of the burned area maps. Differences between the methods were also observed when considering the accuracy along the edges of the burned patches; however, again these were not statistically significant. The findings of our study in a Mediterranean environment clearly demonstrate that, for the selection of the most suitable classification approach, other factors could be given more weight, such as computational resources, imagery characteristics, availability of ancillary data, available software and the analyst's experience. Maybe the most important finding of our work is that the variance imposed by the methods is less than the variance imposed by factors differentiated locally in the three study sites since the between-group variance of the overall accuracy is higher than that of the within groups.