Burn severity estimation from remotely sensed data: Performance of simulation versus empirical models

https://doi.org/10.1016/j.rse.2006.11.022Get rights and content

Abstract

Burn severity is a key factor in post-fire assessment and its estimation is traditionally restricted to field work and empirical fitting from remotely sensed data. However, the first method is limited in terms of spatial coverage and cost effectiveness and the second is site- and data-specific. Since alternative approaches based on radiative transfer models (RTM) have been usefully applied in retrieving several biophysical plant parameters (leaf area index, water and dry matter content, chlorophyll), this paper has applied the inversion of a simulation model to estimate burn severity in terms of the Composite Burn Index (CBI). The performance of the model inversion method was compared to standard empirical techniques. The study area chosen was a large forest fire in central Spain which occurred in July 2005. The model inversion showed the most accurate estimation for high severity levels (for CBI > 2.7, RMSE = 0.30) and for unburned areas (CBI < 0.5, RMSE = 0). In both methodologies, the error associated to CBI from 0.5 to 2.7 was not acceptable (RMSE > 0.7), because it is higher than 25% of the total range of the index. Finally, burn severity maps from both methods were compared.

Introduction

Forest fires are a critical factor of disturbance in worldwide ecosystems. Their effects on soil and plants depend on frequency, fire intensity, and fire residence time, as well as on plant resilience and resistance (Pérez & Moreno, 1998). Moreover, the main consequences of fire on plants and atmospheric emissions depend largely on fire/burn severity. The term “fire severity”, which has a long tradition within the forest fire research community, refers to the combination of soil and overstory effects caused by fire (Brewer et al., 2005, Chappell and Agee, 1996, Doerr et al., 2006, Ryan and Noste, 1985, Turner et al., 1994, Wang, 2002, White et al., 1996). More recently, other authors have used the term “burn severity” to address the same concept (Chuvieco et al., 2005, Chuvieco et al., 2006, Key, 2005, Key and Benson, 2004, Key and Benson, 2005, Parra and Chuvieco, 2005, Patterson and Yool, 1998). This discrepancy of terminology makes comparing map products potentially ambiguous (Miller & Yool, 2002). To clarify these concepts, the analysis of fire effects can be better classified in the context of the fire disturbance continuum (Jain & Graham, 2004), which addresses three different temporal phases: before the fire, during the fire and after the fire. Within this framework, the term fire severity indicates the direct effects of the combustion process and refers to the active fire (direct effects of fire process). In contrast, burn severity identifies the impact of fire on soil and plants when the fire is extinguished, and it is related to the post-fire phase (what is left). The latter definition will be used in this paper.

Burn severity is generally estimated using post-fire field data (Moreno and Oechel, 1989, Pérez and Moreno, 1998), which consider several variables as: depth of char, percentage of tree basal area mortality (Chappell & Agee, 1996), decrease in plant cover (Jain and Graham, 2004, Rogan and Yool, 2001), volatilization or transformation of soil components to soluble mineral forms (Turner et al., 1994, Wang, 2002, Wells and Campbell, 1979), proportion of fine branches remaining on the canopy (Moreno & Oechel, 1989), and degree of canopy consumption and mortality (Doerr et al., 2006, Key and Benson, 2002, Patterson and Yool, 1998, Rogan and Franklin, 2001, Ryan and Noste, 1985, van Wagtendonk et al., 2004). The poor spatial representation associated with field methods and the cost of these approaches make it advisable to use alternative methods.

Remote sensing can be potentially a sound choice to map burn severity, since vegetation removal, soil exposure, changes in soil and vegetation moisture content imply changes in reflectance (Jakubauskas et al., 1990). Indeed, fire-related decreases in chlorophyll content and vegetation moisture lead to decreases in the visible and near-infrared (NIR) reflectance and increases in the mid-infrared (SWIR) reflectance (White et al., 1996).

Since the amount of green biomass destroyed by fires depends upon the burn severity, several authors have found good correlations between vegetation indices, computed from post-fire remotely sensed data, and burn severity (Díaz-Delgado et al., 2003, Doerr et al., 2006, García-Haro et al., 2001, Hammill and Bradstock, 2006, Ruiz-Gallardo et al., 2004, Sunar and Özkan, 2001). The Normalized Difference Vegetation Index (NDVI) has been related to field measurements of burn severity (Chafer et al., 2004, Hammill and Bradstock, 2006, Sunar and Özkan, 2001). NDVI is defined as:NDVI=(ρNIRρRED)(ρNIR+ρRED)where ρNIR and ρRED are the reflectance of near infrared (NIR) and RED bands respectively. However, according to White et al. (1996), a single post-fire band 7 (SWIR) of Landsat Thematic Mapper (TM) showed stronger correlations than NDVI. Likewise, Jakubauskas et al. (1990) used the 7/5 ratio of Landsat Multi Spectral Scanner (MSS) to map the burn area and extract degrees of burn severity. Other authors have found stronger correlations for spectral indices using the NIR and short wave infrared (SWIR) bands rather than the NDVI. Although, these NIR-SWIR indices were originally designed to estimate plant water content (De Santis et al., 2006, Fraser et al., 2000, Gao, 1996, Hunt and Rock, 1989), they have also proved useful to map burnt areas (López García & Caselles, 1991) since burning implies a severe decrease in plant and soil moisture contents. The most effective NIR-SWIR index for burn severity available in the literature is the Normalized Burn Ratio (NBR) proposed by Key and Benson (2002):NBR=(ρ4ρ7)(ρ4+ρ7)where ρ4 and ρ7 are the reflectance of band 4 (NIR) and 7 (SWIR) of Landsat TM respectively. Since burn severity is dependent on the pre-fire vegetation conditions, these authors suggest the use of the temporal difference between pre- and post-fire NBR (ΔNBR) values (Key & Benson, 2002):ΔNBR=NBRPRE − FIRENBRPOST − FIRE

This variable has been proposed as an operational index to estimate burn severity from satellite data.

The post-burn approach (simple NBR) is less expensive than the multi-temporal approach, and reduces the errors caused by differences in geometric correction, in sensor calibration, in sun-sensor geometry, in atmospheric effects and in plant phenology. However, the use of a single post-image, without the pre-burn reference image, leads to difficulties in mapping spectrally similar areas such water and recent burns, or senescent vegetation and older burns (Epting et al., 2005, Garcia and Chuvieco, 2004, Pereira, 1999, Pereira and Setzer, 1993).

The ΔNBR index calculated from Landsat TM and ETM + images have shown very strong correlation with burn severity values estimated in the field in several study cases (Cocke et al., 2005, Epting et al., 2005, Miller and Yool, 2002). A comparison between ΔNBR calculated from Landsat-TM and Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) data, showed very similar results (van Wagtendonk et al., 2004).

Other indices that include the mid-infrared spectral region have also shown high correlations, according to Rogan and Yool (2001) and van Wagtendonk et al. (2004), but generally did not perform as consistently as the NBR index. Similarly, the evaluation of six different approaches for classifying and mapping fire severity using multi-temporal Landsat TM data, performed by Brewer et al. (2005), confirms that the NBR provides a flexible, robust and analytically simple approach.

As well as spectral indices, linear transformation techniques have been used for the multi-temporal mapping of burn severity. Patterson and Yool (1998) compared two linear transformation techniques, the Kauth–Thomas (KT) and principal components (PC) transforms, for mapping fire severity. The KT or “Tasselled Cap” transform is sensitive to fire-induced changes in the moisture content of soil and vegetation and, in this study, produced betters result than the PC transform. Chuvieco (2002) and Caetano et al. (1994) concluded that spectral mixture analysis (SMA) proved to be efficient in detecting the charcoal signal even in lightly burnt areas that kept a strong vegetation signal, a situation that is typically considered to be problematic. SMA was considered advantageous over vegetation index-based methods, due to its improved capability to distinguish burns from other bare or sparsely vegetated areas (Caetano et al., 1996, Díaz-Delgado et al., 2001). This technique was also successful applied by Díaz-Delgado and Pons (1999) and Rogan and Franklin (2001) to carry out the burn severity classification.

The studies previously referred to are based on empirical approaches, which are relatively easy to compute, when a good set of field data is available. However, empirical approaches have also limitations due to the lack of physics introduced in the retrieval technique which reduces their generalization power (Weiss et al., 2000). Alternative approaches are based on radiative transfer model (RTM) techniques. In the forward mode, RTM help understand how the changes in plant biophysical parameters modify the spectral response at both leaf and canopy level, whereas inverse modelling uses spectral signatures as inputs to quantify plant parameters. The latter mode has been extensively used to estimate: leaf area index (LAI) (Fang and Liang, 2003, Koetz et al., 2005), water and dry matter content (Riaño et al., 2005, Zarco-Tejada et al., 2003), and chlorophyll content (Zarco-Tejada et al., 2001). The results of these studies are generally very precise, but the performance of RTM greatly depends on whether the assumptions of the model are properly met for a specific vegetation type and a specific biophysical parameter. There are few examples of the use of RTM in burned land mapping. Roy et al., 2005, Roy et al., 2002 have applied them indirectly for generating the standard burned land product of MODIS, which is based on multi-temporal analysis of BRDF (Bi-directional Reflectance Distribution Function) corrected images. Pereira et al. (2004) used these models to analyse the detectability of the burned signal in surface fires, which is a critical factor in tropical fires. In the specific field of burn severity, only a recent study by Chuvieco et al. (2006) explores the use of RTM (forward mode) to simulate different burn severities. The simulations were performed with a coupled leaf canopy model designed by Kuusk (2001). Several simulation scenarios were analysed, and different spectral indices proposed, underlying the importance of the red-edge (720–760 nm) and NIR-SWIR regions for discriminating burn severity values.

From an operative point of view, the final goal of RTM in remote sensing studies, is to be able to retrieve critical parameters by applying a model inversion technique. This has not yet been applied in burn severity studies and it is the main objective of this paper. The aim is therefore to determinate burn severity from the inversion of the simulation proposed by Chuvieco et al. (2006), in the context of an initial assessment, and to compare the performance of this technique with the empirical models.

Section snippets

Study area

The study area is located in the north-eastern part of the Province of Guadalajara (40–41°N and 1–2°W) in central Spain (Fig. 1). From a geological point of view, this site presents a sequence of limestone and sandstone deposits and the dominant types of soil are Xerochrept, Plexeralf and Xerorthent. The topography is rugged and the altitudes range between 1100 m and 1400 m. Rainfall in the region averages 600–800 mm per year. Maximum and minimum precipitations are recorded in November–December

Results of the field work

Most of the burned area presented very high observed CBI values. Only 28 of 103 field plots have CBI values lower than 2.5, and 32 have CBI higher than 2.9. In all cases the burn severity is higher in the understory than the overstory.

Empirical fitting between spectral indices and CBI values

Table 1 shows the correlations between the CBI of the total plot and a number of variables: reflective bands of the post-fire TM image, spectral indices and results from linear transformation.

The CBI of the total plot correlated most strongly with Greenness

Critical factors in burn severity estimation from remotely sensed data

The original CBI field form proposed by Key and Benson (2004) did not consider the percentage of dead leaves litter nor the FCOV of each stratum, but the results from the model inversion suggest that both variables and their mixing effects are key factors of burn severity estimation from remotely sensed data.

The first one can lead to an underestimation of the severity level of burned plots where high canopy consumption took place. The reason for this is that the spectral signature of dead leaf

Conclusions

Burn severity is a key factor in post-fire assessment, but has normally been estimated from field observations and empirical fittings using remotely sensed data.

The first method is costly in terms of time and both human and economic resources, and has a limited spatial coverage, although is critical in the verification phase of remote sensing methods. Empirical fittings between remotely sensed data and field sampling measurements of burn severity provide easy and generally accurate estimations

Acknowledgements

Special thanks to the Spanish Forest Service staff of the Province of Guadalajara who have been extremely helpful during the fieldwork campaign. The Spanish Ministry of Science and Technology supports Angela De Santis within the FPU Programme framework. Funding for this project was obtained through the Preview European Project (www.preview-risk.com).

References (69)

  • F.A. Kruse et al.

    The spectral image processing (SIPS)—interactive visualization and analysis of imaging spectrometer data

    Remote Sensing of Environment

    (1993)
  • A. Kuusk

    A two-layer canopy reflectance model

    Journal of Quantitative Spectroscopy & Radiative Transfer

    (2001)
  • M.W. Patterson et al.

    Mapping fire-induced vegetation mortality using Landsat thematic mapper data: A comparison of linear transformation techniques

    Remote Sensing of Environment

    (1998)
  • J.M.C. Pereira et al.

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

    Remote Sensing of Environment

    (2004)
  • J.L. Privette et al.

    Inversion of a vegetation reflectance model with NOAA AVHRR data

    Remote Sensing of Environment

    (1996)
  • D. Roy et al.

    Prototyping a global algorithm for systematic fire-affected area mapping using MODIS time series data

    Remote Sensing of Environment

    (2005)
  • D. Roy et al.

    Burned area mapping using multi-temporal moderate spatial resolution data—a bi-directional reflectance model-based expectation approach

    Remote Sensing of Environment

    (2002)
  • J.W. van Wagtendonk et al.

    Comparison of AVIRIS and Landsat ETM + detection capabilities for burn severity

    Remote Sensing of Environment

    (2004)
  • G.G. Wang

    Fire severity in relation to canopy composition within burned boreal mixewood stands

    Forest Ecology and Management

    (2002)
  • P.J. Zarco-Tejada et al.

    Water content estimation in vegetation with MODIS reflectance data and model inversion methods

    Remote Sensing of Environment

    (2003)
  • C.K. Brewer et al.

    Classifying and mapping wildfire severity: A comparison of methods

    Photogrammetric Engineering and Remote Sensing

    (2005)
  • M.S. Caetano et al.

    Assessment of AVHRR data for characterising burned areas and post-fire vegetation recovery

    EARSeL Advances in Remote Sensing

    (1996)
  • M.S. Caetano et al.

    Using spectral mixture analysis for fire severity mapping

  • D.L. Civco

    Topographic normalization of Landsat thematic mapper digital imagery

    Photogrammetric Engineering and Remote Sensing

    (1989)
  • C.J. Chafer et al.

    The post-fire measurement of fire severity and intensity in the Christmas 2001 Sydney Wildfires

    International Journal of Wildland Fire

    (2004)
  • G. Chander et al.

    Revised Landsat-5 TM radiometric calibration procedures and postcalibration dynamic ranges

    IEEE Transactions on Geoscience and Remote Sensing

    (2003)
  • C.B. Chappell et al.

    Fire severity and tree seedling establishment in Abies magnifica forests, Southern Cascades, Oregon

    Ecological Applications

    (1996)
  • P.S. Chavez

    Image-based atmospheric corrections. Revisited and improved

    Photogrammetric Engineering and Remote Sensing

    (1996)
  • E. Chuvieco

    Teledetección Ambiental: La observación de la Tierra desde el Espacio

    (2002)
  • E. Chuvieco et al.

    Use of canopy reflectance models to simulate burn severity levels

  • E. Chuvieco et al.

    Use of a radiative transfer model to simulate the post-fire spectral response to burn severity

    Journal of Geophysical Research-Biosciences

    (2006)
  • A.E. Cocke et al.

    Comparison of burn severity assessments using Differenced Normalized Burn Ratio and ground data

    International Journal of Wildland Fire

    (2005)
  • A. De Santis et al.

    Foliage moisture content estimation from 1-D and 2-D spectroradiometry for fire danger assessment

    Journal of Geophysical Research- Biosciences

    (2006)
  • R. Díaz-Delgado et al.

    Influence of fire severity on plant regeneration by means of remote sensing imagery

    International Journal of Remote Sensing

    (2003)
  • Cited by (154)

    View all citing articles on Scopus
    View full text