NASA Logo, National Aeronautics and Space Administration
LPVS website  banner


lpv logo

Focus Area on Fire Product Validation

Gareth Roberts, Southampton University, UK
Louis Giglio, University of Maryland, USA


Burned Area, Active Fire, and Fire Radiative Power Definition

Fire Disturbance includes Burned Area (BA) as the primary variable, and two supplementary variables: Active Fire (AF) and Fire Radiative Power (FRP), as stated by the Global Terrestrial Observing System (GTOS) in Csiszar et al. (2009). Burned Area is defined as the area affected by wildfires. Active Fire is the location of burning at the time of the observation. Fire Radiative Power is the rate of emitted radiative energy by the fire at the time of the observation.

Burned Area is expressed in units of area such as km2.
Active Fire is often a spatially explicit digital raster map indicating of presence or absence of fire, and can also be provided as spatial coordinate pairs.
Fire Radiative Power is expressed in units of power, such as Watts (W).

Highest Validation Stage Currently Reached for Satellite-derived Fire Products

Validation stage 2 (LPV validation stage hierarchy) - The highest LPV validation stage reached for satellite-derived burned area products. Most burned area products are made publicly available along with a Stage 2 validation.
In the framework of the Fire Disturbance project of the European Space Agency Climate Change Initiative, temporal stability of accuracy was assessed for the currently available products on ten study sites (Padilla et al. 2014b) and MODIS Burned Area product (MCD45) was validated using around 100 Landsat image pairs selected with probability sampling (Padilla et al. 2014a), thus achieving Stage 3 validation.

Validation stage 1 (LPV validation stage hierarchy) - The highest LPV validation stage reached for satellite-derived fire radiative power products. The main challenge to validate fire products is the generation of reference data due to the ephemerity of the phenomenon to be mapped and the current revisiting times of moderate-high spatial resolution observing systems. Therefore, reference data samples are scarce and not coincident with product estimates.

Validation Good Practices

The standard procedure to generated reference data for Burned Area includes the use of pairs of Landsat-class imagery, as described in CEOS LPV good practices guidelines (Boschetti et al. 2009). Cross-tabulation analysis is the most common validation approach, although other regional or patch-level analyses are of interest (Boschetti et al. 2004; Chuvieco et al. 2008; Giglio et al. 2009; Padilla et al. 2014a; Roy and Boschetti 2009). Validation results showed that most BA products have high to moderate commission and omission errors and tend to highly underestimate BA.

Active Fires have been validated with simultaneous observations with higher spatial resolution. This was the case for MODIS (Csiszar et al. 2006; Morisette et al. 2005), on board the Terra satellite that also carries a moderate resolution sensor (ASTER). More recently, Active Fires have been validated with Landsat imagery (Hantson et al. 2013; Schroeder et al. 2008; Tansey et al. 2008). The latter validation results show low commission errors while many fire patches are omitted.

Validation work is limited for fire radiative power since it requires coincident unsaturated radiance measurements (Csiszar et al. 2009). Coincident measurements currently are not routinely available, even opportunistic experiments used BIRD FRP estimates (Zhukov et al. 2006) and airborne thermal imaging systems (Schroeder et al. 2014). A field experiment is planned in Kruger National Park in August 2014 in support of validation activities.

The Fire focus area of CEOS LPV is currently developing a good practices protocol for the validation of satellite-derived active fire and burned area products. The current draft version on burned area validation will be complemented by sampling schemes of reference data and an active fire validation component.

Active Fire and Burned Area Reference Data Sets

  • Globcarbon project, from the European Space Agency: Reference data derived from 72 Landsat image pairs from year 2000.
  • Fire Disturbance project, from the European Space Agency: Reference data derived from 105 Landsat image pairs from year 2008, selected with a stratified random sampling, and from 112 Landsat image pairs, one for almost each year, from 1995 to 2009, at ten study sites.
  • MODIS Burned Area: Reference data derived from 11 Landsat images distributed across southern Africa.

NASA Logo -