The shortcomings of wildfire spread modeling systems and the widespread use of their outcomes in fire management decisions render the evaluation of fire simulation results crucial for model calibration and improvement. This study proposes an exploratory evaluation framework of fire growth simulations using satellite active fire data.

Wildfires in the Mediterranean Basin, are frequent and have large environmental and socio-economic impacts. Projections of future climate point to an increase in frequency and severity of summer heat waves, such that an increase in the number and extent of wildfires is likely. Hence, wildfires impacts and resources required to manage them are also likely to increase in the future.

Spatially explicit wildfires spread models are often used to understand the intertwined relationships between fire, topography, fuel, and weather. More often than not, fire spread modeling systems are used to support fuel and fire management decisions without a proper evaluation of their outputs. Consequently, lack of systematic information on the quality of fire spread predictions can have a considerable impact on those decisions.

The potential of satellite active fire data can be applied in the evaluation of fire growth simulations although this requires an approach that can be objectively applied to a comprehensive number of large wildfires; uses metrics capable of evaluating simulation dynamics; and does not rely on the collection of reference burnt area perimeters.

Based on these needs, the current study proposes an exploratory approach for quantifying the spatial and temporal discrepancies between simulated fire growth and time series of satellite active fire observations.

Portugal is one of the Southern European countries most affected by wildfires. Nine large fires that occurred in central-southern Portugal were selected for this analysis (Fig. 1).

Taking advantage of technology in the fight against wildfires

The FARSITE, one of the most widely used fire propagation simulation systems, predicts fire behavior in spatially heterogeneous terrain and fuels landscapes, under variable weather conditions. FARSITE is based on a semi-empirical fire spread model that relies on Rothermel-based fuel models, and integrates models for surface and crown fire spread, dead fuel moisture and spotting from torching trees. Surface fire growth is simulated as an elliptical wave propagation based on the Huygens‘Principle to model the expansion of a polygonal fire front through time. FARSITE incorporates raster layers of topography and fuels, as well as weather data (temperature, precipitation, relative humidity, wind and cloud cover).

Satellite active fire data were used to determine the fire start and end dates, thus fire event duration, determine ignition location(s) and evaluate temporal and spatial discrepancies between active fire’s observations and simulated fire growth.

Fire data from the MODIS Global Monthly Fire Location Product (MCD14ML, collection 5)was used, which combines the middle-infrared and the thermal bands to detect fires that are burning at the time of overpass, providing information about the location, date, and time of the detected active fires.

Testing current models

Quantification of the spatio-temporal discrepancies of fire growth simulations was based on two novel measures: the Spatial Discrepancy (SpD) and the Normalized Relative Spatial Discrepancy (NRSpD). The SpD provides information on how distant the simulated fire growth is from the active fire(s) detected at a given satellite overpass, and can be integrated for the full individual fire length. The NRSpD is a relative measure useful for comparing simulations of fires of different sizes and durations and comparing time intervals of the same fire.

Researchers analyzed how some of the recognized limitations of MODIS active fire data can potentially affect the proposed satellite-derived fire growth discrepancy measures by focusing on two main issues: underestimation of fire activity and spatial resolution. Fires burning under persistent cloud cover and dense smoke plumes are hard to detect . Still, detection may be possible if the cloud or smoke layer is thin.

READ ALSO: Modeling fire size of wildfires

Some discrepancies are evident

Fire growth simulations were run and its estimates were compared with the satellite active fires position for the same time periods (Fig. 4). Three case studies (COV2, TAV and LL) showed reasonable spatial and temporal correspondence between the observed active fires and the corresponding fire growth simulations. The COV2 simulation covered nearly the entire fire perimeter, while trailing slightly behind compared with the satellite active fire positions. Similarly, for the TAV and LL case studies, fire growth simulations surpassed active fire progression denoting slight model overestimation.

While Fig. 2 maps the temporal distribution of active fires and simulated fire growth along each interval of elapsed time, quantification of the spatial discrepancy between both sources of data for each active fire position is mapped on Fig. 3. There are mainly two classes of spatial discrepancy values depicted in Fig. 3: observations with SpD lower than 1 km and NRSpD values between −0.2 and 0.2 (those with the best spatial agreement); and those observations with SpD N1–2 km, which correspond to those observations with a strong delay of the fire growth simulation (NRSpD values lower than −0.4). The combined information from Figs. 2 and 3 shows a general pattern of increasing discrepancies with time elapsed since ignition. In the southwest region of the COV2 case study, there are some differences between the timing of satellite detection and of simulated fire front (Fig. 2). The spatial discrepancy for most of the active fires is below 1 km, which denotes a good spatial agreement between simulation and active fire data.

In general, the NRSpD values increase towards 0 near the end of the fire’s duration.

Still need to improve

For most of the case studies, fire growth occurred during the initial 50–70% of fire length time, a pattern not followed by the simulations, except for the LL and TAV case studies. Fire growth simulations were delayed for most of the case studies when compared with satellite active fire positions (Fig. 2).

Overall, the SpD values are large and the NRSpD values close to −1, meaning that simulations underestimated fire growth with large spatial discrepancies (N1–2 km).

Potential underestimation of fire activity due to cloud cover and smoke plumes is a problem that affects thermal detections from all sensors, not only MODIS.

The proposed satellite-based evaluation scheme is applicable to any other satellite or airborne thermal dataset. The fusion of existent active fire data sets from existing and/or upcoming sensors, with improved spatial and temporal resolutions will enhance the applicability of satellite active fire data to evaluate fire growth simulations. In this context, VIIRS active fire data have remarkable potential given their higher spatial resolution, smaller footprint deformation and higher detection rates, when compared with MODIS data, particularly for small and low-intensity fires.

Overall, this study proposes a systematic evaluation approach and is a first attempt to show the potential of using MODIS active fire data to evaluate fire spread simulations.

Furthermore, the approach can also be extended to evaluate other spreading processes such as for example flooding, vegetation mortality and oil spills.

Source: Ana C.L. Sá , Akli Benali, Paulo M. Fernandes, RenataM.S. Pinto, Ricardo M. Trigo, Michele Salis, Ana Russo, Sonia Jerez, Pedro M.M. Soares,Wilfrid Schroeder, JoséM.C. Pereira. Evaluating fire growth simulations using satellite active fire data

Main photo credit: Google Earth satellite photo in the Portuguese zone between Alentejo and Algarve