Pollutant emissions from wildand fires can degrade air quality on local, regional, and continental scales.

The Regional Haze Rule, recent tightening of the National Ambient Air Quality Standards for fine particulate matter and ozone, and proposals to regulate greenhouse gas emissions have greatly increased the pressure on land management agencies to address the air quality impact of wildland fire. A major research effort by the Fire, Fuels, and Smoke Program focuses on the air quality impacts of pollutants from wildfires in the western U.S. The main goals of the research are: 1) to support land management agencies in developing strategies to minimize the air quality impacts of wildland fire and 2) assist the forecasting and short-term management of regional air quality. In an effort to achieve these goals, the Missoula Fire Sciences Laboratory (FiSL) has developed a satellite based fire detection system. The fire detection system is based on real-time Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data from the FiSL MODIS direct broadcast (DB) receiving station.

PRINCIPAL INVESTIGATOR

Shawn Urbanski, Physical Scientist

PROJECT OBJECTIVES

• Develop a MODIS-DB single scene burned area mapping algorithm capable of providing near-real-time observations of fire location and burned area
• Validate the MODIS-DB burn area mapping algorithm
• Implement a system to provide fire managers, air quality managers, researchers, and the general public with near-real-time access to fire burned area data

METHODOLOGY

MODIS burned area mapping algorithm

The FiSL operates a MODIS-DB receiving station which was the focal point for the burned area mapping algorithm development and implementation. The receiving station provided a Level 1B data set for each MODIS overpass received. The Level 1B data was processed using the standard MOD14 algorithm (Giglio et al., 2003) rendering the MODIS fire and thermal anomaly products. A single scene burn scar detection algorithm, described below, was applied to the Level 1B data to identify potentially burned pixels.

Burn scar algorithm

The MODIS fire and thermal anomaly algorithm (MOD14) is only capable of detecting fires active during the observation. A burn scar algorithm enables the detection of burned area, providing information of fire activity that occurs between observations. The burn scar algorithm used in this study is an implementation of that described previously in Li et al. (2004). Here we briefly describe our modifications, and detail the implementation to produce a burned area product. Details of the algorithm are provided elsewhere (Li et al., 2004; Hao and Urbanski, 2005).

While the Li algorithm effectively identifies burned pixels, operational use of the algorithm produced an unacceptably large number of false detections. The false detection rate was reduced by implementing the contextual filtering scheme which eliminates burn scar detections if they are not proximate to a recent fire detection (based on the MOD14 algorithm). To be retained, burn scars must be within 5 km of any fire detection from the preceding 10 days.

Fire Perimeters

Fire perimeters are generated following each MODIS overpass. The perimeter generation process combines active fire and single scene burn scar detections, transforming the collection of dissociated points to a set of “fire events”. Fire events are produced exclusively from the MODIS-DB data, without reference to external reporting sources.

Fire perimeter generation is an iterative process. First, current active fire detections are added to existing fire events if they are within 5 km of an event. Fire detections which cannot be associated with an existing fire event are considered new, and a new “event” is created. Next, burn scar detections within 5 km of a fire event are added to that event. After fire events have accumulated new detections, new fire event perimeters are constructed by buffering the detections with a circle of an appropriate diameter (500 m for burn scars, 1 km for active fire detections) about its center point. Using the new fire event perimeters, the final step is to check whether two or more fire events have merged. Fire events are considered to have merged into a single fire event if their perimeters are less than 5 km apart at the point of nearest approach. These perimeters grow with the addition of detections, merge together when appropriate, and become ‘inactive’ after 10 days without any active fire detections.

Burned Area Map Delivery

We have developed a platform to deliver the near-real-time fire data provided by the FiSL MODIS-DB fire detection system. Following each satellite overpass, the data is rapidly processed, providing active fire detections, burn scars and fire perimeters within 40 minutes of the observation. A web feature server allows users to easily download the fire data in a ready to use GIS format. The data available on the web feature server is immediately updated following the processing of data with each MODIS overpass. The platform allows users to select data for a specified region and time period (up to 30 days).

Validation

Fire detection
Performance of our MODIS-DB burned area algorithm in detecting fire occurrence was validated using final incident fire polygons for 1016 fires from 2006 and 2007. A fire event was considered detected if the incident fire polygon was within 1 km of any MODIS-DB fire perimeter. The MODIS-DB fire detection rate classified by MODIS Land Cover Type (Friedl et al., 2002) is given in Figure 1.

Burned area
The ability of our MODIS-DB burned area algorithm to measure fire burned area was evaluated using final incident fire polygons for 373 large fires (> 400 hectare) from 2006 and 2007 as ‘ground truth’. The MODIS-DB final fire event perimeters that were compared against each large fire polygon were selected by visual inspection. The MODIS-DB burned area is plotted against the final incident fire perimeter area in Figure 2. The MODIS-DB burned area is highly correlated (r2 = 0.87) with the ‘ground truth’ burned area. The ordinary least squares linear best-fit line is not significantly different from the 1:1 line (p < 0.01).

REFERENCES

Friedl, M.A., D. K. McIver, et al. (2002). “Global land cover mapping from MODIS: algorithms and early results”, Remote Sensing of Environment 83(1-2):287-302.

Giglio, L., J. Descloitres, et al. (2003). "An enhanced contextual fire detection algorithm for MODIS." Remote Sensing of Environment 87(2-3): 273-282.

Hao, W. M. and S. P. Urbanski (2005) "Automated Forecasting of Smoke Dispersion and Air Quality Using NASA Terra and Aqua Satellite Data". Final Report for Joint Fire Sciences Program Project 01-1-5-3. LINK

Li, R. R., Y. J. Kaufman, et al. (2004). "A technique for detecting burn scars using MODIS data. IEEE Transactions on Geoscience and Remote Sensing" 42(6): 1300-1308.

PROJECT STATUS

The MODIS-DB burned area mapping algorithm has been successfully developed and implemented. Fire data (active fire detections, burn scars, fire perimeters, and burned area maps) have been produced in a quasi-operational mode since the spring of 2006. Fire data have been available, in near-real-time, via the web feature server regularly since June 2008. The MODIS-DB burned area mapping algorithm has been validated. A manuscript reporting the validation study is in preparation and will be submitted by Dec 1, 2008. Additional tasks to be performed:

• Verification of burned area measurement on daily time-scale
• Assessment of burned area detection with respect to burn severity

FUNDING ORGANIZATIONS

NASA North American Carbon Plan
National Fire Plan
Joint Fire Science Program Project 01-1-5-03

PUBLICATIONS AND PRODUCTS

Final Report for JFSP Project 01-1-5-03