3.0 Technical Approach

3.1 Creation of Boreal Forest Fire Emissions Database

The model outlined in Eq. (2) will be used to generate a data base of CO₂ and other greenhouse gas (CH₄, CO₂, and NMHC) emissions from boreal forest fires. The derivation of the individual inputs into this carbon emissions model in Eq. (2) are presented below:

• Area burned (A)

Databases of the locations of the large fires (>100 ha in size) for the North American boreal forest have been created through digitizing the boundaries of these fires from records compiled by the Canadian Forest Service (CFS) (from 1980 to 1994) and the Alaskan Fire Service (AFS) (1950 to 1999) (Murphy et al. 1999). Efforts are currently underway to expand the Canadian database to include all fires back to 1950 and from 1995 through the current fire year. This activity is being funded by a grant from the Canadian Climate Change Action Fund. Both CFS and AFS are updating their databases annually.

While statistics on fires in Russia are available (Korovin 1996; Shvidenko and Nilsson 1999), it is recognized that these data are incomplete (Conard and Ivanova 1998; Kasischke et al. 1999b; Stocks 1991; Shvidenko and Nilsson 1999). In addition, the compilation of a similar database for Russia would represent a very difficult and time-consuming activity. As an alternative, we propose to use satellite imagery collected by the AVHRR and MODIS system to provide the baseline Russian large-fire database. As discussed above in Section 2.1, AVHRR data can be used to locate and map large fires in the boreal forest.

Cahoon et al. (1994, 1996) have already produced fire maps for Russia for 1987 and 1992. Cahoon and Stocks plan to complete a fire map for the years 1980 to 1990 at the NASA Langley Research Center. As part of the proposed study, we will use the algorithms developed by Cahoon to map fire boundaries for 1978 and 1979 and 1990 to 1999 and use the MODIS Fire Data Products to produce maps for 2000 to 2002.

Finally, we will use higher resolution satellite imagery for two purposes. First, we will use historical imagery to provide fire maps for missing fire records that have been identified for Canada and Alaska. Our familiarity with the fire databases for this region indicates that there are some missing data records for several years (e.g., the late 1960's for Alaska). Second, we will use the Landsat imagery to assess the overall accuracy of the AVHRR maps and the fire boundaries contained within the large fire data bases for North America. While it would be possible to conduct an evaluation of fire boundaries for an entire year using Landsat TM/ETM imagery, the cost of such an activity is beyond the funds provided for this project. Instead, we will design a stratified random sample to assess the accuracy of the fire boundary maps contained in large fire databases for both Russia and North America. The first strata will be to select a number of fires within the major ecoregions or ecozones that exist for North America (Bourgeau-Chavez et al. 1999) and Russia (Alexyev et al. 1999). The second strata will be to select fire boundaries from major fire event years that occurred in each region. We will then randomly select a subset of these fire boundaries to examine using Landsat imagery. Unlike other regions (such as savannas), the scars from fires in boreal forests are persistent and visible on satellite imagery for 5 to >10 years after a fire (Ahern et al. 1999; Michalek et al. 1999), and these scars are easily discriminated from non-burned areas. Thus, for this task we will use ETM data collected by Landsat 7, as well as historical Landsat TM imagery. In particular, we plan to make use of the global Landsat data set being purchased by NASA for the early 1990s time period. Based on the accuracy assessment, we will then produce correction coefficients to apply to the areas contained within the large fire databases.

• Temporal Patterns of Fire

Both the CFS and AFS have compiled data records that reflect the seasonal patterns of fire that occur in the North American boreal forests. These records will be consulted to create a temporal distribution of fire for this region. Our goal is to be able to distribute the area burned in each region into bi-weekly or monthly time bins.

For Russia, we will examine the AVHRR data sets to determine the spatial patterns of fire. This may require sub-sampling of a set of AVHRR for different regions to determine the temporal patterns of fire. In addition, we will consult databases that are being created by the Russian Forest Service using thermal imagery collected by AVHRR. Since 1993, Russian scientists have been mapping the spatial and temporal distribution of hot spots detected on AVHRR imagery. Although this approach does not detect and map all fire activity, it does provide a good indication of overall fire activity. Finally, we will also use the Fire Data products from MODIS to analyze patterns of fire in Russia for 2000 to 2002 in a similar way.

• Biomass/Carbon Densities (B_a, C_g)

Data bases presently exist that can be used to map the distribution of aboveground and ground layer biomass/carbon throughout the boreal forest region in North America and Russia. For aboveground biomass, we will use the digital GIS data bases compiled by the Canadian Forest Service (Lacelle et al. 1997; Lowe et al. 1996) and Kasischke et al. (1995a) for North America (as described in Bourgeau-Chavez et al. 1999) and by the St. Petersburg Research Institute of Forests and the Woods Hole Research Center for Russia (Alexyev et al. Birdsey 1999; Schlesinger et al. 1999). To create a carbon density map, we will assume that aboveground biomass is 50% carbon. For ground-layer carbon, we will use the soil carbon maps that have been produced by Lacelle et al. (1997) for North America and by Rozhkov et al. (1996) for Russia. Again, these maps already exist in a digital format.

• Burn Coefficients

Most of our understanding of the patterns of biomass/carbon consumption during fires is based on results of experiments during controlled burns (Stocks and Kauffman 1997, Shvidenko and Nilsson 1999) or through post-burn analysis of fires (Kasischke et al. 1999a). The challenge in estimating greenhouse gas emissions from fires is to take this limited number of field observations and extrapolate them over broad regions. First order estimates of carbon emissions can be calculated by using a single burn coefficient for very large regions (see, e.g., Seiler and Crutzen 1980; Cahoon et al. 1994). Other approaches have included developing models that assume the burn coefficient varies as a function of forest type (French et al. 1999; Kasischke et al. 1995a; 1999a) or as a function of overall fire activity during a specific year (French et al. 1999). More recently, approaches have been demonstrated to use Landsat imagery to estimate carbon release from fires (Kasischke et al. 1999c, Michalek et al. 1999). Michalek et al. demonstrated that the variations in spectral signatures of recent fires in Alaskan boreal forests are strongly correlated with patterns of burn severity, which in turn, can be directly associated with burn coefficients based on field observations.

In this project, we will use a combination of approaches to improve our estimates of burn coefficients for the different boreal forest regions. This approach will combine consulting with Russian and North American forest fire experts and using Landsat imagery to derive fire severity maps for specific regions. We will use the same Landsat imagery that was collected for the fire boundary study discussed above.

We will use a workshop format to initiate the interactions with local fire experts. A workshop in Canada will be organized by the Canadian Forest Service, while a workshop in Russia will be organized in collaboration with international scientific activities already being carried out, such as the IGBP Northern Eurasia Transect Study and the Global Observation of Forest Cover. The Russian workshops will be coordinated by the Russian collaborators on this project.

The focus of these workshops will be two fold: (1) to review the current understanding of the patterns of biomass consumption during fires and to arrive at a consensus among the experts as to average burn coefficients for different forest types/forest regions; and (2) to initiate field studies to provide the ground truth necessary to: (a) produce a burn severity map for different regions based on Landsat imagery; and (b) convert burn severity to burn coefficients. Under this project, we will provide Russian and Canadian fire experts with copies of the Landsat data products and provide some logistical support (to Russian participants) to carry out the field sampling. The information from the workshops and various field components will then be compiled and analyzed.

3.2 Atmospheric Modeling Approach

We will use monthly steady state NEP fluxes from the CASA model as a baseline for our analysis (Randerson, et al. 1997, 1999). Following from Eq. (2), emissions from monthly biomass burning anomalies will be added to the NEP fluxes for the period of 1980 to 1999. The combined NEP -biomass burning time series will then be used to drive the GISS atmospheric transport model used by Randerson in earlier studies (Randerson et al. 1997, 1999). We will compare CO₂ anomalies predicted from the transport model with flask observations (described below) to assess the variance that can be attributed to biomass burning.
 

3.3 Baseline Atmospheric CO₂ Data

The baseline data set to be compared to the boreal forest fire CO₂ emissions will be derived from monthly mean CO₂ concentrations from the NOAA/CMDL flask network. We will utilize data from those high northern latitude observation stations that have a continuous record from 1980, including Mould Bay (76° N), Point Barrow (71° N), Ocean Station M (66° N) and Cold Bay (55° N). These data will be filtered to remove the long-term secular trend (e.g., fossil fuel emissions) using a locally-weighted regression technique (Cleveland et al. 1990). In addition to examining the year-to-year variations in the minimum and maximum values of each data set, we will also calculate the linear rates of change using techniques by Randerson et al. (1997). These data sets have already been produced for the time period of 1980 to 1997 (Randerson et al. 1999); thus, for this study, we will only have to analyze data collected from 1998 to 2002.
 

3.4 Study Timeline

The various activities identified in Sections 3.1 to 3.3 will be carried out in an iterative or phased fashion over a three-year period. This approach is being adopted because many of the activities associated with processing and analyzing remote sensing data sets will take several years to complete. Rather than waiting for the completion of these studies, we will initiate an end-to-end analysis based on preliminary versions of the required input data sets.

During the first iteration (year 1), we will use existing large fire databases from Canada and Russia to estimate CO₂ emissions for the years 1980 to 1989. The area estimates contained in the databases will not be adjusted for biases as discussed in Section 3.1. We will estimate burn coefficients based on values that have been published in the scientific literature. At this time, we will assume a distribution of fires during the growing season based on published information. We will then utilize the atmospheric circulation/mixing models to determine how the boreal forest fire emissions contribute to the magnitude of the atmospheric CO₂ concentration at high northern latitudes, and compare these estimates to seasonal CO₂ variations derived from surface flask samples. During this initial iteration, we will alter various parameters within the biomass burning/carbon emissions model to determine the sensitivity of the resultant variations in the amplitude of the atmospheric CO₂ signature. The parameters to be varied include total area burned, fraction of biomass consumed, and timing of the fires during the growing season.

During the first year of the project we will also begin to compile and process the satellite remote sensing data sets required to address the three accuracy constraints associated with estimating fire carbon emissions. We will create a large-fire database for Russia for the period of 1978-2002 using fire-scar maps derived from AVHRR and MODIS imagery. At this time, we will begin an analysis of the timing of fires in the boreal forest based on a more detailed review of fire records (North America) and AVHRR/Landsat imagery. In addition, we will acquire and begin to process the Landsat TM/ETM imagery to: (1) improve the accuracy of total area burned derived from the large-fire databases; and (2) estimate patterns of burn severity as a function of ecoregion type.

At the beginning of the second year of the project, we will hold our workshops with the Canadian and Russian fire experts and distribute preliminary versions of the Landsat- derived burn severity maps to scientists who will provide field observations. During the second year, we will complete the analysis of the AVHRR imagery collected over Russia and produce our fire boundary map for the period 1978-1999. We will also complete our fire area accuracy assessment based on assessment of Landsat imagery.

Towards the end of the second year of the project, we will conduct a second exercising (iteration) of the carbon emission models using an expanded number of years as well a subset of the improvements in area burned and burn severity provided by analysis of the Landsat imagery. These estimates will then be expanded into atmospheric CO₂ amplitude variations (through the atmospheric circulation/mixing model) and compared to CO₂ flask estimates.

Finally, towards the middle of the third year, we will complete the analyses of all satellite imagery, and again exercise the various models for a final analysis.

The iterative or phased approach proposed for this project has several advantages. First, this approach will allow us to begin to investigate how boreal forest carbon emissions influence the atmospheric CO₂ record without having to wait until completion of the analysis of all the remote sensing data. The initial sensitivity analyses with the end-to-end model will allow us to determine what parts of the carbon emissions model result in the greatest variation in the atmospheric CO₂ amplitudes. This analysis will allow us to focus our efforts on reducing the uncertainties in these parameters. Finally, this approach is flexible enough to allow us to adjust our study to incorporate the results of other investigators. To a large extent, the improvement of the accuracy associated with constraints 2 (actual area of fires) and 3 (patterns of fire severity) is largely one of improved sample size. There is a good chance that other investigations will be initiated over the next few years by investigators from Canada, Russia, Europe or Japan under the auspices of the Global Observation of Forest Cover (GOFC) or the Global Terrestrial Carbon Initiative (TCI) that address these two constraints. Our experimental design allows for the incorporation of the results from additional studies.
 

3.5 Project Outputs

The results from this project will be summarized in a series of papers to be submitted to the appropriate peer-reviewed journals. Particular attention will be paid to the publication of results produced by our Russian collaborators. Access to results of scientific studies carried out in Russia is particularly difficult for western Scientists because most of this research is published in Russian-language journals and reports. We will also present our results at appropriate scientific meetings.

In addition, we will create a website that includes all the output data products generated from this study, including maps of: (1) fire locations for both North America and Russia; and (2) spatial/temporal maps of fire emissions of CO₂, CH₄, CO and non-methane hydrocarbons (NMHC). We also plan to post all the various coefficients from the different models developed under this study so that other investigators conducting similar studies have access to these data. Finally, the web page will contain copies of or linkages to the various carbon density data sets used in this study.