Center for Earth Sciences Environmental Research Institute of Michigan Ann Arbor, Michigan Phone: (313) 994-1200 ext. 2566 FAX: (313) 994-5824 email: french@erim.org submitted for publication AGU Chapman Conference on Biomass Burning and Climate Change revised Jan 1996

Introduction

The large size, remoteness, and temporal variability in occurrence of wildfires in boreal forest regions make remote sensing techniques well suited for monitoring and studying wildfire. The goal of this paper is two-fold: First, to illustrate how different remote sensing systems detect signatures related to wildfires in boreal forests. And second, to demonstrate how information derived from remotely-sensed data can be used to study patterns of carbon flux from boreal forests. In addressing these goals, remote sensing data collected primarily over two test sites in the boreal forest region of Alaska are used. The first site is in an area where a 40,000 ha wildfire occurred in 1990 near Tok, Alaska (61° 21' N latitude, 142° 54' W longitude). The remote sensing data examined over this site include high-altitude aerial photography as well as and digital imagery from the satellite-borne AVHRR, SPOT, and ERS-1 SAR systems. The second site is in an area where two adjacent fires occurred near Circle, Alaska (65° 50' N, 145°W). SPOT multispectral imagery is examined for this test site.

Remote Sensing Signatures from Fire-Disturbed Boreal Forests

State-Wide Fire Mapping Using AVHRR NDVI Data

The NDVI images of Alaska were analyzed by looking at the difference in vegetation index late in the growing season, after wildfires had burned, versus early in the growing season, before the wildfires had begun. Areas where the vegetation index had dropped substantially from early in the season were considered to be affected by fire. Additional information produced from analyzing the delay in vegetation green-up in the NDVI data collected early in the following growing season was added to the results from the within-season difference analysis in order to obtain better estimates the amount of area burned. Sites disturbed by fire have a delayed green-up relative to undisturbed in the spring following the fire, so the boundaries of fire scars are easily seen in the early June images.

Figures 1 and 2 are shown as an example of how the NDVI data was used for a single fire. Figure 1 shows a three-season sequence of AVHRR NDVI composite data from the Tok Alaska study site. Figure 2 shows a fire history and boundary map of the Tok fire created by the Alaska Department of Natural Resources. The Tok fire event was actually a series of fires ignited by lightning strikes during July and August of 1990. The first fire started on 2 July and the last fire was extinguished on 7 August. A visual comparison of Figures 1 and 2 clearly shows that AVHRR is detecting the effects of this fire event. The area estimate for this fire event by the Alaska Department of Natural Resources was 39,661 ha. The digitization of the outer edge of the fire boundary map in Figure 2 produced an estimate of 41,090 ha. Analysis of the AVHRR data resulted in an area burned estimate of 40,200 ha (Kasischke and French, 1995).

The information derived from the analysis of the state-wide AVHRR data was compared to aerial and ground reconnaissance records maintained by the Alaska Fire Service (AFS) (Figure 3). This technique was used to detect 83% of the fires greater than 2,000 ha in size which occurred in 1990 and 1991 and provided an area estimate which 78% of the total area burned during the summers of 1990 and 1991. (Kasischke and French, 1995). Larger fires were detected with higher accuracy, for instance 100% of the fires greater than 10,000 ha were detected, and 88% of area mapped by the AFS was mapped using the AVHRR imagery.

Vegetation Mapping Using High-Altitude Aerial Photography

The initial studies in this region focused on the northern half of the Tok fire event, specifically the area lying within an outwash deposit to the south and west of the Tanana River. This region has little topographic relief, with the exception of 2 to 4 m of relief in remnant stream beds and on terraces adjacent to river flood plains. The soils originated from river sediment and fluvial outwash and consist of very-fine textured silt and clay underlain by coarser-grained sand and gravel. In some areas, the upper soil layers have a higher sand and gravel content, resulting in better drainage and dryer soils.

The drainage conditions determined by the soils are a principal factor in development of the forest ecosystems of this region. On better drained sites, white spruce forests develop. Following disturbance by fire, these dry, warm sites are initially invaded by quaking aspen (Populous tremuloides Michx.), which form in thick dense stands through cloning. Aspen dominates these sites for 50 to 100 years, during which time white spruce (Picea glauca [Moench] Voss) becomes established in the understory and eventually grows to dominate the overstory canopy after about 100 years. In colder, wetter sites with poor drainage, black spruce (Picea mariana [Mill.] B.S.P.) is the dominant canopy tree, becoming established through the release of seeds after fire. The buildup of mosses and dead organic matter in the ground-layer of these sites results in a further cooling of the ground layer, eventually leading to the formation of permafrost in the black spruce forests. On the most poorly-drained soils, the presence of continually-saturated soils further impedes establishment of black spruce, and allows for the establishment of shrub willow (Salix sp.), which in the wettest sites is the dominant aboveground cover.

To generate the pre-burn vegetation cover map for the Tok test site, four color-infrared aerial photographs were used. Boundaries of discrete vegetation types were delineated and digitized into a geographic information system (GIS). Once in the GIS, the four maps were merged and georectified using ground control points identified on a 1:63,360 U.S. Geological Survey topographic map. A vegetation type was assigned to each polygon (patch) in the GIS map based upon a visual interpretation of the photographs. The interpretation was performed using the visual differences in color (tone and shade) and apparent texture. Initially, 23 different forest-cover classes were defined. Several of these classes were combined because of their similarity in terms of the forest-successional chronosequences found in this region to create a total of 7 vegetation classes. Using the fire scar boundary derived from ERS-1 SAR image of 3 May 1993 (see below), the vegetation map was clipped to include only areas which had burned in the 1990 fire. A map of pre-burn vegetation within the area burned was created, and the total area of each vegetation class was computed using the GIS (Figure 4; Table 1).

The map of vegetation cover in Figure 4 illustrates the patchiness which is typical of a boreal forest in interior Alaska. This patchiness is due both to spatial variations in the soil texture as well as past disturbance. The distribution of patch size in this area (Table 2) has important ramifications for using satellite based remote sensing systems. Coarser resolution systems (such as AVHRR with a 1 km/100 ha resolution cell) are not able to discriminate smaller patches based upon their spectral signatures, and individual pixels in fact may represent a spectral mixture of multiple patches. Finer resolution systems, such as SPOT (20 m), Landsat TM (30 m) or ERS-1 SAR are required for spatial discrimination of the forest stands in this region.

Finally, the distribution of forest types in regions of Alaska is an important input for carbon flux models. In a study of carbon released during fires throughout the entire Alaskan boreal forest region, Kasischke et al. (1995b) used a forest-type distribution derived from field data collected in a mountainous region to the north of the Tok site. The distribution of the major forest types in the Tok site (e.g., black spruce forests versus white spruce forests) are very similar to those used by Kasischke et al. (1995b). Thus, the analysis of the aerial photographs validates the previous modeling work for this region.

SPOT Signatures from Fire-Disturbed Sites

Because of these restrictions, the study of fine resolution MSS data over recently fire-disturbed forests in interior Alaska to date has been limited to two images collected by the SPOT system. One scene was collected over the Tok fire site on 22 July 1992. The second scene was collected over a region to the east of Circle, Alaska on 13 August 1991. This scene imaged portions of a fire which occurred in 1990, and the entire area burned by a fire early in the summer of 1991.

Figure 5 presents a false-color composite image of the 1991 Circle image generated from SPOT bands 1, 2 and 3. The area burned in 1991 is clearly visible as the darkest portion of the image. The area burned in 1990 to the east of the 1991 burn is detectable, but less distinct than the 1991 burn. Analysis of the SPOT data showed most of the burn-scar signal was concentrated in the near-infrared channel (band 3). This channel was extracted and a level slice threshold determined based upon visual interpretation of the resultant image. The level slice was used to detected the area burned in the two Circle wildfires in the 1991 SPOT image (Figure 5). Note that there are variations within each burn that may be related to burn severity, but that this variation is generally less than that between the two burns. The band 3 thresholding technique detects the 1990 burn, but to a much lesser extent than the 1991 burn due to deterioration of the signature from vegetation regrowth. Also shown in Figure 5 is the boundary of the 1991 burn recorded by the Alaska Fire Service. While the overall boundary of the 1991 fire is consistent between the AFS records and the SPOT image, significant differences do exist.

From a standpoint of carbon flux measurement, it is important to note that high-resolution data such as that collected by SPOT can be useful in detecting regions within the outer fire boundary which did not burn. Such regions can be small in dimension, yet represent a considerable fraction of the total area burned. In coarse resolution imagery, such as that provided by AVHRR, smaller unburned areas are likely to be mapped as fire scar. For the 1991 fire shown in Figure 5, the area encompassed by the outer edge of the burn boundary mapped by the Alaska Fire Service was 23,224 ha. The area estimate of the burn derived from the AVHRR data was 19,500 ha (Kasischke and French, 1995). The area estimated via interpretation of the SPOT imagery was 16,245 ha. This considerably lower estimate is due to a large extent to the patches of unburned vegetation within the outer burn boundary. The difference is also a result of more precise delineation of the very irregular outer boundary, which is not well defined by aerial reconnaissance or AVHRR.

While SPOT or other high resolution MSS data collected during the same season as the fire can certainly provide a more accurate estimate of total area burned, it is not practical or feasible to collect this type of data over all regions that burn in a given year. However, fine resolution data provides a practical means to "calibrate" coarse resolution data sets, such as AVHRR, in order to estimate the magnitude of the errors encountered using data from the coarse resolution systems.

Figure 6 presents an NDVI image created from SPOT data ([band 3 - band 2]/[band 3 ~ band 2]) collected over the Tok site in July 1992 compared with an NDVI image created from the composite AVHRR data collected during the same time period (see Eidenshink, 1992). In these examples, the 1990 burn scar is not clearly visible in either of the images indicating that, in this case, the utility of NDVI at all resolutions for burn detection decays rapidly beyond the first year after fire (see Figure 1). Note, however, that the spatial variation in vegetation cover in this region is clearly more evident on the SPOT image than on the AVHRR image, reinforcing the fact that proper interpretation of vegetation regrowth patterns on the coarse resolution AVHRR data requires an understanding of the sub-pixel distribution of vegetation cover which are present in the region under study.

Thermal Signatures from Fire-Disturbed Sites

One result of fire in boreal forests is a considerable warming of the ground surface for several years (up to 10 to 20) after the fire. This warming occurs as a result of several factors, including: (1) removal of the tree canopy, which increases solar insolation of the ground surface; (2) a lowering of the surface albedo, which increases the amount of solar energy being absorbed; and (3) removal of some or all of the moss and organic soil, which acts as an insulating blanket causing cool ground temperatures when present (Dyrness et al., 1986; Kasischke et al., 1995a). It has been shown that in a fire-disturbed black spruce forest in interior Alaska, the ground temperature increases by > 5° C the first year after a fire (Viereck and Dyrness' 1979). This phenomena is quite important for post-fire biogenic emission of carbon. Increased soil temperature encourages microbial activity and organic matter decomposition.

Figure 7 presents an AVHRR band 4- (11 m m) derived image of the Tok site illustrating the potential utility of thermal data for monitoring variations in surface temperature in fire-disturbed sites. As an initial look at the ability to use the thermal infrared AVHRR channel for inputs to carbon flux studies, the composite AVHRR data set was used. Although problems in interpretation may arise when working with multiple-date composite imagery, this data set is convenient to use and easily available, so it lends itself well to initial studies of the utility of AVHRR for various applications. Pre- and post-burn "anniversary" band 4 composite images (16-30 June 1990, 1991) were used to explore the potential for post-fire monitoring of burn scars.

The two composite images were normalized to correct for differential ambient and sensor conditions, and a difference image was computed. The difference image was color-level sliced (Figure 8). The areas of high thermal difference compare well to the fire scar boundary at Tok. Using the brightness temperature calibration information provided with the AVHRR data, approximate temperature increases (± 1° C) for the normalized composite brightness values were computed. Within the fire-scar boundary, surface temperature appears to have increased by approximately 5° C on average (with a range of approximately 2° to 12° C). The average brightness temperature increase compares well to results found during ground-based studies (Viereck and Dyrness, 1979). While there are many caveats on the use of composite AVHRR data to monitor surface temperature changes, the results shown in Figure 8 illustrate the potential of using the AVHRR thermal channels to monitor temperature changes if fire-disturbed boreal forests.

Spatial and Temporal Variations in ERS-1 SAR Signatures

Figure 8 presents the series of ERS-1 images collected over the 1990 Tok fire during the summers of 1992 to 1994. These images show how the characteristic radar signature varies both temporally and spatially. They also show that topographic relief influences the utility of single-date ERS-1 SAR imagery for detecting fire-scar signatures. In this case, in the mountainous regions of the fire there is no obvious SAR signature. However, in other locations in Alaska, characteristic radar signatures are visible in mountainous terrain disturbed by fire (Kasischke et al., 1992; Bourgeau-Chavez et al., 1996).

At times when the fire scar is particularly bright in the ERS- 1 imagery, as it is in the early May images, it is possible to accurately delineate the area burned, particularly in the flat outwash plain area. The fire boundary as defined using the ERS-1 data is more accurate than the boundary delineated using the coarse resolution AVHRR system because of the higher spatial resolution of the ERS-1 system. Like the boundary defined with the SPOT data at Circle, the ERS-1 derived fire boundary at Tok takes into consideration areas of unburned forest within the external boundary. The result is a better estimate of area burned than with AVHRR or the Fire Service records. For the Tok fire, the area burned in the southern part of the burn is not detectable on single-date SAR imagery because of the mountainous terrain. However, analyses have shown that the boundary of the entire burn scar is detectable using a color-composite image derived from multiple-date ERS-1 SAR imagery (Bourgeau-Chavez et al., 1996). Areas not burned within the outer boundary were determined using the 3 May 1993 ERS-1 image, while the outer boundary of the entire burn was determined with the three-date composite ERS-1 image. The result is a burn area of 38,000 ha estimated using ERS-1, versus 40,200 ha using the AVHRR imagery. With unburned areas within the burn not taken into consideration, the ERS-1 estimate of the fire is 39,325 ha, which is very close to the AVHRR-derived estimate.

Temporal changes in the ERS-1 radar signatures from Tok are significant (Figure 8). Figure 9 presents monthly temperature averages and precipitation totals from two weather stations located near the Tok fire event. Studies have shown that the inter- and intra-annual variations in the average radar image intensities from the Tok fire event are strongly correlated to soil moisture variations (French, 1994; French et al., 1996), which in turn are dependent on snow melt, soil thawing, precipitation, and soil drainage conditions. Overall, the most distinct radar signatures are observed early in the growing season because melting snow and frozen ground conditions lead to high moisture levels in the upper soil layers. Once the soil thaws out and water begins to drain, then variations in soil moisture are due to temperature and precipitation patterns combined with spatial variations in soil drainage. The radar images are generally brighter during June through August in 1992 than in 1993 and 1994 because of more precipitation in 1992 (there was 5.9 cm or 36% more precipitation in the April to August time period in 1992 than in the same time period in 1993 and 1994. Figure 9a) and cooler temperatures (the average daily high temperature in the April to August time period was 15.2° C in 1992, 17.7° C in 1993, and 18.9° C in 1994. Figure 9b).

Figure 10 illustrates the dependence of radar backscatter on precipitation and soil moisture. Figure 10a plots average precipitation in the 5 days previous to each ERS-1 overpass versus average radar backscatter for the entire Tok fire event (exclusive of mountainous regions). For the early May data, it was assumed that 20% of the moisture present in the snow pack present four weeks prior to the ERS-1 SAR overpass remained in the upper ground layer. The other 80% was assumed to have sublimated or evaporated. With the exception of the data from July 28,1993, the variations in precipitation and snow-pack moisture explain a large percentage of the variation in the radar imagery.

Field measurements were collected in the summers of 1993 and 1994 to find an empirical relationship between soil moisture and ERS-1 radar backscatter (French, 1994; French et al., 1996). Volumetric soil moisture measurements were collected from four different sites on two dates in 1993 and from 5 different sites on two dates in 1994. The results from this analysis show there is a strong correlation between soil moisture and radar backscatter within the Tok fire scar region (Figure 10b).

Regrowth of vegetation has been eliminated as a potential source of radar intensity variation in the Tok fire region because it is so low within the time period studied. Measurements made during the summer of 1994 show that the average aboveground, living biomass levels present in this region are 1.5 t/ha, mostly in herbaceous vegetation. This low level of plant regrowth does not have a great effect on radar backscatter, both in terms of attenuation of the soil moisture signature as well as in direct backscattering.

Recent analysis of the ERS-1 SAR signatures from the Tok burn (Figure 8) when compared with the pre-burn vegetation map (Figure 4) show a relationship between pre-burn vegetation and the spatial and temporal variations in the post-burn SAR signature. For example, the area that appears bright in the 12 August 1992 image (Figure 8) corresponds directly with areas of black spruce/willow and black spruce, while the darker areas correspond to white spruce and aspen dominated areas. These initial observations indicate that further analysis of the relationship should be conducted to understand how pre-burn vegetation patterns effect the post-burn SAR signature. The relationship of soil water status on SAR backscatter is understood, but the role of ecosystem type and burn severity has not yet been thoroughly investigated. Further studies are currently underway to look at these questions in more detail.

Using Remote Sensing Data to Estimate Carbon Flux

Carbon Release During the 1990 Tok Forest Fire

Local history indicates that large-scale wildfires occurred in the Tok region about 100 to 150 years ago; therefore, the forest stands found in the region evolved through succession after these fires. Field measurements have established the average carbon levels of the aboveground or canopy layer (e.g., all aboveground biomass found in trees and shrubs, including living and standing dead) and the ground layer (e.g., biomass found in living and dead mosses, litter, humus, and organic soil). Variations in the carbon density of these components can be described in terms of the environmental gradient controlling forest succession (Figure 11a). The highest canopy-layer carbon densities are found in the white spruce forests present on the dry/warm sites, while the lowest are found in the willow peatlands of the cold/wet sites. Conversely, the highest ground-layer carbon densities are found in the spruce peatlands and the lowest in the white spruce forests. The average carbon densities presented in Figure 11a are based upon field measurements from test stands in the Tok study site. Note that on average, there is much more carbon stored in the ground-layer than in the canopy layer.

From field measurements collected at different test stands at the Tok study site, first-order models of biomass consumption during fire have been developed for the different forest types found in this region. The consumption of the canopy layer biomass (defined as all aboveground biomass) occurs during a flaming stage as the fire front progress. The consumption of the ground-layer biomass occurs during smoldering fires after the flaming stage. Field measurements show that in the canopy layer, the lowest levels of biomass consumption occur in white spruce forests (~15%), while the highest levels (35% to 50%) occur in the black spruce forests and the willow peatlands. This biomass consumption gradient is illustrated in Figure 11b.

Field measurements show that the highest levels of biomass consumption in the ground layer during the Tok fire (a mid-summer fire) occurred in the white spruce sites (70%), while the lowest levels occurred in the black spruce sites (35%). The measurements of ground-layer biomass consumption in the willow peatland are incomplete at this time. For the purposes of this study, because of the wetter ground-layer conditions, it has been assumed that this site has the lowest ground-layer biomass consumption (20%). The ground-layer biomass consumption for a mid-summer fire such as the Tok fire is also illustrated in Figure 11b.

Finally, it is important to consider the timing of the fire relative to the growing season. A large number of fires in boreal forests occur early in the growing season, immediately after snow melt. Because the canopy layer is extremely dry at this time, fires will propagate themselves over extensive areas via burning of the understory vegetation as well as the crowns of the overstory trees. The level of canopy biomass consumption during these fires should be close to the level burned during fires later in the growing season. In the early spring, the ground-layer is essentially frozen, which retards smoldering fires and lessens the amounts of biomass consumed in the ground layer. While actual measurements do not exist, for the purposes of our model, it was assumed that during early spring fires, only 10% of the ground layer is consumed in the dry/warm sites and 5% of the biomass is consumed in the cold/wet sites (Figure 11b).

Figure 11c presents the estimated patterns of carbon release during biomass burning for the different ecosystem types of the Tok study site. The plots in this figure were derived by multiplying the average carbon density (Figure 11a) times the percentage biomass consumption (Figure 11b). These plots show: (1) most of the carbon released during wildfires originates from the ground layer; (2) there is a definite pattern in carbon release as a function of the environmental gradient; and (3) the highest carbon releases are from the black spruce forests of this region. The amount of carbon release from the black spruce forests is 30% to 40% greater than that released from white spruce forests or the spruce/willow peatlands.

Figure 11d presents the estimated patterns of carbon release from biomass burning in early spring fires. Because of the lower levels of consumption in the ground layer, the overall levels of carbon release are about one-third the levels of carbon release during a mid-summer fire.

In order to determine how much carbon was released during the 1990 fire in the outwash plain area of the Tok study site, the map of vegetation type developed from the aerial photographs was used (Figure 4; Table 1). Each type was assigned a position along the environmental gradient depicted in Figure 11 to estimate the amount of carbon present as well as the expected patterns of carbon release during biomass burning. These estimates were summarized to determine the average amount of carbon released during the Tok fire. This approach resulted in an estimate of carbon released during biomass burning of 33.2 t/ha, which is comparable to the estimate of 28.8 t/ha using the model developed by Kasischke et al. (1995b). In contrast, a figure commonly used to estimate carbon release during boreal forests has been 11.3 t/ha (Stocks 1991; Cahoon et al., 1994). If we assume that the fire burned early in the spring, then the lower levels of biomass burning in the ground layer reduces the estimate of carbon released by a factor of three, to 10.1 t/ha.

State-Wide Estimates of Carbon Released During Wildfires Using AVHRR Data

Cahoon et al. (1994) used a simple model to estimate the amounts of carbon released during forest fires in northwest China/southeast Russia in 1987. The areal extent of the fires was determined from AVHRR data. This approach assumed a single value for the amount of biomass consumed in both the aboveground and ground layers during these fires. These estimates were then used as a basis for estimating carbon released.

Kasischke et al. (1995b) used a more detailed model to estimate the amount of carbon released during fires in Alaska in 1990 and 1991. The model was based upon the ecological studies of Viereck et al. (1983, 1986), and divided the forests of this region into two types. It also assumed stand age distribution in the region was determined by the fire frequency and assumed different biomass accumulation and burning patterns for the two forest types.

Studies at finer temporal and spatial scales, such as those being conducted at Tok described previously in this paper, provide the basis for improving the accuracy of models used with coarser-scale remote sensing data, such as AVHRR. Furthermore, the model of Kasischke et al. (1995b) did not assume any seasonal variations in the patterns of biomass burning in Alaskan boreal forests. Such information can easily be determined from AVHRR data.

To illustrate the potential effects of timing of fires on carbon release from boreal forests, the fire history records maintained by the Alaska Fire Service were consulted. Specifically, the start/stop times for all fires > 1,000 ha in 1990 and 1991 were determined from this data base. The total area burned within the state was then divided into three categories: (1) early season burns (prior to 1 June), (2) mid-season burns (1 June through 15 July), and (3) late season burns (after 15 July). Based on this categorization in 1990, 0.2% of the biomass burning occurred in the early season, 19.4% in the mid-season, and 80.5% in the late season. In 1991, 0.3% of the burning occurred in the early season, 30.6% in the mid-season, and 69.1% in the late season.

The biomass burning estimate used by Kasischke et al. (1995b) was for late season fires. To account for seasonal variations in biomass burning, it was assumed the biomass burning level in the early season is 1/3 of the late season level, and the mid-season level is 2/3 of the late season level. Using these figures, it was estimated that a total of 0.0312 Pg of carbon were released during biomass burning in Alaskan boreal forests in 1990 and 1991, which compares to the estimate of 0.0339 Pg by Kasischke et al. (1995b).

Monitoring of Post-Fire Biogenic Emissions

As discussed previously in this paper, fires in boreal forests cause significant changes in the spatial and temporal patterns of soil moisture and soil temperature in the ground layer. For example, Figure 12 presents a soil temperature profile collected in a set of black spruce test sites in the Tok Alaska region in August 1995, five years after a fire occurred. Temperature increases in burned sites are typical for boreal forests and result in significant increases in rates of decomposition and release of carbon dioxide.

A model developed by Bunnell et al. (1976) has been used calculate seasonal patterns of soil respiration in Alaskan boreal forests (Schlentner and Van Cleve, 1985; Bonan and Van Cleve, 1992). Total soil respiration, R. (g CO₂ × m^-2 × h^-1) is calculated as
 

(l)

where M is the average forest floor moisture content (percent by dry weight), T is the soil temperature (°C) at a depth of 15 cm, and the coefficients a_l, a₂, a₃, and a₄ are substrate specific soil properties related to water holding capacity, theoretical respiration rate, and temperature (Bunnell et al., 1976).

Figure 13 presents measurements of CO₂ emission from burned and unburned black spruce stands in the Tok region collected in mid-September 1995. These measurements were collected in the late afternoon within several hours of each other, when the air temperature was 20°C. Very little vegetation was present in the burned stands, therefore it is assumed that all of the respiration is due to microbial respiration (decomposition). For the unburned stand, it is assumed that 20% of the total respiration is due to decomposition (Bonan and Van Cleve, 1992).

Figure 14 presents a plot of soil respiration estimated based on Eq. (1). Model predictions for total respiration and the fraction due to decomposition (20%) are shown along with data points derived based on measurements taken at the Tok site. Soil temperatures were measured at the time of the CO₂ measurements presented above, and estimates of soil moisture were made based on field experience at these sites. The coefficients for Eq. (1), al through a4, were derived empirically from laboratory and field measurements for mature stands of aspen, white spruce, and black spruce (Bunnell et al., 1976), and therefore may not be appropriate for estimation of respiration in fire disturbed sites. The burned test site had an average soil temperate (at a 15 cm depth) of 6.7° C, while the unburned stand had a soil temperature of 3.6° C. The plot shows that the observed CO₂ from the test sites was significantly higher than predicted for decomposition at these two temperatures. However, the relative levels of respiration based upon moisture and temperature match the model results quite well. Additional field and laboratory measurements should allow for refinement of these predictions or development of additional models to estimate respiration in fire disturbed forest sites. The measurements made here, however, show that temperature and soil moisture are key in determination of decomposition, and therefore CO₂ flux, in fire-disturbed sites.

In order to study the patterns of soil respiration in fire disturbed forests requires estimation of the spatial and temporal patterns of soil moisture and temperature. Soil moisture does not have a clear seasonal pattern, but is related to precipitation patterns, which are highly variable, and to soil drainage, which is controlled by mineral soil composition and the presence or absence of permafrost. Even with precipitation records, it would be very difficult to estimate the spatial and temporal patterns of soil moisture to provide inputs to a soil respiration model. Likewise, soil temperature is difficult to measure because of its spatial variability. Remote sensing can be helpful for estimation of moisture and temperature. Because SAR is sensitive to variations in soil moisture in boreal forests (Figures 9 and 11), it can be used to estimate soil moisture. In a similar fashion, soil temperature can be inferred from surface temperatures, which in turn, can be derived from satellite observations using thermal wavelengths.

Conclusions

To a great extent, over the past decade the scientific community has directed most of its attention on biomass burning in tropical and subtropical biomes, primarily savannas and tropical forests. The focus of these studies has been to develop a better understanding of the effects of biomass burning on climatic processes, including the development of methodologies to use satellite-based remote sensing systems to monitor the location, extent and timing of fires in this region. From a scientific perspective, the causes of biomass burning in both savannas and tropical forests are anthropogenic in their sources, but the types of biomass burning and their effects on land cover are distinctly different. The major global change issues being addressed in these regions deal largely with understanding the short and long term effects of fire on atmospheric chemistry and physics. The methodologies in using coarse-resolution satellite systems to monitor biomass burning in savannas and tropical forests are quite similar, with thermal infrared sensors being used to detect and map active, ongoing fires which are typically sub-pixel in size.

It has only been recently that scientists have begun to realize the extent and importance of biomass burning in boreal forests. In studying fires in boreal forests, it is important to realize that both the issues related to and the methods used to monitor biomass burning are dramatically different than those for savannas and tropical forests.

First, most of the area burned in boreal forests is the result of natural, not anthropogenic, fire. In addition, these fires typically cover very large areas, with > 90% of the total area burned occurring if fires greater than 100 square kilometers in size. Second, fire is an important ecological process in boreal forests. To understand the impacts of fire in boreal forests on the global climate requires an understanding of the ecological effects of fire. Third, seasonal and interannual variations in fire in boreal forests are closely tied to seasonal and interannual variations in the patterns of temperature and precipitation. If the projected patterns of global climate change occur, it is probable that a significant percentage of the carbon presently stored in boreal forests will be released into the atmosphere because of the direct and indirect effects of fire; fire occurrence, in fact, will increase significantly (Kasischke et al., 1995a). Thus, the most pressing scientific issue related to fires, boreal forests, and climate change is not the present contribution of biomass burning to atmospheric chemistry and physics, but future affects on the global carbon budget and the atmospheric concentration of greenhouse gases.

Most of the approaches developed for satellite monitoring of biomass burning in savannas and tropical forests are not appropriate for monitoring of fires in boreal forests. The techniques used in tropical and sub-tropical regions evolved out of the need to monitor fire events which are relatively small in size (several hectares in size, at most) and short-term in duration (e.g., several hours in length). Therefore, there is a requirement for daily (or more frequent) sampling of regions where biomass is occurring. In contrast, large fire events in Alaska have a duration of many weeks. During the 1990 and 1991 fire seasons in Alaska, fires 1 to 10 km² in size had an average duration of 44 days. Fires 10 to 50 km² had an average duration of 60 days. And fires >50 km² in size had an average duration of 78 days. In addition, these large fires leave a distinct scar which is clearly detectable through differences in surface reflectance in the visible, near-infrared, and thermal remote sensing bands. Given the fact that a very high percentage of total area burned occurs in large fires which last many days, it is not necessary nor is it desirable to try to monitor fire location and extent on a daily basis. Rather, images collected after the fire is extinguished are ideally suited to map the location and areal extent of the large fire events in boreal forests.

In addition to monitoring the location and extent of fires, for studies of carbon flux it is also necessary to monitor the effects of fire on specific ecological processes in boreal forests. Satellite remote sensors provide a unique means to monitor post-fire surface characteristics which are important in studying these processes, including patterns of fire severity, soil moisture, surface temperature, and vegetation regrowth. This information can be derived from a wide range of remote sensing systems operating over different wavelength regions (e.g., visible, infrared and microwave) and spatial and temporal scales.
 

Acknowledgments

References

Bourgeau-Chavez, L.L., P.A. Harrell, E. S. Kasischke, and N.H.F. French, The detection and mapping of Alaskan wildfires using a spaceborne imaging radar system, Int. J. Remote Sens (in review), 1996.

Bonan, G.B. and K. Van Cleve, Soil temperature, nitrogen mineralization and carbon source-sink relationships in boreal forests, Can. J. For. Res., 22, 629-639, 1992.

Bunnell, F.L., D.E.N. Tait, P.W. Flanagan, and K. Van Cleve, Microbial respiration and substrate weight loss - I: a general model of the influences of abiotic variables, Soil Biol. Biochem., 9, 33-40, 1977.

Cahoon, D.R. Jr., J.S. Levine, W.R. Cofer III, J.E. Miller, P. Minnis, G.M. Tennille, T.W. Yip, B.J. Stocks, and P.W. Heck The great Chinese fire of 1987: a view from space, in Global Biomass Burning: Atmospheric, Climatic and Biospheric Implications, edited by J.S. Levine, pp. 61-65, The MIT Press, Cambridge, Massachusetts, USA, 1991.

Cahoon, D.R., Jr., B.J. Stocks, J.S. Levine, W.R. Cofer III, and J.M. Pierson, Satellite analysis of the severe 1987 forest fires in northern China and southeastern Siberia, J. Geophys. Res., 99, 18,627-18,638, 1994.

Dyrness, C.T., L.A. Viereck and K. Van Cleve, Fire in taiga communities of interior Alaska. Pages 74-86 in K. Van Cleve, F.S.. Chapin III, P.W. Flanagan, L.A. Viereck and C.T. Dyrness, editors. Ecological series vol. 57: Forest Ecosystems in the Alaskan Taiga. Springer-Verlag, New York, USA, 74-86, 1986.

Eidenshink J.C.,., The 1990 conterminous U.S. AVHRR data set, Photogramm. Eng. Remote. Sens., 58, 809-813, 1992.

Flannigan, M.D. and T.H. yonder Haar, Forest fire monitoring using NOAA satellite AVHRR, Can. J. For. Res., 16, 975-982, 1986.

French, N.H.F., The Use of Synthetic Aperture Radar to Measure Soil Water in Fire Disturbed Boreal Forest Ecosystems, University of Michigan M.S. Thesis, Ann Arbor, MI, 55 pp., 1994.

French, N.H.F., E.S. Kasischke, L.L. Bourgeau-Chavez, and D. Barry, Mapping the location of wildfires in Alaskan boreal forests using AVHRR imagery, Int. J. Wildland Fire, 5, 55-61, 1995.

French, N.H.F., E.S. Kasischke, L.L. Bourgeau-Chavez, P. Harrell, and N.L. Christensen, Jr., Monitoring variations in soil moisture on fire disturbed sites in Alaska using ERS-1 SAR imagery, Int. J. Remote Sensing (in review), 1996.

Justice, C., and P. Dowty, IGBP-DIS Satellite Fire Detection Algorithm Workshop Technical Report, IGBP-DIS Working Paper #9, Paris, France, 87 pp., 1994.

Kasischke, E.S., Fire, climate change, and carbon cycling in Alaskan boreal forests, in Biomass Burning and Climate Change, edited by J.S. Levine, (this issue) 1996.

Kasischke, E.S., and French, N.H.F. Locating and estimating the areal extent of wildfires in Alaskan boreal forests using multiple season AVHRR NDVI composite data, Remote Sens. Environ., 51, 263-275, 1995.

Kasischke, E.S., L.L. Bourgeau-Chavez, N.H.F. French, P. Harrell, and N.L. Christensen, Jr., Initial observations on the use of SAR imagery to monitor wildfires in boreal forests, Intern. J. Remote Sens., 13, 3495-3501, 1992.

Kasischke, E. S. , French, N.H.F., Harrell, P., Christensen, N.L., Ustin, S.L., and Barry, D., Monitoring of wildfires in boreal forests using large area AVHRR NDVI composite data, Remote Sens. Environ., 44, 61-71, 1993.

Kasischke, E. S., L.L. Bourgeau-Chavez, and N.H.F. French, Observations of variations in ERS-1 SAR image intensity associated with forest fires in Alaska, IEEE Trans. Geosci. Remote Sensing, 32, 206-210, 1994.

Kasischke, E.S., N.L. Christensen, Jr., and B.J. Stocks, Fire, global warming and the mass balance of carbon in boreal forests, Ecol. Appl. 5, 437-451, 1995a.

Kasischke, E. S., N.H.F. French, L.L. Bourgeau-Chavez, and N.L. Christensen, Jr., Estimating release of carbon from 1990 and 1991 forest fires in Alaska, J. Geophys. Res., 100, 29412951, 1995b.

Kasischke, E.S. L. Morrissey, J.B. Way, N.H.F. French, L.L. Bourgeau-Chavez, E. Rignot, J.\ Steam, G.P. Livingston, Monitoring seasonal variations in boreal ecosystems using multi-temporal spaceborne SAR data, Can. J. Remote Sens., 21, 96-109, 1995c.

Kuhlbusch, T.A.J., and P.J. Crutzen, Black carbon, the global carbon cycle, and atmospheric carbon dioxide, in Biomass Burning and Climate Change, edited by J.S. Levine, (this issue) 1996.

Kurz, W.A., Apps, M.J., Webb, T.M., and McNamee, P.J. The contribution of biomass burning to the carbon budget of the Canadian forest sector: A conceptual model, in Global Biomass Burning: Atmospheric, Climatic and Biospheric Implications, edited by J.S. Levine, pp. 339-344, The MIT Press, Cambridge, Massachusetts, USA, 1991.

Schlentner, R.E. and K. Van Cleve, Relationships between CO2 evolution from soil, substrate temperature, and substrate moisture in four mature forest types in interior Alaska, Can. J. For. Res., IS, 97-106, 1985.

Stocks, B.J., The extent and impact of forest fires in northern circumpolar countries, in Global Biomass Burning: Atmospheric, Climatic and Biospheric Implications, edited by J.S. Levine, pp. 197-202, The MIT Press, Cambridge, Massachusetts, USA, 1991.

Viereck L.A. and C.T. Dyrness, Ecological Effects of the Wickersham Dome Fire Near Fairbanks, Alaska, U.S. Forest Service General Technical Report PNW-90, 71 pp., 1979.

Viereck L.A., C.T. Dyrness, K. Van Cleve, and M.J. Foote, Vegetation, soils, and forest productivity in selected forest types in interior Alaska, Can. J. For. Res., 13, 703-720, 1983.

Viereck L.A., K. Van Cleve, and C.T. Dyrness, Forest ecosystem distribution in the taiga environment, in Ecological Series Vol. 57: Forest Ecosystems in the Alaskan Taiga, ed. by K. Van Cleve, F. S. Chapin III, P.W. Flanagan, L.A. Viereck and C.T. Dyrness, Springer-Verlag, NY, pp. 23-43, 1986.