Manuscript submitted to:
International Journal of Remote Sensing
September 1998
Address for Correspondence:
Dr. Susan L. Ustin
Department of Land, Air, and Water Resources
University of California
Davis, CA 95616
Phone: (916) 752-0621
FAX: (916) 752-5262
email: slustin@ucdavis.edu
Observations of ecological factors necessary to monitor ecosystem change,
such as forest identification, successional stage, canopy closure, and
stand density, at regional scales over periods of years to decades or longer
are difficult if not impossible with current mapping and inventory methods.
The mosaic of forest types across the boreal landscape precludes simple
assessment of vegetation zones. Remotely sensed imagery offers the most
practical opportunity to obtain the necessary information at detailed space
and time scales. Current satellites have not succeeded in providing this
assessment, either because of infrequent overpasses during the growing
season, poor spatial resolution, and spectral resolution or data quality.
However, several new satellite systems offer the potential to improve mapping,
particularly the imaging spectrometers, like NASA’s New Millenium EO-1
satellite to be launched in 1999, DOD’s Warfighter and NEMO, and the Australian
commercial sensor, AIRES to be launched in 2001. The goal of this research
was to evaluate the use of vegetation endmembers derived from spectral
unmixing (Smith et al., 1990) to classify taiga vegetation types. The data
were derived from a NASA airborne high spatial and spectral imaging spectrometer
(Advanced Visible/Infrared Imaging Spectrometer, AVIRIS). This sensor acquires
224 contiguous spectral bands (ca. 10 nm band resolution) in the visible
to short-wave infrared region for each 20 m pixel. We contrast these results
to those obtained from classifying a SPOT (Satellite Pour l’Observation
de la Terre), three-band (visible to near-infrared) scene but at similar
spatial resolution. Classification compared the results of a SPOT classification
by Rignot et al. (1994) and a hybrid DEM and SPOT classification by Ustin
et al. (1994). The high resolution of the SPOT and AVIRIS images allow
us to evaluate the spatial scales and the spectral resolutions needed to
identify forest ecosystems in the boreal region.
The Bonanza Creek Experimental Forest (BCEF), Alaska, contains the four late seral boreal forests described by Van Cleve and Viereck (1981): (1) flood plain white spruce, (2) upland white spruce, (3) upland black spruce and (4) lowland black spruce (Figure 1). There are several mid-successional broadleaf forests (e.g., alder or balsam poplar) and conifer/hardwood mixtures. Upland white spruce succession includes aspen and paper birch phases while black spruce includes only a paper birch phase. Flood plain white spruce starts with invasion by willows and a shrub stage with an alder and a balsalm poplar phase that is absent from the lowland black spruce sequence (Van Cleve and Viereck, 1981; Viereck et al., 1993a). Flood plain white spruce will undergo a transition to lowland black spruce if succession continues undisturbed for an extended period. It is generally agreed that these forest ecosystems are associated with temperature and precipitation gradients, nutrient availability and quality, and fire history (Viereck, 1983; Van Cleve and Viereck, 1981; Van Cleve et al., 1983; Bonan et al., 1992; Van Cleve et al., 1993). Topography, especially, elevation and aspect are critical in determining the potential distribution of these species. Soil factors, primarily drainage, temperatures, and permafrost are significant in controlling species distributions (Bonan et al., 1992). As forests undergo seral sequences, changes in canopy cover and the depth of organic matter in the forest floor alter the energy budget and change a site from permafrost-free to permafrost-dominated (Viereck, 1983; Oechel and Lawrence, 1985).
Broadleaf and conifer forests differ widely in nutrient use, turnover
rates, and fluxes of trace gases (Bonan et al., 1992). Changes in the distribution
or abundance of major forest ecosystems could have significant global implications
for the storage, turnover and release of greenhouse gases (Bonan, 1991;
Kasischke et al., 1995; Pastor and Mladenoff, 1992).
An unpublished digital vegetation map of the BCEF LTER site was assembled from interpretation of 1978 aerial photos and field surveys, and provided by J. Yarie, Department of Forest Sciences, University of Alaska, Fairbanks, AK 99775. The map was assembled in 1992 in collaboration with E.F. Binnian, USGS, EROS Alaska Field Office, Anchorage, AK 99508. Rignot et al. (1994) established that the BCEF map does not reflect changes in the river course since 1978, vegetation changes resulting from the 1983 Rosie Creek Fire (fire boundary provided by Alaska Fire Service, Fairbanks, AK), and changes due to logging and stand development. Nonetheless, this is the most complete dataset available for validating remote sensing of forest composition at the watershed scale. The smallest mapped parcel area was 1 039 m2. The 1993 digital version of this map is an ARC/Info coverage and was used to georeference the SPOT image by co-registering to 21 control points in ARC/Info (ESRI, Redlands, CA). The AVIRIS image was registered to the SPOT image for the area coincident with BCEF coverage. An unpublished digital soils map of the uplands area was obtained from the USDI National Resource Conservation Service (NRCS) in Alaska and input into Arc/Info using the published soil survey maps (USDA, Soil Conservation Service, 1977) of the Goldstream-Nenana Area, Alaska to define the polygon attributes.
AVIRIS images (950614B0201) were acquired over the BCEF LTER site June 14, 1995. Scenes two and three were merged and an area of 10 km x 17 km was analyzed. Sky conditions were mostly clear at the time of acquisition but clouds and cloud shadows were located near the lower right and center of the scene and several smaller clouds near the upper left side of the scene. Data were calibrated to apparent surface reflectance using the ATmosphere REMoval Program (ATREM 2.0, U. Colorado) based on a modified version of the Modtran III radiative transfer atmospheric code. Thirty-eight bands were eliminated (1 - 5, 107-116, 152-169, 220-224), which either had near zero reflectance, low radiometric variability and/or were noisy. A linear mixture analysis was performed in IDL/ENVI 3.0 (RSI, Boulder, CO 80301) using image selected endmembers for clouds, bare soil, green vegetation, dry plant litter, and water/shade. The identity of the endmembers was confirmed using vegetation polygon boundary overlays to characterize them. The first step was to identify cloud pixels using the endmembers. These pixels were masked and a subsequent unmixing on the remaining pixels was performed and these data used in subsequent analyses. A supervised maximum likelihood classification was obtained using the endmember images.
A Satellite Pour l'Observation de la Terre (SPOT-2) multispectral (HRV-2
mode) image of the region was acquired under clear skies on August 28,
1991. The nominal pixel resolution of SPOT and AVIRIS is the same at 20m.
The normalized difference vegetative index (NDVI) was computed using red
and near-infrared bands (3-2)/(3+2). Supervised maximum likelihood classification
procedures cited by Rignot et al. (1994) were followed to reproduce their
results for comparison to AVIRIS results. Band 3 (NIR) was used to map
the 1983 Rosie Creek Fire boundary (6 000 ha wildfire located north of
the Tanana River, Figure 2). The mapped
location of the wildfire boundary shows some significant disagreement with
the registered 1991 SPOT image, demonstrating the difficulty in accurately
field mapping vegetation boundaries. Some vegetation within the boundary
was not burned and this is observed in the image.
Vegetation distribution based on aerial photo interpretations.
The original digital vegetation distribution map of the BCEF provided
by the Forest Service included 100 map categories which were merged into
seven forest classes following a simplified paper-copy scheme provided
by the LTER (Figure 6a). This produced a
forest class map consistent with the principal seral sequences characteristic
of interior boreal forests. This consolidation step primarily involved
merging forests of different stocking density and crown closure into a
single type. Mixed forest classes were consolidated with the dominant forest
type. We did not analyze the effect of species mixtures or degree of canopy
closure on the identification of forest classes so some classification
errors are attributable to stand variability that we did not separate.
The ground-based map included a heterogeneous "non-forest" class composed
of low ericaceous shrubs, mosses/muskeg, or lands of other undefined disturbance
characteristics, including roads and recently logged or burned sites.
Differentiation of boreal forests based on
the BCEF vegetation map and topography
The BCEF data shows the study area to be divided nearly equally between
lowland (54%) and upland (46%) forest ecosystems. The distribution of forests
across elevational zones varies widely. In both upland and lowland forests,
the black spruce successional sequence covers a greater total area then
the white spruce seral sequences (Table 1).
The montane zone black spruce forests are six times the spatial extent
of white spruce but about equally abundant the lowland (flood plain and
alluvial plain) area. In terms of the distribution of late seral stage
forests however, black spruce is only half as abundant as white spruce
in the montane area but is 1.6 times more abundant in lowland forests.
Only a small fraction of the land area is classified as non-forest (17%
lowlands and 14% montane) also indicating that most of this landscape is
at the relatively late seral stages (conifer or mixed conifer classes).
NDVI results
The normalized difference vegetative index (NDVI)
of the study area is shown in Figure 7.
The histogram showed a distinct bimodal distribution in NDVI. Water has
an NDVI near zero so it provided the lower non-vegetation boundary. The
two peaks in NDVI were apparent in the histogram. High NDVI values were
observed to occur in BCEF mapped deciduous forest types and low values
in the conifer types with mixed broadleaf and conifer stands in the intermediate
values. Forest NDVI values show distinct spatial patterns that are highest
in the uplands areas and in the riparian zone near the Tanana River.
The forest assignments were evaluated by establishing training sites for each forest type in the ground-based map (Figure 6a). In general the forest patterns on the classified SPOT image are correctly distributed (Figure 6b), however, significant differences with the base map are apparent. In the uplands region, the white spruce and paper birch communities appear more fragmented and patch size is smaller than the mapped units. In contrast, the flood plain and alluvial plain region has much less diversity, with most of the area identified as black spruce and greatly underestimating the non-forest and hardwood species. The Rosie Creek wildfire scar is identified as black spruce, presumably resulting from confusion between low vegetation cover and typical black spruce values. Even in the flood plain region, balsam poplar and alder are underestimated, despite apparent NDVI patterning that is observed on Figure 7.
The endmember images.
The endmember abundance images derived from the linear unmixing of
the 1995 AVIRIS pixel spectra are shown for green vegetation, dry vegetation,
bare soil, and shadow endmember (Figure 8). The shadow endmember was essential
to mapping the cloud shadows within the scene. The green vegetation endmember
image shows vegetation recovery since the 1991 SPOT image. The dry vegetation
endmember shows that plant litter is significantly more abundant in the
wildfire region than in surrounding unburned areas. Bare soil/gravel is
apparent on the unpaved road, creeks, gravel bars, and in the Tanana River.
The latter is due to both the shallowness and the rock powder/sediment
in the meltwater. The shadow endmember image also shows high values in
the river and in other water bodies (such as the oxbow north of the river
near the center of the image), indicating that most incident photons are
absorbed. The false color image of the vegetation endmember (red), dry
vegetation (green) and soil (blue) is shown in Figure
6c. Many of the vegetation patterns that are mapped in the ground-based
map (Fig. 6a) are readily observed without
further classification. The most striking difference is the greater richness
of pattern within the AVIRIS mapped units than is found in the BCEF map.
The map units field checked by Rignot et al. are shown superimposed over the AVIRIS image in Figure 4b. While the patterns show considerable correspondence, there are some obvious differences. The changes in the Tanana River during the time between creation of the vegetation map and the image makes precise registration impossible. In other cases, the BCEF polygons include significant heterogeneity, showing the mismatch between the mapping scale and the image resolution. Other errors may be attributed to either location errors in the original BCEF mapping or to changes in forest boundaries that have resulted from growth or logging.
Figure 5b shows the aspect distribution of forest types in the BCEF after applying a cloud mask to match the AVIRIS pixels. Figure 5c shows the aspect distribution of forest types based on the AVIRIS classification, which are seen to closely follow the distribution patterns in the BCEF map (Figure 5a and 5b) but differ in their abundance. Table 1 shows that the relative distributions of white spruce and black spruce successional forests are consistent with the BCEF map in both lowlands and upland habitats, but indicate greater coverage of white spruce in uplands areas and less in lowlands areas (from 6.2% to -9.6%). The classified AVIRIS image is shown in Figure 6d. There is a close spatial match between the AVIRIS classified image and the BCEF map.
Overall we found reasonably good agreement on an absolute pixel basis between the two maps with much of the disagreement due to the differences in spatial scale between ground-based polygons and the raster images (Table 2). If the more heterogeneous non-forest vegetation class is omitted, the overall pixel accuracy for agreement between the base map and the classified AVIRIS map is 74%. Classification accuracy varied with topographic zone as shown in Table 2. In general the accuracy is highest for the most common forest types.
To evaluate the AVIRIS map, a supervised maximum likelihood classification (MLC) was done using the selected pure stands described by Rignot et al. (1994), as training sites on the three SPOT bands. We were obtained an 86% pixel accuracy compared to the 83% reported by Rignot et al. using their training sites along the Tanana River and a mask covering the same area. Their MLC was shown to be highly accurate when the analysis was confined to the flood plain zone from which the training sites were derived. It had higher accuracy then our AVIRIS classification for this region (83% vs. 67%), since our results were based on an analysis of the entire site. However, once their MLC was expanded to include the alluvial plain using the same vegetation categories, their classification accuracy decreased to 30% while ours remained at 69%. The decline in accuracy of the MLC in this zone is due to the spatial misregistration of vegetation patterns south of the Tanana River, the changes in vegetation within the Rosie Creek Fire (which has partially regrown by 1995), and spectral class identification. The significant loss of accuracy whenexpanding from the flood plain zone to the alluvial plain zone was unexpected, as the same forest types occur and topography does not affect radiance values. When the MLC analysis is extended to the full image area the SPOT accuracy declines to 43% in contrast to the overall AVIRIS accuracy of 74%.
Ustin et al. (1994) developed a rule-based classification for this site using a combination of elevation, slope, aspect and NDVI to classify the same SPOT image for this scene. Nonetheless, their overall accuracy was only 43% for a seven-class map. Therefore, despite similarity in spatial patterns in the classified SPOT image, more accurate mapping tools are still needed. Based on these results, imaging spectrometry data has considerable potential to improve forest maps in the boreal region.
Errors are also evident in the "ground-truth" vegetation map. This is
partly due to changes since the original photography was acquired in 1978.
Further, some level of subjective decisions occurs in defining and locating
the boundaries and in identification of forest and mixed forest communities
in the BCEF map. Therefore, the true AVIRIS accuracy may be better then
the minimum estimate from this analysis, which is well within confidence
for existing boreal forest mapping. It is evident that the ground-based
map is composed of fewer but larger polygons relative to the image-based
predictions. Moody and Woodcock (1994) have shown scale dependent errors
are a function of the original spatial resolution of the map and the fine-scale
patterns of the class. Because forest succession cycles average 50-100
years (Viereck 1983), the most recent analysis, based on the 1978 photography,
would identify a forest type that is actually 20% to 25% further into a
seral fire cycle then indicated on the ground-based map. The Rosie Creek
fire occurred within the mapped region in 1983 and is clearly evident in
the SPOT data but is not readily observed in AVIRIS data five years later.
This fire (Figure 2) primarily burned the
mid-elevation white spruce forest.
Knowledge of the spatial extent and location of these forest ecosystems is important for understanding climate-biosphere feedbacks because they have markedly different patterns of carbon storage, exchanges of carbon dioxide and water, and in the cycling of other trace gases e.g., methane (Bonan, 1991). Both location, spatial extent, and total area of each vegetation type were relatively accurately predicted in this analysis.
Production of the AVIRIS forest map provides information to examine the distribution of the four seral sequences and to characterize the fraction of the landscape in each type (Table 2). The site is roughly equal between lowland and upland forest ecosystems but the distribution of forest types varies widely. If we assume the seral forest sequences in Figure 1, and allow half the paper birch to be in each upland spruce type, we can estimate the distribution of these seral communities. This is a reasonable assumption, given the aspect distribution (Figure 5). In the lowland forests, the black spruce successional sequence exhibits greater total area then white spruce seral sequences while the reverse is true in the uplands area. The lowland black spruce forests are 1.5 times the spatial abundance of lowland white spruce. Table 2 shows that 90% of the lowland black spruce forest is at a late seral stage while half of the flood plain white spruce is at a late seral stage. This partially reflects the loss of white spruce forests in the Rosie Creek Fire. For the upland forests about equal proportions of the black spruce and white spruce forests are at a late successional stage (79 and 84%, respectively). The most interesting observation is the striking difference in the percent of the forest in late seral stage between upland and lowland forests, which are opposite in pattern for white and black spruce. Only a small fraction of the area is classified as non-forest (5.4%) also indicating that most of this landscape is at the relatively late seral stages (conifer or mixed conifer classes). Depending on the number of years a forest will remain in this initial phase of secondary succession, the actual turnover rates for these forests probably exceeds 100 years. Possibly the differences in proportion of late seral stages in upland and lowland forests indicates a difference in the types of fires and the extent of combustion in these forests.
Most commission errors arose by mis-classification into a related forest type, e.g., white spruce rather then black spruce. From a global climate perspective, considerable information about carbon flux conditions within the forests could be gained only by a general knowledge of the classes of forests. For example, Bonan et al., (1992) found that simply differentiating between forests and tundra was sufficient to define climate feedbacks in boreal forests. Thus, this analysis should produce sufficiently detailed information about ecosystem structure and dynamics that is needed for improving both climate and ecological assessments.
Acknowledgments. This research was performed under NASA EOS grant
NAS-31359 and NAGW-2636 to the U. California Davis. The analyses were performed
on DEC Alpha computers provided under a Sequoia 2000 grant (Cooperative
Research Agreement #1243) and a NASA Center of Excellence Grant (NRA-97-05).
We wish to recognize the ground-based vegetation maps and SPOT image used
in this study were contributed by Dr. JoBea Way (Radar Sciences Group,
Jet Propulsion Laboratory), Dr. Les Viereck (Institute of Northern Forestry,
Fairbanks, AK) and Dr. J. Yarie (U. Alaska Fairbanks, AK). The Alaska SCS
is gratefully acknowledged for providing their provisional soils map, and
the Alaska Fire Service for the Rosie Creek Fire map, and the USGS for
the DEM.
BONAN, G.B., 1991, Atmosphere-biosphere exchange of carbon dioxide in boreal forest, J. Geophys. Res. 96, 7301-7312.
BONAN, G.B., POLLARD, D., and THOMPSON, S.L., 1992, Effects of boreal forest vegetation on global climate, Nature, 359, 716-718.
BONAN,G.B., and VAN CLEVE, K., 1992, Soil temperature, nitrogen mineralization, and carbon source-sink relationships in boreal forests. Can. J. For.Res.Rev. Rev. Can. Rech. For. 22, 629-639.
BONAN, G.B., CHAPIN, F.S., THOMPSON, S.L., 1995, Boreal forest and tundra ecosystems as components of the climate system. Climatic Change, 29, 145-167.
BONAN, G.B., 1995, Land-atmosphere interactions for climate system models: Coupling biophysical, biogeochemical, and ecosystem dynamical processes. Remote Sens. Environ. 5, 57-73.
CARD, D.H., 1982, Using known map category marginal frequencies to improve estimates of thematic map accuracy. Photogramm. Eng. and Remote Sens., 48, 431-439.
CHAPIN, F.S. III, HOBBIE, S.E., SHAVER, G.R., 1997, Impacts of global change on composition of artic communities: implications for ecosystem functioning, Ecol. Studies 124, 221-228.
CONGALTON, R., 1991, Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sens. Environ., 37, 35-46.
DAVIS, F.W., GOETZ, S., 1990, Modeling vegetation pattern using digital terrain data, Landscape Ecol. 4, 69-80.
FRENCH, N.F.H., KASISCHKE, E.S., JOHNSON, R.D., BOURGEAU-CHAVEZ, L.L., FRICK, A.L., and S.L. USTIN, 1996, Using multi-sensor satellite data to monitor carbon flux in Alaskan boreal forests, Levine, J.L., (editor),Biomass Burning and Climate Change, MIT Press, Cambridge. pp. 808-826.
HALSEY, L.A.,VITT, D.H., ZOLTAI, S.C., 1995, Disequilibrium response of permafrost in boreal continental western Canada to climate change. Climatic Change, 30, 57-73.
HOUGHTON, J.T., JENKINS, G.J., and EPHRAUMS, J.J., 1990, Climate Change - The IPCC Scientific Assessment. Cambridge University Press, Cambridge, England, 365 pp.
KASISCHKE, E.S., CHRISTENSEN, N.L. JR., and STOCKS, B.J., 1995, Fire, global warming and the carbon balance of boreal forests, Ecol. Appl.5, 437-451.
MOODY, A., and WOODCOCK, C.E., 1994, Scale-dependent errors in the estimation of land-cover proportions: implications for global land-cover datasets, Photogramm. Eng. Remote Sens. 60, 585-594.
MYNENI, R.B., KEELING, C.D., TUCKER, C.J., ASRAR, G. and R.R. NEMANI, 1997, Increased plant growth in the northern high latitudes from 1981 to 1991. Nature, 386, 698-702.
OECHEL, W.C. and LAWRENCE, W.T., 1985, Taiga, In: Chabot, B.F. and Mooney, H.A. (eds), Physiological Ecology of Northern American Plant Communities, Chapman and Hall, NY, pp 66-94.
OECHEL, W.C., VOURLITIS, G.L., HASTINGS, S.J., BOCHKAREV, S.A., 1995, Change in arctic CO2 flux over two decades - Effects of climate change at Barrow, Alaska. Ecol. Appl., 5, 846-855.
OECHEL, W.C., COWLES, S., GRULKE, N., HASTINGS, S.J., LAWRENCE, B., ROUDHOMME, T., RIECHERS, G., STRAIN, B., TISSUE, VOURLITIS, G., 1994, Transient Nature of CO2 fertilization in arctic tundra. Nature, 371, 500-503.
OECHEL, W.C., 1995, Global Change and Arctic Terrestrial Ecosystems, Ecological Studies 124, Springer, New York, 439p.
PASTOR, J. and MLADENOFF, D.J., 1992, Southern boreal-northern hardwood forest border. In: Shugart, H.H., Leemans, R.L., and Bonan, G.B., A Systems Analysis of the Global Boreal Forest, Cambridge University Press, NY, pp, 216-240.
RIGNOT, E., WILLIAMS, C., WAY, J.-B., and VIERECK, L., 1994, Mapping of forest types in Alaskan boreal forests using SAR imagery, Trans. Geosci and Remote Sens. 32, 1051-1059.
SMITH, M.O., S.L. USTIN, J.B. ADAMS, and A.R. GILLESPIE, 1990, Vegetation in deserts I. A regional measure of abundances from multispectral images. Remote Sensing of Environment. 29, 1-26.
SOIL CONSERVATION SERVICE, 1977, Soil survey of Goldstream-Nenana Area, Alaska. USDA SCS, 45p. +21 maps.
Takhtajan, A., 1986, Floristic Regions of the World. University of California Press, Berkeley, CA.
U.S.D.A. Soil Conservation Service, Soil survey of Goldstream-Nenana Area, Alaska, 45p + 22 plates, July, 1977.
USTIN, S.L., L-H. SZETO, Q.-F. XIAO, Q.J. HART, and E.S. KASISCHKE, 1994, Vegetation mapping of forested ecosystems in Central Alaska. in Proceedings International Geoscience and Remote Sensing Symposium IGARSS '94. August 8-12, 1994 California Institute of Technology.
VAN CLEVE, K. and VIERECK, L.A., 1981, Forest succession in relation to nutrient cycling in the boreal forest of Alaska. In: West, D.C., Shugart, H.H. and Botkin, D.B. (eds), Forest Succession: Concepts and Applications, pp. 185-211, Springer-Verlag, NY.
VAN CLEVE, K., OLIVER, L., SCHLENTNER, R., VIERECK, L.A. and DYRNESS, C.T., 1983, Productivity and Nutrient Cycling in Taiga Forest Ecosystems, Can. J. For. Res., 13, 747-766.
VAN CLEVE, K., DYRNESS, C.T., MARION, G.M., ERICKSON, R., 1993, Control of soil development on the tanana river floodplain, interior Alaska. Canadian Journal of Forest Research-Journal, 23, 941-955.
VIERECK, L.A., VAN CLEVE, K., ADAMS, P.C., SCHLENTNER, R.E., 1993a, Climate of the Tanana River floodplain near Fairbanks, Alaska, Can. J. For. Res. Rev. Can. Rech. For. 2, 899-913.
VIERECK, L.A., 1983b, The effects of fire in black spruce ecosystems of Alaska and northern Canada, in: Wein, R.W and MacLean, D.A. (eds.), The Role of Fire in Northern Circumpolar Ecosystems. Wiley, NY, pp 201-220.
VIERECK, L.A., DYRNESS, C.T., FOOTE, M.J., 1993, An overview of the vegetation and soils of the floodplain ecosystems of the Tanana River, interior Alaska, Can. J. For. Res. Rev. Can. Rech. For. 23, 889-898.
WAY, J.B., McDONALD, K., PAYLOR, E., KARAS, G., CHERNOBIEFF, S., and BIERY, N, 1992, Collected data of the Bonanza Creek Experimental Forest, Alaska, vol. I and II, AirSAR and in situ Measurements March 1988, SEASAT SAR July – October, 1978, NASA/Jet Propulsion Laboratory.
WEBER, M.G., FLANNIGAN, M.D., 1997, Canadian boreal forest ecosystem structure and function in a changing climate: impact on fire regimes. Environ. Rev. 5, 145-166.