Skip booklet index and go to page content

Technical Thematic Report No. 17. - Monitoring ecosystems remotely: a selection of trends measured from satellite observations of Canada

Indicators of Fragmentation: Forest Density and Forest Edge Density

To allow for a meaningful discussion of landscape changes over time, changes in the total area of various land cover classes (for example, the first section of this report starting on page 3) should be discussed in conjunction with changes in the pattern of land cover classes (such as homogeneity). This discussion fits within the theme of fragmentation. Habitat fragmentation has played a significant role in conservation research since the 1970s (Haila, 2002; Manning et al., 2004). A widely accepted definition of fragmentation (or the process of fragmentation) is the division of contiguous habitat into smaller pieces (Fleishman and Mac Nally, 2007). Fragmentation results in a decrease in connectivity, an increase in edge density, and an increase in isolation of remnant areas. Analyses of fragmentation impacts are complex as the different components of fragmentation are often confounded (Haila, 2002; Lee et al., 2002; Manning et al., 2004) and results are highly species dependent (Flaspohler et al., 2001; Ries and Sisk, 2004; Villard et al., 2007).

The completion of a high resolution (30 m) map of Canada’s forested region circa 2000 by the Canadian Forest Service under the joint Earth Observation for Sustainable Development (EOSD) initiative has provided a detailed dataset of Canada’s forest cover. This dataset has enabled Canadian Forest Service, in conjunction with the University of British Columbia, to derive spatial statistics on Canadian forest cover at a number of resolutions (Wulder et al., 2008b). Of the fragmentation statistics reported by Wulder et al., (2008b), two are presented here by ecozone+ for ESTR: (1) the proportion forested pixels within a 1 km2 analysis unit (that is, forest density); and (2) the length of all edges between forest and non-forest pixels within each 1 km2 analysis unit (that is, forest edge density).

These analyses should be considered as broad-brush, entry-level fragmentation data sets. As this analysis is currently for one time period, the results are primarily a description of the patterns of forest within each ecozone+.


The EOSD land cover maps were produced from 30 m Landsat data collected between 1999 and 2002 from May to October (90% obtained within one year of the 2000 target date) (Wulder et al., 2008a). Each pixel was classified in one of 23 categories. Further discussion of the classification process can be found in Wulder et al., (2008a).

Detailed methodology can be found in Wulder et al. (2008b). The ”proportion of forest area” (referred to as ”forest density” in this report) and ”forest edge density” metrics are summarized by ecozone+ here. Regions classified as ‘Treed’ in the EOSD dataset (>10% tree cover (Wulder and Nelson, 2003)) were reclassified as ‘Forest’ and all other regions were reclassified as ‘non-forest’ for the purpose of these analyses (Wulder et al., 2008a). Forest density represents the proportion of forested pixels (30 m resolution) within each 1 km2 analysis unit while the forest edge density metric represents the total length of all of the edges between forested pixels and non-forested pixels within a 1 km2 analysis unit.

Quality checks and limitations

Pre-existing data are often used for validation of remotely sensed data sets (for example, the Satellite Database for the Land Cover of Canada database was used to validate the land cover change maps outlined in the first section of this report on page 3) (Wulder et al., 2007). The EOSD data set from which the forest and edge density calculations were drawn was validated using airborne video data collected specifically for the validation process of this data set. A 31,000 km2 sub-sample of the EOSD data set was analyzed on Vancouver Island in the Pacific Maritime Ecozone+ (Wulder et al., 2007) with a systematic stratified random sampling approach outlined in Wulder et al. (2006). The coniferous treed class (a sub classification of “Forest”), which made up 71% of the area sampled was found to have a classification accuracy of 86%.

A distinct visual seam is noted at the Ontario-Quebec border in the forest edge density map. While not large numerically, this seam introduces an element of caution into the interpretation; greater confidence should be placed in differences highlighted within a single province than between provinces. This seam could also reflect a difference in forest management policy between the two provinces.

In the interpretation of the following results, it is also important to note that these metrics are not a measure of human influence; they are a measure of landscape characteristics which may result from either natural or anthropogenic processes. For example in wetland or alpine areas, forest edge density will be relatively high due to the natural heterogeneity of the landscape, not as a result of human influences (Wulder et al., 2008b).


Results of the forest density analysis are presented in Figure 12 and the results of edge density analysis are presented in Figure 13. These results are discussed by ecozone+ below.

Figure 12. Forest density in the forested region of Canada circa 2000.


Long Description for Figure 12

This map shows forest density within the forested region of Canada circa 2000 as the proportion of forest ranging from 0 to 100%. Forest density is generally highest in the south of the region, with a broad transition zone in the north along the boundary with the arctic tundra and a narrower transition zone around Canada’s prairie region. Some higher altitude areas also show lower forest densities.

Forest density is the proportion of 30 m2 pixels that are forested in each 1 km2 analysis unit.

Source: derived from the EOSD land cover of 2000 data set created by CFS (Wulder et al., 2008b) 

Figure 13. Forest edge density in the forested region of Canada, circa 2000.


Long Description for Figure 13

This map shows forest edge density within the forested region of Canada circa 2000 as values ranging from 0 to 800 m/km2. Similar to forest density, forest edge density is generally lowest in the south of the region, with higher values along a broad transition zone in the north along the boundary with the arctic tundra. Higher altitude areas also show higher forest edge densities near and above the treeline.

Forest edge density is the total length of all edges between forested pixels (30 m resolution) and non-forested pixels within a 1 km2 analysis unit.

Source: derived from the EOSD land cover of 2000 data set created by CFS (Wulder et al., 2008b)

Taiga Plains

There is a large variation in latitude, elevation, and climate in the Taiga Plains Ecozone+. It is an area of frequent large wildfires and thus the vegetation is made up of a patchwork of fire scars at different stages of regeneration. Forest density in much of this ecozone+ was found to be greater than 50% (Figure 12). Lower forest densities were found immediately south of Great Slave Lake, the uplands near Norman Wells, in a large burn scar west of Lac la Martre, and in portions of the lower reaches of the Mackenzie Valley. The forest edge density in the Taiga Plains Ecozone+ was found to be higher than more southerly forests, with a typical value of 250 m/km2, increasing to 500 to 600 m/km2 in the area of tundra surrounding the eastern foothills of the Mackenzie Mountains (Figure 13).

Taiga Shield

The western portion of the Taiga Shield Ecozone+ is approximately two-thirds forest and one-third tundra. The majority of the forest area in this ecozone+ was found to have a forest density greater than 50% while areas of tundra and fire scars were found to have forest densities of 30% or less (Figure 12). Forest edge density in the western portion of the ecozone+ was typically around 250 m/km2 (Figure 13). A large area just north of Great Slave Lake was found to have edge densities up to 650 m/km2, likely corresponding to a transition region between forest and tundra land covers.

The eastern portion of the ecozone+ is approximately one-third forested and two-thirds tundra. As in the west, the forested portion includes many fire scars. Forest density greater than 50% was found for the majority of the forested area in the eastern portion of the ecozone+, while forest density closer to 30% or less was found in areas of tundra and fire scars (Figure 12). In contrast to the western portion, the forest density decreased to less than 10% at the northern extreme of this ecozone+, suggesting the eastern portion of the ecozone+ includes more severe growing conditions than the western portion. As in the west, typical values of forest edge density were around 250 m/km2 with a few small areas in the northern extreme of this ecozone+ with forest edge densities up to 650 m/km2 (Figure 13). These may be associated with areas of poor drainage and a greater density of wetlands.

Boreal Shield

Forest is the dominant land cover in the Boreal Shield Ecozone+ with numerous fire scars of varying ages. Many of the areas of lower forest densities corresponded to young fire scars (Figure 12). Forest edge density, while low throughout this ecozone+ (0 to about 150 m/km2), was slightly greater in the north, particularly in regions most affected by wildfires (Figure 13). Notable in this ecozone+ was the large area of low forest density and high forest edge density surrounding the nickel smelter in Sudbury.

Atlantic Maritime

The Atlantic Maritime Ecozone+ is relatively small with a dominant land cover of mixed forest. Agricultural land has replaced the forest in much of Prince Edward Island, the Annapolis Valley of Nova Scotia, and the Saint John valley of New Brunswick. High forest density was found for most of this area (Figure 12). Of note is an area of low forest density found in the region of Cape Breton Highlands National Park. This area of low density may be the result of forests within the park being left to regenerate naturally after severe spruce-budworm outbreaks in the 1970s and early 1980s while forests outside of the park were salvaged-logged, though this hypothesis has not been confirmed.

The density of forest edges in the Atlantic Maritime Ecozone+ was found to be low (Figure 13), ranging from 0 to about 150 m/km2 throughout much of the region. Edge density increased to about 270 m/km2 in the Northern Cape Breton highlands.

Boreal Plains

Considerable area in the Boreal Plains Ecozone+ has been converted to agriculture, as evidenced in the lower density of forest found around the periphery of the Prairies Ecozone+ and in the Peace River area associated with the towns of Dawson Creek, Grand Prairie, Peace River, and Valleyview (Figure 12). A very large wildfire that burned about 2,000 km2 north of Whitecourt in 1998 also corresponded to an area of low forest density. Areas with high elevation and irregular terrain including the uplands of Riding Mountain National Park, Duck Mountain Provincial Park, and the Porcupine Hills and Pasquia Hills typically exhibited high forest density.

The density of forest edges generally ranged from 0 to about 150 m/km2, with some areas up to 250 m/km2 (Figure 13). A visual comparison with the original EOSD land cover map (before simplification to ‘forest’ and ‘non-forest’) within the Boreal Plains Ecozone+ showed that the density of forest edges in homogeneous uncut forest was generally below 100 m/km2, while areas of forest containing many cutblocks had forest edge densities ranging between 100 and 200 m/km2. In many areas that are not subject to logging, natural edges, such as contact with water bodies and wetlands, produced edge densities between 100 and 200 m/km2.

Taiga Cordillera

The dominant land cover of the Taiga Cordillera Ecozone+ is shrub at lower elevations and tundra in the Mackenzie and Stikine Mountains and the Canadian portion of the Brooks Range. A large area of this ecozone+ was found to have very low to low forest density (Figure 12) due to large, essentially treeless upland areas. At lower elevations, most areas contained scattered and open forests rather than continuous, high density forests. A significant area with relatively high forest density was found, however, in the lowlands between the Mackenzie Mountains and the Brooks Range. Forest edge density was generally low throughout the ecozone+ (Figure 13) (0 to about 150 m/km2). Some areas, however, had edge densities approaching 300 m/km2, particularly in the lowlands to the west of Fort Good Hope.

Boreal Cordillera

The Boreal Cordillera Ecozone+ exhibits strong altitudinal zonation of its vegetation, from dense conifer forests in the valleys, progressing to shrubs and stunted trees to tundra and finally to snow and ice on the peaks. The forest density was found to be quite heterogeneous (Figure 12), but generally correlated with altitude gradient with the highest forest densities at the lowest elevations. Forest edge density was typical of other boreal forest regions (Figure 13), ranging from 0 to about 250 m/km2.

Pacific Maritime

The vegetation in the Pacific Maritime Ecozone+ exhibits strong altitudinal zonation. Forest density was found to be very high in the Queen Charlotte Islands, on Vancouver Island, and at lower elevations along the Pacific coast (Figure 12). Forest edge density increased with increasing altitude, reaching values as high as 650 m/km2 reaching the altitudinal treeline. This high edge density may also result from the geometry of the mountains themselves; several drainage channels within a single 1 km2 analysis unit would result in high forest edge densities.

Montane Cordillera and Western Interior Basin

The Western Interior Basin and Montane Cordillera ecozones+ are largely indistinguishable based on forest and forest edge density and are summarized together here. Growing conditions for forests are ideal in these ecozones+, with the exception of the dry interior valleys of the Western Interior Basin Ecozone+. Forest density was found to be strongly skewed toward high-density classes in both ecozones+, with more analysis units with very low forest density found in the Western Interior Basin (Figure 12) as a result of the open forests of the dry interior valleys. The area of mountain pine beetle mortality was not yet visible in these landscape metrics as of the year 2000. The northernmost regions of the Montane Cordillera Ecozone+ were found to have large areas with very high forest density, likely due to an absence of logging.

In these ecozones+, the density of forest edges was inversely correlated with forest density, areas with lower forest density typically had higher forest edge density (Figure 13), likely a result of the fragmentation from industrial logging in these two ecozones+.

Hudson Plains

The vegetation of the Hudson Plains Ecozone+ is dominated by conifer forest, with a belt of shrubland and wetland covering a poorly drained band toward the south of the ecozone+, and a predominance of tundra vegetation in Wapusk National Park to the north of York Factory. Forest density was found to be low in much of the Hudson Plains Ecozone+ (Figure 12) as a result of the many small bodies of open water and non-forested wetlands. The variable landscape was also reflected in the forest edge density (Figure 13), with a large part of the ecozone+ showing forest edge densities in the range of 250 m/km2.

Newfoundland Boreal

Forest density in the Newfoundland Boreal Ecozone+ was found to be greater than 50% for most of the ecozone+ (Figure 12). Two large patches of low density forest found east of Corner Brook and Deer Lake on the Buchans Plateau appear to be old fire scars. A large area of low density forest centered on Bay du Nord Wilderness Reserve also appears to be an old burn scar.

Forest edge density in this ecozone+ was found to be relatively high (Figure 13): greater than 350 m/km2 for most of the area classified as shrub; and greater than approximately 600 m/km2 in the Long Range Mountains. This suggests a very heterogeneous forest/shrub matrix, with many edges.


As the EOSD dataset is currently only available for the 2000 time period, the main result of this analysis is descriptive. The full-resolution forest density image shows numerous areas of lower forest density, particularly smaller patches affected by industrial logging, and larger patches affected by forest fires. Industrial-scale forestry also appears to increase the edge density from less than 100 m/km2 in uncut areas of homogeneous mature forest to a range between 100 and 200 m/km2 in areas with an extensive pattern of cutblocks. The edge density is also higher in the northern portion of the Boreal Shield Ecozone+ where the forest becomes less continuous and prone to larger fires.

The utility of this dataset is currently limited to describing the characteristics of forests at a point in time. A time series, produced every ten years for example, would enable monitoring of trends in the forest.