Technical Thematic Report No. 17. - Monitoring ecosystems remotely: a selection of trends measured from satellite observations of Canada
Normalized Difference Vegetation Index 1985 to 2006
Earth observation researchers developed the Normalized-Difference Vegetation Index (NDVI) early in the era of satellite observations and it is now a popular and robust indicator of the amount and vigour of green vegetation. NDVI measures the contrast between the reflectance of red solar radiation, which is absorbed by chlorophyll, and the reflectance of near-infrared (NIR) solar radiation, which is reflected by the internal structure of leaves. NDVI most directly represents gross primary photosynthesis (Tucker, 1979; Sellers 1985 and Myneni et al. 1995 in Pouliot et al., 2009) and can be used as a proxy of green leaf area (Myneni et al., 1998). The NDVI is calculated for a given area using the following formula:
Long description for the NDVI formula
NVDI = (NIR - RED) divided by (NIR + RED)
NIR = intensity of near-infrared radiation reflected
Red = intensity of red radiation reflected
NDVI values range from -1 to +1, with clouds, water, and snow resulting in negative values, barren areas of rock and soil resulting in values close to zero, and densely vegetated areas resulting values close to 1. Different types of vegetation also have different characteristic NDVI values. For example, early-succession broadleaf vegetation has a higher NDVI value than late-succession conifers (Myneni and Williams 1994 in Pouliot et al., 2009).
There are several different global NDVI datasets compiled from Advanced Very High Resolution Radiometer (AVHRR) sensors. Different processing of AVHRR data and/or analyses during different time periods may show different or even contradictory NDVI trends (Alcaraz-Segura et al., 2010). The dataset used here was developed by CCRS (Pouliot et al., 2009) from 1 km resolution AVHRR data (Latifovic et al., 2005). They analyzed trends in annual peak NDVI during a 22 year period (1985 to 2006) and their results are examined at an ecozone+-level here. Changes in NDVI are discussed as a proxy for changes in primary productivity.
Pouliot et al. (2009) developed a complete and rigorous approach for processing the newly developed 10-day composite 1 km AVHRR satellite data and analyzing the output. The data were spatially averaged to 3 km resolution and annual peak growing season values were calculated by averaging the three highest NDVI values from all July-August composite images. Significance of the trend analyses conducted for each location was assessed at the 95% confidence level using the Mann-Kendall test. Only locations with significant trends are shown in the maps. Detailed methods are described in Pouliot et al. (2009).
This CCRS dataset has improved corrections and a higher resolution than the Global Inventory, Monitoring, and Modeling Studies dataset, currently the most widely used dataset of global NDVI.
Significant trends (p < 0.05) are summarized here by ecozone+ and visually compared to the 1995 land cover map (see Land cover change on page 3) to facilitate discussion of NDVI trends relative to the land cover.
Pouliot et al. (2009) also examined the influence of climate and land cover change on the observed NDVI trends. The influence of climate was examined by calculating a Climate Trend Impact Index (CTII) for each region based on correlation between gridded monthly temperature and precipitation data (Mitchell 2005 in Pouliot et al., 2009) and annual peak NDVI. An analysis of the influence of land cover change on NDVI was conducted for regions where Landsat time series data (Regions 5, 6, 8 in Figure 9) or Census of Agriculture data (Region 7 in Figure 9) was available. These results are presented here by ecozone+.
Figure 9. Study regions used to address the impacts of climate forcing and land cover on NDVI trends.
Long Description for Figure 9
This map shows the study regions used by Pouliot et al. (2009) to examine the impacts of climate forcing and land cover on observed Normalized Difference Vegetation Index (NDVI) trends. Nine different regions are shown (1-Taiga Plains, 2-Arctic in Nunavut, 3-Arctic in northern Quebec, 4-Boreal Shield/Taiga Shield in Labrador, 5-Boreal Shield in southern Quebec, 7-Prairies, and 8-Pacific Maritime). Regions 1 to 4 are considered ‘northern’ while Regions 5 to 8 are considered ‘southern’.
Regions 1 to 4 are considered ‘northern’ while Regions 5 to 8 are considered ‘southern’.
Source: adapted from Pouliot et al. (2009)
Quality checks and limitations
In areas where soil is visible, soil reflectance can affect the NDVI value. The appearance of soil changes based on moisture content and, as a result, soil moisture influences NDVI in areas where there is low to moderate vegetation with visible soil (Huete and Jackson, 1987; Huete and Jackson, 1988). Another limitation of NDVI is that it becomes less sensitive to changes in greenness as its value increases. For example, a larger increase in greenness is required to increase NDVI by the same increment the closer the value is to 1 (Gilabert et al., 1996; Santin-Janin et al., 2009).
Two quality checks were conducted on the 1 km AVHRR NDVI dataset used in this analysis (Pouliot et al., 2009). The first quality check involved a quantitative comparison of this NDVI dataset with an NDVI dataset derived from high resolution (30 m) Landsat data for two tundra regions (Region 1 in the Taiga Plains Ecozone+ and Region 3 in the Arctic Ecozone+, see Figure 9). The NDVI trends from the 1 km AVHRR NDVI dataset corresponded well with the Landsat data with a mean absolute error of 4.5% in Region 1 and 6.8% in Region 3.
The second quality check involved examining NDVI trends in all areas burned between 1960 and 2004 (forest fire databases from Fraser et al., 2004 and Zhang et al., 2004a,b in Pouliot et al., 2009) within the eight regions to verify that the expected dynamic of burning and regrowth was observed. NDVI trends were all found to be negative in areas recently affected by fire (1994 to 2004), likely a result of reduced vegetation present. Trends were positive in areas affected by fires from 1980 to 1990 as regeneration would have dominated. Trends were generally positive or close to zero in areas affected by fire prior to 1980 (1960 to 1980). These results suggest reasonable agreement between expected and observed NDVI trends.
Twenty-two percent of Canada’s land area showed a significant positive NDVI trend (p < 0.05) while only 0.5% showed a significant negative NDVI trend (Figure 10). There was a small area in the Prairies Ecozone+ near the Alberta, Saskatchewan, and United States borders with a negative trend. In addition, a number of patches with negative trends were found in the western Boreal Shield and Taiga Shield ecozones+, the southern Taiga Plains Ecozone+, and the western-central Montane Cordillera Ecozone+ corresponding to areas of mountain pine beetle kills.
Figure 10. Significant (p <0.05) trends in peak annual NDVI from 1985 to 2006.
Long Description for Figure 10
This map shows the significant trends in peak annual Normalized Difference Vegetation Index (NDVI) from 1985 to 2006. Twenty-two percent of Canada’s land area showed a significant positive NDVI trend (p < 0.05) while only 0.5% showed a significant negative NDVI trend. While broadly distributed across the country, the largest positive trends were found in regions of arctic tundra and taiga, alpine tundra, the Pacific coast, and the eastern prairies. The Newfoundland Boreal Ecozone+ had the greatest percent of its area showing positive trends with positive trends over nearly 41% of the Ecozone+. There was a small area in the Prairies Ecozone+ near the Alberta, Saskatchewan, and United States borders with a negative trend. In addition, a number of patches with negative trends were found in the western Boreal Shield and Taiga Shield Ecozones+, the southern Taiga Plains Ecozone+, and the western-central Montane Cordillera Ecozone+, corresponding to areas of mountain pine beetle kills.
Source: Adapted from Pouliot et al. (2009)
While broadly distributed across the country, the largest positive trends were found in regions of arctic tundra and taiga, alpine tundra, the Pacific coast, and the eastern prairies. The Newfoundland Boreal Ecozone+ had the greatest percent of its area showing positive trends with positive trends over nearly 41% of the ecozone+ (Figure 10, Table 1).
The analysis of climate forcing by Pouliot et al. (2009) showed moderate linear relationships for all regions in Figure 9 with temperature variables more strongly correlated with NDVI trends in northern regions (Regions 1 to 4) and precipitation more strongly correlated in the southern regions (5 to 8). The CTII showed that trends in NDVI for northern regions were more strongly influenced by climate than southern regions.
Results of the analysis of the impact of land cover change show that the positive NDVI trend in the south is driven primarily by changes in land cover, particularly on Vancouver Island in the Pacific Maritime Ecozone+ where vigorous succession following logging has led to increasing NDVI , and in the Prairies Ecozone+ where increases in NDVI correspond with increases in cropland area (Pouliot et al., 2009).
|Ecozone+||Area with Trend (km2)||Area with ¯ Trend (km2)|
|Arctic||2,281,558 (12.2%)||3,303 (0.1%)|
|Taiga Plains||116,163 (22.7%)||7,470 (1.5%)|
|Taiga Shield||415,278 (36.5%)||8,730 (0.8%)|
|Boreal Shield||335,205 (21.0%)||14,742 (0.9%)|
|Atlantic Maritime||33,408 (16.5%)||720 (0.4%)|
|Mixedwood Plains||15,876 (13.8%)||666 (0.6%)|
|Boreal Plains||130,554 (20.8%)||3,861 (0.6%)|
|Prairies||157,491 (35.1%)||1,116 (0.2%)|
|Taiga Cordillera||118,449 (35.3%)||189 (0.1%)|
|Boreal Cordillera||103,833 (23.8%)||594 (0.1%)|
|Pacific Maritime||63,864 (32.2%)||387 (0.2%)|
|Montane Cordillera||122,391 (29.8%)||4,527 (1.1%)|
|Hudson Plains||16,713 (4.9%)||405 (0.1%)|
|Western Interior Basin||16,713 (30.1%)||1,035 (1.9%)|
|Newfoundland Boreal||43,290 (40.9%)||0 (0%)|
|All of Canada||1,967,535 (22.3%)||47,745 (0.5%)|
Areas in the Arctic Ecozone+ with notable increases in NDVI include the northern portion of Banks Island, the Dundas and Sabine peninsulas of Melville Island, the south shore of Bowman Bay on Baffin island, the area along the northwestern shore of Hudson Bay, and the Labrador Peninsula within the Arctic Ecozone+, particularly the lower elevations bordering Ungava Bay (Figure 10). All of these areas (within the Arctic Ecozone+) correspond to tundra vegetation.
The CTII calculated for Regions 2 and 3 (Figure 9 on page 19) by Pouliot et al. (2009) revealed that NDVI trends in these regions of the Arctic Ecozone+ were strongly influenced by climate. In general, NDVI in northern regions (Regions 1 to 4, Figure 9) was negatively correlated with precipitation and positively correlated with temperature (Pouliot et al., 2009).
Olthof et al. (2008) examined NDVI trends in a portion of the Arctic Ecozone+ corresponding roughly to Region 3 in Figure 9 using the same AVHRR NDVI dataset analyzed here along with high resolution (30 m) Landsat data. They found that lichen-dominated communities had consistently lower NDVI trends than vascular plant dominated communities, though all showed increasing trends. This is consistent with ground studies (for example, Arft et al., 1999; Sturm et al., 2001; Hollister et al., 2005; Tape et al., 2006; Walker et al., 2006) and was attributed to increasing vigour and biomass of vascular plants and some impairment of lichen growth due to drying (Olthof et al., 2008).
An extensive area of strong NDVI increase was found in the northern region of this ecozone+, corresponding to a large area of conifer forest north of Great Bear Lake to the east of the Mackenzie Valley (Figure 10). A similar but smaller patch of increasing NDVI was found in the lower Mackenzie Valley. Further south, areas of increasing NDVI were more isolated. An area of decreasing NDVI was found west of Great Slave Lake which does not correspond to any recent burns.
Region 1 (Figure 9) in the climate forcing analysis by Pouliot et al. (2009) is within the Taiga Plains Ecozone+. The CTII calculated for this region revealed that NDVI is strongly influenced by climate, with a higher CTII than any other region analyzed. In general, NDVI in northern regions (Regions 1 to 4) was negatively correlated with precipitation and positively correlated with temperature (Pouliot et al., 2009).
Olthof et al. (2008) examined NDVI trends in a portion of the Taiga Plains Ecozone+ corresponding roughly to Region 1 in Figure 9 using the same AVHRR NDVI dataset analyzed here along with high resolution (30 m) Landsat data. They found that lichen-dominated communities had consistently lower NDVI trends than vascular plant dominated communities, though all showed increasing trends. This is consistent with ground studies (Arft et al., 1999; e.g. Sturm et al., 2001; Hollister et al., 2005; Tape et al., 2006; Walker et al., 2006) and was attributed to increasing vigour and biomass of vascular plants and some impairment of lichen growth due to drying (Olthof et al., 2008).
NDVI trends increased over a sizeable area in the northeastern portion of the ecozone+ (Figure 10). Increases were most pronounced south of Ungava Bay, which is dominated by tundra vegetation, and in southern Labrador, which is dominated by coniferous forest and shrubland. The area between these two “hotspots” also exhibited a positive, but less pronounced trend during this period. The positive trend identified south of Hamilton Inlet and Lake Melville in Labrador has not been identified in most other studies except for Slayback et al. (2003) whose analysis suggests that the increases in NDVI trends in this area were recent. A sizeable area of increasing NDVI trends was also found in the northwestern portion of the Taiga Shield. This area is predominantly covered with coniferous forests, but shrub and tundra vegetation are also found there.
Region 4 in the climate forcing analysis by Pouliot et al. (2009) is contained within the eastern portion of the Taiga Shield Ecozone+. The CTII calculated for this region revealed that NDVI there was strongly influenced by climate, though not as strongly as in the Taiga Plains and Arctic ecozones+. In general, NDVI in northern regions (Regions 1 to 4, Figure 9) was negatively correlated with precipitation and positively correlated with temperature (Pouliot et al., 2009).
The area of significant positive NDVI trend observed in southern Labrador continued into the eastern Boreal Shield Ecozone+ in Quebec, predominantly in areas of coniferous forest, but also in shrubland (Figure 10). The positive trend in this region has not been identified in most other studies, except for Slayback et al. (2003) whose analysis suggests that the increases in NDVI trends in this area are recent. Isolated areas of increasing NDVI were found further west in central Quebec with the greatest increases in areas of shrubland. Further west, a significant patch of mixedwood forests just north of Lake Superior showed increasing trends, as well as a similar patch just west of Lake Nipigon. Further west still, there were many isolated patches of positive NDVI trends found and a smaller number with negative NDVI trends. NDVI trends in this area are related to the dynamic process of wildfire and regeneration that is common in the western portion of the Boreal Shield Ecozone+. Not all increases in NDVI found in this region were the result of post-fire regeneration. In an analysis of recently burned and unburned sites using the same AVHRR NDVI dataset within the boreal forest region of central Canada, Alcaraz-Segura et al. (2010) found NDVI increases in all recently burned (since 1984) sites and 50% of unburned sites analyzed.
Regions 4 and 5 (Figure 9) in the climate forcing analysis by Pouliot et al. (2009) correspond to the northeastern and southcentral Quebec portions of the Boreal Shield Ecozone+. The CTII calculated for these regions revealed that NDVI in the northeastern region of the Boreal Shield was strongly influenced by climate, while NDVI in the southcentral Quebec region of the Boreal Shield was less influenced by climate. In general, NDVI in northern regions (Regions 1 to 4) was negatively correlated with precipitation and positively correlated with temperature while regions in the south (Regions 5 to 8) were positively correlated with precipitation and negatively correlated with temperature (Myneni and Williams, 1994 in Pouliot et al., 2009).
NDVI increased significantly in areas of mixed forest along the Gaspé Peninsula and on most of Cape Breton Island (Figure 10). Increasing trends may be associated with commercial logging that has increased the proportion of broadleaf trees but more detailed studies are necessary to test this hypothesis. It is important to note that deciduous and mixed deciduous forests, which make up a large portion of this ecozone+, have NDVI values close to the saturation point (Gilabert et al., 1996; Santin-Janin et al., 2009; Myneni and Williams, 1994 in Pouliot et al., 2009) (see Quality checks and limitations on page 19) which makes detection of small changes in NDVI difficult.
This ecozone+ is dominated by human modification of the landscape through urban development, extensive agriculture, and commercial logging. A few areas of negative NDVI trends west of Toronto were found and are likely associated with urban development. Extensive areas of positive NDVI trends were found mostly in areas of agricultural land (Figure 10). A more detailed analysis is necessary to attribute a cause and interpret ecological significance.
Significant NDVI trends were extensive but scattered (Figure 10). Much of the area found with positive trends was in agricultural areas, as well as some patches of strong positive trends in the forest and shrubland south and west of Lake Athabasca. Two small patches of strong negative NDVI trends appear to be associated with the Athabasca oil sands development.
A small area of the Prairies Ecozone+ in southern Alberta between Pakowki Lake and the Saskatchewan border showed a strong negative NDVI trend. Much of the remainder of this ecozone+ showed significant positive NDVI trends, particularly pronounced in Alberta west of Lethbridge and in Saskatchewan west of Moose Jaw (Figure 10).
In an arid area like the Prairies Ecozone+ moisture plays a major role in the value of NDVI, because the greenness of vegetation in this ecozone+ is very sensitive to the amount and timing of precipitation. It is possible that the increase in greenness that followed the drought of 2000 to 2002 (Bonsal and Regier, 2007) may be responsible for the positive trend in NDVI. Increasing NDVI trends in the Saskatchewan portion of this ecozone+ were found to be highly correlated with increasing cropland area, suggesting that land cover is an important driver of NDVI trends in this ecozone+ (Pouliot et al., 2009).
Increases in NDVI in this ecozone+ have also been shown by Slayback et al. (2003), Zhou et al. (2001), and Tateishi and Ebata (2004).
NDVI increased significantly in the area of shrub and tundra vegetation south of the Mackenzie Mountains, and to the west of the Mackenzie Valley (Figure 10).
An extensive but patchy area of increasing NDVI was found in the central region of this ecozone+, corresponding to an area of patchy conifer, shrub, and tundra vegetation (Figure 10). Although a CTII was not calculated within this ecozone+, it is likely the NDVI trends are correlated with climate as significant warming (particularly in the winter) is occurring in this ecozone+ (Zhang et al., 2011) and most of the Boreal Cordillera remains as intact wilderness (ESTR Secretariat, In Prep.).
NDVI increased over extensive areas within this ecozone+ (Figure 10). In particular, most of Vancouver Island showed a significant NDVI increase. Increases in NDVI trends on Vancouver Island (Region 8 in Figure 9) were found to be highly correlated with changes in land cover. This increase in NDVI is likely a result of logging followed by vigorous succession (Pouliot et al., 2009).
NDVI increased over much of the ecozone+, particularly at higher elevations in areas dominated by shrub and tundra vegetation (Figure 10). At lower elevations, increases in NDVI were found primarily in areas of mixedwood forest. These low elevation areas of increasing NDVI may represent mixtures of mature forest and cutblocks at early successional stages which have higher NDVI values than mature conifer forest. Negative trends in central British Columbia corresponded to an area of known mountain pine beetle damage that has been affecting the area since approximately 1994 (BCMF, 2003 in Pouliot et al., 2009). The extent of the area with negative NDVI trends (Figure 10) is smaller than the corresponding area of mountain pine beetle kill due to the resolution of AVHRR data, local variations in damage severity, and the position of the disturbance event in the time series. More recent damage (2003 to 2006) can be missed because these points may be seen as outliers in the robust trend analysis (Pouliot et al., 2009).
Relatively little of this ecozone+ showed significant NDVI trends. Those areas where NDVI increased were in the lowland portion of the ecozone+ dominated by wetlands (Figure 10).
Western Interior Basin
NDVI increased in this ecozone+ in areas of mixed forest and may result from regeneration after extensive forest harvesting (Figure 10). NDVI decreased significantly over approximately 2% of its area, scattered throughout the ecozone+ in areas that are primarily classified as conifer forest. The cause of these negative trends is not known. Negative NDVI trends may be an indication of drying within the ecozone+, though an analysis of the Palmer Drought Severity Index for this ecozone+ from 1950 to 2006 found no significant changes during this period (Zhang et al., 2011).
The greatest proportion of area with increasing NDVI was found in this ecozone+ (41%) (Figure 10 and Table 1). Much of northcentral Newfoundland showed an increase in NDVI. This appears to be centered to the south of the town of Grand Falls-Windsor. This is an area of extensive shrub and poor forest cover. It is possible that warming climate conditions are enabling this climate-limited vegetation to increase in density and vigour.
Several long-term studies of NDVI derived from data from the AVHRR sensor on the NOAA polar orbiting weather satellites have shown statistically significant increases in the north from Alaska to Ungava Bay during various periods from the 1980s to present (Myneni et al., 1997; e.g. Los et al., 2000; Kawabata et al., 2001; Zhou et al., 2001; Slayback et al., 2003; Goetz et al., 2005). The scientific consensus attributes these changes to the effects of climate change, particularly climate warming. More detailed studies have shown increases in herb and shrub vegetation (Arft et al., 1999; e.g. Sturm et al., 2001; Hollister et al., 2005; Tape et al., 2006; Walker et al., 2006; Olthof et al., 2008; Olthof and Pouliot, 2010) which would in turn cause long-term increases in NDVI. Together with in-situ observations, the NDVI trends provide supporting evidence that the effects of climate change are already occurring in northern regions.
Further south, areas of increasing trends in NDVI warrant further study as NDVI trends vary with the time period analysed and the dataset used. While no studies show extensive areas of decreasing NDVI, some studies using the 8 km resolution GIMMS AVHRR dataset (Goetz et al., 2005; Bunn and Goetz, 2006) identified more patches of negative NDVI trends in boreal forests, attributed to potential factors such as fire, drought stress, nutrient limitation, or insect and disease damage, than found in this study. A comparison between these two datasets showed that the GIMMS dataset may be biased towards negative trends (Alcaraz-Segura et al., 2010). As these results indicate a potential for reduced primary productivity in boreal forest areas (affecting the carbon balance), it is important to follow up on the extent to which this is occurring.
In conclusion, this study indicates a real greening of the north that is likely related to climate change, the normal burn-and-regeneration cycle in the boreal forest, and a possible greening related to a change of the forest age distribution in the commercial forest zone. Greening in the settled agricultural and urban areas is also observed and warrants more study.
- Date Modified: