Satellite Imagery

Introduction and Methods

Vegetation productivity of an area can be estimated using satellite imagery. Satellite sensors measure the amount of red (R) and near infrared (NIR) radiation reflected by the surface of the earth. Since plant leaves absorb red radiation and reflect near infrared radiation, the ratio of NIR to R reflectance values at a given location is directly related to vegetation productivity at that location. Many vegetation indices have been developed based on this relationship. One of the most common is the Normalized Difference Vegetation Index (NDVI); the Enhanced Vegetation Index (EVI) is calculated similarly, but adjusts for the influence of bare soil and aerosol scattering in the atmosphere. Values generally range from 0 to 1 over land, with values closer to 1 indicating highly productive vegetation and values closer to 0 indicating bare ground.

We used 250-m resolution vegetation index data obtained from the MODIS sensor on NASA’s Terra satellite to quantify recent changes in vegetation productivity across the range of the Bathurst caribou herd. Data are available from February 2000 to the present. Although the MODIS sensor takes an image of the entire surface of the earth every day, images are provided as 16-day “composites”, meaning each image contains the highest quality pixels obtained in a given 16-day period. Compositing is primarily done to eliminate cloudy pixels.

We opted to use EVI values in our analysis of vegetation change, downloading all available MODIS data from across the entire range of the Bathurst caribou herd. Composite images still contain a small number of poor quality pixels, so we downloaded NASA’s accompanying pixel quality dataset. We then mosaicked the EVI and pixel quality images from each composite period together to form a single seamless image.

To analyze changes in vegetation productivity over time, we extracted annual maximum EVI values from each pixel as estimates of peak vegetation productivity in each year (Figure 1). However, time series’ of vegetation indices are frequently noisy due to differences in view angle, illumination patterns, aerosol content in the atmosphere, and cloud cover between composite periods. To correct for this, we weighted the influence of measured EVI values based on the pixel quality rating associated with them. Good-quality EVI values were assigned a weight of 1.0, medium-quality values (e.g., pixels with a steep view angle) were assigned a weight of 0.5, and poor-quality values (e.g., partially cloud-covered or snow/ice-covered) were assigned a weight of 0.1.

Figure 1: An 18-year time series of raw EVI values (blue line) taken from a single 250-m MODIS pixel and corrected for image quality (red line). Circle size is indicative of pixel quality. Note that the abnormally high values in 2010 and 2013 are the result of poor quality pixels and that their influence was substantially downweighted. Also note that a fire occurred in this pixel during the 2014 growing season, resulting in a sharp decline in maximum EVI values followed by a steady increase thereafter.

We performed linear regression analyses on the time series curves to determine the direction (slope) and magnitude of change in annual maximum EVI values from 2000 to 2017. Positive slopes indicate an increase in vegetation productivity (greening), negative slopes indicate a decrease in vegetation productivity (browning), and a value of zero indicates no change. To ensure that any trends in EVI values we observed were not due to disturbance, we omitted waterbodies and areas in which a fire had occurred within the last 60 years.

Results

Vegetation productivity within the Bathurst caribou herd’s range has generally increased over the past 18 years. Within their annual range, greening trends (i.e. positive slopes in annual maximum EVI) occurred in 82% of all pixels since the year 2000, while browning trends occurred in the remaining 18% of pixels. In their calving grounds, greening occurred in 74% of pixels and browning in the remaining 26%. However, most of these trends (83% in their annual range and 89% in their calving grounds) were not statistically significant at the p = 0.05 level.

After excluding pixels with non-significant trends, the percentage of greening pixels increased from 82% to 94% and 74% to 95% in the herd’s annual range and calving grounds, respectively (Figure 2; Figure 3). In other words, the vast majority of browning trends were not statistically significant. Most of the significantly greening pixels in the herd’s annual range had slopes between 0.002 and 0.004 EVI/year and were located near the centre, or along the eastern border of their range between approximately 63 and 65°N. There were also three smaller clusters of significantly greening pixels in their calving grounds; two along the northern border between 109 and 111°W, and one near the centre.

Figure 2: Statistically significant trends in MODIS annual maximum EVI values from 2000 to 2017. Green colours indicate positive trends, while orange and brown colours indicate negative trends. White areas indicate fire scars or pixels with statistically insignificant trends at the p = 0.05 level, both us which were excluded from analysis. The calving grounds and annual range boundaries were defined based use distributions derived from 1996 to 2015 collar data.

Figure 3: Area (in km2) of statistically significant trends in annual maximum EVI values from 2000 to 2017 within the Bathurst caribou herd’s annual range (left) and calving grounds (right). Note that the y-axes are scaled differently and that the total areas with positive trends (greening) and negative trends (browning) in each location are represented by the two right-most bars.