Do recent NDVI trends demonstrate boreal forest decline in Alaska?

We used the 1940–2018 Alaskan fire perimeter database produced by the Alaska Fire Service (AFS) and acquired from the Alaska Interagency Coordination Center (https://fire.ak.blm.gov/incinfo/aklgfire.php) to identify sample locations. We subsequently determined the actual age of each stand with tree cores, so errors in the fire perimeter database and the lack of fire history information before 1940 would not affect our analysis.

June-August Collection 1, Level 1 Landsat 5, 7, and 8 Surface Reflectance and Brightness Temperature images with less than 30% cloud cover for WRS2 Paths 66–72 and rows 14–16 were downloaded from the USGS (https://espa.cr.usgs.gov; 658 images total; 604 images after 1998) after reprojection at 0.0002695° resolution (30 m latitude, variable longitude). The images were masked for snow, cloud and cloud shadow, and NDVI was calculated, homogenized across instruments and the summer median calculated (see details in Goulden and Bales 2019). Images were not masked for standing water. There was some residual effect of the Landsat 7 Scan Line Corrector (SLC) failure despite masking, presumably due to systematic alignment of gaps and resulting differences in scene availability. Subsequent analyses focused on 1999–2018 due to large gaps in the earlier record with the lack of a consistent downlink station in Alaska. Simple linear regressions were calculated from the resulting NDVI summer medians over the 20 years time period. Trends with a significant non-zero slope were calculated using a two-tailed t-test, at p ≤ 0.1 (slopes were considered significant for α ≤0.05 or ≥0.95).

We surveyed 102 forest stands across interior Alaska, spanning a distance of over 425 km (figure 1) in August 2017 and 2018. The sites were selected before visiting the field to include locations with and without a fire since 1940. Recently burned sites were selected to span a range of years since fire, while sites without a recent fire were selected to include a range of Landsat NDVI trends.

Zoom In Zoom Out Reset image size

Sites were further screened to avoid large topographic and land use gradients and to allow efficient access. Two perpendicular 60 m transects were established at each site, and stand characteristics were measured at 10 m intervals along each transect. Measurements included canopy cover, percent species composition of upper and lower canopies (measured as proportional cover of species over total area on interval), average height of upper and lower canopies, ground cover, standing and downed deciduous stems, and standing and downed evergreen stems. Any evidence of disturbance (fire, permafrost thaw, beetle damage, logging, or other human impacts) was noted. Fraction deciduous was subsequently calculated from the proportion of deciduous cover in the upper canopy over the total canopy coverage.

LAI (leaf area index) data were collected at each site using LAI-2000 plant canopy analyzer (LI-COR). Eight LAI-2000 measurements were taken under the canopy in each 30 m interval along the 60 m perpendicular transects and corresponding above canopy measurements were taken in the nearest canopy clearing. The measurements were then processed with the internal LI-COR LAI software to provide four estimations of LAI per site. The LAI-2000 is known to underestimate LAI in coniferous stands, so species specific corrections for black spruce (Picea mariana (Mill.) B.S.P.) were applied to black spruce site data (Chen 1996, Law et al 2008). Corrections for white spruce (Picea glauca (Moench) Voss) are not available, and these stands were frequently mixed with deciduous species; we consequently were unable to correct for clumping in the white spruce stands and note that LAI may be underestimated at these sites. The mean LAI was then calculated from the four estimations and used as the site level LAI.

To estimate tree age of each site, multiple tree cores were collected from representative trees of each species in the upper canopy (2–4 trees per species) with 4.3 mm outer diameter increment borers (Haglöf). Tree cores were mounted, sanded manually, and rings were counted using Lignovision software (Rinntech). The average tree age was subsequently determined by taking the mean ring count of cores analyzed. For recently burned and early fire recovery sites the site age was determined from the year of fire and confirmed by tree cores when possible.

The sites visited were subsequently sorted into five successional stages based on stand age and majority (>50%) species composition of the upper canopy: (1) Recently burned (fire after 2005; ≤13 years since fire), (2) Early fire recovery (burned between 1980–2005; 13–38 years since fire), (3) Mid-to-late successional, deciduous dominated stand, (4) Mid-to-late successional, white spruce dominated stand, and (5) Mid-to-late successional, black spruce dominated stand.

Each successional stage had a distinct forest structure, and a characteristic NDVI trend (figure 2 and table S1 (available online at (stacks.iop.org/ERL/15/095007/mmedia))). Recently burned stands (age 0–13 years) had charred, dead trees with occasional live, evergreen trees in the upper canopy that survived the fire, and deciduous saplings in the lower canopy (figure 3). Herbaceous plants, moss, and charred material covered most of the ground. Early fire recovery stands (age 13–38 years) had upper canopies of evergreen trees that had survived fire, along with deciduous species such as paper birch (Betula papyrifera Marshall) and quaking aspen (Populus tremuloides Michx.), and lower canopies of birch and aspen, and mostly herbaceous ground cover (figure 3). Standing and fallen dead trees from the fire were still visible. Mid-to-late successional deciduous stands (age 30–180 years) had dense upper canopies of birch and aspen, and lower canopies of birch, aspen, and alder (Alnus sps.), with progressively more white spruce or black spruce in older stands (figure 3). The ground cover of deciduous stands consisted of bare ground and litter, herbaceous plants and small percentages of moss. Mid-to-late successional white spruce stands (age 50–200 years) often had multiple generations of white spruce in the upper canopy, with varying percentages of deciduous species in the upper canopy, and lower canopies of white spruce with dense thickets of alder (figure 3). Mid-to-late successional black spruce (age 30–200 years) had sparse black spruce in both the upper and lower canopies, and occasionally small percentages of birch and willow (Salix sp.) in the upper and lower canopies. Thick layers of moss, lichen, and herbaceous plants covered most of the ground in mid-to-late successional black spruce stands (figure 3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

There was a visible mosaic of increasing and decreasing NDVI trends across interior Alaska, which was clearly linked with recent fire history (figures 4 and S1). Burn scars from recent decades show consistent behavior in NDVI trends, with trend perimeters typically matching fire perimeters. Timing of fire largely controlled NDVI trend direction and intensity, with recent burns (2018–2005) showing strong NDVI decreases and burns from 2005–1980 showing NDVI increases (figures 4, S1 and S2). Trend significance inside burn scar perimeters was variable, with some pixels showing significant trends (p-value < 0.1), and others insignificant within the same burn scar (figure 4(B)).

Zoom In Zoom Out Reset image size

There was heterogeneity in NDVI trends in the absence of recent fires, with some undisturbed patches showing increasing NDVI and others showing decreasing NDVI. These undisturbed patches often showed smaller magnitude trends than areas with reported fires, though 17.2% of these trends were significant (p-value < 0.1) (figure 4(B) and S2). Moreover, the unburned locations with significant trends were clustered in relatively homogenous patches (figures 4(B) and S1(b)), and were not intermixed with pixels of opposite slope as would be expected with a random distribution, implying that NDVI trends in the absence of recent fire may reflect ecological variation.

Our data allowed us to create a 200 years long chronosequence of the boreal forest NDVI successional trajectory (figure 5), and to calculate the resulting 20 years chronosequence of NDVI trends related to recent fires (since 1940) (figure 6). Stands with recent fire displayed large ranges in NDVI due to the rapid changes in forest structure and composition following fire, but overall had the lowest average NDVIs (figure 5(A)). Fire disturbance caused a sharp drop in NDVI due to the loss of vegetation, followed by rapid recovery of herbaceous and deciduous vegetation leading to strong NDVI increases (figures 3 and 5(B)). NDVI continued to increase until around 30 years since fire (figure 5(A)), at which point many stands displayed maximum fraction deciduous in the upper canopy (figure 3). The immediate effect of fire and subsequent recovery were large drivers of NDVI change in stands <30 years old (figures 5(B) and 6).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

After stands reached 30 years of age, NDVI typically stabilized and displayed more gradual changes. Mid-to-late successional deciduous stands collectively had the highest summer median NDVI, followed by white spruce, then black spruce stands (figure 5(A)). With increasing age, deciduous stands exhibited weak NDVI decreases and black spruce stands of all ages showed weak NDVI increases (figures 2 and 5(B)). White spruce stands displayed both decreasing and increasing trends, with younger white spruce stands typically showing decreases, and older white spruce stands showing increases (figures 2 and 5(B)).

The relationship between successional stage and NDVI can be explored further by examining the relationships between fraction deciduous in the upper canopy and NDVI, and LAI and NDVI. We found that NDVI was positively correlated with fraction deciduous in the upper canopy (r² = 0.63) and with LAI (r² = 0.15) (figure 7). Deciduous stands by definition had the highest fraction deciduous and highest NDVI. These stands showed marked variation in LAI, which was correlated with differences in NDVI, especially in sparsely vegetated, recent burns. Compared to deciduous stands, white spruce stands had a smaller fraction deciduous, a smaller range in LAI, and lower NDVI. Here, upper canopies typically consisted of a mixture of white spruce and deciduous species and an increasing fraction deciduous generally increased NDVI (figures 3 and 7(A)). Black spruce stands had the lowest NDVI, a relatively small range in LAI, and were less likely to have deciduous species in the upper canopy. Overall, fraction deciduous in the upper canopy was a stronger predictor of NDVI than LAI.

Zoom In Zoom Out Reset image size