Downslope Movement of Montane Forest Ecotones in

advertisement
Downslope Movement of Montane Forest Ecotones in
Northeastern US in Spite of Warming Between 1984 and 2011
#GC23E-0684
Jane R. Foster1 and Anthony W. D’Amato1
1Department
jrfoster@umn.edu
damato@umn.edu
of Forest Resources, University of Minnesota
1. Introduction
3. Methods
4. Results
Ecotones occur where distinct forest types meet across a climatic gradient and form transition zones. In the
Comparing Landsat TM Images
We were able to fit logistic models to Landsat NDMI data from at least one image date for 182 of 227
mountains, ecotones are compressed over steep vertical gradients. As the climate warms in New England,
Accurate detection of montane forest ecotone change requires differences between images to be minimized. Landsat reflectance needs to
potential topographic subsets (80%) in the Whites and 93 of 145 potential subsets in the Greens, representing
USA, high elevation forests dominated by spruce and fir are expected to recede upslope, with deciduous
be corrected in mountainous regions for differences in atmosphere, snow, clouds, solar incidence angles and shadows, and deciduous
areas of 94,927 ha and 23,920 ha, respectively. At the scale of individual mountain slopes (45-1000’s ha),
species like sugar maple moving up behind them (Beckage et al. 2008, Iverson et al. 2008). The rate of this
elevational range shift remains in question, as forest turnover can take decades or longer.
leaf-off phenology. We carefully matched Landsat surface reflectance images phenologically and corrected them for topographic
illumination differences using C-correction (Teillet et al. 1982). Spring leaf-off images show the full extent of both overstory and
understory coniferous tree canopies, while fall imagery represents mainly overstory trees. We show analyses for both seasons (Table 1).
Figure 1. Peak in the White Mountains, NH, with an
Ecotone
movement occurred in the understory, visible in the spring, with less movement in the overstory, indicated by
Year
ecotone (dashed lines) from high elevation boreal forest
Spring
1991
2010
Fall
1984
2000
2011
of spruce (Picea rubra) and fir (Abies balsamea) to
northern hardwood forest dominated by sugar maple
maple
hardwood forest
(Acer saccharum).
To detect elevational shifts in montane ecotones and attribute them to climate, one should observe changes at
spatial scales over which climate is correlated. Yet many empirical studies are severely limited in spatial
Day of
Year
Date
(Month-Day)
133
137
05-13
05-17
290
294
284
10-16
10-20
10-11
fall imagery (Table 2).
Green
Mountains
White
Mountains
Ecotones
Figure 3. Study area.
Figure
4. Aspect subsets.
subsets of
Figure
4. Topographic
Elevation area sampled
similar aspect for the White Mountains.
for local ecotone models shown in dark blue (650-1000 m).
extent. A frequently cited example of the rate and direction of montane ecotone change reported that forests
dominated by boreal species receded upslope in the Green Mountains, VT, at an alarming rate of 2.13 to
and lower ecotone boundaries moved downslope an average of -2.1 and -1.5 m year-1 in the Green
Mountains between 1991 and 2010, and -1.2 to -0.4 m year-1 in the White Mountains. Much of the
Table 1. Landsat scenes.
spruce/fir
boreal forest
montane forest ecotones moved both up and down in elevation, or were stable (Figure 8). On average, upper
Deriving Sample Areas – Contiguous Mountain Slopes
Burke
Mountain
We subdivided the elevation zone by aspect to create topographic subsets of NE or SW
2.77 m year-1 from 1965 to 2005 (Beckage et al. 2008). Yet this rate was calculated from aerial photos for
exposure and at least 500 pixels (45 ha). Subsets ranged from 45 to 3897 ha (median 132
narrow transects (6 m wide) on only two peaks. Other empirical studies have reported increases in conifer
ha). For each subset, we extracted surface reflectance for forested Landsat pixels that were
cloud and snow free in all five dates and balanced data across the elevation range. We used
greenness in the White Mountains (Vogelmann et al. 2012) and increases in P. rubra and A. balsamea
vegetation indices (VI) as our proxies for montane boreal species abundance (Normalized
basal area in Hubbard Brook over the same time period (van Doorn et al. 2011). An expanded analysis of
Difference Vegetation Index (NDVI), and Normalized Difference Moisture Index (NDMI)).
ecotone location and movement across the northeastern region is needed.
Figure 7. Violin plots show the distribution of upper (green) and lower (orange) ecotone boundary elevations
for the boreal-hardwood forest ecotone in the Green Mountains (a) and the White Mountains (b). Dark dashed lines
show the trend between weighted means of boundary locations detected in spring leaf-off conditions (understory
visible to sensor). Dotted grey lines show boundary locations detected in fall conditions (understory less visible to
sensor, 1984, 2000, and 2011). Data represented by violin plots varies by year as shown by the number of models
(n) that could be fit in each Landsat date. Lower ecotone boundary distribution is offset slightly for clarity.
Table 2. Ecotone elevation change in the Green Mountains.
a.
Ecotone Edge Elevations
ecotone difference*
season
1984
1991
2000
2010
Ecotone elevation
range
650-1000 m
(A.S.L)
Mount
Figure 2. Elevation range over which
Washington (1917 m) montane forest ecotone occurs (yellow
lines) in the Green and White
N
Mountains of northeastern U.S.A.,
draped on 3D Landsat image of Mount
Washington, NH. Leaf-off Landsat
image from October 11, 2011, shows
coniferous vegetation in dark red
tones. Local abundance of highelevation boreal species varies
ski slopes
Fall
logistic model fits. Landsat RGB
images with ecotone boundaries
lower ecotone boundaries (Figure 5). We drew
(yellow) (a) and NDMI images (c)
b.
random draws from the posterior distribution
N
yr-1
(m)
upper
-38.87
54
2010
1991
19
-2.05
lower
-29.30
55
2010
1991
19
-1.54
upper
-18.71
31
2000
1984
16
-1.17
lower
-16.59
33
2000
1984
16
-1.04
upper
24.45
48
2000
2011
-11
-2.22
lower
-8.16
49
2000
2011
-11
0.74
edges (blue).
NDMI
of ecotone fits, based on parameter mean and
low Æ
high
c.
ecotone elevations among years using Tukey’s
5. Conclusions
Landsat change detection showed that coniferous boreal species have moved downslope over the past 20-
HSD.
30 years in the Green and White Mountains of the northeastern US, in contrast to smaller scale studies that
throughout elevation band, which
Hypothetical transects
Spring
show (b) ecotone edges from
inflection points (max and min of the 2nd
derivative) of the fitted curve as the upper and
would be hard to quantify with narrow
N
subsets year1 year2 years
Burke, VT. Dashed vertical lines
spatially correlated errors. We defined the
SD and compared posterior distributions of
(m)
Green Mountains
Figure 5. Landsat data and
model fits for example subset, Mt.
package ‘nlme’ in R. Models accounted for
edge
2011
We fit 4-parameter logistic models to predict
NDMI as a function of elevation with the
difference
weighted
Figure 8. White Mountains 1991-2010
Fitting Models & Deriving
1984
1991
2000
2010
2011
found the opposite trend (upward movement of the ecotone). Most of this change represents spread of
transects alone. Hypothetical transects
Validation with Forest Inventory Data from Hubbard Brook
spruce and fir in the understory layer, as shown by larger changes in spring imagery in comparison to fall.
illustrate this variability.
We compared Landsat reflectance for locations of forest inventory plots from Hubbard Brook Experimental Forest (Figure 6a)
At the scale of individual mountain slopes (45-1000’s ha), montane forest ecotones have moved both up and
2. Objectives
(Schwarz et al. 2003). Relative basal area (RBA) of boreal species measured in 400 m2 plots (1995-1998) agreed well with spring
leaf-off Landsat vegetation indices (NDMI for 900
m2
pixels shown) from 1991 (Figure 6b). This confirms that NDMI is a
suitable proxy for boreal species abundance from which to model ecotone boundary locations.
I. Systematically model and map the boreal-hardwood forest ecotone from leaf-off Landsat imagery across
the entire elevational zone (650 and 1000 m A.S.L) for the Green and White Mountains (Figure 2 and 3),
including named peaks and valleys from prior publications: Mt. Abraham, Mt. Bolton and Camel’s Hump
(Siccama 1974, Beckage et al. 2008), Hubbard Brook, Crawford Notch, “The Bowl”, etc.
stress in these species, other factors must be having a stronger effect on recent forest dynamics than climate
change . We hypothesize that the observed process represents recovery of spruce and fir forests to
a.
b.
b
b.
Figure 6. Hubbard
historically lower ecotone elevations, following spruce dieback and decline in the 70’s and 80’s, and human
Brook plots that overlap
resource extraction over the prior century. These results highlight the importance of analyzing forest
with modeled
topographic subsets (a.).
Boreal species
II. Calculate the distribution of elevational changes in the montane ecotone at local and regional scales,
abundance (relative
accounting for model uncertainty, from 1984 to 2011.
basal area (RBA)) vs.
Landsat vegetation index
III. Model the dependence of ecotone elevation on Latitude, slope, and aspect (not reported here).
down, but the overall average is down. As this is opposite to the response predicted by increasing climate
(NDMI) (b.).
processes at appropriate spatial scales and considering competing drivers of forest change.
References
Beckage et al. (2008) Proc Nat Acad Science 105:4197-4202.
Schwarz et al. (2003) Ecology 84:1862-1878.
Siccama (1974) Ecol Monogr 44:325–349.
Teillet et al. (1982) Can J of Remote Sensing 8: 84-106.
Van Doorn et al. (2011) Can J For Res 41:1369-1379.
Vogelmann et al. (2012) Remote Sens of Environ 122:92-105..
Photo: Steve Loynd
From Loon Mountain looking NE.
Download