1

1) Develop climate-based land-cover categories (cliomes) for
Alaska and western Canada using down-scaled gridded historic climate data from the Scenarios Network for Alaska and Arctic Planning (SNAP) and cluster analysis
2)Link the resulting cliomes to land cover classes, and define each biome by both climate and ecosystem characteristics.
3)Couple these cliomes with SNAP’s climate projections, and create predictions for climate-change-induced shifts in cliome ranges and locations.
4) Use the results to identify areas within Alaska, the Yukon and NWT that are least likely to change, and those that are most likely to change over the course of this century.
2

Follow-up to Connecting Alaska Landscapes into the Future Project
 Broader spatial scope
 More input data
 Clustering methodology
3
3




 Eliminates false line at US/Canada border


Creates groups with greatest degree of intra-group similarity and inter-group dissimilarity
Gets around the problem of imperfect mapping of vegetation and ecosystem types
 Allows for comparison and/or validation against existing maps of vegetation and ecosystems
4

Area of Canada selected for cluster
analysis. Selected area is lightly shaded, and the unselected area is blue. The red line includes all ecoregions that have any portion within NWT. Limiting total area improves processing capabilities.
5

6





Calculated concurrence of 15 models with data for 1958-2000 for surface air temperature, air pressure at sea level, and precipitation
Used root-mean-square error
(RMSE) evaluation to select the
5 models that performed best for
Alaska and northwestern Canada
Focused on A1B, B1, and A2 emissions scenarios
Downscaled course resolution GCM data to 2km
GCM output (ECHAM5)
Figure 1A from Frankenberg st al., Science, Sept. 11, 2009
7

150
145
140
135
130
125
120
115
110
105
100
Ice breakup dates for the Tanana (left) and Yukon
(right) Rivers for the full recorded time periods.
Days are expressed as ordinal dates. A statistically significant trend toward earlier thaw dates can be found for both rivers. y = -0.1825x + 128.69
R² = 0.1982
125
120
115
110
105
100
150
145
140
135
130 y = -0.0579x + 133.71
R² = 0.1371
8

Example of a dendrogram.
Clusters can be created by cutting off this tree at any vertical level, creating (in this case) from one to
29 clusters.



Cluster analysis is the statistical assignment of a set of observations into subsets so that observations in the same cluster are similar in some sense.
It is a method of “unsupervised learning” – where all data are compared in a multidimensional space and classifying patterns are found in the data.
Clustering is common for statistical data analysis and is used in many fields.
9






The dissimilarity matrix describes pairwise distinction between objects.
The algorithm PAM computes representative objects, called medoids whose average dissimilarity to all the objects in the cluster is minimal
Each object of the data set is assigned to the nearest medoid.
PAM is more robust than the well-known kmeans algorithm, because it minimizes a sum of dissimilarities instead of a sum of squared Euclidean distances, thereby reducing the influence of outliers.
PAM is a standard procedure
10





For Alaska, Yukon, and BC, SNAP uses 1961-1990 climatologies from
PRISM, at 2 km
For all other regions of Canada SNAP uses climatologies from CRU, at 10 minutes lat/long (~18.4 km)
In clustering these data, the differences in scale and gridding algorithms led to artificial incongruities across boundaries.
The solution was to cluster across the whole region using CRU data, but to project future climate-biomes using PRISM, where available, to maximize resolution and sensitivity to slope, aspect, and proximity to coastlines.
CRU data and SNAP outputs after PRISM downscaling
11





Choice is mathematically somewhat arbitrary, since all splits are valid
Some groupings likely to more closely match existing land cover classifications
How many clusters are defensible?
How large a biome shift is “really” a shift from the conservation perspective?
Sample cluster analysis showing 5 clusters, based on
CRU 10’ climatologies. This level of detail was deemed too simplistic to meet the needs of end users.
Sample cluster analysis showing 30 clusters, based on
CRU 10’ climatologies. This level of detail was deemed too complex to meet the needs of end users, as well as too fine-scale for the inherent uncertainties of the data.
12

0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 25 30 40 50
Number of Clusters Returned
Mean silhouette width for varying numbers of clusters between 3 and 50.
High values in the selected range between 10 and 20 occur at 11, 17, and
18.
13

Eighteen-cluster map for the entire study area. This cluster number was selected in order to maximize both the distinctness of each cluster and the utility to land managers and other stakeholders.
14

Cluster certainty based on silhouette width. Note that certainty is lowest along boundaries
.
15

Describing the clusters:
-30
-40
20
10
0
-10 winter
-20 spring summer fall 7
8
5
6
3
4
1
2
9
10
11
16
17
18
12
13
14
15
Mean seasonal temperature
by cluster. For the purposes of this graph, seasons are defined as the means of 3months periods, where winter is December, January, and February, spring is
March, April, May, etc.
16

Describing the clusters:
2500
2000
2249
1500
1000
500
117 174
198 206 243
274 281
355
284
561
390 420
586
0
857
474
545
443
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Precipitation by
cluster. Mean annual precipitation varies widely across the clustering area, with
Cluster 17 standing out as the wettest.
Cluster 17
17

70
110
90
230
210
190
170
150
130
Describing the clusters:
3000
2500
2000
1500
Days above freezing
Length of above-freezing season and
GDD by cluster. Days above freezing were estimated via linear interpolation between monthly mean temperatures. Growing degree days (GDD) were calculated using
0 ° C as a baseline.
1000 Growing
Degree
Days
500
1200
50
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 cluster
0
1000
800 total for months with mean temperature below freezing
600 Warm-season and cold-season
precipitation by cluster. The majority of precipitation in months with mean temperatures below freezing is assumed to be snow (measured as rainwater equivalent).
400
200
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Clusters total for months with mean temperature above freezing
18

Describing the clusters:
http://land cover.usgs.gov/nalcms.php
GlobCover
2009
North American
Land Change
Monitoring System
(NALCMS 2005)
Created 2/4/11 3:00 PM by
Conservation Biology Institute
Alaska Biomes and Canadian
Ecoregions.
AVHRR Land cover, 1995
19

Cluster
Number
1
2
16
17
13
14
11
12
8
9
6
7
3
4
5
10
15
18
AVHRR
Canadian and
Alaskan
Ecoregions GlobCover NALCMS
Open shrub
Open shrub
Open shrub
Closed Shrubland
Open shrub
Closed Shrubland
Woodland
Wooded Grassland
Woodland
Grassland
Woodland
Woodland
Northern Arctic
Southern Arctic
Alaska Arctic
Alaska Arctic
Southern Arctic
Taiga Shield
Taiga Plain
Boreal Cordillera
Alaska Boreal
Western Tundra
Taiga Shield
Taiga Plain
Sparse (<15%) vegetation
Sparse (<15%) vegetation evergreen forest (>5m)
Open (15-40%) needleleaved deciduous or evergreen forest (>5m)
Sparse (<15%) vegetation
Sparse (<15%) vegetation
Open (15-40%) needleleaved deciduous or evergreen forest (>5m) barren lands polar or subpolar grassland lichen moss
Sparse (<15%) vegetation
Sparse (<15%) vegetation polar or subpolar grassland lichen moss polar or subpolar grassland lichen moss polar or subpolar grassland lichen moss Sparse (<15%) vegetation
Sparse (<15%) vegetation
Sparse (<15%) vegetation
Open (15-40%) needleleaved deciduous or polar or subpolar grassland lichen moss subpolar taiga needleleaf forest temperate or subpolar needleleaf forest temperate or subpolar shrubland temperate or subpolar shrubland subpolar taiga needleleaf forest temperate or subpolar needleleaf forest
Open shrub
Evergreen Needleleaf Forest
Evergreen Needleleaf Forest
Taiga Cordillera
Montane
Cordillera
Boreal Plain
Sparse (<15%) vegetation
Open (15-40%) needleleaved deciduous or evergreen forest (>5m)
Open (15-40%) needleleaved deciduous or evergreen forest (>5m)
Open (15-40%) needleleaved deciduous or evergreen forest (>5m) Evergreen Needleleaf Forest Boreal Shield
North Pacific
Bare Ground Maritime
Grassland Prairie
Sparse (<15%) vegetation
Closed to open (>15%) herbaceous vegetation (grassland, savannas or lichens/mosses) barren lands temperate or subpolar needleleaf forest cropland temperate or subpolar needleleaf forest temperate or subpolar needleleaf forest cropland
Comparison of cluster-derived cliomes with existing land cover designations.
This table shows only the highestpercentage designation for each land cover scheme. Colorcoding helps to distinguish categories.
20

Modeled cliomes for the historical baseline years, 1961-1990. As in all projected maps, Alaska and the Yukon are shown at 2km resolution based on PRISM downscaling, and the Northwest Territories are shown at 18.4 km resolution based on CRU downscaling.
21

Future
Projections
Original 18 clusters
Projected cliomes for the five-model composite, A1B
(mid-range ) climate scenario.
Alaska and the Yukon are shown at 2km resolution and
NWT at 10 minute lat/long resolution .
22

2060’s
2090’s
2000’s
2030’s
Future
Projections
Projected cliomes for the A2 emissions scenario.
This scenario assumes higher concentrations of greenhouse gases, as compared to the
A1B scenario.
2000’s
2030’s
Projected cliomes for the B1 emissions
scenario. This scenario assumes lower concentrations of greenhouse gases, as compared to the
A1B scenario.
2060’s
2090’s 23

Future
Projections
Original 18 clusters
Projected cliomes for
single models. The five
GCMs offer differing projections for 2090.
24

Future
Projections
Projected change and resilience under three
emission scenarios. These maps depict the total number of times models predict a shift in cliome between the 2000’s and the
2030’s, the 2030’s and the
2060’s, and the 2060’s and the 2090’s. Note that number of shifts does not necessarily predict the overall magnitude of the projected change.
25

Discussion:
1.00
Bare Ground
0.90
Cropland
Grassland  Comparison with existing land cover designations
 Assessment of which shifts are most significant in terms of vegetation communities
 Linkages with species-specific research
 Habitat characteristics/requirements
 Dispersal ability
 Historical shifts
0.80
Open Shrubland
0.70
Closed Shrubland
0.60
Wooded Grassland
0.50
Woodland
0.40
Mixed Forest
0.30
0.20
0.10
Deciduous
Broadleaf Forest
Evergreen
Needleleaf Forest
Water (and
Goode's interrupted space)
0.00
10 11 12 13 14 15 16 17 18 1 2 3 4 5 6 7 8 9
Dominant AVHRR land cover types by
cluster number. All land cover categories that occur in 15% or more of a given cluster are included.
26

Discussion:


Changes are unlikely to happen smoothly and spontaneously, and are certainly not going to happen instantly



Seed dispersal takes time
Changes to underlying soils and permafrost take even longer
In many cases, intermediate stages are likely to occur when climate change dictates the loss of permafrost , a new forest type, or new hydrologic conditions
Even in cases when biomes do shift on their own, they almost never do so as cohesive units



Trophic mismatches are likely
Invasive species may have greater dispersal abilities than native ones
It may become increasingly difficult to even define what an “invasive species” is
27

Discussion:
 Identification of refugia
 Identification of vulnerable species/areas
 Collaboration and dialogue between modelers and field researchers
 Selecting focus of future research
 Shift from “preservation” to
“adaptation”
Toolik Lake
Catherine Campbell http://www.polartrec.com/expeditions/changingtundra-landscapes/journals/2008-07-22
Brian Bergamaschi
(USGS) sampling wells at Bonanza
Creek LTER site. http://hydrosciences.color
ado.edu/research/govt_par tners.php
28

 All project inputs and outputs are available to the public
 The final report (full report, main text only, or appendices only) can be downloaded here: http://www.snap.uaf.edu/project_page.php?projectid=8
 Maps and data are also available in GIS formats; contact SNAP for further information
(nlfresco@alaska.edu)
29

The US portion of this study was made possible by the US Fish and Wildlife Service, Region 7, on behalf of the Arctic
Landscape Conservation Cooperative (LCC), with Karen Murphy as project lead and assistance from Joel Reynolds and
Jennifer Jenkins (USFWS). The Canadian portion of this study was made possible by The Nature Conservancy Canada,
Ducks Unlimited Canada, Government Canada and Government Northwest Territories, with Evie Whitten as project lead. Data and analysis were provided by the University of Alaska Fairbanks (UAF) Scenarios Network for Alaska and
Arctic Planning (SNAP) program and Ecological Wildlife Habitat Data Analysis for the Land and Seascape Laboratory
(EWHALE) lab, with Nancy Fresco, Michael Lindgren, and Falk Huettmann as project leads. Further input was provided by stakeholders from other interested organizations.
We would also like to acknowledge the following organizations and individuals:








Karen Clyde, Government YT
David Douglas, US Geological Survey
Evelyn Gah, Government NWT
Lois Grabke, Ducks Unlimited Canada
Troy Hegel, Government YT
James Kenyon, Ducks Unlimited Canada
Wendy Loya , the Wilderness Society
Lorien Nesbitt , Déline Renewable Resources Council
Thomas Paragi, Alaska Department of Fish and Game
Michael Palmer, TNC
Scott Rupp , SNAP
Brian Sieben, Government NWT
Stuart Slattery, Ducks Unlimited Canada
Jim Sparling, Government NWT
30