Predicting Future Potential Climate-Biomes

for the Yukon, Northwest Territories, and Alaska

Goals1) 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.

BackgroundFollow-up to Connecting Alaska Landscapes into the Future Project

Broader spatial scopeMore input dataClustering methodology

3

Improvements over Phase IExtended scope to northwestern CanadaUsed all 12 months of data, not just 2Eliminated pre-defined biome/ecozone

categories in favor of model-defined groupings (clusters)Eliminates false line at US/Canada borderCreates groups with greatest degree of intra-

group and inter-group dissimilarityGets around the problem of imperfect mapping of

vegetation and ecosystem typesAllows for comparison and/or validation against

existing maps of vegetation and ecosystems

Sampling Extent

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.

 

Historical Climate Trends:Ice Breakup DataIce 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 + 128.69R² = 0.1982

100

105

110

115

120

125

130

135

140

145

150

1962

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

y = -0.0579x + 133.71R² = 0.1371

100

105

110

115

120

125

130

135

140

145

150

18

96

19

01

19

06

19

11

19

16

19

21

19

26

19

31

19

36

19

41

19

46

19

51

19

56

19

61

19

66

19

71

19

76

19

81

19

86

19

91

19

96

20

01

20

06

20

11

Methods: cluster analysisCluster 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.

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.

Methods: SNAP climate models

SNAP is a collaborative network of the University of Alaska, state, federal, provincial, and local agencies, NGOs, and industry partners.

Its mission is to provide timely access to scenarios of future conditions in Alaska and the Arctic for more effective planning by decision-makers, communities, and industry.

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, B2, and A1 emissions scenarios

Downscaled course resolution GCM data to 2km

SNAP data based on CRU historical datasets and IPCC Global Circulation (GCM) models

GCM output (ECHAM5)Figure 1A from Frankenberg st al., Science, Sept. 11, 2009

Methods: Partitioning Around Medoids (PAM)

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

Resolution limitations 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

How many clusters?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.

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 25 30 40 50

Ave

rage

Sih

ouett

e W

idth

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.

How many clusters?

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.

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

Describing the clusters: temperature

Mean seasonal temperature by cluster. For the purposes of this graph, seasons are defined as the means of 3-months periods, where winter is December, January, and February, spring is March, April, May, etc.

-40

-30

-20

-10

0

10

20

winter spring summer fall

Mea

n Se

ason

al Te

mpe

ratu

re (°

C)

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

Describing the clusters:precipitation

0

500

1000

1500

2000

2500

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

117 174 198 206 243 274 281355 284

561390 420

586

857

474 545

2249

443Tota

l Ann

ual P

recip

itatio

n(ra

inw

ater

equ

ivale

nt, m

m)

Precipitation by cluster. Mean annual precipitation varies widely across the clustering area, with Cluster 17 standing out as the wettest.

Cluster 17

Describing the clusters:growing degree days, season length, and snowfall

0

500

1000

1500

2000

2500

3000

50

70

90

110

130

150

170

190

210

230

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Gro

win

g de

gree

day

s

Day

s ab

ove

free

zing

cluster

Days above freezing

Growing Degree Days

0

200

400

600

800

1000

1200

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Tota

l pre

cipi

tatio

n, m

m (

rain

wat

er e

quiv

alen

t)

Clusters

total for months with mean temperature below freezing

total for months with mean temperature 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.

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).

Describing the clusters: existing land classification

http://land cover.usgs.gov/nalcms.php

North American Land Change Monitoring System (NALCMS 2005)

AVHRR Land cover, 1995

Created 2/4/11 3:00 PM by Conservation Biology Institute

GlobCover 2009

Alaska Biomes and Canadian Ecoregions.

Cluster Number AVHRR

Canadian and Alaskan

Ecoregions GlobCover NALCMS

1 Open shrub Northern Arctic Sparse (<15%) vegetation barren lands

2 Open shrub Southern Arctic Sparse (<15%) vegetationpolar or subpolar grassland

lichen moss

3 Open shrub Alaska Arctic Sparse (<15%) vegetationpolar or subpolar grassland

lichen moss

4 Closed Shrubland Alaska Arctic Sparse (<15%) vegetationpolar or subpolar grassland

lichen moss

5 Open shrub Southern Arctic Sparse (<15%) vegetationpolar or subpolar grassland

lichen moss

6 Closed Shrubland Taiga Shield Sparse (<15%) vegetationpolar or subpolar grassland

lichen moss

7 Woodland Taiga Plain Sparse (<15%) vegetationsubpolar taiga needleleaf

forest

8 Wooded Grassland Boreal CordilleraOpen (15-40%) needleleaved deciduous or

evergreen forest (>5m)temperate or subpolar

needleleaf forest

9 Woodland Alaska BorealOpen (15-40%) needleleaved deciduous or

evergreen forest (>5m)temperate or subpolar

shrubland

10 Grassland Western Tundra Sparse (<15%) vegetationtemperate or subpolar

shrubland

11 Woodland Taiga Shield Sparse (<15%) vegetationsubpolar taiga needleleaf

forest

12 Woodland Taiga PlainOpen (15-40%) needleleaved deciduous or

evergreen forest (>5m)temperate or subpolar

needleleaf forest

13 Open shrub Taiga Cordillera Sparse (<15%) vegetation barren lands

14 Evergreen Needleleaf ForestMontane Cordillera

Open (15-40%) needleleaved deciduous or evergreen forest (>5m)

temperate or subpolar needleleaf forest

15 Evergreen Needleleaf Forest Boreal PlainOpen (15-40%) needleleaved deciduous or

evergreen forest (>5m) cropland

16 Evergreen Needleleaf Forest Boreal ShieldOpen (15-40%) needleleaved deciduous or

evergreen forest (>5m)temperate or subpolar

needleleaf forest

17 Bare GroundNorth Pacific

Maritime Sparse (<15%) vegetationtemperate or subpolar

needleleaf forest

18 Grassland Prairie

Closed to open (>15%) herbaceous vegetation (grassland, savannas or

lichens/mosses) cropland

Comparison of cluster-derived cliomes with existing land cover designations. This table shows only the highest-percentage designation for each land cover scheme. Color-coding helps to distinguish categories.

Baseline maps

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.

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 .

Future Projections

2000’s

2030’s

2060’s

2090’s

Projected cliomes for the A2 emissions scenario. This scenario assumes higher concentrations of greenhouse gases, as compared to the A1B scenario.Projected cliomes for the B1 emissions scenario. This scenario assumes lower concentrations of greenhouse gases, as compared to the A1B scenario.

2000’s

2030’s

2060’s

2090’s

Future Projections

Projected cliomes for single models. The five GCMs offer differing projections for 2090.

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.

Future Projections

Discussion: Interpreting results

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 abilityHistorical shifts

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Bare Ground

Cropland

Grassland

Open Shrubland

Closed Shrubland

Wooded Grassland

Woodland

Mixed Forest

Deciduous Broadleaf Forest

Evergreen Needleleaf Forest

Water (and Goode's interrupted space)

Dominant AVHRR land cover types by cluster number. All land cover categories that occur in 15% or more of a given cluster are included.

Discussion: Real-world limitations of modeled resultsChanges are unlikely to happen smoothly and spontaneously,

and are certainly not going to happen instantlySeed dispersal takes timeChanges to underlying soils and permafrost take even longerIn 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 unitsTrophic mismatches are likelyInvasive species may have greater dispersal abilities than

native onesIt may become increasingly difficult to even define what an

“invasive species” is

Discussion: Management implications

Identification of refugiaIdentification of vulnerable

species/areasCollaboration 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/changing-tundra-landscapes/journals/2008-07-22

Brian Bergamaschi (USGS) sampling wells at Bonanza Creek LTER site.

http://hydrosciences.colorado.edu/research/govt_partners.php

Accessing project documents and dataAll project inputs and outputs are available to

the publicThe 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)

AcknowledgmentsThe 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 , SNAPBrian Sieben, Government NWTStuart Slattery, Ducks Unlimited CanadaJim Sparling, Government NWT