Predicting Future Potential Climate-Biomes for the Yukon, Northwest Territories, and Alaska

Predicting Future Potential Climate-Biomes for the Yukon, Northwest Territories, and Alaska

1Goals1)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 analysis2)Link the resulting cliomes to land cover classes, and define each biome by both climate and ecosystem characteristics.3)Couple these cliomes with SNAPs 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

BackgroundFollow-up to Connecting Alaska Landscapes into the Future ProjectBroader spatial scopeMore input dataClustering methodology

33Improvements 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 similarity 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

4Sampling 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.

5Methods: SNAP climate modelsSNAP 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.

6Calculated concurrence of 15 models with data for 1958-2000 for surface air temperature, air pressure at sea level, and precipitationUsed root-mean-square error(RMSE) evaluation to select the 5 models that performed best for Alaska and northwestern CanadaFocused on A1B, B1, and A2 emissions scenariosDownscaled course resolution GCM data to 2kmSNAP 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 7777A primary initial science activity has been focused on developing baseline climate scenarios for Alaska and the Arctic. This work has involved:

An assessment of all the global climate models utilized in the most recent 4th assessment report of the IPCC (Intergovernmental Panel on Climate Change). The assessment process allowed us to identify the 5 models that work best in Alaska, the Arctic, and the Northern Hemisphere. We have focused on the 3 primary emission scenarios (B1, A1B, and A2) for each of the 5 best models. And we have downscaled the model output to management and policy relevant scales.

The details of all these activities can be found on our website.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.

8Methods: 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.

9Methods: 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 minimalEach 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 procedure10Resolution limitationsFor Alaska, Yukon, and BC, SNAP uses 1961-1990 climatologies from PRISM, at 2 kmFor 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 downscaling11How many clusters?Choice is mathematically somewhat arbitrary, since all splits are validSome groupings likely to more closely match existing land cover classificationsHow 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

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?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: temperatureMean 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.

16

Describing the clusters:precipitation

Precipitation by cluster. Mean annual precipitation varies widely across the clustering area, with Cluster 17 standing out as the wettest.Cluster 1717Describing the clusters:growing degree days, season length, and snowfall

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 0C 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). 18Describing the clusters: existing land classification

http://land cover.usgs.gov/nalcms.phpNorth 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. 19

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.20Baseline 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.

21

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 ProjectionsOriginal 18 clusters22

2000s2030s2060s2090sProjected 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.

2000s2030s2060s2090s

Future Projections23

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

Future Projections

Original 18 clusters24

Projected change and resilience under three emission scenarios. These maps depict the total number of times models predict a shift in cliome between the 2000s and the 2030s, the 2030s and the 2060s, and the 2060s and the 2090s. Note that number of shifts does not necessarily predict the overall magnitude of the projected change.Future Projections25Discussion: Interpreting resultsComparison with existing land cover designationsAssessment of which shifts are most significant in terms of vegetation communitiesLinkages with species-specific research Habitat characteristics/requirementsDispersal abilityHistorical shifts

Dominant AVHRR land cover types by cluster number. All land cover categories that occur in 15% or more of a given cluster are included.26Discussion: 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 conditionsEven 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

27Discussion: Management implicationsIdentification of refugiaIdentification of vulnerable species/areasCollaboration and dialogue between modelers and field researchersSelecting focus of future researchShift 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

28Accessing 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=8Maps and data are also available in GIS formats; contact SNAP for further information (nlfresco@alaska.edu)29AcknowledgmentsThe 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 YTDavid Douglas, US Geological SurveyEvelyn Gah, Government NWTLois Grabke, Ducks Unlimited CanadaTroy Hegel, Government YTJames Kenyon, Ducks Unlimited CanadaWendy Loya , the Wilderness SocietyLorien Nesbitt , Dline Renewable Resources CouncilThomas Paragi, Alaska Department of Fish and GameMichael Palmer, TNCScott Rupp , SNAPBrian Sieben, Government NWTStuart Slattery, Ducks Unlimited CanadaJim Sparling, Government NWT

30