Warming climate and changes in Alaska’s temperate rainforest: combining predictive modeling with monitoring Tara M. Barrett 1,Greg Latta 2, Paul E. Hennon.

Warming climate and changes in Alaska’s temperate rainforest: combining predictive

modeling with monitoring

Tara M. Barrett1 ,Greg Latta2, Paul E. Hennon3, Bianca N.I. Eskelson2, Hailemariam Temesgen2

1 Unaffiliated2 Oregon State University3 Pacific Northwest Research Station

Distribution of western hemlock trees (Tsuga heterophylla)

Western hemlock, along with other temperate rainforest tree species (Sitka spruce, western red cedar, yellow cedar) reaches its northern limit in the Gulf of Alaska region.

Hemlock dwarf mistletoe (Arceuthobium tsugense subsp. tsugense) is a small plant that parasitizes western hemlock (Tsuga heterophylla) trees.

Brooms (branch deformation)

Bole deformation

Chlorosis

Both western hemlock and western hemlock dwarf mistletoe distributions are at the northern part of their geographic range in

Alaska

Map by Dustin Wittwer

Population size of host (Tsuga heterophylla) and parasite (Arceuthobium tsugense subsp. tsugense) by elevation

0-30

31-60

61-91

92-121

122-

152

153-

182

183-

213

214-

243

244-

274

275-

304

305-

335

336-

365

366-

396

397-

426

427-

457

458-

487

488-

518

518-

548

549-

579

580-

609

610-

640

641-

670

671-

701

702-

731

732-

762

762+

0

10000000

20000000

30000000

40000000

50000000

60000000

70000000

-

5

10

15

20

25

Trees with mistletoe Western hemlock trees Percentage of mistletoe infection

Elevation (meters)

Mis

tle

toe

infe

cti

on

(p

erc

en

t)

1950 1960 1970 1980 1990 2000 2010 (8)

(6)

(4)

(2)

-

2

4

6

8

10

12 Average annual temperature 1954-2008

AnnetteJuneauKodiakYakutatHomerAnchorageTalkeetnaGulkana

Year

Degr

ees C

elsiu

s

Projections are for further increases in temperature over the next half century

Map from SNAP modeling group, Univ. of Alaska Fairbanks (www.snap.uaf.edu)

Climate envelope models

Distribution is limited by a climate envelope outside of which a species cannot survive

For prediction, assumes distribution across climate-space remains the same, but spatial distribution shifts as climate changes

empirical – not process based habitat only, does not predict actual presence many different approaches:

GAM, neural nets, customized models logistic modeling Most Similar Neighbors imputation Random Forest imputation

Prediction modeling for hemlock dwarf mistletoe and western hemlock

Methods

1. Reviewed literature for climate-related mechanisms that might explain mistletoe distributions and abundance.

2. Created climate variables that corresponded to possible explanatory mechanisms.

3. Tested predictive models against field data.4. Predicted future distribution

Western dwarf mistletoe seed and holdfast

Reviewed previous research:

• Extreme minimum winter temperatures reduce seed viability.• Snow reduces seed establishment and germination.• Rain reduces seed establishment and germination?• Spring frosts damage pollen viability• Fall frosts damage fruit

• Life cycle takes longer to complete in Alaska (e.g., 12 yrs) than British Columbia (e.g. 5 yrs)

• Transplanted to 120 m higher elevation, fruits never matured

Predictive Variables Used

Growing season variables:GDD (growing degree days above 0 C)RADIANS (modeled solar radiation)

Low winter temperaturesMINTEMP (min of mean min monthly temp)MINTEMPSD (standard deviation of mean min temps)

PrecipitationSNOW (modeled precipitation as snow)RAIN (mean annual precipitation - SNOW)

Autumn freezesMINFALLTEMP (mean min Sept. temp)

Spring freezesMINSPRINGTEMP (mean min April temp)

OtherSLOPEET (modeled evapotranspiration)CMI (modeled precip – evapotranspiration)

8

-12Mean annual temperature (C)

Used 1961-1990 PRISM climate data from Oregon State Univ.: a spatial model of climate normals (monthly temperature and precipitation) to develop climate variables

PRISM data was 2 km resolution, so rescaled using 2 dimensional linear interpolation for precipitation and geographically weighted regression for temperature

Results – hemlock dwarf mistletoeSensitivity (correct prediction of presence)

MSN Random Forests Logistic

Development data set 29 37 38

Validation data set 24 20 37

Specificity (correct prediction of absence)

MSN Random Forests Logistic

Development data set 92 98 93

Validation data set 90 96 93

Predicted proportions

MSN Random Forests ActualDevelopment data set 10.5 5.6 10.0

Validation data set 11.0 5.5 8.9

All three methods (Random Forests, Most Similar Neighbors, and Logistic modeling) did fairly well at predicting the current range of hemlock and mistletoe

Methods

1. Reviewed literature for climate-related mechanisms that might explain mistletoe distributions and abundance.

2. Created climate variables that corresponded to possible explanatory mechanisms.

3. Tested predictive models against field data.4. Predict future distribution

We used downscaled GCM composites created by the Scenarios Network for Alaska Planning (SNAP 2011). The composite models were made from

the MPI ECHAM5, the GFDL CM2.1, the 324 MIROC 3.2 (medres), the UKMO HADCM3, and the CCCma CGCM3.1 models

which had been chosen based on relatively good performance in a review of GCMs for Alaska and Greenland by Walsh et al. (2008).

PRISM + (GCM_future – GCM_present) = Predicted

Scenarios Network for Alaska Planning [SNAP]. 2011. Alaska climate datasets online. Available from www.snap.uaf.edu/downloads/alaska-climate-datasets

Walsh, J.E., Chapman, W.L., Romanovsky, V., Christensen, J.H., and Stendel, M. 2008. Global climate model performance over Alaska and Greenland. J. Clim. 21:6156-6171.

Future Climate Models

Approach A1B A2 B1

MSN 137 141 118

RF 112 113 107

Logistic 134 136 115

Scenario

Prediction for 2100 western hemlock habitat as a percent of present (=100), bias adjusted

Approach A1B A2 B1

MSN 577 596 416

RF 384 449 374

Logistic 724 757 571

Scenario

Prediction for 2100 dwarf mistletoe habitat as a percent of present (=100), bias adjusted

BUT …

Habitat is not presenceTrees migrate slowly

eg, Sitka spruce on Kodiak island, roughly 1 mile per century (per Griggs, 1930s)

Dwarf mistletoe is dioecious; seeds can only travel (beyond a few dozen meters) with the help of birds or mammals

BUT …

Western red cedar Yellow cedar

Weste

rn hemlock

(556)

Sitka

spru

ce (5

45)

Mountai

n hemlock (4

32)

Yellow-ce

dar (266)

Weste

rn re

dcedar

(141)

Shore pine (1

30)-6%

-4%

-2%

0%

2%

4%

6%

Net change in live-tree carbon by species between 1995-2003 and 2004-2008, southeast and southcentral Alaska

Chan

ge a

s a

perc

ent o

f ini

tial (

1995

-200

3)

carb

on m

ass

Numbers in parenthesis indicate number of forested plots with this species present

Forested lands, excludes National Forest wilderness and Glacier Bay NP, only trees >= 5” d.b.h.

Summary

• Plot level imputation is promising for climate/host/parasite mapping as it can be used for (1) range mapping (2) area affected estimates and (3) impact estimates

• For this case study: Both Most Similar Neighbors and Random Forest do poorly at predicting presence/absence of western hemlock dwarf mistletoe at the plot level.

• Most Similar Neighbors does moderately well at predicting abundance and distribution.

• Random Forests does well at predicting distribution, but tends to underestimate abundance

Summary

For future predictions:

Climate Envelope Models may be most useful for• discussing why we think they are wrong• formulating hypotheses to test

Long-term monitoring using stable protocols really needs to accompany predictive modeling

For more information:

Barrett, T.M.; Latta, G.; Hennon, P.E.; Eskelson, B.N.I.; Temesgen, H. 2012. Host-parasite distributions under changing climate: Tsuga heterophylla and Arceuthobium tsugense in Alaska. Canadian Journal of Forest Research 42:4:642-656

Barrett, T.M. 2011. Change in forests between 1995-2003 and 2004-2008. In Forest Service Pacific Northwest Research Station General Technical Report PNW-GTR-835.