The Birds & The Trees: Potential Responses of Eastern

Forests to Climate Change

Susan L. StoutUSDA Forest Service Research & Development, Irvine, PA

With help from Louis Iverson, Anantha Prasad, Steve Matthews, and Matt Peters

The Birds & the Trees

Quick look at some evidence of changing climate in PA with thoughts about adapting to uncertaintyIntroduction to USFS Climate Change Modeling EffortsExamples of outputs from PAThoughts about using these models in a statewide adaptation plan

Climate Change – We’re Already Seeing the Effects

Fire -Fire season are coming earlier and lasting longer. Fires are burning hotter and bigger and more damaging and dangerous to people and property.

Insects -

Both the natives and the invaders—are spreading more rapidly than ever. The winter cold isn’t knocking them back. They are killing more trees and making the fire danger even worse.

Water -

Warmer winters are affecting our water supplies. Snowpacks

are thinner and melt earlier, so water runs off from the forest earlier in summer. Droughty forest soils makes trees more vulnerable to fire and insects.

Presenter

Presentation Notes

http://climvis.ncdc.noaa.gov/cgi-bin/cag3/hr-display3.pl

The Future is Very Uncertain

How to translate this to trees and birds?

Look for environmental data sets that areSpatially explicit & widespreadQuality controlled

Model the relationships between these data and today’s climateUse global circulation models to project future data distributionsForest Inventory & Analysis Tree DataBreeding Bird Survey Bird Data

www.nrs.fs.fed.us/atlas

•• FOREST INVENTORY (US Forest Service)–– 37 states east of 100th meridian•-

134 tree taxa

•-

103,488 plots, ~1 plot per 2400 ha of forest•-

2,938,518 tree records

•• PROCESS– Extract latest FIA plot data by State–

Calculate Importance Value (IV) based on

number of stems & basal area– Aggregate points to 20 x 20 km polygons

• OUTPUT–

Importance Value (IV) for 134 tree species, by

20 km cell

Forest Inventory and Analysis

Available online: Prasad and Iverson 2003

Quercus alba

Environmental Predictor Variables

Soil PropertyBD Soil bulk density (g/cm3) CLAY

Percent clay (< 0.002 mm size) KFFACT Soil erodibility

factor, rock fragments NO10 % soil passing sieve No. 10 (coarse) NO200

%soil passing sieve No. 200 (fine) OM

Organic matter content (% by wt) ORD Potential soil productivityPERM Soil permeability rate (cm/hour) PH Soil pH ROCKDEP Depth to bedrock (cm) ROCKFRAG

% weight of rock fragments 8-25 cm SLOPE Soil slope (%) of a soil component TAWC Total avail water capacity(cm

to 152)

Land Use and FragmentationAGRICULT Cropland (%) FOREST

Forest

land (%) FRAG

Fragmentation Index NONFOREST

Non-forest land (%)

ClimateAVGT Mean annual temperature (deg. C) JANT Mean Jan temperature (deg. C) JULT Mean July temperature (deg. C) TMAYSEPT

Mean May-Sept. temperaturePMAYSEPT Mean May-Sept precipitationPPT Annual precipitation (mm) JANJULDif

Difference temp Jan/Jul

ElevationELV_CV Elevation coefficient of variation ELV_MAX Maximum elevation (m) ELV_MEAN

Average elevation (m)ELV_MIN

Minimum elevation (m) ELV_RANGE

Range of elevation (m)

Soil ClassALFISOL

Alfisol (%) ARIDISOL

Aridisol (%)ENTISOL

Entisol (%)HISTOSOL

Histosol (%)INCEPTSOL Inceptisol (%) MOLLISOL Mollisol (%) SPODOSOL Spodosol (%) ULTISOL Ultisol (%) VERTISOL

Vertisol

(%)

••Response variable: FIAResponse variable: FIA--derived importance values by 20 kmderived importance values by 20 km

TJuly< 16.5

PPT < 750

pH > 6 MElev > 500Alfisol < 15

0.5n=1800

6.5n=300

35n=200

2.4n=85 15

n=15063

n=95•A single (best) predictor is selected to split the data

•Additional best predictors are selected for each subset of data, thus creating ‘branches’

of a ‘tree’

Regression Tree Analysis (RTA)

Iverson and Prasad 1998 Ecological Monographs

•At the bottom are a terminal nodes that contain the predicted value of species importance

•These values are then mapped

TJuly< 16.5

PPT < 750

pH > 6 MElev > 500Alfisol < 15

0.5n=1800

6.5n=300

35n=200

2.4n=85 15

n=15063

n=95

Regression Tree Analysis (RTA)

Highly suited for distributional mapping where different variables operate

at different geographic regions –

can map predictor-rules driving the distribution.

TJuly

< 16.5 & PPT < 750 &PH <= 6

Iverson and Prasad 1998 Ecological Monographs

Importance Valuesfor 134 Tree

Species

(Response Variables)

38 Variables:• Climate• Soil• Elevation• Land-use• Landscape Frag-

mentation(Predictor Variables)

DISTRIBModel

ModelPredictedCurrent

FIACurrent

DISTRIB

DataManipulation & Analysis

Single Decision Tree

Helps understandrelationships, and map drivers

30

Bootstrap sampling for each “tree” – Use subsets of data

Bagging Trees

100

0

Bootstrap sampling + randomized subset of predictors for each tree

Random Forests

use 30 “trees” to assess variability among individual tree models - a measure of model reliability

best for prediction without overfitting

Prasad, A. M., L. R. Iverson, and A. Liaw. 2006. Newer classification and regression tree techniques: bagging and random forests for ecological prediction. Ecosystems 9:181-199.

Model ReliabilityNot all species models are equal –

need to

know about “model confidence”

for each species:

Factors used in model reliability score:

•R2

equivalent of the Random Forest model

•A statistic comparing prediction to actual data•An assessment of predictor stability and consistency using the 30 “bagged”

“trees”

0.1 0.3 0.5 0.7 0.9Model Reliabilty Score

0

1000

2000

3000

4000

5000

Are

a, c

ells

•Worst 10% •Best 10%

•Models are not created equal.•If range is small, model is less reliable as are predictions of extinction

Schwartz, Iverson, Prasad, Matthews, O’Connor 2006 Ecology

ModelPredicted

Future

GCM Climate Variables

Swap

Importance Valuesfor 134 Tree

Species

(Response Variables)

38 Variables:• Climate• Soil• Elevation• Land-use• Landscape

(Predictor Variables)

DISTRIBModel

ModelPredictedCurrent

FIACurrent

DISTRIB

DataManipulation & Analysis

Important!

With these models, we are predicting potential suitable habitat

by year 2100.

We are NOT predicting where the species will be at that time, as great lag times are involved in tree species migrations.

Important!

With these models, we are predicting potential suitable habitat by year 2100. We are NOT predicting where the species will be at that time, as great lag times are involved in tree species migrations.

Use Global Circulation Models and Emissions Scenarios from

IGPCCUse IGPCC Global Circulation Models:

HadCM3GFDLPCMAverage of all three

Use 2 IGPCC Emissions ScenariosA1f1 (High) –

note that current global track is

higher than thisB1 (Low)

Losers for PA Wilds

Winners for PA Wilds

Adaptation Implication: Oak regeneration more important now, potentially easier in the future

Forest Types fromcombinations of species.

What about Birds?

Similar Approach, using Breeding Bird Atlas dataAdditional piece is ability to link birds, where significant, to specific tree spp. or mixes of tree spp.

Predicted Change

Comm.Name Cur_Prd PCM Lo GCM3 Lo GCM3 Hi Hadley Hi

Yellow Warbler 272.9 ‐43.5 ‐99.9 ‐213.7 ‐223.5

Savannah Sparrow 124.9 ‐77.6 ‐99 ‐124.7 ‐124.7

Song Sparrow 291.6 ‐29.3 ‐93.4 ‐221.4 ‐229.7

House Finch 232 ‐39.7 ‐85.5 ‐158 ‐166.7

Baltimore Oriole 264.5 ‐41.1 ‐85 ‐155.1 ‐162.9

American Redstart 172.4 ‐43.7 ‐75.8 ‐104.6 ‐106.9

Ring‐necked Pheasant 102.7 ‐48.5 ‐68.7 ‐77.6 ‐67.2Swamp Sparrow 73.9 ‐49.4 ‐65 ‐70.5 ‐71.1Hermit Thrush 72.4 ‐53.8 ‐63.6 ‐68.2 ‐68.5Black‐throated Green 

Warbler 96.3 ‐41.2 ‐60.1 ‐77.4 ‐77

Birds Likely to Decrease Sharply

Birds Likely to IncreasePredicted Change

Comm.Name Sci.Name Cur_Pr d

PCMlo_ D

GCM3lo _D

GCM3H i_D

HadHi_ D

Northern Bobwhite

Colinus

virginianus 53.7 99.6 166.6 224.6 225

Yellow-

breasted Chat Icteria

virens 89.3 99.1 141.7 157.9 144.4Orchard Oriole Icterus

spurius 62.9 82.4 125 134.3 102.7White-eyed

Vireo Vireo griseus 58.8 81.6 115.4 133.5 130.8Summer

Tanager Piranga

rubra 8.5 78.1 136.5 203.4 204.3Red-bellied

WoodpeckerMelanerpes

carolinus 127 76.8 106.2 117.9 111.4

Carolina WrenThryothorus

ludovicianus 105 73.4 90.1 86.4 91.6

Blue Grosbeak

Guiraca

caerulea 7.5 69.5 130.8 210.4 203

Blue-gray Gnatcatcher

Polioptila

caerulea 112.8 69.3 99.4 105.6 94.9

Yellow-billed Cuckoo

Coccyzus

americanus 141.9 67.3 100.9 128.9 126.4

Weaknesses of DISTRIB1.

Limited in scope to modeling the potential current/future suitable habitats –

not their actual

future distributions. SHIFT begins to address this issue.

2.

Does not account for many biologic attributes and disturbance factors (partially addressed later).

3.

FIA data are spatially sparse so that fine-scale analyses are not usually appropriate –

20 x 20 km is

about right. 4.

Depends on a decent sample size (>~50 cells), so not great for rare species.

5.

Assumes equilibrium with environment.6.

There likely are better predictors that could be used.7.

Not all species have their entire ranges captured with IVs (Canada, West US).

Strengths 1 DISTRIB1.

FIA samples are statistically sound and non-biased2.

Analysis and prediction based more on core of distribution via IVs, not the range edges or just presence/absence maps that are more susceptible to error

3.

Extremely robust non-parametric statistical tools using ensemble “tri-model”

approach4.

The reliability of individual species models can be evaluated5.

RF is stable predicting into novel environments 6.

Can use different variables/parameters to describe primary drivers in different parts of its geographic setting

7.

Accounts for reality in that a particular species exists where it is, in spite of all legacies over decades and centuries. It therefore integrates over historic disturbances and climatic phenomena.

Strengths 2 DISTRIB

8.

Need not be parameterized with a large suite of variables that are imperfectly known or cannot be adequately generalized for a species throughout its range.

9.

Can rank among species for the most vulnerable to change (mean center changes).

10.

Can produce ranked lists of species that may be in greatest risk or likely to have sufficient suitable habitat for future management.

11.

Models mostly agree with trends observed via FIA plots (Woodall et al paper).

.13o

North***.46o

North ***

Compared the distributions of Compared the distributions of ““seedlingsseedlings””

(trees with (trees with dbhdbh

≤≤

2.5 cm) with 2.5 cm) with distributions distributions of of larger trees. For larger trees. For ““northernnorthern””

species, mean latitude of species, mean latitude of seedlings was 20 km north seedlings was 20 km north of of mean latitude of larger trees.mean latitude of larger trees.

% of potential new suitable species habitat occupied in 100 yearLoblolly Persimmon Sweetgum Sourwood S. Red Oak

CCC>2% 8.4 2.7 11.6 12.7 7.6>20% 1.5 0.8 2.2 2.4 2.0>50% 0.6 0.6 1.2 1.0 1.2HAD>2% 9.9 3.8 14.7 8.2 11.5>20% 3.2 1.3 5.1 2.2 4.1>50% 1.6 0.9 3.0 0.9 2.5

Shift Output Summary

Migration Potential into New Suitable Habitat in 100 yrs

So, DISTRIB accounts for edaphic

barriers or facilitators to migration, along with defining the suitable habitat, while SHIFT accounts for dispersal into a fragmented landscape.

Modifying Factors We have model outputs showing tendencies towards gaining or losing under climate change

Many other factors come in to play to determine final outcomes

Can we rate these other factors for relative positive or negative impacts, along with some assessment of uncertainty?

We would like to generate a scoring system to help evaluate modifications

Acer rubrumRed MapleAceraceae -- Maple familyRussell S. Walters and Harry W. Yawney

Fire topkill Red maple is very sensitive to fire injury, and even large trees can be killed by a fire of moderate intensity.

Document sources:

Silvics

Manual

Plants Database

Climate Change Tree Atlas

Forest Service Fire Effects Information System

Arriving at the base broad scale numbers

Modifying Factor ScoresFor selected PA Wilds species

Median = 0, scaled -3 to +3

Species may do better than modeledSpecies may do worse than modeledDepends also on species mix

-3

-2

-1

0

1

2

3

blackcherry

easternhemlock

sweetbirch

blackwillow

green ash black oak sugarmaple

white oak hackberry red maple

DisturbanceBiological

Advice to Managers 11.

Pay attention to the reliability of each species model –

and regardless, there still will be errors!

2.

Less common species are more prone to error.

3.

Edge boundaries are ‘fuzzy’, both now and in future –

core areas of higher IVs are more indicative

4.

Use these models as guidelines for regional trends –

they are not appropriate for stand level management without the regional context

5.

Use modifying factors to help understand model outputs

6.

Concentrate on the factors you can do something about

Advice to Managers 27.

But if you abide by these caveats, and you live in the Eastern US, you can use these to:

a.

Learn which species are in, or could be in, your location now

b.

Learn which environmental factors are likely driving species’

suitable habitat, e.g., which are most susceptible to climate drivers

c.

Learn what species are most and least likely to have their habitats move, and how much

d.

Learn which species could incur the most risk under climate change

e.

Learn which species could become newly suitable for your location (from the south)

f.

With SHIFT, learn where potential colonization could occur within 100 yrs

g.

Identify which factors are most likely to modify model outputs, and which ones you might be able to do something about

More Contributions with Research and Mangers working together to manage

under climate changeDevelop and model “what if”

mitigation

and adaptation strategiesMitigation –

where to best sequester, minimize emissionsAdaptation –

create resistance to change or improve resilience to change (e.g., maintain diversity)

Design landscapes to be more sustainable

Enhance connectivityRestoration of habitatsIs translocation (assisted migration) a possibility?Protection of refugia?

Use adaptive management approaches while considering multiple scenarios

Hoegh-Guldberg, O. et al. 2008. Science 321: 345-346.

Assisted Migration or not?

Thank you for your attention!

Web site for most data presented today:

Little’s

boundariesFIA data grouped by 20x20 km cellClimate change atlases Species-environment data for 134 treesPdfs

of related paperswww.nrs.fs.fed.us/atlas

For free hard copy of atlases or reprints:

liverson@fs.fed.us

Thanks to USDA FS Northern GlobalChange Program for support