Predictive modeling of vegetation distributions

Predictive modeling of Predictive modeling of vegetation distributionsvegetation distributions

Symposium on Bioinformatics: Temporal and Spatial Symposium on Bioinformatics: Temporal and Spatial Syntheses of Vegetation DataSyntheses of Vegetation Data

International Association of Vegetation ScienceInternational Association of Vegetation Science4949thth Annual Meeting, Palmerston North, New Zealand Annual Meeting, Palmerston North, New Zealand

12-16 Feb 200712-16 Feb 2007

Janet FranklinJanet FranklinVegetation Science & Landscape Ecology LaboratoryVegetation Science & Landscape Ecology Laboratory

Department of BiologyDepartment of BiologySan Diego State UniversitySan Diego State University

AcknowledgementsAcknowledgements

US National Science Foundation (0452389) US National Science Foundation (0452389) Geography & Regional Science ProgramGeography & Regional Science Program

Jennifer Miller, West Virginia UniversityJennifer Miller, West Virginia University Robert Taylor, US National Park Service, VTM data Robert Taylor, US National Park Service, VTM data

championchampion Tom Edwards, Mike Austin, Kim van Neil and many Tom Edwards, Mike Austin, Kim van Neil and many

others…others…

OutlineOutline IntroductionIntroduction

– What is Species Distribution Modeling (SDM)?What is Species Distribution Modeling (SDM)?– What is special about vegetation data?What is special about vegetation data?– Framework for SDMFramework for SDM

The Data Model and Vegetation DataThe Data Model and Vegetation Data1)1) Sample designSample design

2)2) Response variableResponse variable

3)3) Explanatory environmental variablesExplanatory environmental variables

4)4) ScaleScale

What are species distribution modelsWhat are species distribution models??

Quantitative models of species-Quantitative models of species-environment relationships…environment relationships…

……used to predict the occurrence of used to predict the occurrence of a species for locations where a species for locations where survey data are lacking (interpolate survey data are lacking (interpolate biological data in space)biological data in space)

– Species abundance or presenceSpecies abundance or presence– Habitat suitabilityHabitat suitability– Realized nicheRealized niche

What do you need?What do you need?

datadata on species occurrence in on species occurrence in geographical space geographical space

mapsmaps of environmental variables of environmental variables A A modelmodel linking habitat linking habitat

requirements to environmental requirements to environmental variables variables

A way to produce a map of predicted A way to produce a map of predicted species occurrence -- species occurrence -- GISGIS

Data to Data to validatevalidate the predictions the predictions

The DataThe Data

Elevation, Quercus pacifica Presence (n=131), Absence (n=797)

Potential Solar Radiation (winter solstice)

Channelislandsrestoration.com

Probability of Species PresenceProbability of Species Presence

WhyWhy make spatial predictions of make spatial predictions of species distributionsspecies distributions??

Conservation planning Conservation planning – Reserve designReserve design– Impact assessmentImpact assessment– Land and resource managementLand and resource management

Climate changeClimate change Invasive speciesInvasive species Ecological restorationEcological restoration Population viability analysisPopulation viability analysis Modeling community dynamicsModeling community dynamics

What is Special About Vegetation What is Special About Vegetation Databases and Databanks?Databases and Databanks?

++ Lots of itLots of it++ Multiple species (community)Multiple species (community)++ Presence Presence andand absence, abundance absence, abundance++ Plants not (usually) (very) cryptic or Plants not (usually) (very) cryptic or

mobilemobile

-- May come from multiple surveys May come from multiple surveys-- Time periods may varyTime periods may vary-- Protocols may varyProtocols may vary

- - May lack locational precisionMay lack locational precision

Wieslander California Vegetation Type Wieslander California Vegetation Type Mapping Survey -1930sMapping Survey -1930s

18,000 plots state-wide18,000 plots state-wide1481 Southern California shrubland plots1481 Southern California shrubland plots

400-m400-m22, 233 species, 233 species (http://vtm.berkeley.edu/)(http://vtm.berkeley.edu/)

San DiegoSan Diego

Los AngelesLos Angeles

Framework for Modeling Species Framework for Modeling Species DistributionsDistributions

““Any mechanistic process model of ecosystem Any mechanistic process model of ecosystem dynamics should be consistent with a static, dynamics should be consistent with a static, quantitative and rigorous description of the same quantitative and rigorous description of the same ecosystem” (Austin 2002, p. 112)ecosystem” (Austin 2002, p. 112)

EcologicalEcologicalModelModel

DataDataModelModel

EmpiricalEmpiricalModelModel

The Data ModelThe Data Model

““Theory and decisions about how Theory and decisions about how the data are sampled and the data are sampled and measured”measured”

1.1. Sampling in space and timeSampling in space and time

2.2. Response variableResponse variable

3.3. Predictor variablesPredictor variables

4.4. Spatial scaleSpatial scale ResolutionResolution ExtentExtent

Sampling in Vegetation SurveysSampling in Vegetation Surveys

-- Not always Not always probability-basedprobability-based

But…But…

++dense data can dense data can be sampledbe sampled

++can supplement can supplement with random with random samplesample

Yucca brevifolia Alliance Pr/Abs

Response Variable in Vegetation Response Variable in Vegetation SurveysSurveys

Presence or abundance of all plant Presence or abundance of all plant species makes it possible tospecies makes it possible to– Model speciesModel species– Model communitiesModel communities

Predict (species) first, then classifyPredict (species) first, then classifyClassify or ordinate (community) first, then Classify or ordinate (community) first, then

predictpredict(review of modeling communities by Ferrier and Guisan (review of modeling communities by Ferrier and Guisan 2006 2006 J. Appl EcolJ. Appl Ecol 43:393-404) 43:393-404)

Date from John T. Curtis. Figure from Gurevitch et al. The Ecology of Plants

SDM is direct gradient analysisSDM is direct gradient analysis

Resource utilization function

Fundamental vs. realized niche

Model species first, then classify Model species first, then classify communitycommunity

Vegetation continuum, composition varies Vegetation continuum, composition varies continuously, individual species responses to continuously, individual species responses to

gradientsgradients (Austin 1998 AMOB 85:2)(Austin 1998 AMOB 85:2)

Ferrier et al. 2002, Biodiv. & Conserv Ferrier et al. 2002, Biodiv. & Conserv 11:230911:2309

Classify first, then modelClassify first, then model ““Predictive Vegetation Modelling” Predictive Vegetation Modelling”

(Franklin 1995 Progr Phy Geogr)(Franklin 1995 Progr Phy Geogr)

Yucca brevifolia Alliance Pr/Abs

Ordinate and model together (CCA)Ordinate and model together (CCA) Oregon coastal ranges, forest (800 Oregon coastal ranges, forest (800

plots, multiple surveys and agencies)plots, multiple surveys and agencies)(Ohmann and Gregory 2002 Can J For Res)(Ohmann and Gregory 2002 Can J For Res)

Classify or ordinate first, then modelClassify or ordinate first, then model(or classify and model together)(or classify and model together)

Classify first, then model starts with Classify first, then model starts with indirectindirect gradient analysis of gradient analysis of communitiescommunities

Classify/ordinate and model Classify/ordinate and model environment together is environment together is directdirect gradient analysis of communitiesgradient analysis of communities

Summary – Vegetation Surveys Summary – Vegetation Surveys and Databanks…and Databanks…

Are large datasets, often Are large datasets, often geographically comprehensivegeographically comprehensive++ Can overcome some sampling problems Can overcome some sampling problems

++ New modeling methods robust to data New modeling methods robust to data qualityquality

Summary – Vegetation Surveys Summary – Vegetation Surveys and Databanks…and Databanks…

Usually include P/A or abundance of Usually include P/A or abundance of all plant speciesall plant species++ P/A data yield powerful species models P/A data yield powerful species models

? Community composition data may be ? Community composition data may be underutilized in vegetation modellingunderutilized in vegetation modelling

Thank you!Thank you!Questions?Questions?

What do we really want?What do we really want?

Plant Distributions: Primary Environmental RegimesPlant Distributions: Primary Environmental Regimes

Guisan & Guisan & Zimmerman Zimmerman (2000)(2000)

Predictor Variables for Vegetation Predictor Variables for Vegetation ModellingModelling

Solar RadiationSolar Radiation

Slope CurvatureSlope Curvature

Scale in Species Distribution ModelingScale in Species Distribution Modeling

Biogeographical scaleBiogeographical scale– Point observationsPoint observations– Lots of themLots of them– Not from designed surveysNot from designed surveys– Presence only, atlases, collectionsPresence only, atlases, collections– Resolution of analysis 10x10-50x50 kmResolution of analysis 10x10-50x50 km– Many to oneMany to one

Ecological scaleEcological scale– Scale of data collection 10Scale of data collection 1022-10-1033 m m22

– Probability sample designsProbability sample designs– Resolution of analysis 10x10 to Resolution of analysis 10x10 to

1000x1000 m 1000x1000 m – One to oneOne to one

McPherson et al. (2006)McPherson et al. (2006)

http://geochange.er.usgs.gov/sw/impacts/biology/veg_chg_model/http://geochange.er.usgs.gov/sw/impacts/biology/veg_chg_model/

Biogeographical ScaleBiogeographical Scale

Assessment of Potential Future Vegetation Changes Assessment of Potential Future Vegetation Changes in the Southwestern United Statesin the Southwestern United States

Robert S. Thompson, Katherine H. Anderson,, Patrick J. BartleinRobert S. Thompson, Katherine H. Anderson,, Patrick J. Bartlein

Scale in Species Distribution ModelingScale in Species Distribution Modeling

Biogeographical scaleBiogeographical scale– Point observationsPoint observations– Lots of themLots of them– Not from designed surveysNot from designed surveys– Presence only, atlases, collectionsPresence only, atlases, collections– Resolution of analysis 10x10-50x50 kmResolution of analysis 10x10-50x50 km– Many to oneMany to one

Ecological scaleEcological scale– Scale of data collection 10Scale of data collection 1022-10-1033 m m22

– Probability sample designsProbability sample designs– Resolution of analysis 10x10 to Resolution of analysis 10x10 to

1000x1000 m 1000x1000 m – One to oneOne to one

Channelislandsrestoration.com

Ecological ScaleEcological Scale

Species Modeling Studies (23) Circle size - number of species

0.0001

0.001

0.01

0.1

1

10

100

1000

10000

100 1000 10000 100000 1000000 10000000 100000000

Extent (km2)

Re

solu

tion

(km

2)

Biogeographical Biogeographical scalescale

Ecological scaleEcological scale

Summary – Vegetation Surveys Summary – Vegetation Surveys and Databanks…and Databanks…

Plant distributions primarily Plant distributions primarily controlled by light, heat sum, water controlled by light, heat sum, water and nutrientsand nutrients++ Tools and data exist for mapping Tools and data exist for mapping

environmental gradients related to environmental gradients related to these primary regimesthese primary regimes

Summary – Vegetation Surveys Summary – Vegetation Surveys and Databanks…and Databanks…

Modeling and spatial prediction at Modeling and spatial prediction at biogeographical or ecological spatial biogeographical or ecological spatial scalescale++ Coarse-scale modeling can overcome Coarse-scale modeling can overcome

locational errors in historical surveyslocational errors in historical surveys

-- But limited to coarse-scale predictors But limited to coarse-scale predictors (climate, not terrain)(climate, not terrain)

Conceptual model of geographical dataConceptual model of geographical data(Goodchild 1994)(Goodchild 1994)

FieldField: geographical space is a : geographical space is a multivariate vector field where multivariate vector field where variables can be defined and variables can be defined and measured at any locationmeasured at any location– ElevationElevation– Vegetation typeVegetation type

EntityEntity: empty geographical space : empty geographical space contains objectscontains objects– TreeTree– Species occurrenceSpecies occurrence– Fire perimeterFire perimeter

The Species Data ModelThe Species Data Model

In species distribution modeling we In species distribution modeling we start with entities…start with entities…– observations of species occurrenceobservations of species occurrence

and end with fieldsand end with fields– Maps of probability of occurrenceMaps of probability of occurrence

What do we really want?What do we really want?

San Diego County is 11,721 kmSan Diego County is 11,721 km22

San Diego Bird Atlas:San Diego Bird Atlas:http://www.sdnhm.org/research/birdatlas/yellowwarbler.htmlhttp://www.sdnhm.org/research/birdatlas/yellowwarbler.html