Depicting uncertainty in wildlife habitat suitability models using Bayesian inference and expert opinion Southwest Regional GAP Project Arizona, Colorado, Nevada, New Mexico, Utah IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Eco Lee O’Brien Natural Resource Ecology Laboratory Colorado Sate University, Fort Collins, CO
34
Embed
Depicting uncertainty in wildlife habitat suitability models using Bayesian inference and expert opinion Southwest Regional GAP Project Arizona, Colorado,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Depicting uncertainty in wildlife habitat suitability models using Bayesian
inference and expert opinion
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, UtahUS-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Lee O’BrienNatural Resource Ecology Laboratory
Colorado Sate University, Fort Collins, CO
Acknowledgments
This project was funded by the USGS,
National Gap Analysis Program
I would also like to thank…David Theobald, Natural Resource Ecology Laboratory
Ken Burnham, Fishery and Wildlife Department at Colorado State University
Fritz Agterberg, Geological Survey of Canada
Donald Schrupp, Colorado Division of Wildlife
…and the species experts who agreed to be “guinea pigs” for the project: Brad Lambert, Lauren Livo, Erin
Muths, Rick Scherer, Tanya Shenk and Michael Wunder.
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Project Goals
Develop alternative for “absolute” predictions of habitat suitability
Quantify expert reviews of wildlife habitat suitability models
Compile and depict the cumulative uncertainty in wildlife habitat suitability models
Easily update models as new data become available
Honestly relate the “state of knowledge” about predicted habitat distributions to natural resource planners and managers
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Wildlife Habitat Suitability Models
Expert models based upon Wildlife Habitat Relationships (WHR)
Usually binary, without indication of strengths or certainty of relationships
Examples from Colorado Gap Analysis Project (Schrupp et al. 2000)
GIS Layers - Land cover - Elevation - Range limits- Distance to water - Soils
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Why Bayesian Inference ?
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
P(S|E) = P(S) * P(E|S)
P(E) Where: P(S) = probability of habitat being suitable (prior probability) P(E) = habitat element probabilities (for suitable and unsuitable habitat)P(E|S) = probability of habitat elements given suitable habitat
(averaged across elements and experts)P(S|E) = probability of habitat being suitable given habitat elements
(posterior probability)
The revision of orderly opinion in light of relevant new information
Allows the combination of empirical and knowledge-based data
Method is transparent and straightforward; species experts, and natural resource planners and managers can fully understand and interpret
In Bayesian framework probabilities are measures of uncertaintyBayes’ Theorem
Methods
Use best available data (literature and expert) to build habitat suitability models
Have species experts review model parameters and provide opinions on the certainty of the habitat relationships
Re-code raster GIS data layers to create probability surfaces
Combine habitat probability surfaces by averaging expert probabilities for each corresponding pixel
Use Bayes’ Theorem to combine the expert probabilities with the prior model to create a posterior probability surface, which depicts the uncertainty in the predicted distribution of suitable habitat
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Mountain Plover Example
Example of method incorporating expert opinion into the Colorado Gap Analysis habitat suitability model for the mountain plover (Charadrius montanus)
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Wildlife Habitat Suitability Model
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Prior Probability Surface
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Model Review
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Tools used to review habitat relationships and ranges, and collect expert opinion
Developed in ESRI ArcView and MS Excel
Range Review Tool
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Elicitation by Species Experts
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
You are asked to review the range maps and add your opinion about the range of the species, by selecting hydro-units and providing a value for how certain you are that the species habitat can be found in the selected hydro-units. The value entered should be between 0 and 1 inclusive, with 0 meaning that you are absolutely certain species habitat does not occur in the hydro-unit and 1 meaning that you are absolutely certain that the species habitat does occur in the hydro-unit. A value of 0.5 would indicate that you are not certain whether the species habitat occurs in the hydro-unit or not. The value should reflect both your knowledge about the particular species and how certain you are that suitable habitat occurs in a particular hydro-unit.
1) 0.5 is “non-informative” probability value = “I don’t know”
2) modeling distribution of suitable habitat; not species occurrence
3) two types of uncertainty: habitat relationships & knowledge about species
Range Probability Surface
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Habitat Relationship Review Tool
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Colorado GAP Land Cover Map
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Land Cover Probability Surface
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Digital Elevation Model
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Elevation Probability Surface
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Bayes’ Inference Calculation
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Prior Probability Surface
WHR Probability Surfaces
Range
Elevation
Land cover (x2)
Posterior Probability Surface
P(S)
P(E|S)
P(S|E)
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Probability of Habitat Suitability for Mountain
Plover
Model Comparisons
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Prior “Absolute” Model
Posterior Probability Model
Land Cover Classification Accuracy
Acknowledged spatial and classification inaccuracies in land cover map
Identify per cover class via some accuracy assessment procedure
Accuracy assessment for Colorado land cover map (Reiners et al. 2000) included a fuzzy assessment of classification accuracy (i.e., degrees of “rightness” and “wrongness” - Gopal and Woodcock 1994)
“RIGHT” fuzzy assessment converts nicely into probabilities (certainty)
Multiply habitat suitability probability map and land cover certainty map
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
There was not enough data to assess accuracy of some land cover classes, these were assigned an “un-informative” probability of 0.5
There was an unknown level of uncertainty added by using air-videography interpretation as “truth” to assess classification accuracy
Need robust accuracy assessment to produce reliable certainty map
Caveats
Land Cover Classification Accuracy Surface
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Uncertainty in Mountain Plover Wildlife Habitat Suitability Model with Additional Uncertainty from Land Cover
Classification
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Uncertainty Comparisons
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Model Uncertainty
Model Uncertainty with Land Cover Classification Uncertainty
Distance to Water Coverage
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Distance to Water Habitat Relationship as Probability
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Probability of Habitat Suitability for Boreal Toad
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Probability of Habitat Suitability for Boreal Toad Combining Several Expert
Reviews
Patch Size as Probability for Lynx Model
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Lynx Model Comparisons
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Lynx Model
Lynx Model with Patch Size Probability
Findings
The expert reviewers who I contacted agreed with the utility of the project, were willing to participate and quickly learned the procedures for quantifying their certainty of the habitat relationships
It took an average of 1 hour per species for range and model reviews
The reviews were done in workshops or the tools were given to experts to do reviews on their own (need ESRI ArcView and MS Excel); each method had advantages and disadvantages
Needed robust accuracy assessment of land cover classes to assign reliable uncertainty contributed by this layer
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
Conclusions
Depicts accumulated uncertainty in habitat suitability models
Provides a way to incorporate knowledge from many species experts
Provides a way to incorporate uncertainty of land cover classification
Provides a way to incorporate new modeling elements and reveal the additional associated uncertainty
Provides an easy way to update models with new information
Relates “state of knowledge” about predicted suitable habitat distribution
Southwest Regional GAP Project
Arizona, Colorado, Nevada, New Mexico, Utah
US-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology
This procedure…
Does not…
Address uncertainty from scale inconsistencies or cartographic errors
Predict species occurrence
Usefulness for Gap Analysis
Provides a way to incorporate species expert knowledge into models
“Honest” depiction of uncertainty in predicted habitat distributions
Time and effort involved per review is reasonable
The resulting continuous surface probability map would have to be divided into categories to be used in gap analysis (e.g., areas with probabilities over 0.75 could be considered “suitable” habitat and used in the analysis of ‘gaps’ in networks of conservation lands)
The habitat suitability surfaces can be used in other “what if” planning scenarios and used to direct future habitat analysis
Verifying models vs. showing current “state of knowledge” ?Southwest Regional GAP
ProjectArizona, Colorado, Nevada, New Mexico,
UtahUS-IALE 2004, Las Vegas, Nevada: Transdisciplinary Challenges in Landscape Ecology