View
213
Download
0
Tags:
Embed Size (px)
Citation preview
R. Scott McNay, Wildlife InfometricsR. Scott McNay, Wildlife InfometricsRandy Sulyma, BC Min. of ForestsRandy Sulyma, BC Min. of Forests
Building and Validating Building and Validating Bayesian ModelsBayesian Models
Identification of Mountain Identification of Mountain Goat Winter Range in North-Goat Winter Range in North-
central BCcentral BC
AcknowledgementsAcknowledgements
Funding from the BC Min. of EnvironmentFunding from the BC Min. of Environment Other participants included:Other participants included:
R. EllisR. Ellis D. Fillier, S. Gordon, L. Vanderstar, D. Heard, G. D. Fillier, S. Gordon, L. Vanderstar, D. Heard, G.
Watts, D. Wilson, J. Vinnedge, B. Brade, R. Watts, D. Wilson, J. Vinnedge, B. Brade, R. MacDonaldMacDonald
Line Giguere, Robin McKinleyLine Giguere, Robin McKinley Concepts and ideas:Concepts and ideas:
the last workshop in Chasethe last workshop in Chase subsequent discussions, most notably, B. subsequent discussions, most notably, B.
Marcot & S. Wilson, C. Apps Marcot & S. Wilson, C. Apps
Modeling ContextModeling Context
from Bunnell 1989from Bunnell 1989
Uses:Uses: PredictionPrediction
ManagementManagement Implications Implications
of of Predictions?Predictions?
ExplanationExplanation ResearchResearch Why?Why?
EmpiricalMechanistic
TheoryMechanistic
EmpiricalCorrelative
TheoryCorrelative
FrequencyProbability
Conceptual
RationaleRationale
General, portable modelGeneral, portable model Management & researchManagement & research Prediction & explanationPrediction & explanation Minimal resources to developMinimal resources to develop
Little information from FSJLittle information from FSJ Insufficient resources to develop Insufficient resources to develop
empirical or other more traditional empirical or other more traditional approachesapproaches
Limited Time FrameLimited Time Frame
Study AreasStudy Areas
Adjacent MUsAdjacent MUs Similar but not Similar but not
the samethe same Preliminary Preliminary
model already model already builtbuilt
Simple UWR ModelSimple UWR Model
= Spatial relationships of cells processed/defined in a GIS.
Typically a distance function from escape terrain.
S1: Escape Terrain
S2: UWR
I1: Slope I2: Aspect/Solar I3: Elevation
AR: I: Alpine and Rocks (FC)
RGH: Roughness from DEM LCC: S: Landcover Condition
BA: I: Bare Areas (BTM)
ET: S: Escape Terrain
SC: S: Spatial Configuration
PET: I: % of ET within Effective Forage
ETE: S: Escape Terrain Effectiveness
SL: AI: Solar Loading (Whr/m2)
EA: I: Amount (ha) of ET - 2.0 ha Neighbourhood
AEFT: I: Amount of Effective Forage Terrain (ha)
FWDB: AI: Forage-weighted Distance Buffer
A: S: Smooth / Good slope
EP: S: Winter Range Preference
SLP: I: Slope (Degrees) from DEM
FTE: S: Forage Terrain Effectiveness
= Spatial relationships of cells processed/defined in a GIS
GoatsGoats
AreaArea GoatsGoats RelocationsRelocations
OspikaOspika 2323 232232
OsilinkaOsilinka 1818 184184
Akie Akie (Pesika)(Pesika)
1414 3939
Model Construction ResultsModel Construction Results
Primary indicators:Primary indicators: AccuracyAccuracy
100% relocations100% relocations Maximize coverageMaximize coverage
PrecisionPrecision 100% alpine100% alpine Maximize area decreaseMaximize area decrease
OspikaOspika OsilinkOsilinkaa
PesikaPesika
LocLoc AtAt LocLoc AtAt LocLoc AtAt
11 2929 8282
5858 7777
22 2222 9393
6666 6767
33 6969 5656 3333 4848
Overall Results:Overall Results:
Basic model of EPBasic model of EP 65% relocations 65% relocations
coveredcovered 57% reduction in alpine 57% reduction in alpine
Spatially generalized Spatially generalized model (nearest model (nearest neighbor algorithm)neighbor algorithm)
92% relocations 92% relocations coveredcovered
70% reduction in alpine70% reduction in alpine
Model TestingModel Testing Random sample approach applied.Random sample approach applied. Aerial reconnaissance completed.Aerial reconnaissance completed. Data collected to evaluate/verify Data collected to evaluate/verify
both input parameters, and both input parameters, and summary results.summary results.
Given funding and timing Given funding and timing constraints, not possible to constraints, not possible to evaluate some of the spatial evaluate some of the spatial relationships.relationships.
AR: I: Alpine and Rocks (FC)
RGH: Roughness from DEM LCC: S: Landcover Condition
BA: I: Bare Areas (BTM)
ET: S: Escape Terrain
SC: S: Spatial Configuration
PET: I: % of ET within Effective Forage
ETE: S: Escape Terrain Effectiveness
SL: AI: Solar Loading (Whr/m2)
EA: I: Amount (ha) of ET - 2.0 ha Neighbourhood
AEFT: I: Amount of Effective Forage Terrain (ha)
FWDB: AI: Forage-weighted Distance Buffer
A: S: Smooth / Good slope
EP: S: Winter Range Preference
SLP: I: Slope (Degrees) from DEM
FTE: S: Forage Terrain Effectiveness
Model Testing ResultsModel Testing ResultsObserved
Preferred Equivocal Avoided Total
Preferred 31 13 6 50
Equivocal 0 0 0 0
Avoided 0 0 7 7
Total 31 13 13 57
Correct Classification Rate = 67%False Positive Error Rate = 73%False Negative Error Rate = 0%Κ = 0.50τ = 0.32
Modele
d
Model Testing ResultsModel Testing Results
Correct Classification Rate = 89%False Positive Error Rate = 46%False Negative Error Rate = 0%Κ = 0.64τ = 0.79
UWRUWR OtherOther TotalTotal
UWRUWR 4444 66 5050
OtherOther 00 77 77
TotalTotal 4444 1313 5757
Observed
Modele
d
DiscussionDiscussion
Were we able to restrict our search Were we able to restrict our search for UWR sufficiently yet remain for UWR sufficiently yet remain accurate?accurate?
Was it important that we were not Was it important that we were not strictly analytical in our Bayesian strictly analytical in our Bayesian learning?learning?
Was our test protocol appropriate Was our test protocol appropriate given the project goal?given the project goal?