PhD Defense

2014-08-14

Marine Conflicts

human uses vs. endangered species

Marine Spatial Planning

Spatial Decision Support System

Data from Many Platforms

Questions (themes)

How to combine data from many platforms to best predict distribution and abundance of species? (disperate data)
How do these distributions change over time, seasonally and trending with climate change? (distributions and time)
What environmental covariates best predict where and when these animals are distributed? (distributions and environment)
How do we effectively capture and integrate uncertainty for these distributions into decision making? (uncertainty)
Once we can best describe the distribution of these species in space and time, how can we integrate this information into spatial decision frameworks? (decision frameworks)
- for siting
- for routing

Chapters: Data to Decision

Robust, Dynamic Distribution Models
- Combine plane and boat observation data (disperate data)
- Distance from Gulf Stream (distributions and environment)
- Predicting with Uncertainty in Measurement and/or Gap-filled Environment (uncertainty)
Predicting Seasonal Migration (distributions and time)
Probabilistic Range Maps (disperate data, decision frameworks, uncertainty)
Decision Mapping (decision frameworks, uncertainty)
Conservation Routing (decision frameworks)

1. Robust, Dynamic Distribution Models

Baseline SDM: Boat + Plane

Dynamic Variables from Satellite

Eddies from AVISO Fronts from Pathfinder / GHRSST

Dynamic Vars from Ocean Models

Hybrid Coordinate Ocean Model (HYCOM)

pros: 1/12 °, cloud-free, 3D, forecast
cons: modeled, since 2003, physical only

Forecast Whales using Ocean Models

using ROMS to Oct, Nov (predicting from July)

Robust Comparison

Presence < Presence-Absence < Density?
Regression (GLM, GAM) vs Machine Learning (Maxent, BRT)
Predictive Performance vs Ecological Interpretation
Space Time Lags
Scaling Effects

Space vs Time

Processes Observations

2. Predicting Seasonal Migration

Grey Whale Migration

longest mammalian migration on planet
Eastern North Pacific gray whales make a mammoth 20,000 km (12,400 mile) round trip between their southern breeding grounds off Baja California, Mexico and their northern feeding grounds off Alaska and the Beaufort Sea.
April - November: Arctic feeding grounds
[October - February: migrates south]
December - April: Mexican breeding grounds
[February - July: migrates north]
In the early winter, they move south to breed in the warm, shallow lagoons along the Mexican coast. The most popular breeding lagoons are San Ignacio lagoon, Scammon's lagoon, and Magdalena Bay, on the Pacific coast of Baja California, Mexico. Around February, the grays migrate north to feed in Arctic waters (western Beaufort Sea and Bering Sea), northwest of Alaska. A few - mainly younger - whales make a shorter journey north from Mexico, stopping off along the coastline stretching between northern California, Oregon, Washington State, USA, and British Columbia, Canada. Some feeding behaviour has been observed in all parts of the range, and around Vancouver Island, British Columbia, Canada, grays are present year-round.

Migration Model

3. Probabilistic Range Maps

Range Map

Eubalaena glacialis

IUCN Range Maps Applied

AquaMaps Environmental Envelope

Relative Environmental Suitability

Ocean Health Index: Species

Probabilistic Range Maps

Combine:

\(Y\): Occurrences for presence-only observation
\(R\): Range map from expert opinion
\(E\): "Effort"" proxy from all "Cetacea" occurrences

…

Probabilistic Range Maps

Combine:

…

Environment:
- \(sst\): sea-surface temperature
- \(depth\): bathymetric depth
- \(d2shore\): distance to shore

Bayesian State-Space Model

\[ \operatorname{p}(\boldsymbol{\lambda}, \boldsymbol{\beta}, \sigma^2, z | \boldsymbol{y}, \boldsymbol{W}, \boldsymbol{E}, \boldsymbol{R}) \alpha \\ \operatorname{Pois}(\boldsymbol{y}, \boldsymbol{E} \boldsymbol{\lambda}) \\ \operatorname{N_5}(\operatorname{ln} \boldsymbol{\lambda}, \boldsymbol{W} \boldsymbol{\beta}, \sigma^2 \boldsymbol{I_5}) \\ \operatorname{Bin}(\boldsymbol{R} │ 1, 1 - exp(-z \boldsymbol{\lambda}) )^{.5} \\ \operatorname{N_5}(\boldsymbol{\beta} │ \boldsymbol{\beta}_p, \boldsymbol{V}_p ) \\ \operatorname{IG}(\sigma^2 │ s_1, s_2) \]

prior densities:

\(N_5(\boldsymbol{\beta}_p, \boldsymbol{V}_p)\)
\(IG(s_1, s_2)\)

hyperparameters:

\(\boldsymbol{\beta}_p = [0,0,0,0,0]\)
\(\boldsymbol{V}_p = 1000 \boldsymbol{I}_5\)
\(s_1 = s_2 = 0.1\)

Right Whale Estimated Range

4. Decision Mapping

Habitat

Error

Weights

decision	p(0-0.5)	p(0.5-1)
go	0	3
no go	1	0

Weights associating a decided action with the integrated probability of encounter as a simple step function.

Weights

decision	p(0-0.5)	p(0.5-1)
go	0	3
no go	1	0

conservation risk of operating (go) and encountering whales, vs.
opportunity loss of not operating (no go) and not encountering whales

Integrated Probabilities

Cost Maps per Decision

Decision Map

Per pixel, choose decision which minimizes risk-loss function.

5. Conservation Routing

Vessel Routes

Integrate Marine Mammal Distributions

Using density spatial models for 9 marine mammal species in BC from Raincoast surveys 2004-2008
Composite risk map derived per pixel (\(i\)) across \(n\) species (\(s\)) by summing relative density (\(z_i\)), which is based on the pixel values' (\(x_i\)) deviation (\(\sigma_s\)) from mean density (\(\mu_s\)), and multiplying by conservation status (\(w_s\))

\[ z_{i,s} = \frac{ x_{i,s} - \mu_s }{ \sigma_s } \] \[ Z_i = \frac{ \sum_{s=1}^{n} z_{i,s} w_s }{ n } \]

Conservation Status (\(w_s\))

Composite Risk Map

Routes for Oil Tankers

Routes for Cruise Ships

Further Application

Boston Harbor Rerouting for Right Whales

Global Traffic

Acknowledgements

…

Backup Slides

…