Objectives

1
PREDICTIVE HABITAT MODELS FOR MARINE
MAMMALS http//serdp.env.duke.edu Andy Read,
Pat Halpin, Larry Crowder, David Hyrenbach, Ben
Best, Michael Coyne, Sloan Freeman, Caroline
Good, Song Qian
Duke University Marine Laboratory
Landscape Ecology
Lab Duke University

• Objectives
• Provide Navy users with objective and flexible
  software tools to analyze marine mammal habitats
  
• Problem Statement

• Approach
• Design an objective and flexible Spatial Decision
  Support System (SDSS) software for Navy users
• Technical Overview
• Data Sources

• Model Building
• Develop models to predict marine mammal
  distributions using concurrent environmental data
  
• Type of Models

• Pilot Project
• Contrasting habitat models from different
  statistical approaches
• Focal Species
• Sighting Data

• Progress
• We have made substantial progress during the
  first of our 4-year project (April November,
  2004)
• Modeling Workshop
• Data Compilation

• Selected the
  sperm whale
  as the subject
  of the pilot
  analyses because it is susceptible to
  human-caused sound pollution
• Cosmopolitan range
• Deep-diving species

• The Navy must conduct training exercises to
  maintain readiness
• There are presently no analytical tools with
  which to predict the distribution of marine
  mammals at spatial and temporal scales of
  relevance to these exercises
• We propose to develop a novel analytical toolkit
  which will allow the Navy to predict the
  distribution of marine mammals

• Convened planning workshop at Duke University
  (June) to review existing marine mammal habitat
  models and to select promising modeling
  approaches
• Product
• Proceedings of the planning meeting will be
  submitted for publication in a scientific journal
  (Ecological Modeling) by January 1st

• Our project considers a broad array of
  statistical and modeling approaches

Spatial Decision Support System (SDSS)
(Image Garth Mix)

• Generalized Linear Model with
• second order polynomial terms
• (b) Generalized Additive Model with smoothed
  spline functions
• (c) Classification Tree (CART)
• (d) Environmental Envelope models
• (e) Canonical Correspondence
• (f) Bayesian models (Aspinal 1992)
  model a posterior presence probability for focal
  species (pp), as a function of its a priori
  occurrence (ppp / ppa)

• Combined four datasets from the OBIS-SEAMAP
  system

• The development and testing of marine mammal
  habitat models requires concurrent distribution
  and oceanographic data

• As of November 29, OBIS-SEAMAP has compiled 100
  datasets, with approximately 300,000 records of
  marine mammals
• Product
• Online software tools to view, subset, and
  integrate these sightings with remotely sensed
  environmental data are publicly available at
• http//seamap.env.duke.edu

• Review existing approaches to the modeling of
  marine mammal distribution, as predicted by the
  structure of the marine environment
• Design a novel, hierarchical framework for
  analyzing marine mammal distributions across
  annual, seasonal and synoptic time frames

User
Effort 23,093 km Sightings 115
Biological
Hierarchical Modeling Framework
OBIS-SEAMAP

• Describe how a binary (presence / absence)
  response variable is associated with a set of
  explanatory variables
• All seven variables significant
• Mapping potential habitat

Physical
Source Guisan Zimmermann, 2000. Predictive
habitat distribution models in ecology

• Have undertaken a pilot analysis using data from
  New England and Mid-Atlantic region, following
  similar steps outlined here. Pilot project will
  be completed by January 1st
• Product
• During year-2, we will undertake a full
  analysis of marine mammal habitats within U.S.
  Atlantic E.E.Z. We will submit a manuscript
  describing our modeling efforts for publication
  in an ecological journal (Ecological
  Applications) by the end of 2005

• Statistical approach entails reiterative process
  of model formulation, calibration, and evaluation

Model Fitting / Evaluation Framework

• Develop and test novel models of marine mammal
  distribution
• Assemble a Spatial Decision Support System that
  will allow Navy users to
• access archived oceanographic and cetacean
  distribution data for different localities and
  time periods
• update these climatological data sets with
  real-time sightings and oceanographic data
  products
• analyze marine mammal model outputs across
  multiple forecasting time and space scales

• Model output calculated for oceanographic
  conditions on Aug. 5-12 1998

• Habitat 11 probability threshold
  (ROC curve)

Area under the curve 0.86
Source Guisan Zimmermann, 2000. Predictive
habitat distribution models in ecology