Extrapolating with density surface models Laura Mannocci Workshop - - PowerPoint PPT Presentation

▶
extrapolating with
SMART_READER_LITE
LIVE PREVIEW

Extrapolating with density surface models Laura Mannocci Workshop - - PowerPoint PPT Presentation

Dec 14, 2023 299 likes •853 views

Extrapolating with density surface models Laura Mannocci Workshop on spatial models for distance sampling - Oct 2015 - Duke Case study Extrapolating cetacean densities into the unsurveyed high seas of the western North Atlantic Laura

slide-1
SLIDE 1

Extrapolating with density surface models

Workshop on spatial models for distance sampling - Oct 2015 - Duke

Laura Mannocci

slide-2
SLIDE 2

Extrapolating cetacean densities into the unsurveyed high seas

  • f the western North Atlantic

Laura Mannocci, Jason J Roberts, David L Miller, Patrick N Halpin

Case study

slide-3
SLIDE 3

Extrapolating cetacean densities into the unsurveyed high seas

  • f the western North Atlantic

Laura Mannocci, Jason J Roberts, David L Miller, Patrick N Halpin

Case study

“Here be dragons”

slide-4
SLIDE 4
  • Observers, crews, funding agencies, and everyone responsible for conducting the

surveys:

  • Many people who shared surveys, provided advice, and reviewed results:

Suzanne Bates, Elizabeth Becker, Tim Cole, Peter Corkeron, Andrew DiMatteo, Megan Ferguson, Karin Forney, Lance Garrison, Tim Gowan, Jim Hain, Phil Hammond, Jolie Harrison, Christin Khan, Anu Kumar, Erin LaBrecque, Claire Lacey, Gwen Lockhart, Bill McLellan, Dave Miller, Richard Pace, Debi Palka, Andy Read, Vincent Ridoux, Rob Schick, Sofie Van Parijs, Gordon Waring, Amy Whitt and many others…

  • Our funders:

Acknowledgements

slide-5
SLIDE 5

sanctuaries.noaa.org us.whales.org http://timzimmermann.com

Fisheries Ship traffic Military sonars

INTRODUCTION

slide-6
SLIDE 6

sanctuaries.noaa.org us.whales.org http://timzimmermann.com

To evaluate the impacts of these human activities on cetacean populations in the high seas, we need density estimates

Ship traffic Military sonars Fisheries

INTRODUCTION

slide-7
SLIDE 7

Large regions of the high seas have never been surveyed

Kaschner et al. 2012

INTRODUCTION

slide-8
SLIDE 8

NAVY Atlantic Fleet Testing & Training Area

Our goal: to produce the most reliable density estimates for all cetacean species in the U.S. Navy AFTT area

INTRODUCTION

slide-9
SLIDE 9

Our goal: to produce the most reliable density estimates for all cetacean species in the U.S. Navy AFTT area

?

U.S. surveys only covered a fraction of the AFTT area  extrapolate carefully

EEZ

? ? ? ?

NAVY Atlantic Fleet Testing & Training Area

?

INTRODUCTION

slide-10
SLIDE 10

Our goal: to produce the most reliable density estimates for all cetacean species in the U.S. Navy AFTT area

INTRODUCTION

?

U.S. surveys only covered a fraction of the AFTT area  extrapolate carefully

EEZ

? ? ? ?

NAVY Atlantic Fleet Testing & Training Area

?

slide-11
SLIDE 11

To extrapolate carefully, we:

(1) Built models with environmental covariates only

MATERIAL AND METHODS

slide-12
SLIDE 12

Environmental covariates with a broad range of values sampled by the surveys

MATERIAL AND METHODS

slide-13
SLIDE 13

Spatial covariates

  • Latitude, Longitude

Environmental covariates with a broad range of values sampled by the surveys

MATERIAL AND METHODS

slide-14
SLIDE 14

Spatial covariates

  • Latitude, Longitude

Physiographic covariates

  • Depth
  • Slope
  • Distance to shore
  • Distance to isobaths

Environmental covariates with a broad range of values sampled by the surveys

MATERIAL AND METHODS

slide-15
SLIDE 15

This is what would happen if we use distance to shore as a covariate:

Predicted density map for beaked whales Aberrant predictions Surveyed Not surveyed MATERIAL AND METHODS

slide-16
SLIDE 16

This is what would happen if we use distance to shore as a covariate:

Predicted density map for beaked whales Aberrant predictions Dangerous extrapolation beyond the covariate values sampled by surveys MATERIAL AND METHODS Surveyed Not surveyed

slide-17
SLIDE 17

Spatial covariates

  • Latitude, Longitude

Physiographic covariates

  • Depth
  • Slope
  • Distance to shore
  • Distance to isobaths

Physical covariates

  • Sea surface temperature
  • Distance to SST fronts
  • Sea level anomaly

Environmental covariates with a broad range of values sampled by the surveys

MATERIAL AND METHODS

slide-18
SLIDE 18

Spatial covariates

  • Latitude, Longitude

Physiographic covariates

  • Depth
  • Slope
  • Distance to shore
  • Distance to isobaths

Physical covariates

  • Sea surface temperature
  • Distance to SST fronts
  • Sea level anomaly

Biological covariates

  • Chlorophyll concentration
  • Primary productivity
  • Biomass / production of

zooplankton and micronekton (SEAPODYM outputs)

Environmental covariates with a broad range of values sampled by the surveys

MATERIAL AND METHODS

slide-19
SLIDE 19

To extrapolate carefully, we:

(1) Built models with environmental covariates only (2) Incorporated surveys from relevant ecological biomes in the North Atlantic

MATERIAL AND METHODS

slide-20
SLIDE 20

MATERIAL AND METHODS

slide-21
SLIDE 21

MATERIAL AND METHODS

slide-22
SLIDE 22

MATERIAL AND METHODS

slide-23
SLIDE 23

Increase the coverage of ecological biomes encompassed by the AFTT area

MATERIAL AND METHODS

slide-24
SLIDE 24

To extrapolate carefully, we:

(1) Built models with environmental covariates only (2) Incorporated line transect surveys from relevant ecological biomes in the North Atlantic (3) Fitted parsimonious models

“Fit” Simplicity MATERIAL AND METHODS

slide-25
SLIDE 25

Limited the degrees of freedom of smooth functions to mitigate overfitting and avoid reproducing the detailed patterns present in the data

MATERIAL AND METHODS

slide-26
SLIDE 26

Limited the degrees of freedom of smooth functions to mitigate overfitting and avoid reproducing the detailed patterns present in the data

Limited degrees of freedom MATERIAL AND METHODS

slide-27
SLIDE 27

Limited the degrees of freedom of smooth functions to mitigate overfitting and avoid reproducing the detailed patterns present in the data

Overfitted Limited degrees of freedom MATERIAL AND METHODS

slide-28
SLIDE 28

Limited the degrees of freedom of smooth functions to mitigate overfitting and avoid reproducing the detailed patterns present in the data

Limited degrees of freedom

Limited the number of covariates to help understand the primary environmental drivers of cetacean abundances

MATERIAL AND METHODS Overfitted

slide-29
SLIDE 29

Limited the degrees of freedom of smooth functions to mitigate overfitting and avoid reproducing the detailed patterns present in the data

MATERIAL AND METHODS Limited degrees of freedom

Limited the number of covariates to help understand the primary environmental drivers of cetacean abundances  Better generalize predictions to unsurveyed areas

Overfitted

slide-30
SLIDE 30

In total, we modeled 29 cetacean taxa

NOAA NMFS

Sei whale Striped dolphin

RESULTS

slide-31
SLIDE 31

Sei whale

Summer model

RESULTS

slide-32
SLIDE 32

Sei whale

Surveys: EC GOM CAR MAR

Summer model

RESULTS

slide-33
SLIDE 33

Sei whale

Surveys: EC GOM CAR MAR

Summer model

RESULTS

Predictors: Expl Dev 38.5% Depth Sea level anomaly Sea surface temperature Production of micronekton

slide-34
SLIDE 34

Predicted densities (individuals. 100 km-2) Coefficient of variation

Surveys: EC GOM CAR MAR

Sei whale

Summer model

RESULTS

Predictors: Expl Dev 38.5% Depth Sea level anomaly Sea surface temperature Production of micronekton

slide-35
SLIDE 35

Predicted densities (individuals. 100 km-2) Coefficient of variation

Predictors: Expl Dev 38.5% Depth Sea level anomaly Sea surface temperature Production of micronekton

SST Depth SLA SST Depth SLA

Surveys: EC GOM CAR MAR

Sei whale

Summer model

RESULTS

slide-36
SLIDE 36

Striped dolphin

Year-round model

RESULTS

slide-37
SLIDE 37

Striped dolphin

Surveys: EC GOM CAR MAR EU

Year-round model

RESULTS

slide-38
SLIDE 38

Striped dolphin

Surveys: EC GOM CAR MAR EU

Year-round model

RESULTS

Predictors: Expl Dev 57% Depth Production of micronekton Chlorophyll concentration Distance to SST fronts

slide-39
SLIDE 39

Predicted densities (individuals. 100 km-2) Coefficient of variation

Surveys: EC GOM CAR MAR EU

Striped dolphin

Year-round model

RESULTS

Predictors: Expl Dev 57% Depth Production of micronekton Chlorophyll concentration Distance to SST fronts

slide-40
SLIDE 40

RESULTS Predicted densities (individuals. 100 km-2) Coefficient of variation

Predictors: Expl Dev 57% Depth Production of micronekton Chlorophyll concentration Distance to SST fronts

CHL & DFronts CHL CHL & DFronts CHL

Surveys: EC GOM CAR MAR EU

Striped dolphin

Year-round model

slide-41
SLIDE 41

CAVEATS

slide-42
SLIDE 42

Strong assumptions on the shapes of cetacean-environment relationships beyond the sampled covariate ranges

Possible underestimation of sei whale abundance in cold northern waters Example: sei whale

?

CAVEATS

slide-43
SLIDE 43

Predictions less reliable in certain areas

North Atlantic gyre with lower CHL in summer Polar waters with colder SST in winter Log CHL in June (mg.m-3) SST in February (°C) CAVEATS

slide-44
SLIDE 44

CAVEATS

Lack of data for evaluating model predictions in the high seas

Qualitative assessment of predictions with presence only data from the literature:

slide-45
SLIDE 45

CAVEATS

Lack of data for evaluating model predictions in the high seas

Prieto et al. 2012

Tracks of sei whales tagged in the Azores

Qualitative assessment of predictions with presence only data from the literature:

Hydrophones from the Navy SOSUS

Clark and Gagnon 2004

slide-46
SLIDE 46

APPLICATIONS

slide-47
SLIDE 47

APPLICATIONS

These density estimates will be entered in the Navy Acoustic Effects Model to estimate potential incidental ‘takes’ of marine mammals in the AFTT area

Incidental ‘takes’

slide-48
SLIDE 48
  • As new survey data become

available, we plan to continuously update and refine our models to provide the most accurate estimates in the AFTT area PERSPECTIVES

slide-49
SLIDE 49

PERSPECTIVES

  • As new survey data become

available, we plan to continuously update and refine our models to provide the most accurate estimates in the AFTT area

  • The incorporation of surveys

from the North Atlantic gyre and polar waters would greatly improve the models

slide-50
SLIDE 50

Thank you for your attention!

slide-51
SLIDE 51
slide-52
SLIDE 52

Sei whale

slide-53
SLIDE 53

Striped dolphin

slide-54
SLIDE 54

Two-stage density surface modeling

Material and Methods

(1) Fit detection functions and estimate abundance on segments (2) Fit a GAM with estimated abundance as the response and segment area as the offset

Nj = 𝑠=1

𝑆𝑘 𝑇𝑠𝑘

g(0) 𝑞𝑘 𝐹 𝑂𝑘 = 𝐵𝑘 exp[β0 +

𝑙

𝑔𝑙(𝑨𝑘𝑙)]

Rj number of observations in segment j Srj size of the rth group in segment j pj probability of detection on segment j g(0) probability of detection on the trackline Nj is assumed to follow a Tweedie distribution The offset Aj is the area of segment j fk are smooth functions of the covariates zjk β0is the intercept