Outline Building Delphinid Habitat Models with Passive Acoustic Monitoring Data Data Collection Model Challenges True absence vs. false absence Data sampling differences Temporal scale of datasets Case Study Pacific whitesided dolphins Melissa S. Soldevilla1, Erin M. Oleson1, Megan Ferguson2, Sean M. Wiggins1, John A. Hildebrand1 2 1 Study Area Species Frequency (khz) Southern California Bight Risso s Dolphins 5 (Grampus griseus) Currents California Current Southern California Eddy Southern California Countercurrent High Productivity Coastal upwelling Topography Submesoscale eddies Human Activity Shipping Fisheries Naval bases after Hickey et al 1992 Instrumentation Highfrequency Acoustic Recording Southern California Bight Package (HARP) PC Pacific Whitesided Dolphins (agenorhynchus obliquidens) 1 Click Type B Frequency (khz) 1 Discussion 39 31 25 22 5 4 37 26 22.5 Time (hours) 1. 27.5 22 Click Types A & B.5 Time (hours) 1. HARP Data Analysis Identify click bout start and end times Classify to species Assign onehour time bins as P / A SBC 2 khz sample rate 16 bit 1.9 TB data storage CAT Up tosnb 55 SCI days continuous recording TB onger deployments possible with duty cycle??? 1
Dolphin Occurrence Absence = not vocalizing? OR Presence Animals present Animals vocalizing HARP sampling ow ambient noise. Acoustic detection Absence True absence Animals not vocalizing HARP not sampling High ambient noise conditions mask sound Missed detection Calls are behavior dependent Not necessarily vital function May not be produced by all population members May vary in temporal production May vary in spatial production e.g. whale song But these do not hold for dolphin echolocation Echolocation Variability Echolocation Variability Spinner Dolphins Kaneohe Bay, HI Bottlenose Dolphins Sarasota Bay, F Risso s Dolphins Southern California Bight, CA Temporal Spatial Norris, K. S., Wursig, B., Wells, R. S., and Wursig, M. (1994). The Hawaiian Spinner Dolphin. University of California Press, Berkeley. Nowacek, D. P. (25). "Acoustic ecology of foraging bottlenose dolphins (Tursiops truncatus), habitatspecific use of three sound types," Mar Mammal Sci 21, 58762. Assumptions Risso s and Pacific whitesided dolphin click most when foraging Absence = not sampling? OR Dolphins need to forage daily Quality foraging habitat can be represented by the amount of time echolocating dolphins spend in it Absences due to high noise or observer are not biased by habitat conditions Some periods sampled continuously Some periods on sampling schedule Problems: How to compare between schedules? What is missed? 2
Sampling Schedule Detection Probability Sampling Schedule Detection Probability Subsample all data to same sampling schedule 5 / 1 5 / 15 5 / 2 Subsample continuous data at each sampling schedule Calculate probability of detection Calculate correction factor for sampling schedules B A Detection probability Duration (min) Risso's PWS A PWS B 5.97.95.97 1.93.89.93 15.88.83.88 Correction factor 5 1.3 1.6 1.3 1 1.8 1.13 1.8 15 1.13 1.21 1.13 Temporal Resolution: Autocorrelation & Matching datasets Dolphin sampling Minimum of hours per day Hours per week best Environmental sampling Variable scales Daily Weekly Monthly Quarterly Environmental Data HARPs SST Chl Satellite Telemtry Data Summary of Data Included in Model Pros Readily available to management Appropriate temporal scale Weekly scale ideal for coverage and autocorrelation Cons No depth information Cannot sample zooplankton Cannot sample nekton Dependent variable Hours per week with dolphin clicks present Correction factors HARP sampling schedule correction Hours per week with recordings Independent variables Satellite Telemetry (SST & Chl) Upwelling Index Temporal variables Week and unar Phase index 3
Can environmental conditions be used to predict dolphin click bout occurrence? Predictor Terms Generalized Additive Model (GAM) og (Hours / week with click bouts present) + Offset variables = Model Selection Forwardbackward stepwise algorithm Akaike s Information Criterion (AIC) Intercept + f1(sst) + f2 (SST CV) + f3 (Chl) + f4 (Chl CV) + Region* f5(week) + f6(unar Phase) + f7(upwelling) Case Study: Pacific Whitesided Dolphins Crossvalidation & Predictions Pseudojackknife method 8 % training data 2 % test data 5 models total Average Squared Prediction Error (ASPE) Crossvalidation Model Comparison Moon Phase Chl residual Chl CV 367.9 5.31 Model 2 563.13 5.39 Model 3 695.75 5.65 s2 Model 4 529.54 111. Population structure Model 5 57.25 3.57 Two populations in SCB % Models including term 6 6 4 8 1 1 Week Upwelling Model 1 Squid Mesopelagic fish Epipelagic schooling fish Northeast Pacific Distribution SST CV ASPE Prey types Cool temperate waters Baja California to Alaska SST mean AIC Distribution Baja California CA / OR / WA From CMS / GROMS Pacific whitesided dolphins Type B Model Fits Observed Data and Model Predictions Habitat characteristics Homogenous 18oC waters Upwelling Fallwinter season Partial Hours/Week with Clicks (oc) 82 % Explained deviance Environmental Variables 4
Case Study Conclusions Fallwinter season, increased upwelling, and homogenous 18oC waters arge confidence intervals at high SST and SSTCV High explained deviance compared to visual studies Modeling Summary Modeling hours per week with calls provides quantitative measure of habitat importance Calculating detection probability effective for handling variability in sampling schedule Ensure appropriate temporal resolution for all model variables and consider autocorrelation Issues Future work Clicks are also used in communication and navigation Click production and behavior comparison Models represent time spent in habitat, but do not indicate number of animals Dolphins may produce additional unknown/unidentified click types Mechanisms Prey availability? Predator avoidance? Competitive interactions? Quantify animals based on calls Develop models with insitu data for comparison Include measure of prey abundance Include competing species and predators Expand spatial coverage Acknowledgements Questions? 5
Time Series Comparison Pacific whitesided occurrence, SST and Chl Santa Catalina Island, CA 6