E N D
2. 2/39 The Harbor Seals Strike Back? – After the 1989 Exxon/Valdez oil spill S. Bhotika (Ecology, UF), B. Chan (Math, Cornell),
E. Heestand (EEOB, OSU), M. Madan (Stat, MUN),
H. Zhang, Ph.D. (EEOB, OSU)
3. 3/39 Outline Introduction
Prince William Sound
Exxon Valdez oil spill
Harbor Seals
Dataset
Statistical Techniques
Exploratory Analysis
Regression Models
Hierarchical Bayesian Model
Discussion
4. 4/39 Prince William Sound (PWS), Alaska Dense populations of marine mammals
Sea otters, killer whales, sea lions, seals
Large populations of sea birds
Rich herring and salmon fisheries
5. 5/39 Exxon/Valdez oil spill 3/24/1989 -12:04am
11 mil gallons of oil spilled into PWS
1900 km coastline contaminated
Ecological impact:
250,000 seabirds
28,000 sea otters
300 harbor seals
250 bald eagles
22 orcas
Billions of salmon and
herring eggs
6. 6/39 Harbor seals, Phoca vitulina richardsi Haul out onto land to rest, molt, and give birth
7. 7/39 Harbor seals in PWS Do not appear to avoid oil at haulout sites or in the water.
Do not appear to move between sites (oiled/un-oiled).
Seals exposed to oil exhibit lethargic behavior and are easily approached.
8. 8/39 Objective Assess the trends of harbor seal populations over time in PWS after the Exxon/Valdez oil spill.
9. 9/39 Goals Estimate trends in the study area as a whole
Estimate trends at individual sites
Estimate the effects of covariates on seal counts
10. 10/39 Dataset Harbor seal populations monitored at 12 haulout sites
Sites visited 7 to 10 times annually (1990 – 2002)
Counts conducted by aerial survey
Data collected within 2 hrs of low tide
Poor weather or a rapidly rising tide prohibited data collection at times (with the aid of 7-power binoculars), from a single-engine fixed-wing aircraft (Cessna185) at altitudes of 100-300m.
(with the aid of 7-power binoculars), from a single-engine fixed-wing aircraft (Cessna185) at altitudes of 100-300m.
11. 11/39 Dataset SITE: 1 to 12
YEAR: 1990 - 2002
DATE: Number of days starting Aug. 1st
TIME: Time of day (in hrs)
TIDEDEV: Time relative to low tide (in hrs)
SEALS: Number of observed harbor seals
x.UTM, y.UTM: Coordinates of the haulout sites
12. 12/39 Statistical Techniques Exploratory data analysis
General trends
Spatial correlation
Regression models
Linear regression
Poisson regression
Hierarchical Bayesian model
13. 13/39 Exploratory analysis – General trends
14. 14/39 Exploratory analysis – General trends
15. 15/39 Exploratory analysis – General trends
16. 16/39 Exploratory analysis – General trends
17. 17/39 Exploratory analysis – Spatial correlation Dataset was organized by Site and Trip Date
58 trips on which all sites were visited
Cor. Matrix computed, Pearson’s value = 0.26
18. 18/39 Exploratory analysis – Spatial correlation 6 data points available for each year
Pearson’s value = 0.811
Decreasing trend of correlation over time
Cannot detect any spatial correlation
19. 19/39 Exploratory analysis – Spatial correlation Variogram is used to fit spatial correlation
Omni-directional
No spatial correlation is observed
20. 20/39 Conceptual Model
21. 21/39 An approximation
22. 22/39 Linear regression Categorize data
DATE: August, September
TIDEDEV: (-8, -1], (-1, 0], (0, 1], (1, 8)
TIME: [0,4), [4, 8), …, [20, 24)
23. 23/39 Results – Linear regression YEAR, SITE, MONTH, TIDEDEV are significant
R2 = 0.593, Adj. R2 = 0.581
24. 24/39 Poisson regression
25. 25/39 Results – Poisson regression YEAR, SITE, MONTH, TIDEDEV are significant
Akaike information criterion (AIC) = 16145
26. 26/39 Results – Poisson regression
27. 27/39 Results – Poisson regression Applied the same regression to each site
Green: pos. trend, White: neg. trend, Red: unclear
28. 28/39 Hierarchical Bayesian (HB) Model
29. 29/39 Remarks on the HB Model TIME was removed
Quadratic form for TIDEDEV was used
Reflects a concave effect
Poisson regression with quadratic term was also performed with AIC = 16163 > 16145
Diffuse prior distributions were used for all parameters
30. 30/39 Results – HB Model
31. 31/39 Results – HB Model
32. 32/39 Results – Poisson vs. HB
33. 33/39 Summary of Results Similar results were obtained from Poisson regression and Hierarchical Bayesian model
Significant factors: Site, Year, Date, Tidedev
Harbor seal population trends in PWS:
declining at sites 2, 3, 4, 6, 7, 10, 12
increasing at sites 11
not distinguishable at sites 1, 5, 8, 9
No observable spatial correlation, from either the correlation matrix or the variogram
34. 34/39 Discussion – Overall trend Expected: seal populations declining after the oil spill and increasing and/or stabilizing over a period of time.
Results: a slight decrease / undetectable
Possible reasons:
initial decline following spill (1989-1990) not captured in our dataset
after cleanup efforts, lingering oil may affect seal health but not seal counts
seal populations were already declining before the oil spill, since the 1970s
35. 35/39 Discussion – Site specific trends Expected: seal populations declining more at oiled sites than at un-oiled sites.
Results show
7 sites are declining
1 site is increasing
4 sites are indistinguishable
No pattern for oiled vs. un-oiled sites
Possible reasons:
Water is connecting all sites (especially in PWS-enclosed body of water)
36. 36/39 Discussion – Effects of covariates Over the study period, data were collected earlier in the year (seals may haul out less later in the year - molting period and breeding season over)
Tidedev (More seals haulout closer to low tide)
37. 37/39 Discussion – Study Limitations No seal counts prior to the spill
All sites were not visited on some of the trips
Sites not visited randomly (affects which sites observed at low tide)
Sites visited earlier in the year over time
High variability in seal observations at site over small time period
Limited number of sites
Limited covariates in the dataset: site area, human traffic, resources, predators
38. 38/39 Acknowledgements Hongfei Li
Kate Calder
MBI / OSU
Alaska Department of Fish & Game
39. 39/39 References
Ver Hoef, J. and Frost, K.J. (2003) A Bayesian hierarchical model for monitoring harbor seal changes in Prince William Sound, Alaska. Environ. And Ecol. Stat. 10, 201 – 219.
Frost, K.F., Lowry, L.F., Sinclair, E., Ver Hoef, J., and McAllister, D.C. (1994) Impacts on distribution, abundance, and productivity of harbor seals. In Marine Mammals and the Exxon Valdez, T.R. Loughlin (ed.), Academic Press, Inc., San Diego. pp. 97 – 118.
Garrison, W. May/June 1994. Watchable Wildlife – Harbor Seals. http://www.dfg.ca.gov/watchable/seals.html. Accessed on July 24, 2006.
Kinkhart, E. and Pitcher, K. Mar 2005. Harbor Seal. Alaska Dept of Fish and Game. http://www.adfg.state.ak.us/pubs/notebook/marine/harseal.php. Accessed on July 24, 2006.
NOAA/HMRAD. Oil Spill Case History 1967 – 1991. Sept 1992. http://response.restoration.noaa.gov/book_shelf/26_spilldb.pdf. Accessed on July 24, 2006.
ValdezScience.com. Valdez Science: An Environmental Update. 2004. http://www.valdezscience.com. Accessed on July 24, 2006.
Alaska Stock, LLC. 2006. Prince William Sound http://www.alaskastock.com/Prince_William_Sound_Photos.asp. Accessed on Aug 2, 2006.