Exploratory Spatial Analysis Norma
Download as PPT, PDF1 like570 views
The document summarizes research on detecting and predicting locations of illegal oil discharges from vessels off Canada's Pacific coast. It finds spatial autocorrelation between detected oil spills and marine traffic densities of different vessel types through statistical analysis. Based on these results, it proposes using spatially autoregressive and geographically weighted regression models to predict locations of illegal oil discharges based on vessel movement patterns and surveillance efforts, in order to help transportation authorities better target their monitoring programs.
Exploratory Spatial Analysis Norma
- 1. Exploratory spatial analysis of illegal oil discharges detected off Canada’s Pacific Coast Norma Serra Rosaline Canessa Patrick O’Hara Stefania Bertazzon Marina Gavrilova ICCSA Conference 2008
- 2. What are illegal vessel-source oil discharges? Photo courtesy of Gov. of Canada Photo courtesy of P. O’Hara Mystery Spill
- 3. The ecological impact in marine ecosystems of small-scale chronic-levels of oil pollution is shown to be greater than large accidental oil spills; especially for seabird populations. ~ 80% global population of Cassin’s Auklets in BC ~ 74% global population of Ancient Murrelets ~ 56% global population Rhinoceros Auklet In British Columbia there is a rich diversity of seabirds at least 62 different species, including major proportions of world breeding populations: ~ 25% global population of Marbled Murrelets
- 4. National Aerial Surveillance Program (NASP) Primary tool of Transport Canada for preventing and monitoring illegal oil pollution within Canada’s Economic Exclusive Zone, which generally extents 200 nautical miles (370 km) out from its coast . The Vancouver Sun. February 16 th , 2008
- 5. Research Goal Specify a multivariate regression model to predict where illegal oil spills are more likely to occur, to provide NASP with a decision making tool to maximize the effectiveness of their surveillance program. In order to identify a valid model, in necessary to explore the spatial properties of our data: a) Spatial Heterogeneity b) Spatial Dependence or Spatial Autocorrelation (SAC)
- 6. Study Area Maximum bounding area covered by surveillance flights off the west coast of Vancouver Island, British Columbia. 1
- 7. Data sources From NASP : 53 detected oil spills in BC from 1997 to 2006 Surveillance flight path data in BC from 1997 to 2006. From Marine Communications Traffic Services - Vessel Traffic Operations Support System: Relative marine traffic densities in BC were estimated for different vessel type groups, from data collected in 2003.
- 8. Marine shipping traffic in Canada’s West Coast Bulk carrier
- 9. Marine shipping traffic in Canada’s West Coast Oil tanker
- 10. Marine shipping traffic in Canada’s West Coast Tug vessel
- 11. Marine shipping traffic in Canada’s West Coast Fishing vessels
- 12. Marine shipping traffic in Canada’s West Coast Cruise ship
- 13. Methods Global measures of spatial autocorrelation examine the nature and extent of the dependence within model variables and produce a single value for the entire data set. The two most common global measures: I ~ + 1 Positive SAC I ~ - 1 Negative SAC I = 0 No SAC Moran’s I Geary's c 0<C< 1 Positive SAC C > 1 Negative SAC C = 1 No SAC
- 14. Methods (Cont.) Local measures of spatial association – Local Moran’s I identify the location and spatial scale of aggregations of unusual values, such as clusters of high values (hot spots) and low values (cold spots) and, because it generates an autocorrelation index for each data site, this can be mapped providing additional information about the pattern under study. E.g. LISA Cluster Map (GeoDA TM ) Outlier Outlier Cluster Cluster high values surrounded by low neighboring values High - Low low values surrounded by high neighboring values Low – High low values surrounded by neighbors of similar low values Low – Low high values surrounded by neighbors of similar high values High – High
- 15. Results: Spatial autocorrelation – Global indices 0.871 (0.01) 0.158 (0.01) Oil spills counts 0.455 (0.01) 0.518 (0.01) Cruise ship movement counts 0.371 (0.01) 0.553 (0.01) Tug vessel movement counts 0.395 (0.01) 0.553 (0.01) Carrier vessel movement counts 0.369 (0.01) 0.589 (0.01) Oil tankers movement counts 0.249 (0.01) 0.783 (0.01) Fishing vessel movement counts 0.172 (0.01) 0.805 (0.01) NASP flight counts Geary's c ( p -values) Moran's I ( p -values)
- 16. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of surveillance flight counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 17. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of detected oil spills counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 18. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of carrier vessel movement counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 19. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of oil tanker movement counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 20. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of tug vessel movement counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 21. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of fishing vessel movement counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 22. Results (Cont.): Spatial pattern description - LISA Cluster Maps LISA cluster map of cruise ships movement counts. Significance test using 999 permutations and applying a significance filter of p ≤ 0.01.
- 23. Discussion Our exploratory analyses indicate that there is a positive spatial autocorrelation within datasets for all variables (dependent and independent). LISA cluster maps allowed the visualization of significant aggregations of high values of oil spill counts, surveillance effort and shipping traffic. To some extent LISA cluster maps show the effect of deterrence. For example, oil spill clusters were found in areas of Barkley Sound, whereas clusters of high counts of surveillance flights and marine vessels were mainly found in the entrance to the Strait of Juan de Fuca.
- 24. Conclusion (I) Based on our results: Our variables exhibit spatial dependencies and non-stationarities, therefore, spatially autoregressive (SAR) and geographically weighted regression (GWR) models are the most suitable to predict spatial patterns of illegal oil discharges based on marine commercial vessel type and movement patterns, while controlling for the spatial distribution of surveillance flights.
- 25. Conclusion (II) Based on our results: We propose to calibrate a multivariate regression model, combining a spatial autoregressive and a local specification. Our model will estimate oil spill probabilities based only on areas poorly covered by the current NASP flight zone, In order to provide effective and reliable recommendations for the planned surveillance program expansion.
- 26. Explore the degree of correlation between detected oil spills and marine traffic densities for different vessel types (i.e. fishers, tugs, cargo vessels, tankers, cruise ships and ferries). Incorporate results from O’Hara et al. * that defines the relationship between oil spill detection probabilities and surveillance effort. Explore oil spill probabilities using a multivariate approach that will include variables such as season, distances between oil spills and shore and distances to nearest port or marina, and other characteristics that define the vessel type (i.e., flag state, inbound vs. outbound) Future research * O’Hara, P.D., Serra-Sogas, N., Canessa, R., Keller, P., Pelot, P.: Estimating oil spill rates and deterrence based on aerial surveillance data in Western Canadian marine waters. Marine Pollution Bulletin, (submitted) (2008).
- 27. Thank you for your attention Any questions? Acknowledgements: