Brooke Wikgren, New England Aquarium, with Kerry Lagueux

Brooke Wikgren, New England Aquarium, with Kerry Lagueux

Relative Distribution of Baleen Whales in the Gulf of Maine: An innovative approach to mapping relative species distribution using an ordinary kriging interpolation method

Brooke Wikgren, New England Aquarium

Kerry Lagueux, New England Aquarium

With a growing demand for marine spatial planning to mitigate conflicts between existing and future ocean uses, determining the relative spatial distribution of marine animals has become increasingly important. Traditional distribution analyses based on survey sightings can create highly variable spatial data and is greatly dependent on survey effort. To help account for this, a methodology was created incorporating survey effort with the sightings data resulting in an index termed sightings-per-unit-effort (SPUE) and involves assigning calculated SPUE values to spatially explicit gridded cells based on latitude and longitude. Mapping SPUE species distributions by gridded cells is a widely used practice; however, SPUE data is often sparse and can be difficult to interpret. In addition, many marine animals are highly migratory, and mapping based on arbitrarily defined cells can be unrealistic as these animals are not constrained to the grid cell they were sighted in. In response to this, we have developed a kriging methodology to smooth the relative species density and fill in un-sampled areas with predicted values based on spatial autocorrelation. The resulting interpolated surface provides a more realistic species distribution that is easier to interpret and provides a visually enhanced dataset for mapping purposes. For this exercise annual SPUE data for all baleen whales was used to compare the more traditional gridded cell mapping method with that of our improved interpolated method.

Sightings per Unit Effort:

Marine animal distributions are commonly mapped based on survey sightings data. A major issue, however, with spatial distribution and habitat-use patterns based on raw sightings data is that the species distributions are usually biased by the distribution of survey coverage (“effort”). To overcome this possible bias, survey effort is quantified and sighting frequencies are corrected for differences in effort, producing an index termed sightings-per-unit-effort (SPUE). The units are numbers of animals sighted per unit length of survey track. SPUE values are computed for consistent spatial units and can therefore be mapped or statistically compared across areas, seasons, years, etc. Development of this method was begun during the Cetacean and Turtle Assessment Program (CETAP) (1982), and it has been used in a variety of analyses (Kenney and Winn, 1986; Winn et al., 1986; Kenny, 1990; Hain et al., 1992; Shoop & Kenney, 1992; Kraus et al., 1993; Pittman et al., 2006).

The SPUE method requires partitioning the study area into a regular grid based on latitude and longitude. The selected grid size is a compromise between resolution (smaller cells) and sample sizes (larger cells) and can only be determined after preliminary examination of the available survey data. Studies have used cells ranging from 1’ x 1’ to 10’ x 10’. For this project 5’ x 5’ cells were used. Aerial and shipboard survey tracks were broken down into grid cells and their lengths computed. Sightings were also assigned to cells and the numbers of sightings per species were summed by cell. The number of animals in each cell was divided by the effort value and multiplied by 1,000 to avoid small decimal values, creating a SPUE index in units of animals per 1,000 km of survey track. This analysis was done on custom programs in SAS 9.3.1 (SAS Institute, Inc., Cary, NC).

Mapping SPUE by their corresponding gridded cells is a common practice (e.g., Shoop & Kenney, 1992; Kraus et al., 1993), however, the SPUE data is often sparse and can be difficult to interpret. This can be especially true when mapping highly migratory species who are not physically constrained within the grid cell they were sighted in. To overcome this issue, a kriging methodology was developed to smooth the relative density and expand values outside the constrained grid cell boundaries. The resulting interpolated surface creates a more realistic species spatial distribution that is easier to interpret and more visually appealing.

Interpolation Methods:

Sightings per Unit Effort (SPUE) data, as described above, was provided from the Right Whale Consortium (RWC) database and consists of sightings and survey effort from 1978 to 2009 that has been compiled from multiple agencies and organizations into a single database. The SPUE results were summarized to grid cell center points and presented in dbase files containing the species; SPUE calculation; latitude and longitude of 5’ x 5’ cell centerpoint; season (annual, spring, summer, and autumn, and winter); number of animals, and kilometers of trackline effort. For this exercise, annual data for a species grouping of all baleen whales was used. The dbase file was imported into ArcGIS using the latitude and longitude of the 5’ x 5’ centerpoint locations in the WGS 1984 geographic coordinate system. The file was exported into an ArcGIS point feature class inside a File Geodatabase. The feature class was projected into UTM Zone 19 North, North American Datum 1983 for the kriging interpolation. The resulting point dataset was a regular spaced grid of points with SPUE values for the annual distribution for all baleen whales.

ArcGIS’ Geostatistical Analyst was used to create an ordinary kriging interpolation of all baleen whales distributions from the SPUE point data. Kriging is a geostatistical method that builds on mathematical and statistical models of spatial autocorrelation and uses these relationships to create an interpolated surface (ESRI 2010). Spatial autocorrelation is the tendency of locations closer together to be similar in values, (Bolstad 2008) and this relationship can be modeled using a semivariogram during the kriging process. The semivariogram is determined by plotting the average semivariance between all points used in the model at increasing distances. There are many parameters to specify when modeling the semivariogram function, these include: type of model to fit (i.e. Guassian or Exponential), lag distance (distance to bin data) and total number of bins (lags). The semivariogram model is used to weight the points in the search neighborhood to determine the spatial prediction. The Guassian model was the best fit model to our data points and was the model used in the interpolation. The average nearest neighbor of the SPUE points was used as the lag size and the default, 12, was used for the number of lags. We used a smoothing neighborhood for our predictions which adjusts the distance weights determined from the model using a sigmodial function away from the prediction location up to a distance equal to 2 times the Major Semiaxis (ESRI 2010, Gribov and Krivoruchko 2004). A smoothing factor of 1 (the maximum) and a major and minor semiaxis of 20 km to include at least 6 points into our calculations was used for our predictions. The smoothing neighborhood decreased the interpolated surface values by approximately a factor of 10, however, the distributions match the overall SPUE point distributions and the interpolated surface provides the relative abundance for all baleen whales.

The final model of all baleen whales annual distribution was exported from an ArcGIS Geostatistical Layer to an ArcGIS raster grid with a 250 meter cell size. Any negative values as a result of the interpolation were reclassified to zero. The raster grid was mapped and symbolized on a stretched scale representing baleen whales relative SPUE.

Sources:

Bolstad, Paul. 2008. GIS Fundamentals. 3rd. White Bear Lake: Eider Press.

ESRI. 2010. ArcGIS 10 Documentation,

Gribov A., Krivoruchko K.. 2004. Geostatical Mapping with Continuous Moving Neighborhood. Mathematical Geology 36, no. 2: 1-15.

Hain, J. H. W., M. J. Ratnaswamy, R. D. Kenney, and H. E. Winn. 1992. The fin whale, Balaenoptera physalus, in waters of the northeastern United States continental shelf. Report of the International Whaling Commission 42: 653–669.

Kenney, R. D. 1990. Bottlenose dolphins off the northeastern United States. Pp. 369–396 in: S. Leatherwood and R. R. Reeves, eds. The Bottlenose Dolphin. Academic Press, San Diego, CA.

Kenney, R.D. and H.E. Winn. 1986. Cetacean high-use habitats of the northeast United States continental shelf. Fishery Bulletin 84: 345– 357.

Kraus, S. D., R. D. Kenney, A. R. Knowlton, and J. N. Ciano. 1993. Endangered Right Whales of the Southwestern North Atlantic. Final Report, Contract No. 14-35-0001-30486. U. S. Department of the Interior, Minerals Management Service, Herndon, Virginia. 69 pp.

Pittman, S., B. Costa, C. Moy, D. Wiley, and R. D. Kenney. 2006. Cetacean distribution and diversity. Pp. 265-326 in: T. Battista, R. Clark, and S. Pittman, eds. An Ecological -178-

Characterization of the Stellwagen Bank National Marine Sanctuary Region: Oceanographic, Biogeographic, and Contaminants Assessment. NOAA Technical Memorandum NCCOS 45. Center for Coastal Monitoring and Assessment, NOAA National Centers for Coastal Ocean Science, Silver Spring, MD.

Shoop, C. R., and R. D. Kenney. 1992. Distributions and abundances of loggerhead and leatherback sea turtles in northeastern United States waters. Herpetological Monographs 6: 43–67.

Winn, H. E., C. A. Price, and P. W. Sorensen. 1986. The distributional ecology of the right whale Eubalaena glacialis in the western North Atlantic. Reports of the International Whaling Commission, Special Issue 10: 129–1