Habitat-based cetacean density models for the U.S. Atlantic and Gulf of Mexico (2015 Version)

Developed by Marine Geospatial Ecology Laboratory / Duke University

Northeast Fisheries Science Center / NOAA Fisheries, Southeast Fisheries Science Center / NOAA Fisheries, Department of Biology and Marine Biology / UNC Wilmington, Virginia Aquarium & Marine Science Center

Marine Geospatial Ecology LaboratoryNOAA National Marine Fisheries ServiceDept. of Biology and Marine Biology, UNC WilmingtonVirginia Aquarium & Marine Science Center

Contact:
jason.roberts@duke.edu
published

Frequently Asked Questions

Surveys Included in the Analysis

Were the NOAA Atlantic Marine Assessment Program for Protected Species (AMAPPS) 2010-2014 surveys included?

No. Although our models include surveys conducted during 2010-2014 by UNC Wilmington and the NOAA North Atlantic Right Whale Sighting Survey (NARWSS), the AMAPPS surveys were not included. We finalized our models in January 2015. During the three years leading up to that point, we stood ready to incorporate the AMAPPS surveys as soon as NOAA was ready to provide them. But NOAA did not start releasing them to us until February 2015. In January 2016, we received the final batch of surveys through 2013 but as of this writing, in February 2016, we had not received surveys from 2014 yet.

In 2016, we plan to update our Atlantic models to include AMAPPS as well as additional surveys contributed by current and new collaborators. Please contact us for more information.

Were the Cetacean and Turtle Assessment Program (CeTAP) surveys from 1978-1982 included?

No. Although these data may have been sufficiently compatible with our methodology to be included, they were very old. One important factor in our decision not to use them was that we would not have been able to obtain environmental covariates for these data because they predated most available satellite remote sensing data.

Were aerial surveys of Cape Cod Bay by the Center for Coastal Studies included?

No. Those surveys did not collect the data needed to estimate distance from the survey tracklines to sighted animals and therefore could not be used with our methodology.

Were aerial surveys for right whales in the southeast U.S. included?

Surveys conducted by UNC Wilmington (Bill McLellan and colleagues) were included. These surveys covered U.S. Navy study areas in Florida and North Carolina, as well as a large portion of southeast continental shelf during the period 2005-008. Surveys conducted by the Florida Fish and Wildlife Research Institute (FWRI) and organizations in Georgia and South Carolina were not included. We hope to incorporate these into an update of our Atlantic models planned 2016.

Were NOAA surveys from the Gulf of Mexico after 2009 included?

No. Surveys conducted after 2009 occurred after the Deepwater Horizon oil spill. NOAA has declined to contribute these post-oil spill surveys to our analysis. Thus our models are of cetacean density before the oil spill occurred. At this time, we do not have an opinion about how density may have changed since 2009.

Seasonality

Why do some taxa have seasonal models but not others? Why do some taxa have monthly density estimates while others have a single year-round estimate? Don’t all cetaceans exhibit seasonal patterns in their distributions?

For each taxon we modeled, we considered seasonality in two ways: whether it changes its relationship to the environment during different parts of the year, and whether it exhibits a different spatial distribution during different parts of the year.

For the first question, we were concerned with species such baleen whales that could exhibit different environmental preferences during different parts of the year, such as when they are on cold, productive feeding grounds vs. warm, calm breeding grounds far from potential predators. To account for these differences, we sought to fit separate statistical models to each season. For each taxon, we reviewed the literature and examined the available survey effort and sightings. We split the year into taxon-specific seasonal models when all of the following were true:

  1. The literature suggested that the taxon exhibits seasonality in which its relationship to the environment is expected to be different during different parts of the year.

  2. We had sufficient survey effort and sightings to model at least one of the seasons effectively.

  3. The spatial pattern in the sightings resembled the expectation described by the literature.

If any of these conditions were false, we fit a single "year-round" model. The Season column of Supplementary Table S4 of Roberts et al. (2016) tabulates what was done for each taxon.

We considered the second question--whether the taxon exhibits a different spatial distribution during different parts of the year--separately from the first question, to allow for the possibility that a species moves seasonally while maintaining the same general relationship to the environment. Accordingly, we produced monthly predictions of species density when all of the following were true:

  1. There was evidence in the literature of the taxon shifting its distribution seasonally.

  2. We had sufficient survey coverage, both spatially and temporally, to detect the shift.

  3. The spatial pattern in the sightings and the resulting monthly predictions resembled the expectation described by the literature.

If any of these conditions were false, we produced a single, static “year-round” prediction. The Prediction resolution column of Supplementary Table S4 of Roberts et al. (2016) tabulates what was done for each taxon.

These conditions are similar to those for the seasonal modeling question above, but this does not imply that taxa for which we produced monthly predictions also always had multiple seasonal models. For example, with bottlenose dolphins on the east coast, there is evidence that some stocks migrate north in summer and back south in the winter. But unlike with baleen whales, which inhabit distinctly different habitats during their annual feeding/breeding cycle, we were not able to locate sufficient evidence suggesting that the relationship between bottlenose dolphins and the environmental predictors available to us would be distinctly different at different times of the year. Thus we fit a year-round model for bottlenose dolphins. This model included dynamic predictors, including sea surface temperature, which has been suggested as a factor that may constrain bottlenose dolphin distribution on the east coast. The resulting monthly predictions of the year-round model resembled the pattern described in the literature.

As discussed in Roberts et al. (2016), this approach allowed us to scale model complexity and the temporal resolution of predictions to the available knowledge and data. For well-known species for which we had sufficient data, we could produce seasonal models and monthly predictions. For poorly-known species or those for which we lacked data, we resorted to a more conservative approach of a single year-round model and prediction.

How do I find out whether seasonal models were used for a taxon, and what those seasons were?

Supplementary Table S4 of Roberts et al. (2016) tabulates what was done for each taxon. In the taxon-specific, region-specific Supplementary Reports that accompany the downloadable density rasters, the narrative text that appears at the very top of the Density Models section discusses our rationale for the taxon’s seasons.

Why are monthly predictions not offered for any Gulf of Mexico taxa?

Unlike the east coast, in the Gulf of Mexico there are no taxa that are documented to undertake large seasonal migrations. For example, the only baleen whale that occurs regularly in the Gulf of Mexico is Bryde’s whale; little is known about this population and it was sighted so rarely that a seasonal model could not be considered. Equally important, the survey coverage in the Gulf of Mexico was patchy in space and time, both seasonally and across years, making seasonal movements hard to detect.

Given evidence of seasonal migrations and additional survey coverage, particularly in fall and winter, it might be eventually appropriate under our methodology to offer monthly predictions for some species in the Gulf of Mexico.

Guilds, Species, and Stocks

Why are some models for individual species and others for guilds of several species (e.g. beaked whales)?

Most of our models were of individual species, but several models grouped two or more species together into a "guild". We created guilds when sightings of ecologically-similar species were ambiguous--e.g. observers reported many sightings of “beaked whale” rather than of "Blainville's beaked whale", "Gervais' beaked whale", and so on--and we could not confidently classify the ambiguous sightings into individual species based on habitat variables or group size. Species we modeled as guilds include beaked whales, Kogia species, and pilot whales on the east coast. (Only short-finned pilot whales inhabit the Gulf of Mexico, so a guild was not necessary there.) The Methods section of Roberts et al. (2016) describes this process in more detail.

How do I estimate density for individual species modeled as a guild, such as beaked whales?

One possibility would be to consult the NOAA stock assessment reports (SARs), which sometimes offer abundance estimates for individual species that we grouped into guilds, and prorate our density estimate of the guild into individual species estimates according to ratios of abundances from the SARs. But there are potentially two problems with this:

  1. The SAR estimates are often made only from unambiguous sightings for which the observer reported a full species identification. You must be willing to assume that the abundance ratios for unambiguous sightings hold for the ambiguous sightings.

  2. Species within a guild may exhibit habitat preferences. For example, on the east coast, short-finned pilot whales tend to occur in more southerly warm water, and long-finned in more northerly cold water, with some mixing between them in the middle. Simply prorating our model will not account for spatial differences in distribution between the species.

If you require per-species density for members of a guild for a U.S. regulatory process, you should consult with the regulatory agency to develop a mutually-agreeable approach to this problem. Feel free to involve us in that discussion.

How do I estimate density for individual stocks of bottlenose dolphins?

Bottlenose dolphins exhibit complex stock structure. NOAA has identified estuarine, coastal, and oceanic stocks in both the Gulf of Mexico and along the U.S. east coast. Some of these stocks overlap. Some are relatively resident while others exhibit seasonal movements. For more information, please see the NOAA stock assessment reports (SARs).

Developing per-stock estimates for bottlenose dolphins was beyond the scope of our project. In general, our models estimate the combined density of coastal and oceanic stocks but exclude estuarine stocks. At the time of this writing, NOAA defined coastal and oceanic stocks for the Gulf of Mexico using static bathymetric contours and geographic boundaries. These stocks did not overlap. In principle, it is possible to split our Gulf-wide density surface according to these contours and boundaries and obtain per-stock density estimates. On the east coast, such a splitting is probably not possible because NOAA did not define the stocks using static contours and boundaries, and some stocks are believed to overlap.

For more discussion of this problem, please see the Supplementary Reports available with our bottlenose dolphin models. If you require per-stock density estimates for a U.S. regulatory process, you should consult with the regulatory agency to develop a mutually-agreeable approach to this problem. Feel free to involve us in that discussion.

Rare Species

Why do some species have "stratified models" in which density is constant across large areas?

This occurred when there were too few sightings for us to confidently model density from habitat variables. Instead, we estimated mean density (the number of individual animals present divided by area effectively surveyed) from the surveys conducted in the area in which the species was likely to occur, and applied that density estimate uniformly throughout the modeled area. This is sometimes known as a "standard line transect density estimate" or a "stratified model".

I need to estimate takes of a rare species for an MMPA permit application. This species is so rare that NOAA surveys only sighted a few groups of them in the past decade or more. Can I simply assume that one group will be taken, rather than modeling takes from the density estimates?

No. Making that assumption would not account for the difference in how much surveying NOAA did compared to how much potentially harmful activity you're proposing to conduct. To illustrate the problem with this assumption, consider a hypothetical activity in which all cetaceans within a certain distance of a moving vessel moving would be "taken". Assume this distance was about the same as the effective distance within which observers on a NOAA survey vessel could sight a group of cetaceans. Let's set aside whether this is realistic, as well as factors influencing detectability, the problem of animals being submerged and not able to be seen, and other complications. Let's further assume that NOAA conducted 50,000 km of shipboard survey effort over a 20 year period and only sighted five groups of a rare species. Given all this, how many groups are likely to be taken during the proposed hypothetical activity?

It depends on how long the hypothetical vessel would be conducting this activity. Speaking very roughly, if the vessel planned to transit a similar distance to NOAA (50,000 km), we would expect five groups to be taken, since the probability of taking a group per kilometer transited would be about the same as a NOAA observer sighting a group. If the vessel only planned to transit 10,000 km, we would expect one group to be taken.

But what if the hypothetical vessel conducted its activity over a single year? NOAA only sighted five groups over 20 years. Isn't this species so rare that we are unlikely to encounter even one of them in a given year? Doesn't it mean that this species is usually absent from the study area, and only appears in very sparse numbers every few years?

In general, no, these assumptions cannot be made. Without a better understanding of the species' distribution, we cannot say whether it is always present but in low density or present sporadically but in higher density. But in either case, it is incorrect to assume that only one group will be encountered without accounting for how much harmful activity would occur. The proper way to estimate takes is to first determine the area in which they will occur and the rate or probability of take occurrence, and then apply the these to the density surfaces. There is no difference in this procedure between rare and common species. You should treat them the same way.

But isn't statistical confidence in the density estimates very low for species sighted so infrequently?

Yes, statistical confidence is lower for these species. We include estimates of the coefficient of variation (CV) and other uncertainty statistics with our density estimates, and uncertainty is indeed higher for rarely sighted species. You can propagate this uncertainty into your own modeling process, if desired.

Total Abundance Estimates

Where can I find total abundance estimates?

The column of Supplementary Table S4 of Roberts et al. (2016).

How do these abundance estimates compare to those in NOAA stock assessment reports (SARs)?

Our objective was to produce the most accurate density and abundance estimates we could, given the information available. To assist you with comparing our estimates to NOAA's, the Abundance Estimates section of our taxon-specific Supplementary Reports tabulates our estimates with NOAA's from the past few SARs. When comparing our estimates to those from the SARs, you should bear in mind:

  • The extent over which our estimates were made often differs somewhat from NOAA's. You can determine the extent of our estimates from the rasters we offer for download. Bear in mind that this extent does not always span the entire east coast study area (take note of the % Area covered column Supplementary Table S4 of Roberts et al. 2016). For NOAA's estimates, you can consult their technical reports or peer-reviewed publications for diagrams or definitions of their study areas.

  • The months over which our estimates were made often differ from NOAA's. NOAA's estimates are almost always from surveys conducted in summer. Our summary abundance estimates, given in Supplementary Table S4 of Roberts et al. (2016), are year-round or seasonal averages usually made from temporally-dynamic habitat-based models. If a species is more abundant in summer, NOAA's summer estimate might be substantially higher than our year-round average even though our model yields similar results to NOAA's in summer months. For a detailed example of this, please see the Discussion section of our Supplementary Report for east coast pilot whales.

  • We may have used different estimates of availability and perception bias (the g(0) parameter) than NOAA. These estimates can strongly influence total estimated abundance. Please see Roberts et al. (2016) for a discussion about these biases. The g(0) Estimates section of each taxon's Supplementary Report documents the estimates we used.

  • In the Gulf of Mexico, NOAA did not attempt to address availability or perception bias (i.e. they assumed that g(0)=1). As a result, NOAA underestimates abundance of cetacean most species in the Gulf of Mexico. The degree of underestimation can be particularly pronounced for species that have long dive times and thereby exhibit large availability biases, such as beaked and sperm whales. Please see the Discussion section of our taxon-specific Supplementary Reports for the Gulf of Mexico for further treatment of this problem.

Duke 2015 Models vs. Other Models

How do the Duke 2015 models compare to the 2007 Navy OPAREA Density Estimates (NODEs) models?

The Duke 2015 models were conceived of and funded as a replacement for the NODEs models. The NODEs models are obsolete and should no longer be used. Please see this presentation for a detailed list of improvements the Duke 2015 models provide over the NODEs models. You are welcome to contact us if you have any questions.

Characteristics of the Downloadable Density Rasters

What is the projection and cell size?

The rasters use an Albers equal area projection configured to minimize error for the U.S. east coast and Gulf of Mexico. The datum is WGS 1984. The cell size is 10 km.

What are the units of the rasters?

This is discussed in the README.PDF files that accompany the rasters.

References

Roberts JJ, Best BD, Mannocci L, Fujioka E, Halpin PN, Palka DL, Garrison LP, Mullin KD, Cole TVN, Khan CB, McLellan WM, Pabst DA, Lockhart GG (2016) Habitat-based cetacean density models for the U.S. Atlantic and Gulf of Mexico. Scientific Reports 6: 22615. doi: 10.1038/srep22615. (Download)