By Leila Lemos, Ph.D. Student, Department of Fisheries and Wildlife, OSU
Have you ever heard of statistical modeling? What about Hierarchical Bayes Models?
Hard words, I know…
Modeling is when known data (previously collected) is analyzed using sophisticated computer algorithms to look for patterns in these data. Models can be very useful for filling in data gaps where and when no sampling occurred. Hierarchical Bayes model is a type of statistical model that hierarchically integrates the observed data to estimate parameters. This type of model can analyze long-term data from individual animals to predict into data gaps and inform us about population dynamics.
When studying wild animals we often only collect data from brief and random encounters. Therefore, many researchers struggle with the reconstruction of possible pathways that could connect different sightings of wild animals to determine where, when and how the animal was doing in between sightings.
For instance, consider an animal that was observed in healthy condition at one sighting but in a subsequent sighting it was in poor health. How can we estimate what happened to this animal between sightings? Can we estimate where, when and how health deteriorated?
This is where the modeling comes in! It is a powerful tool used by many researchers to fill in gaps in our scientific knowledge using data that we do have. We use these ‘known data’ to estimate patterns and determine probabilities. The hierarchical Bayes model is a type of modeling that can be used to estimate the probability of pathways between known events. Schick et al. (2013) used hierarchical Bayes models to estimate the many factors that impact whale health and survivorship including distribution and movement patterns, true health condition of the individual and survival rates.
Modeling is very advantageous when studying aquatic animals like dolphins and whales that are very hard to spot since they spend a higher proportion of their lives submerged than above water. Also, sea conditions can hamper visual detection.
Schick et al. (2013) analyzed decades of data from photo-identifications of North Atlantic right whale resightings along the east coast of North America. They assessed different information from these pictures including body condition, infestation of cyamids, presence of fishing gear entanglements, rake marks and skin condition. The authors also used information of age and calving of the individuals. A model using these data was constructed and a more complete scenario of health and movement patterns of individuals and the populations were estimated. Survival rates of each individual were also estimated using this model. This is an example of a well-informed model and is important to notice that a model is only as good as the data you put into the model.
Using this model, Schick et al. documented variations in annual spatial distribution patterns between sexes (Fig. 1). For example, females arrive earlier to the BOF region than males, and have greater estimated transitions to SEUS region at the end of the year. It is also possible to see that there is a lack of information for the region MIDA, characterizing another advantage of modeling since it can highlight areas where effort should be increased.
(NRTH: North region; BOF: Bay of Fundy; JL: Jeffreys Ledge; GOM: Gulf of Maine; RB: Roseway Basin; NE: Northeast; GSC: Great South Channel; MIDA: Mid-Atlantic; and SEUS: Southeastern US).
Source: Figures 5 and 6 from Schick et al. 2013.
When the model is applied to individual whales, the authors were able to estimate survival and health rates across the whale’s life-span (Fig. 2). Whale #1077 was a rarely seen adult male, with a sparse sighting history over 25 years. The last sighting of this whale was in 2004 when its health status was poor due to a poor body condition. According with his condition in the last sighting, the model predicted a high decrease in his health over time and since the whale was not seen for more than six years, so was presumed dead, following the standards set by the North Atlantic Right Whale Consortium.
Source: Figure 11 from Schick et al. 2013.
As I begin data collection for my thesis project to examine gray whale health along the Oregon coast in relation to ocean noise and inter-annual variability, I am considering how to apply a similar modeling approach to enhance our understanding of what influences individual gray whale health and also connect pathways between our resightings.
The marine environment is constantly changing, across space and over time. Therefore, distinguishing what contributes most significantly to whale stress levels can be very challenging. However, through a model we may be able to decipher the contributions of several factors to individual stress among the many parameters we are monitoring: ocean noise, prey availability, environmental patterns, season, sex, age, geographic area, reproductive status and body condition.
Marine ecology is a complex world, and sometimes complex models are needed to help us to find patterns in our data! Once estimates of these ecological processes are created and different hypotheses are explored, information can then be provided to conservation and environmental management to aid decision making, such as defining thresholds of ambient ocean noise levels in the vicinity of baleen whales.
Bibliographic Reference:
Schick RS, Kraus SD, Rolland RM, Knowlton AR, Hamilton PK, Pettis HM, Kenney RD and Clark JS. 2013. Using Hierarchical Bayes to Understand Movement, Health, and Survival in the Endangered North Atlantic Right Whale. PLOS ONE 8(6):e64166.
