What did we learn and what we are doing now

We learned a tremendous amount in our experiment sampling a regional scale butterfish habitat model with the fishermen; about the capabilities and limitations of our model; about what we know about fish in the ocean; and about what good fishermen know. We are pretty sure we are among a very few people on the planet to have sampled a dynamic habitat model nowcast at the extent of a regional ecosystem.  We think even fewer have so formally integrated the knowledge of great fishermen into evaluation survey design as well as the construction of the model itself.

Map of the track of the FV Karen Elizabeth (dashed line) on our model evaluation cruise performed in early December 2011 superimposed on a nowcast from the butterfish habitat model we made with the fishermen.  Blue areas indicate where habitat “quality” was poor and fish should be absent, yellow and red areas where habitat “quality” was high and many fish could be present. The three canyon hotspots, Atlantis/Veatch, Wilmington and Norfolk are labeled on the map. The pictures at the top from left to right are a sample of 7000 pounds of butterfish netted at a station Chris Roebuck, the captain, chose based on his understanding of butterfish ecology. Middle: The computer screens on the FV Karen Elizabeth for hydroacoustics and underwater temperature sensors that allowed us to visualize concentrations of organisms in relation to temperature fronts at the bottom. Temperature data collected by the underwater robot glider enroute from New Bedford to the Shelf break MARACOOS sent out to sample near us.

Our survey design was structured by the model output (Habitat model and movie of the model) and the logistical constraints of trying to sample the entire mid-Atlantic Bight Continental Shelf break over 7 days in December using a single fishing boat.  Our model indicated there were hotspots at Atlantis, Wilmington, and Norfolk canyons.  We confirmed these hotspots over the telephone with fishermen before the survey to see whether the model was in the ballpark.  We used these canyons as focal points for evaluation sampling.  Since the habitat model also indicated that butterfish concentrations varied by daylight and nighttime hours,  we sampled a pixel of ocean predicted to be “good habitat” and a pixel of “bad habitat” from the model at each canyon during the day and again at night.  In addition, Chris Roebuck the fishermen picked a 3^rd station in each set where he believed butterfish would be located based upon his experience.

Sampling design we used in our cruise on the FV Karen Elizabeth with Captain Chris Roebuck to evaluate our butterfish habitat model. We sampled locations our model predicted would be good and bad habitat, as well as a location Chris thought would be good habitat, during both the day and night, in the vicinity of 3 canyon hotspots in the mid-Atlantic Bight.

Using this design we learned a huge amount about what we know about butterfish based upon analysis of data at broad scales, and what fishermen like Chris know based on their time on the water fishing very precisely in areas where they think fish probably live. The work we are doing now is based on three of the most important lessons we learned:

1) Our models accurately captured butterfish responses to ocean features like temperature fronts and upwelling zones at spatial scales of 10s to 100s of kilometers.  However, there were habitat features at finer spatial scales of centimeters to kilometers important to butterfish that we couldn’t capture using the data we had available to us.  We made the model using data from the NOAA fisheries surveys of the Northwest Atlantic Continental shelf.  The average distance between the stations in those surveys is about 20 kilometers (11 nautical miles).  As a result, we are unable to detect with certainty species responses to ocean features at scales smaller than about twice that distance or about 40 kilometers.  So our models and approach are useful, but for problems at relatively coarse spatial scales where the size of the pixel is about 40 kilometers.  The models have similar limits in temporal resolution. Each year the NOAA shelf wide surveys are conducted during the spring and fall.  Our model captures dramatic dynamic features of butterfish habitat but at relatively coarse temporal scales.  So as ecosystem scientists, habitat ecologists and oceanographers, we can use our data to make models of the ways dynamic ocean features define habitats and drive changes in patterns of species distribution over relatively broad scales of space and time.

2) In contrast, fishermen like Chris Roebuck have an extremely fine scale knowledge of associations of animals with ocean features we can’t capture using traditional fishery independent surveys. Their livelihoods depend on close observations of fine scale ocean features associated with variations in the oceans plumbing and other properties that control phytoplankton production, the concentration of plants and animals in the food web and the feeding interactions resulting in the mortality of some organisms and growth of others.  Just those sorts of features associated with places and times they catch fish.  These are scales where mechanisms operate that create the broad scale patterns and relationships we ecosystem scientists detect in our data sets. By working together with the fishermen we can try to connect the broad scale patterns we see in our data and models over seasons and decades and the mechanisms the fishermen observe at fine spatial scales on a daily basis.  This integration should benefit all of us in our search for cause and effect and in a science more likely to reflect the true realities of the ocean.  All of us agree that ocean management needs to be based on just that kind of science.

3) We also found during our experiment with Chris showed us some habitats inshore in shallow water and offshore in deep water that are ecologically important to the animals but not accessible to the big ships used for regional fish surveys.

Fisherman, Ecologists and Oceanographers and Ecosystem scientists understand the dynamics of the seascape and its effects on marine animals at different space-time scales.  Broad scale analyses, models and understanding are great for generating hypotheses about mechanisms driving patterns.  Fine scale studies and knowledge are essential for discovering the causal mechanisms behind the broad scale patterns.  By working together we can construct a better, mechanistic understanding of the ways in which the oceans and its physics regulate the biology of populations living in it.

We are now working toward several new goals with the fishermen’s help based upon the information we gathered in our collaborative model building and evaluation.  1) We will try to construct next generation habitat models that allows us to estimate how much inshore habitat and offshore habitat is sampled on the traditional fisheries surveys because of the size of the vessels used and logistic constraints.  And we are asking whether or not the amounts of habitat sampled on the surveys is increasing or decreasing systematically over time, possibly as a result of climate change.  2) We are trying to develop a way to decide how to sample and estimate the sizes of fish populations when the habitats the animals are using in the ocean are changing shape, size and moving very quickly.  That is a very hard problem.
We are now working toward several new goals with the fishermen’s help based upon the information we gathered in our collaborative model building and evaluation.  

To achieve these goals we are working with fishermen and others to construct a set of next generation habitat models.  The models we used in the “Butterfish Smackdown” were statistical and we projected them in time and space using observations of surface ocean features from MARACOOS.  Fish live in the water, not on top of it, so we want our next set of models to be projected based upon conditions underwater.  We would also like to move beyond statistical habitat models toward mechanistic models based upon first principles of the physiology and ecology of the animals.  If we could do that well, our models might perform better for nowcasting and short term forecasting than purely statistical models.  Or we could hedge our bets, construct both types of models, and use an ensemble approach in a similar way to the weather forecasters.