The “Butterfish Smackdown” IV. Our modeling approach

Butterfish are generally more abundant in trawl collections on the bottom during the day than at night based on our generalized additive habitat model constructed with the fisherman using NOAA fisheries independent biological surveys of the Mid Atlantic Bight.  The response of butterfish abundance on the bottom to solar elevation reflects a classic diel vertical migration many animals undertake into surface waters at night and to deeper waters during the daytime.

T-13days

We are using a statistical approach to develop a butterfish habitat model with the squid fisherman.  Ecological models come in flavors ranging from mental, to statistical, to mechanistic.  Mental models are educated guesses about how we think something like a butterfish works, while mechanistic models are formal mathematical or logical descriptions put in terms of vital rates and interactions of constituent parts.  A mechanistic butterfish habitat model might have equations describing habitat preferences based upon relationships between growth and food and temperature, predators and the availability of refuges in the water column including those achieved by vertical migration, as well as swimming and sensory capabilities that limit the animals ability to detect and reach safe habitats also good for growth at more distant locations.  Tradeoffs like trading a little growth for survival can might also be included in a mechanistic model.  Our approach lies halfway between and uses statistical techniques and data, in this case catches of butterfish and ocean habitat data, to describe relationships between butterfish abundance and habitat characteristics.  We are, however, primarily using our mental models to determine the relationships to explore with our statistical models.  Mental models are often discounted in science.  But good fisherman have accurate mental models of the behaviors of fish in the ocean—otherwise they go broke.  Ecologists and oceanographers also have accurate models of the ocean and its animals in their heads too, but they may emphasize different relationships. We have combined forces to build the best possible statistical habitat model for butterfish, which is the hypothesis we will test in our field evaluation experiment 13 days from now.  If the test goes well we can begin to work together to develop mechanistic models of habitat relationships for butterfish and other species.

Our modeling approach involves building a statistical species niche model for butterfish using NOAA fisheries independent biological surveys and ocean observations.  Our statistical modeling is guided by hypotheses (=mental models) of fisherman, ecologists and oceanographers about the way butterfish respond to characteristics of the ocean.   The resulting statistical niche model will be used to project butterfish habitat in space and time as a “nowcast”.  In our evaluation experiment we will use this “nowcast” to test the validity of our butterfish niche model.

So what are the limitations of our approach?  Firstly, we may fail to consider important butterfish habitat relationships that scientists and fisherman don’t fully appreciate yet.  We have tried to avoid this by doing preliminary analyses using most of the habitat data available to us.  Many of the other limitations to our approach have to do with the data available to us. We are using surface observations of the ocean measured with satellites and high frequency radar sensors to describe the habitat of fish that live in the water column. We are also using infrequently collected seasonal bottom trawl surveys to statistically model the habitat of an animal that spends allot of time up in the water column. And we are making predictions outside of the period of the data we used to "train" the model.  The day is coming when we will be able to use acoustic measurements of the animals and 3 dimensional oceanographic models to describe distributions of butterfish in the water column in relation to subsurface habitat features, but we are not quite there yet.  Our models also assume that the best habitat is located where the animals are most abundant as measured by trawls.  There are a number of reasons why this assumption can be incorrect. The abundance of fish in trawls cannot be used to distinguish migration corridors that many animals move through quickly from areas in which fewer individuals take up longer term residency because habitat resources support their long term requirements for survival and growth.  Habitat models based on abundance also assume that organisms evaluate habitat quality accurately, without perceptual and movement constraints, and therefore reach abundances at equilibrium with habitat carrying capacity without any time delays.  This is probably rarely the case, particularly in the Autumn in regions like the Mid-Atlantic Bight where important habitat features are extremely variable in time and space and many of the animals are highly migratory.  So there are all sorts of reasons not to do this.  But science never happens unless you take a risk.  It’s just good to be aware of all the assumptions involved in the approach you take from the outset.

The “Butterfish Smackdown” III. On the vital role of the integrated ocean observation system IOOS

Index of surface current divergence/convergence (upwelling/downwelling) calculated from MARACOOS High Frequency Radar Data in the Mid Atlantic Bight.  Inset is a statistical relationship showing that long fin inshore squid (Loligo pealeii) become more abundant as the index of divergence/upwelling increases.  The index of divergence is a calculation of how often the velocity of the water moving up or down higher than a threshold.  The “soil” (nutrients) sinks in the ocean while the “trees” (phytoplankton) are tiny and live fast, reproducing extremely quickly so "they" remain as long as possible in the sunlit surface water where they can photosynthesize.  Upwelling is therefor the essential process that brings the “soil” to the “trees” of the ocean.  We believe squid may be more abundant near areas of strong divergence where the upwelled nutrients fuel the phytoplankton based food webs upon which they and their prey depend.

T-15 days

Our collaboration with the fisherman to develop and test the prototype of a regional scale operational statistical habitat model for a keystone marine species would not be possible without an integrated ocean observation system (IOOS).  The real time ocean observations and models that MARACOOS provides is essential.  But the project would also not be possible without the atmosphere of collaborative networking characteristic of the regional IOOS association.

The task of monitoring and forecasting a marine ecosystem is so big, complex and expensive, that it is impossible unless a diverse group of experts and interests share their resources and expertise to do it.  The collaborative and diverse culture of the IOOS provides a neutral ground for government scientists such as myself and fisherman,who are often at odds, to get together with experts in operational oceanography such as Josh Kohut (Rutgers University) and Matt Oliver (University of Delaware) to work to develop an open source networked marine habitat science at the scale of the whole ecosystem.  Our approach reflects the new way of doing things that digital hammers and networked nails make possible. It is exactly the technologically driven loose collaboration Clay Shirky from NYU’s Interactive Telecommunications program believes is changing the way society works (1)

There is something very beautiful about applying this approach to natural resource science; we are working to develop scientific solutions to avoid “tragedies of the commons” using the strength of the creative commons.  The IOOS “culture” makes this possible and I believe this is far more important for ecosystem science than the ocean observations, although the observations are essential too. 

As a marine habitat ecologist I have often felt relegated to the “left field” by fisheries scientists.  I think there are two reasons for this.  One is that many of us haven’t been doing our science at large enough scales to be relevant to applied problems at the whole ecosystem level.  Secondly we have borrowed our scientific approach from terrestrial landscape ecology and have a hard time remembering that fish and other marine animals live in the water.  This sounds silly, but I believe it is an extremely difficult problem to overcome.  Building a seascape ecology that explicitly considers how the properties and flow of seawater forms habitats requires that we largely abandon the landscape perspective we have developed millions of years of successful evolution in terrestrial environments.  We need to develop an alternative framework appropriate for the ocean that is alien and inhospitable to us. Our terrestrial landscape perspective is an ecological bias and a hindrance to thinking about habitats in the sea.  However, the data and operational oceanography IOOS provides us can help by making regional scale descriptions of the processes of the oceans fluid available. 

Here are two examples. Josh Kohut is an expert in measuring surface currents using high frequency radar.  He and Laura Palamara have used the HF radar to construct the index vertical current flow depicted in the map at the top of the post. We usually think about currents flowing horizontally but they also flow vertically and move things, including the nutrients plants need, up and down. These flows are critical because habitats in the ocean can be thought of as nodes where the pipes that bring essential ecosystem building blocks from sometimes distant sources connect in one place and time. The habitat isn’t the node.  The habitat is the sources of essential building blocks, the “pipes” that deliver them, as well as the nodes where the concentrations are high enough to support organisms and their interactions. Connectivity is also important in terrestrial landscapes, but gravity and the thinness of the air makes the stuff that falls to the ground more or less stay put.  The flux of materials across landscapes is slower and many more supporting process probably occur locally in habitats on land than they do in the sea.  The HF radar begins to describe the pipes for us and we can use it to look “upstream” for the sources.  But in addition to describing the oceans “pipes”, the IOOS also allows us to consider the movements of important features of ocean habitats.  Since many important habitat feature in the sea are defined by the fluid, ocean habitats are often much “faster” than habitats on land.  Fronts between water masses are defined by changes in temperature and other properties over short distances.  They are important habitat features in the sea because nutrients, organic matter, and organisms tend to flow toward them on currents.  Fronts are where the thin soup of the ocean thickens; many animals are concentrated near fronts. Matt Oliver and his student  have classified satellite ocean color data to identify fronts and have made a movie ocean fronts in the Northwest Atlantic from 2002-2010.

These are the kinds of regional scale data that records the dynamics of the ocean that we can now consider in our ocean habitat models. In addition since these observations are being processed in near real time we can incorporate them into our models to make near real time “nowcasts” of habitats. 

1) Clay Shirky 2010. Cognitive Surplus: Creativity and Generosity in a Connected Age. Penguin Press

The “Butterfish Smackdown” II. Panic mode

Conceptual diagram of the experimental evaluation we are about to perform of "nowcasts" from the habitat model we made with the fisherman.  Our model will be coupled to MARACOOS ocean observations to make 4  hour "nowcasts" which will be sent to the  F/V (fishing vessel) "Karen Elizabeth". The "Karen Elizabeth" will sample butterfish in areas where the model predicts habitat is located and where it is not located.  We will then send those locations, times, and catches from the boat back to shore.  Well...that is the idea.

T-17 days 

It’s Thanksgiving Day and we are 17 days away from sailing from Rhode Island on Chris Roebuck’s, FV "Karen Elizabeth" to sample nowcasts of the butterfish habitat model we made with the fisherman.  Our current panic has to do with finding a way to send the model nowcasts 200 miles offshore every day in a form we can actually use on the ship to figure out precisely where to fish to test the model (If anyone has a solution better than passenger pigeons please don’t hesitate to tell us).  

We also found out yesterday morning that we don’t actually have a research permit allowing us to fish anywhere. We will therefor have to sample on Chris Roebuck’s fishing permit. This worried us allot because along with the constraints of the fishing permit, fisherman are a secretive bunch. As a rule they don’t disclose where they fish and how much they catch except to very best friends.  And of coarse, we want to transmit our fishing locations and catches back to the lab so our experiment occurs in real time on the shore, as well as on the ship 200 nautical miles away. But during yesterday afternoon’s conference call we asked Chris,  whether we could send the locations and catches during fishing back to the lab. “Sure” he replied with no hesitation.  We asked again since we expected “no” and an arduous negotiation. We didn’t believe our ears.  “Sure, I don’t see a problem with that”.  Now that reduced the anxiety a little bit.  

But that was just my anxiety and Josh’s. Laura Palamara who merged my original butterfish habitat model with the fisherman models into a working collaborative model I have only tweaked, now carries the anxiety of having to clean up and automate the badly written nowcasting code I have written for a different computing platform. (I am a habitat ecologist, not a programmer and I make terrible computer programs).  And Josh Kohut who has a true gift for pulling and keeping all the collaborators working happily  together, as well as being an expert at ocean physics, is struggling to keep all the feral cats, including myself, happy in the house.  His anxiety level is probably the highest. 

And we still haven’t found the solution for sending data back and forth across 200 miles of open ocean without breaking the bank.  And that’s exactly the infrastructure upon which this entire experiment depends.  

So this is research science, and with fisherman to boot. When the papers are written and presentations are made it all looks so perfectly thought out and executed.  Or at least we try very hard to make it look like that.  But behind the scenes and just before the “showtime” of the experiment, it’s nearly a train wreck and could very well become one.  Real research science is all about finding out something you don't know using methods you’ve dreamt up which may not actually work.  What else should we expect? 

The “Butterfish Smackdown” I. A Little Background

Butterfish habitat model "hindcast" for October 18, 2010 in the mid-Atlantic Bight.  This model will be used in "nowcast" mode in a field evaluation we will perform with a squid fisherman in December  this year.

We are currently in panic mode preparing for a field test of a habitat model for butterfish we built collaboratively with squid fisherman from Cape May, New Jersey, Narragansett, Rhode Island, and Montauk, New York.  Our “Butterfish Smackdown’s” were meetings held over the past year in which fisherman and scientist competed, in a friendly way, and then collaborated to build the statistical model we are about to test.  Our goal was to develop a statistical model we could couple to near real time ocean observations (MARACOOS) to make a regional scale “nowcasts” of butterfish habitat.  If effective, these “nowcasts” might allow fisherman to reduce catches of butterfish as they fish for squid.  This bycatch problem is hard to solve, because when the animals are offshore during the winter and early spring about 4 of 10 trawl samples containing squid also contain butterfish.

Juvenile butterfish hanging out among the tentacles of lions mane jellyfish

Butterfish (Peprilus triacanthus) are small silvery fish that live in the water column and mostly in the coastal ocean from eastern Newfoundland to Florida and the Gulf of Mexico. In the Northwest Atlantic, they migrate from oversummering areas inshore and to the northeast, to the south and offshore to the edge of the continental shelf in the mid-Atlantic Bight during the late fall.  Butterfish are one of the few marine animals that eat jellyfish and salps (also jelly like organisms), as well as other types of zooplankton. The juveniles also seem use jellyfish to avoid predators by hiding among their tentacles.  If this is the case, they have developed a nearly perfect strategy to overcome the “life-dinner principle”  (1).  The principle states that its good to get your dinner, but its much more important to run for your life if you are about to become somebody else’s dinner.  I.e. growth is good but nothing reduces fitness faster than getting killed and eaten.  Butterfish mature at about a year of age and have an extremely high reproductive rate. Minimum population doubling time is estimated to be about 15 months.

We have been focusing our habitat modeling efforts on butterfish and other species that play central roles in the Mid Atlantic Bight Ecosystem.  Butterfish, along with longfin squid, are prey for many to apex predators in the Bight like tuna’s and are therefor important for the transfer of energy from lower trophic levels to upper trophic levels in the food web (2). We just recently published an article demonstrating that ocean observing systems like MARACOOS are useful for describing the coastal ocean habitats of keystone animals such as butterfish (3)

But we wanted our efforts to extend beyond an article in an obscure academic journal and try too operationalize our science by making a useful tool for marine ecosystem management.   We thought involving fisherman in our current project could improve the accuracy of our models.  Fisherman are, of coarse, great practical marine ecologists.  And it has been particularly fascinating to try to understand the physics and ecology in their insights into the sea and the animals they hunt in it.. all of us who are involved in the project share a deep love for the sea.


Collaborators in “Smackdown's” up to this time have included:

Danny Axelsson, 'H&L Axelsson'

Lars Axelsson, 'H&L Axelsson'

Eleanor Bochenek, Rutgers University
Jason Didden, Mid Atlantic Fisheries Council
Greg DiDomenico, Garden State Seafood
Kyle Goodwin, SeaFreeze LTD
Glenn Goodwin, SeaFreeze LTD
Steven Gray, University of Hawaii
Jimmy Harris,  'Trawler Abracadabra'
John Hoey, NEFSC Cooperative Research
Olaf  Jensen, Rutgers University
Josh Kohut, Rutgers University
John Manderson, NEFSC, Behavioral Ecology
Geir Munson, SeaFreeze LTD
Matt Oliver, University of Delaware
Laura Palamara, Rutgers University
Chris Roebuck, SeaFreeze LTD
Wayne Reichle, Lund's Fisheries, Inc
Joel Sonnen, SeaFreeze LTD

1) Dawkins R, Krebs JR (1979) Arms Races between and within Species. Proceedings of the Royal Society of London Series B, Biological Sciences 205:489-511

2) Link J, Overholtz W, O'Reilly J, Green J, Dow D, Palka D, Legault C, Vitaliano J, Guida V, Fogarty M, Brodziak J, Methratta L, Stockhausen W, Col L, Griswold C (2008) The Northeast U.S. continental shelf Energy Modeling and Analysis exercise (EMAX): Ecological network model development and basic ecosystem metrics. Journal of Marine Systems 74:453-474

3) Manderson J, Palamara L, Kohut J, Oliver MJ (2011) Ocean observatory data is useful for regional habitat modeling of species with different vertical habitat preferences. Marine Ecology Progress Series 438:1-17