What's Hard About Making a Model?

As promised during last week's blog, I am going to focus this week on describing a challenge associated with making spatial models. This is a great week for this topic, as we have been busy preparing for upcoming events, like our conference and field work, so it hasn't been a week of new tasks. 

If you've been reading the blog for the past few months, you will remember that the project we recently finished used a temporal modeling framework-that is, we modeled everything at one place but evaluated changes across time. The temporal framework involves a researcher providing spreadsheets of forcing data (environmental characteristics like temperature, salinity) that inform how organisms survive in their environment. Our models are forecast models, which means that we train our model on existing biological data--how many of each type of marine organism we have per year, how many of those animals are removed by fishing pressure, how many animals get accidentally removed by fishing and then die before they can make it back to the environment--and then we let the model evaluate what the future looks like given model scenarios. These model scenarios for our project involved various Bonnet Carré Spillway operations. 

Now that we are creating a spatiotemporal framework, we need to consider how the organisms use the physical space of the Mississippi Sound and Bight, which creates some challenges. For the previous project I made response curves for temperature and salinity for all our animals, where were based on a combination of laboratory studies, literature searches, and data from field monitoring efforts for the organisms in our model. The field monitoring efforts are almost always reports from state agencies that record data on the number of marine species caught during fishing efforts and the environmental data from where the fishing occurred. While the field monitoring data are quite useful for evaluations of salinity and temperature tolerances, we are now working to incorporate depth as a factor in the model. The fishing efforts are highly biased regarding depth because while the depth of the seafloor is not the same every time they go fishing because they are fishing in different locations, they are deploying an otter trawl, which samples at the sea floor. Even if the fish is slightly above the sea floor, if it gets caught in the net, the team will record the depth of the sea floor because that's the depth of the trawl. Additionally, since the state agencies are not fishing in the rivers or in close proximity to the shore, there are model groups that may not be represented by these fishing efforts (like largemouth bass, sunfishes, many juvenile fishes that frequent rivers or nearshore environments). One solution is to evaluate local knowledge including local fishing histories, perhaps reported sightings of organisms, to narrow in on an appropriate depth range for these organisms. I will note that the depth responses for some of our animals gets quite difficult because while we may know the minimum and maximum depths that an observer has seen an animal, we may struggle to determine its optimum depth range.

As an alternative, I have seen some teams use distance to shore as a qualifier, rather than depth, though this metric is also based on field sightings and fishing efforts, which doesn't necessarily eliminate the potential bias. One solution that does help is relying on depth when necessary but finding alternative geographic or spatial features that may help explain distributions. For instance, if you are modeling a coastline that has mangroves, you could include a mangrove proximity variable to reflect the refuge mangroves provide for juvenile fishes. Perhaps an oyster reef proximity variable, since oysters cannot survive if there is no substrate on which they can settle.

I really enjoy the head-scratching and puzzle aspects of this work, though. When Kim and I meet to talk about these problems we get to talk through the pros and cons of multiple solutions and avenues of inquiry, which makes this work exciting. Stay tuned for next week's blog, where I will focus on oyster research across North America, as a precursor to my conference presentation.