Project Overview

Summary:

Overview

In January 2016, Dr. Steven Saul at Arizona State University was awarded an RFP-V grant of $971,900 to lead the GoMRI project entitled, “Avoiding Surprises: Understanding the Impact of the Deepwater Horizon Oil Spill on the Decision Making Behaviors of Fishers and How This Affects the Assessment and Management of Commercially Important Fish Species in the Gulf of Mexico Using an Agent-Base” consisted of approximately 6 research team members (including students).

 

The Deepwater Horizon oil spill disrupted the livelihoods of many individuals living along the coast of the Gulf of Mexico, ranging from those in the tourism industry to those who fish the Gulf’s waters for a living. Many of those in the fishing industry, and the sectors that depend on it had to modify their operations (i.e. alter their fishing locations, target species, gear used, or trip duration) in the months after the spill due to spatial closures restricting access to potentially polluted waters. Some of the fishing effort during this time was redirected towards assisting with the cleanup efforts associated with the oil spill. This re-tasking had a direct effect on fishing catch and effort in 2010, and perhaps beyond, depending on whether behaviors that were modified due to the oil spill were maintained in the years ahead or if there was a return to the original behavioral patterns that existed before the incident.

 

To assess the status of commercially important fish stocks in the Gulf of Mexico, the National Marine Fisheries Service relies heavily on information on fish catch and fishing effort that is compulsorily provided by the fishing industry to the government. This information is used to estimate trends in fish abundance over time and serves as inputs to tune the fish population models that are used to establish fishing regulations, such as annual catch limits. At the present time, it is not well understood how the oil spill closures affected the catch of fish and the amount of time/effort fishers needed to use to catch those fish. As a result, it has been difficult for the National Marine Fisheries Service to use the 2010 year of data as a proxy for the trends in abundance that year due to the substantial behavior changes that occurred in the fishing fleet. A biased index of abundance could affect the abundance estimates and the estimated catch limit trajectories stock assessment models provide for future years, as recruitment in future years is dependent on the biomass available in previous years, which is in turn, affected by the fishing mortality that year. Such biases could result in socioeconomic losses to the fishing community by either triggering unnecessary reductions in catch, or conversely increases in catch under conditions where biomass is actually reduced.

 

To improve our understanding of these dynamics, the goal of this project was to develop a spatially explicit bioeonomic model of some the most important commercial fishery species and the fleets that harvest them in the Gulf of Mexico. The project continues the work initiated by the PI and his collaborators, whom have developed a spatially explicit model for the West Florida shelf that incorporated the behavior of four reef fish species (red grouper, gag grouper, red snapper and mutton snapper) and two commercial fishing fleets (handline and longline). The new model expanded the geographic extent to represent the entire Gulf of Mexico, eight commercially important reef fish species (red grouper, gag grouper, red snapper, mutton snapper, gray triggerfish, vermilion snapper, tilefish, and yellowedge grouper), and the two primary commercial fishing fleets that catch these species: handline and longline vessels.

 

The model incorporates the effects of oil pollution on the survival of juvenile and adult fish and the reduction in recruitment caused by impacts of oil on spawner fitness and larval survival. The model also explores the effects of the large spatial closed areas on fish catch and fish population dynamics. Alternative responses to the spill will also be evaluated to understand the scope of the possible effects of different sizes of oil spills on the recovery of these populations. Beyond exploring scenarios about the effects of the spill on the fish and fishing fleet, the model can also be used in the future to explore the biological and socioeconomic impacts of alternative management strategies being proposed.

 

As of December 31, 2019, this project’s research resulted in 12 scientific presentations, and 9 datasets being submitted to the GoMRI Information and Data Cooperative (GRIIDC), which are/will be made available to the public. The project also engaged one master’s level student, one PhD level student, and a postdoctoral scholar over its award period. Significant outcomes of this project’s research according to GoMRI Research Theme are highlighted below. 

• Theme 5, behavioral, socioeconomic, environmental risk assessment: Behavioral discrete choice binomial and multinomial logistic economic models fit to logbook data found that most fishers who prior to the accident operated in areas that were closed to fishing during the oil spill, either fished in alternative locations that were not closed to fishing, or did not fish during the months when spatial closures were in place. We hypothesized that due to this change and disruption, fishers might have permanently changed their behavior to fish in different locations. Analysis of data collected in years after the oil spill and spatial closures occurred, determined that people reverted back to fishing in the same locations they used to fish prior to the oil spill. One caveat to this is the same year that the oil spill occurred (2010), a new fisheries management system was implemented called an Individual Transferable Quota, which radically changed the way commercial fishing captains make decisions. This primarily changed the decision of when people go out fishing and did not as much affect the decision as to where people fish. However, it cannot be discounted as a confounding effect, that influenced where people decided to fish. This work will be submitted for publication.
  
  
• Theme 4, technology developments: As part of this project, we developed a novel statistical workflow to calculate the spatial distribution of the fish species we incorporated in the simulation model. The workflow was a way to develop robust spatial distribution maps from sparse, spatially unbalanced, zero inflated field observations. The approach developed and applied a novel statistical approach for simulating the sampling approach and estimating likelihood to develop labeled data (a technique to expand the data that goes into models). We removed outliers using a combination of Bayesian and frequentist approaches, then used the labeled data to fit an ensemble of machine learning models. Finally, we applied model averaging approaches to develop the final predicted maps. This work is under review for publication and represents a significant advance in the field of species spatial distribution modeling.
  
  
• Theme 4, technology developments: Simulated fish migration improved upon an existing fish migration algorithm that the principal investigator developed some years ago. The improvements to the fish movement algorithm incorporated important components that enhanced the realism of how fish ontogenetically migrate. This included applying turning angles to determine direction, an improved way to select movement probabilities (which improved computational efficiency). Though the concepts used to develop the movement algorithm are not themselves new and are borrowed from the terrestrial ecology literature, there are not many migration models for fish, and as such we plan to publish our approach to contribute it to the scientific literature.
  
  
• Theme 5, behavioral, socioeconomic, environmental risk assessment: An individual transferable quota (ITQ) system was coded into the simulation model. Only a few fisheries models explicitly represent an individual transferable quota system. We developed ours based on a similar one that was developed for an agent-based model of reef fish in Australia. We plan to use the simulation model to study the impact of implementing the ITQ system on the fish and the fleet by modeling the time period before and after the ITQ policy was implemented. We will publish these findings in the scientific literature.
  
  
• Theme 5, behavioral, socioeconomic, environmental risk assessment: The agent-based modeling approach we used represented fish as individual agents, thus generating approximately 85 million objects that the computer needs to track. Each agent represents multiple fish using a scaling parameter that can be changed to improve computational run times and remain within the bounds of the computers random access memory (RAM). This pushes the boundaries of what is computationally tractable, and also pushed the limits of the agent-based modeling library we used (called Multi-agent Simulation of Neighborhoods – MASON). Despite the large computational overhead, a 20-year simulation will only take half a day or less to run. A lot of time and effort was spent by the project team to develop computer code for the simulation model that is highly efficient. This included developing the code to run in parallel on multiple processors and developing creative ways to speed up the processing of the simulation. In addition to parallelization, some additional techniques we coded included strategically updating variables only when needed (rather than every time step), using particular number types to reduce the RAM memory used, and selecting from inverse cumulative probability distributions. 

 

Challenges Under Resolution

 

The model is complete however it is still being debugged and sensitivity tested, and as such, we are behind schedule on the delivery of scenario runs, the final deliverable for this project. There are two reasons for this: health issues that the PI was handling and incomplete work by project team members.

 

Dr. Steven Saul, the project director, had some serious health issues in 2018 and 2019 which slowed progress on this project. The health issues included a diagnosis of severe sleep apnea, which caused respiratory complications from the disorder. This included frequent respiratory illness and flu infections, leading to severe asthma attacks, which required high doses of medication (prednisone) followed by several months of recovery during which my workload was greatly impacted. In addition, on two occasions throughout the duration of this grant, Dr. Saul was hospitalized due to pneumonia complications, also followed by a month or two of recovery after the fact. The sleep apnea is currently being treated by using a continuous positive airway pressure (CPAP) machine and a mandibular advancement device at night. While Dr. Saul’s health issues are starting to resolve, this diagnosis greatly affected his health and his productivity, slowing progress on this grant. Now that the condition is being treated, Dr. Saul’s health issues have started to improve, and his productivity has greatly increased.

 

From collaborating with the post-doctoral researcher, Dr. Saul was under the impression that the simulation model was running properly, and that the biological and behavioral processes were working as intended. The post-doctoral researcher showed Dr. Saul output that indicated this was the case. However, the various components of the model (i.e. biology, human behavior, etc.) had not been properly tested and checked for accuracy. For example, fish are maturing before they turn age one instead of at the right time, fish recruitment was not working properly, fish were not moving properly spatially, and fishing vessels in the western Gulf are still not fishing properly. Once the postdoctoral scholar took a new position, he no longer had time to assist with addressing these issues. Most issues are due to minor coding syntax errors (i.e. improper number type conversions, event timing issues, etc.).

 

To resolve these issues and get simulation scenarios up and running, we are implementing the corrective action of partnering with the Decision Theater (https://dt.asu.edu/), a computational modeling center at Arizona State University. Through this partnership, a team of computer scientists is currently reviewing the model code, identifying, and correcting the issues. Once the issues are fixed, the scenarios are already developed and are ready to be run. It is important to note that my postdoctoral research did some novel work with the computer code of the model to make it run very efficiently. This work was critical to the success of this project and applies an approach that is less frequently used to evaluate the dynamics of fish, fishers, and their environment.

 Proposal Abstract - RFP-V PI Steven Saul