Of Fish and Fishermen: Using Human Behavior to Improve Marine Resource Management

Daniel Andrés Ovando

Abstract

People around the world depend on the ocean for their livelihoods and cultural identity. Properly done, marine resource management can help communities balance their extractive needs with the importance of maintaining healthy ecosystems. But, limited data and understanding often inhibits our ability to effectively manage our interactions with the sea, threatening both food security and ecological integrity. My research uses simulation modeling and quantitative methods to demonstrate how integrating data and theories of human behavior with ecological information can improve our understanding and management of marine ecosystems. For my first project, I ask whether we can use satellite data on the behavior of fishermen to predict the abundance of fish, and find that the answer is “sometimes, but…”. My next line of research shows that the region-wide conservation and fishery effects of Marine Protected Areas may be smaller, more variable, and harder to detect than we thought, and demonstrate an empirical approach for estimating these regional MPA effects in the Channel Islands National Marine Sanctuary. Lastly, I present a novel approach for using local historic economic information, together with biological data, to improve the ability of communities to estimate the health of their fishery.

Introduction

My research uses modern data science tools to integrate economics and fisheries science to improve the management of natural marine resources. The research in this dissertation presents three broadly different projects all related by this common thread. Each of these chapters mixes data and theories of human economic behavior with fisheries science though quantitative methods, including machine learning and Bayesian inference, to demonstrate how integration of human behavior can (or cannot) help us understand and manage fisheries.

My first chapter asks, can we use fine-scale data on fishing effort to gain understanding of the abundance of fishes in space and time? To accomplish this, we pair novel data provided by Global Fishing Watch (described in Kroodsma et al. (2018)) detailing the date, location, and nature of large-scale fishing effort all around the globe with data from fishery independent research surveys to determine if the effort data can predict fish abundance. We find that while effort data can be used to predict fish abundance, environmental data such as water temperature can do the same thing better, and that models fitted to one location are not easily exportable to a new location. All materials needed to replicate this analysis can be found here

The second chapter turns to the use of Marine Protected Areas (MPAs) in providing conservation and fishery benefits. MPAs have a long history in marine resource management, and increasingly are looked to to provide benefits not only inside their borders but also to the waters surrounding them. There is a large amount of general theory addressing the question of when and how much we should expect MPAs to provide regional-scale conservation and fishery benefits, along with a broad literature of modeling designed to test a few theories at a time or support planning in a specific place. We created a simulation tool that integrates across the critical components of this literature to provide a new and comprehensive view of the expected regional-scale conservation and fishery impacts of MPAs. Our results demonstrate that even while controlling for critical drivers such as the size of the MPA network and the pre-MPA depletion of the fishery, the region-wide effects of MPAs are highly variable, and in many cases relatively small. We show that human behavior is one of the most critical drivers of the expected regional effects of MPAs. This has important implications for MPA monitoring programs. To demonstrate this, we pair the simulation analysis with an empirical assessment of the regional effects of a network of MPAs placed in the Channel Islands National Marine Sanctuary in 2003. Our results both present a strategy for estimating regional effects of MPAs in the real world, and match closely with the expectations generated by our simulation analysis. All materials needed to replicate this analysis can be found here

The third and final chapter integrates economic theory and data into the fisheries stock assessment process. Stock assessments are statistical models that estimate critical population parameters such as fishing mortality rates using data such as catch-per-unit-effort and/or the distribution of fish lengths observed in samples from fishery catches. We demonstrate how using economic data and theory such as open-access dynamics along with data on profit per unit effort, prices, cost, labor, and technology, can improve the ability of stock assessment models to provide accurate estimate of fishing mortality rates in a data-limited context. We also present a simulation testing tool for examining model performance and helping users decide which model to use under what circumstances. Our results both open a new field of data for stock assessment and improve the ability of local stakeholders to include their historic knowledge of a fishery’s economic development into the assessment process. All materials needed to replicate this analysis can be found here

This dissertation makes use of a number of computing packages without which the results would be much poorer and much delayed. All analyses were based in the R programming language (R Core Team 2018). However, while data processing and plotting were in done in R, I also made extensive use of Stan (Carpenter et al. 2017), interfaced with using the rstan, rstanarm, and brms packages, along with Template Model Builder (TMB, Kristensen et al. 2016). Code throughout the project makes extensive use of the tidyverse suite of packages which made life exponentially easier (thank you dplyr and ggplot2!), and the caret package (Kuhn 2008) as an interface for machine learning tools. This dissertation was written in bookdown, adapted to match the UC Santa Barbara dissertation template through my package gauchodown, which was made possible by numerous contributors but in particular the original work of thesisdown and huskydown. The appearance of all plots are based on the excellent “opinionated” themes presented in hrbrthemes.

References

Kroodsma, D.A., Mayorga, J., Hochberg, T., et al. (2018) Tracking the global footprint of fisheries. Science 359, 904–908.

R Core Team (2018) R: A Language and Environment for Statistical Computing.

Carpenter, B., Gelman, A., Hoffman, M.D., et al. (2017) Stan : A Probabilistic Programming Language. Journal of Statistical Software 76.

Kristensen, K., Nielsen, A., Berg, C.W., Skaug, H. and Bell, B.M. (2016) TMB : Automatic Differentiation and Laplace Approximation. Journal of Statistical Software 70.

Kuhn, M. (2008) Building Predictive Models in R Using the Caret Package. Journal of Statistical Software 28.