This repository contains data and code for the paper:
Rosenberg, A. A., K. M. Kleisner, J. Afflerbach, S. C. Anderson, M. Dickey-Collas, A. B. Cooper, M. J. Fogarty, E. A. Fulton, N. L. Gutiérrez, K. J. W. Hyde, E. Jardim, O. P. Jensen, T. Kristiansen, C. Longo, C. V. Minte-Vera, C. Minto, I. Mosqueira, G. C. Osio, D. Ovando, E. R. Selig, J. T. Thorson, J. C. Walsh, and Y. Ye. 2017. Applying a new ensemble approach to estimating stock status of marine fisheries around the world. Conservation Letters. https://doi.org/10.1111/conl.12363
The superensemble method used to derive the estimates is described in this paper:
Anderson, S. C., A. B. Cooper, O. P. Jensen, C. Minto, J. T. Thorson, J. C. Walsh, J. Afflerbach, M. Dickey-Collas, K. M. Kleisner, C. Longo, G. C. Osio, D. Ovando, I. Mosqueira, A. A. Rosenberg, and E. R. Selig. 2017. Improving estimates of population status and trend with superensemble models. Fish and Fisheries. https://doi.org/10.1111/faf.12200
The superensemble method combines estimates from up to 4 individual catch-only models. The individual models are described and fit as part of this report:
Rosenberg, A. A., M. J. Fogarty, A. B. Cooper, M. Dickey-Collas, E. A. Fulton, N. L. Gutierrez, K. J. W. Hyde, K. M. Kleisner, C. Longo, C. V. Minte-Vera, C. Minto, I. Mosqueira, G. C. Osio, D. Ovando, E. R. Selig, J. T. Thorson, and Y. Ye. 2014. Developing new approaches to global stock status assessment and fishery production potential of the seas. FAO Fisheries and Aquaculture Circular, Rome, Italy. http://www.fao.org/docrep/019/i3491e/i3491e00.htm
The generated data file data-generated/ensemble-estimates.csv (link to raw data, link to table) is the primary output of interest. It contains the following:
stock The FAO stock name
CMSY The CMSY method estimate of B/Bmsy
mPRM The modified panel regression method estimate of B/Bmsy
COMSIR The COMSIR method estimate of B/Bmsy
SSCOM The SSCOM method estimate of B/Bmsy
ensemble_method The version of the superensemble model used. full_ensemble Is the full ensemble including all 4 individual estimates. fao_cmsy, for example, excludes CMSY because it did not converge or was not available.
ensemble contains the superensemble estimate of B/Bmsy.
The raw input data files are in the folder data-raw. This folder can be mostly ignored and is there so that the R code is reproducible.
The file fit-ensemble.Rmd generates the ensemble model based on a simulated data set and fits this model to the FAO data.
All estimates are for the average of the last 5 years of data: 2006-2010.
The superensemble estimates are stochastic. They come from random forest models. Estimates will vary in their second decimal place and occasionally their first decimal place from run to run with different random seeds. The included version is from the run used for the paper. The values here have been rounded to 3 decimal places but even this suggests far more precision than is present in the results.
Any catch-only stock status model may not be particularly accurate on a stock-by-stock basis — especially given that the individual models were not tuned for the unique circumstances of each stock in the paper cited above. Here they were used to determine aggregate status across multiple stocks. For individual stock estimates, these catch only models should only be used if a fuller stock assessment with addition sources of data cannot be performed.
The training dataset was based on 3 life-history types, and may not be reliable for a wider range of species. The superensemble was trained on the simulated fish stock data described in the Rosenberg et al. (2014) FAO technical report cited above.