Skill Testing Oxygen Data for Distribution Modeling of Marine Species

Details

• Journal Title:
  Fisheries Oceanography
• Personal Author:
  Indivero, Julia ; Anderson, Sean C. ; Barnett, Lewis A. K. ; Pohl, John E. ; Rohan, Sean K. ; Siedlecki, Samantha ; Ward, Eric J. ; Essington, Timothy E.
• NOAA Program & Office:
  NMFS (National Marine Fisheries Service) ; AFSC (Alaska Fisheries Science Center) ; AKR (Alaska Region) ; NWFSC (Northwest Fisheries Science Center)
• Description:
  Spatial models that identify statistical relationships between environmental conditions and species distributional data are commonly used in fisheries research to evaluate habitat suitability and predict distributional shifts, such as those driven by changing ocean temperature and oxygen levels. However, a lack of environmental data—particularly dissolved oxygen—at the same temporal and spatial resolution as biological data can limit these analyses. We evaluate the ability to predict bottom dissolved oxygen via imputation and extrapolation and with biophysical oceanographic models in the northeastern Pacific Ocean (Aleutian Islands, Eastern Bering Sea, Gulf of Alaska, British Columbia, and California Current). Specifically, we measure predictive skill compared to in situ observations (measured concurrently with bottom trawl data) for (1) predictions from an empirical statistical model fit to integrated dissolved oxygen observations and (2) a commonly used dynamical oceanographic model estimate of oxygen, the Global Oceanographic Biogeochemistry Hindcast (GOBH). Lastly, we evaluate how estimation and interpretation of a species distribution model are impacted by the use of different oxygen data sources. For year‐out cross‐validation, we find that the empirical statistical model predicts bottom dissolved oxygen for fish catch sampling events with relatively high accuracy in only certain regions (California Current and British Columbia) (root mean squared error [RMSE] ~ 16–30 μmol kg−1). Prediction skill was more than two times lower in Alaska regions that did not have extensive data (around < 0.075 observations per square kilometer), and this approach would likely not provide sufficiently accurate oxygen values for SDMs in these regions. The Copernicus Global Oceanographic Biogeochemistry Hindcast had a substantially lower prediction skill than the integrated statistical predictions (RMSE ~30–90 μmol kg−1). When applied to species distribution models, the estimated dissolved oxygen thresholds differed by 20–50 μmol kg−1 when fit to different dissolved oxygen data sources. We focus on oxygen in the northeastern Pacific, yet our approach is generalizable to other variables and systems. We recommend increased attention to validating oceanographic models when operationalized to fisheries applications and evaluating the robustness of conclusions to environmental covariate data sources.
• Source:
  Fisheries Oceanography (2025)
• DOI:
  https://doi.org/10.1111/fog.70005
• ISSN:
  1054-6006 ; 1365-2419
• Format:
  pdf
• Publisher:
  Wiley
• Document Type:
  Journal Article
• Rights Information:
  Accepted Manuscript
• Compliance:
  Submitted
• Main Document Checksum:
  urn:sha-512:78787de1909f7c4486095a8f7d03404a2d7e3d7b030d76ed32879f63ed14ec995a13d275005a59fa08532c944a79dd758ffcc2fd6f858511c6e72191d5935929
• Download URL:
  https://repository.library.noaa.gov/view/noaa/71471/noaa_71471_DS1.pdf
• File Type:
  [PDF - 3.18 MB ]