Research ArticleECOLOGY

Trophic signatures of seabirds suggest shifts in oceanic ecosystems

See allHide authors and affiliations

Science Advances  14 Feb 2018:
Vol. 4, no. 2, eaao3946
DOI: 10.1126/sciadv.aao3946
  • Fig. 1 Seabird TP declined broadly from 1890 to the present.

    (A to H) Individual seabird species TP series (blue lines) generated from CSIA-AA in feathers. (I) All species TP unweighted mean ensemble (red line). Lines are median model output; shaded area is the 95% credible interval. Circles plot point estimates of the TL computed from the observed stomach contents (29) and forage species MTL (15, 50); thick black lines are SE. Dotted horizontal line is the baseline TP value calculated in 1890. TP declines overall from 4.1 to 3.8 during this period, with the decline from 1950 to the present being twice the rate as before. Where most species showed declines (B to E, G, and H; 63%), two fluctuated over time (D and F; 25%) and one remained remarkably constant (A; 13%). CSSIA, compound specific stable isotope analysis; RTTR, red-tailed tropicbird.

  • Fig. 2 Seabird prey species and their commercial landings.

    (A) Taxa represented in seabird diets observed from stomach contents (29) for all prey comprising >5% diet volume. This yields forage proportions, and we list MTL (parentheses) for each group. Squids, goatfish, flying fish, and jacks are the dominant prey (volume, 65%). (B) Commercial landings reconstructed (58) from spatially explicit catch records for the dominant prey taxa above, available from 1950 to the present. Y axes are square root–transformed to aid visualization.

  • Fig. 3 Diet reconstructions show a shift from fish to squids.

    Bayesian mixing model outputs that reconstruct diet composition longitudinally using CSIA-derived TP (Fig. 1) and prey item TL (fig. S6) for individual species (A to H) and averaged across species (I). We combined four taxa into two diet groups due to TL similarities. The resulting groups encompass six taxa comprising the majority (>80%) of diet volume (29).

  • Fig. 4 Ecomorphology reveals unique structure and foraging strategies.

    Wing tracing silhouettes (A) and various aerodynamic metrics for seabirds in this study (B and C). Observed foraging distances in the North Pacific, plotted as (D) northern latitudinal limits and (E) absolute distances from breeding areas. LAAL have a relatively high wing area per body mass (B) and relatively high aspect ratio per wing loading (C), enabling extended forage bouts at sea (D and E). Conversely, BRBO has a relatively high wing loading per wing aspect (C), limiting its pelagic flights (D and E). Species color codes are retained through all panels.

  • Fig. 5 Random forest variable importance and partial dependence plots.

    (A) Ranked predictors by measure of the variable importance. Variable importance is the measure by which the model mean square error (MSE) is reduced if the variable is randomly permuted. (B) Partial dependence plots for the top 12 predictors in order of variable importance. The plots show the TP response relative to a predictor when all other predictors are mediated for. Yellow subpanels are ecomorphological traits, green subpanels are reconstructed fishery predictors, and blue subpanels are climate inputs. (C) Partial dependence surface plots; all color designations hold except for the red subpanel, which is a cross-sectional interaction. Notable (B) partial dependence plots are the high TLs observed for species with low wing loading, years of low jacks and scads catch, periods of high squid capture, and periods of high NPGO index values. On the surface plots (C), relatively high TPs are observed for species with low wing loading and high forage distances, periods of high SST, and lower PDO index values.

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/4/2/eaao3946/DC1

    Additional Methods

    Additional Results

    fig. S1. Mean absolute TP change.

    fig. S2. TP data simulation raw output.

    fig. S3. TP by measure of phenylalanine and glutamic acid only.

    fig. S4. Incorporated climate indices.

    fig. S5. Informative Dirichlet priors.

    fig. S6. Hierarchal mixing model source distributions.

    fig. S7. Mixing model output with confidence bands.

    fig. S8. Faceted graphs of reconstructed SAU fishery catch time series.

    fig. S9. Partial dependence plots of random forest model output with FAO data.

    fig. S10. Partial dependence plots of random forest model output with SAU data.

    fig. S11. Individual conditional expectation plots.

    fig. S12. Partial dependence surface plots.

    fig. S13. Full random forest variable importance output.

    table S1. Mean absolute TP change.

    table S2. Top 100 linear mixed-effects models.

    table S3. Top 10 linear models by species.

    table S4. Random forest model performance.

  • Supplementary Materials

    This PDF file includes:

    • Additional Methods
    • Additional Results
    • fig. S1. Mean absolute TP change.
    • fig. S2. TP data simulation raw output.
    • fig. S3. TP by measure of phenylalanine and glutamic acid only.
    • fig. S4. Incorporated climate indices.
    • fig. S5. Informative Dirichlet priors.
    • fig. S6. Hierarchal mixing model source distributions.
    • fig. S7. Mixing model output with confidence bands.
    • fig. S8. Faceted graphs of reconstructed SAU fishery catch time series.
    • fig. S9. Partial dependence plots of random forest model output with FAO data.
    • fig. S10. Partial dependence plots of random forest model output with SAU data.
    • fig. S11. Individual conditional expectation plots.
    • fig. S12. Partial dependence surface plots.
    • fig. S13. Full random forest variable importance output.
    • table S1. Mean absolute TP change.
    • table S2. Top 100 linear mixed-effects models.
    • table S3. Top 10 linear models by species.
    • table S4. Random forest model performance.

    Download PDF

    Files in this Data Supplement:

Navigate This Article