Supporting Information
Total Page：16
File Type：pdf, Size：1020Kb
Supporting Information Asch 10.1073/pnas.1421946112 SI Text Relationships Between Phenology, Climate, Oceanic Conditions, and Precision of Estimates of Phenological Change and Evaluation of Bias Ecological Traits. The effects of three basin-scale climate oscillations due to Gaps in Sampling. A sensitivity analysis based on Monte Carlo (e.g., ENSO, PDO, and NPGO) and three local environmental simulation was conducted to quantify observation error related to variables (e.g., SST, zooplankton volume, and coastal upwelling) on noise from variations in larval abundance and the sparse, seasonal fish phenology were examined. ENSO events were characterized sampling resolution of CalCOFI. First, the seasonal distribution of a with the Oceanic Niño Index (ONI) produced by the National simulated larval species was modeled with a Gaussian curve where Centers for Environmental Prediction (www.cpc.ncep.noaa.gov/ June 15 was initially the mean date of larval occurrence and there products/analysis_monitoring/ensostuff/ensoyears.shtml). This index was a 1-mo SD. This standard deviation implied that 95% of larvae was selected because it defines the onset and termination of El would be observed within ±2 mo of the mean. The Gaussian curve Niño and La Niña on a monthly basis. The PDO index was obtained was multiplied by 10 to more realistically approximate larval from the Joint Institute for the Study of the Atmosphere and Ocean abundance in the CalCOFI region. Variations in larval abundance (57) (research.jisao.washington.edu/pdo/PDO.latest). The NPGO that could affect the calculation of seasonal central tendency (CT) index was developed and made available by Emanuele Di Lorenzo were stochastically generated from a coefficient of variation (CV) at the Georgia Institute of Technology (30) (www.o3d.org/npgo/ ranging between 0.05 and 0.60. Twelve CVs separated by 0.05 in- npgo.php). SST and mesozooplankton displacement volume were tervals were used. This range was selected based on the empirical measured at CalCOFI stations where ichthyoplankton samples 3 CV of monthly larval abundance for representative species sam- were collected. Gelatinous organisms with biovolumes >5cm pled by CalCOFI (e.g., Argentina sialis, Engraulis mordax, Lipolagus were excluded from measurements of zooplankton volume (41). ochotensis, Scomber japonicus, Tetragonurus cuvieri). Next, shifts in Zooplankton volume from 1969 to 1977 was multiplied by a cor- phenology at rates of 0–12 d/decade were simulated over six de- rection factor of 1.366 to account for changes in net type and tow cadal time steps. For each time step, the CT was calculated from depth (80). The monthly Bakun upwelling index from 33°N and monthly mean values extracted from the underlying continuous 119°W was obtained from the Environmental Research Division Gaussian distribution. This process was repeated 1,000 times for at the Southwest Fisheries Science Center (83) (www.pfel.noaa. each CV and rate of phenological change. The 90% and 95% CIs gov/products/PFEL/modeled/indices/upwelling/upwelling.html). This of CT were calculated based on these simulated values to assess the index estimates offshore Ekman volume transport based on at- precision of this estimate of phenological change. mospheric pressure fields. At 33° N and 119° W, the Bakun up- For the largest CV considered, the 90% and 95% CIs of CT welling index overestimates offshore transport due to a discontinuity always fell within ±2.8–3.1 and ±3.3–3.6 d/decade, respectively, of in the atmospheric pressure gradient related to the presence of the simulated change in phenology (Tables S3 and S4). The 90% coastal mountains (83). Nevertheless, this index is still correlated CIs corresponded closely to the minimum rate of change observed with upwelling measured with the QuikSCAT scatterometer at among species categorized as displaying earlier or later phenology r = 0.6 (86). Also, the CT of QuikSCAT upwelling over the (i.e., −2.8 d/decade for earlier species; 3.0 d/decade for later CalCOFI region during the 2000s fell within the 95% CI of the species). This finding confirmed that, despite the coarse, sea- Bakun upwelling index CT from this decade (Fig. S4). sonal sampling resolution of CalCOFI, this study was able to obtain Each climate index was partitioned into three categorical reliable estimates of which species exhibited shifts in phenology. variables for use in ANOVA. The ONI was divided into La Niña, A second potential source of error stemmed from the fact that El Niño, and neutral periods. PDO categories included (i)a no CalCOFI surveys were conducted in May, September, or negative PDO between 1951 and 1976; (ii) a positive PDO be- December during the 2000s. These data gaps could skew esti- tween 1977 and 1998; and (iii) a second, negative PDO in 1999– mates of CT from this decade and bias interdecadal trends. To 2002 and 2007–2008 (87). The years 2003–2006 were excluded evaluate this bias, I recalculated CT after removing these months because there were insufficient data to construct a monthly time from the entire time series so that all decades would have an series of larvae during this second positive PDO. Positive, neu- identical monthly distribution of data. The first principal compo- tral, and negative NPGO periods were defined on a monthly nent of this new dataset accounted for 29.9% of variance and basis according to whether this index had a value ≥0.5, between continued to show a progression toward earlier phenology (Fig. 0.5 and −0.5, or ≤−0.5, respectively. These thresholds were se- S2B). However, phenological advancement was no longer evident lected so that there were an approximately equal number of when examining the mean CT for all 43 species (Fig. S2D). For months in each NPGO category. Data on each fish phenophase the three phenology groups, decadal trends in CT were extremely were aggregated into bins corresponding to different phases of similar regardless of whether the missing months were included or the climate indices. CT anomalies of the fish phenophases were excluded from the analysis, indicating resiliency to this potential then calculated for each phase of the climate indices. Initially, source of bias (Fig. S2 E–J). In addition to being able to detect crossed, two-way ANOVAs were performed using climate index divergent trends obscured by the mean, this is another reason why and phenology group as independent variables. Because signifi- this analysis principally focused on the three phenology groups cant interactions were observed between most phenology groups when investigating long-term, ecological changes. and climate indices, separate one-way ANOVAs were performed Measurements of CalCOFI SST and mesozooplankton dis- to evaluate climate effects on each phenology group. Lilliefors’ placement volume were also unavailable during May, September, test, Bartlett’s test, and plots of residuals were used to ascertain and December in the 2000s. Again, these months were removed whether the assumptions of normality, homoscedasticity, and from the time series in its entirety, and CT was recalculated to independence of residuals were met (82). Lilliefors’ test in- detect potential biases resulting from gaps in sampling. Stepwise dicated that the assumption of normality was violated for the multiple regressions between oceanic variables and phenology ANOVA examining ENSO effects on the earlier phenology groups produced broadly consistent results regardless of whether group. As a result, a nonparametric Kruskall–Wallis test was the full dataset or the bias corrected dataset was used (Table S5). performed in lieu of ANOVA. If an ANOVA or Kruskall–Wallis As a result, Fig. 3 only presents results from the full time series. test revealed a significant climate effect, Tukey–Kramer multiple Asch www.pnas.org/cgi/content/short/1421946112 1of9 comparison tests were used post hoc to determine which climate formed to detect major modes of variability among ecological phases exhibited significant phenological differences. variables. Categorical variables (i.e., taxonomic order, adult fish CalCOFI SST, CalCOFI zooplankton volume, and the Bakun habitat, cross-shore distribution, biogeographic affinity, and upwelling index were selected as the local environmental vari- fishing status) were converted into dummy variables before PCA ables to be examined due to known interannual-to-decadal var- (82). Each variable was then standardized based on its mean and iations in their seasonality (19, 21, 22, 25) and their potential to SD to ensure that all variables were given equal weight in the affect fish phenology through physiological and trophic pathways. PCA (82). Correlations between the first two principal compo- The CT of these variables was calculated decadally for years nents of the ecological characteristics and changes in larval fish between 1951 and 2008. Unlike upwelling and larval fish abun- CT were assessed. The PCA revealed three groups of fishes with dance, measurements of SST and zooplankton volume never similar ecological characteristics (Fig. 4A). Phenological trends approached zero during any month of the year. This pattern over time were examined for each of these groups. weakened the weighting of these variables by month when cal- This study also explored whether species that underwent shifts culating the CT, resulting in a CT skewed toward the middle in phenology were more likely to change their geographic range in of the year. To increase the influence of monthly weights, response to fluctuating climatic conditions. This analysis entailed a the minimum monthly mean value (e.g., 13 °C for SST and 3 3 comparison between the results of this study and those of Hsieh 38 cm /1,000 m of seawater strained for zooplankton volume) et al., who investigated climate-related shifts in the distribution of was subtracted from the decadally averaged time series before cal- 35 of the species examined here (Table S2) (50, 51). One-tailed culating the CT. After computing the CT of each variable, its SE binomial tests were used to evaluate whether there was a >50% was estimated with a bootstrap approach (82).