Interannual Variability and Long-Term Change in Pelagic Community Structure Across a Latitudinal Gradient
Author: Alexandra C. Cabanelas
LTER Sites: CCE, NGA, PAL, NES (cross-site analysis)
Working group: Pelagic Community Structure
Created: August 2023 | Last updated: March 2026
This repository is part of a cross-site LTER Pelagic Synthesis Working Group: Interannual variability and long-term change in pelagic community structure across a latitudinal gradient. The working group brings together four marine LTER sites spanning a wide latitudinal gradient to compare how pelagic communities respond to environmental variability at annual and longer time scales:
| Site | Description |
|---|---|
| NES - Northeast U.S. Shelf | Rapidly warming temperate shelf in the northwest Atlantic Ocean. Supports productive fisheries and dynamic planktonic food webs shaped by strong seasonal forcing. |
| CCE - California Current Ecosystem | Eastern boundary upwelling system off the U.S. West Coast. High spatial variability in ocean conditions supports diverse assemblages across trophic levels. |
| NGA - Northern Gulf of Alaska | Subarctic coastal system with strong freshwater influence from glacial runoff. Characterized by a large spring bloom of large-celled phytoplankton and diapausing zooplankton, transitioning to a smaller-bodied summer community. |
| PAL - Palmer, Antarctica | Rapidly warming polar system west of the Antarctic Peninsula. Sea ice seasonality is a dominant driver of food web structure across trophic levels from phytoplankton to marine mammals. |
The working group is pursuing three interconnected projects:
- Normalized Biomass Size Spectra (NBSS): characterizing how zooplankton biomass is distributed across body sizes at each site
- Trophic Amplification: testing whether climate-driven biomass declines are amplified at higher trophic levels and lower latitudes
- Double Integration Hypothesis (DIH) = this repository: examining how cumulative atmospheric forcing drives long-term variability in pelagic populations
This repository contains the cross-site analysis pipeline for Project 3, spanning all four LTER sites. Equivalent pipelines for Projects 1 and 2 are maintained in separate repositories.
This repository tests the Double Integration Hypothesis (DIH) proposed by Di Lorenzo & Ohman (2013) across four marine LTER sites. The DIH posits that marine populations respond to stochastic atmospheric forcing through cumulative integration whereby physical climate signals are first integrated by ocean dynamics and then again by biological populations. This mechanism can produce apparent low-frequency variability and state-like changes in biological time series even when the underlying atmospheric forcing is white noise.
The analysis applies this framework comparatively across contrasting ecosystems spanning a wide latitudinal gradient, asking whether double-integration dynamics are detectable in long-term biological datasets and whether their expression varies across sites.
Key analytical steps include:
- Normalizing biological and physical time series into anomalies
- Performing integration of normalized biological time series
- Computing autoregressive (AR) coefficients for biological and physical driver time series
- Bootstrapping correlation analyses to assess the robustness of relationships between integrated biological signals and large-scale climate indices (e.g., PDO, MEI, ONI, AMO)
Di Lorenzo, E., & Ohman, M. D. (2013). A double-integration hypothesis to explain ocean ecosystem response to climate forcing. Proceedings of the National Academy of Sciences, 110(7), 2496–2499. https://doi.org/10.1073/pnas.1218022110
🚧 This section will be updated - in progress
| Site | Dataset | Source |
|---|---|---|
| NES | ||
| CCE | ||
| NGA | ||
| PAL |
Physical drivers (large-scale climate indices):
| Index | Description | Source |
|---|---|---|
| PDO | Pacific Decadal Oscillation | |
| MEI | Multivariate ENSO Index | |
| ONI | Oceanic Niño Index | |
| AMO | Atlantic Multidecadal Oscillation |
🚧 Script names and structure are actively evolving as the analysis develops. This will be updated to reflect the final pipeline.
lter-double-integration/
├── scripts/
│ ├── 00_packages.R # load all required packages
│ ├── ARcoef_driver_SITE.R # AR(1) coefficients for physical drivers
│ ├── ARcoef_bio_SITE.R # AR(1) coefficients for biological time series
│ ├── integration_SITE.R # single and double integration of biological anomalies
│ └── bootstrap_SITE.R # bootstrap correlations
├── data/
│ ├── raw/ # input files per site (not tracked by git)
│ └── processed/ # intermediate checkpoints (not tracked by git)
├── outputs/ # figures and final tables
├── .gitignore
└── README.md
Replace SITE with the target ecosystem: CCE, PAL, NGA, or NES.
🚧 Final script order will be confirmed once the analysis pipeline is complete. The steps below reflect the current working structure.
- Clone this repository
- Place raw data files in
data/raw/(see Data Sources above) - Open R and set your working directory to the project root, or open the
.Rprojfile - Restore the package environment (see Dependencies):
install.packages("renv")
renv::restore()- Run scripts per site in the following order:
source("scripts/00_packages.R")
source("scripts/ARcoef_driver_SITE.R") # AR(1) for physical drivers
source("scripts/ARcoef_bio_SITE.R") # AR(1) for biological time series
source("scripts/integration_SITE.R") # single and double integration
source("scripts/bootstrap_SITE.R") # bootstrap correlation analysisBiological and physical time series are standardized into anomalies prior to integration.
Autoregressive coefficients of order 1 are estimated separately for biological time series and physical driver indices. These coefficients are used to parameterize the bootstrapping procedure, preserving the autocorrelation structure of the observed data when generating surrogate time series.
Normalized biological anomalies are integrated (cumulatively summed) to produce single- and double-integrated time series. Under the DIH, the double-integrated biological signal should exhibit stronger correlations with climate indices than the raw or single-integrated signal.
Bootstrapped surrogate time series are generated using the estimated AR(1) coefficients. Correlations between integrated biological signals and physical driver indices are computed across bootstrap replicates to assess whether observed correlations exceed what would be expected by chance given the autocorrelation structure of the data.
🚧 R version and full package list will be added once the environment is finalized.
This project uses renv for package version management. To restore the exact environment:
install.packages("renv")
renv::restore()If you use this code, please cite: