Date Published: June 18, 2018
Publisher: Public Library of Science
Author(s): Michael R. Stukel, Moira Décima, Thomas B. Kelly, Suzannah Rutherford.
Oceanographic field programs often use δ15N biogeochemical measurements and in situ rate measurements to investigate nitrogen cycling and planktonic ecosystem structure. However, integrative modeling approaches capable of synthesizing these distinct measurement types are lacking. We develop a novel approach for incorporating δ15N isotopic data into existing Markov Chain Monte Carlo (MCMC) random walk methods for solving linear inverse ecosystem models. We test the ability of this approach to recover food web indices (nitrate uptake, nitrogen fixation, zooplankton trophic level, and secondary production) derived from forward models simulating the planktonic ecosystems of the California Current and Amazon River Plume. We show that the MCMC with δ15N approach typically does a better job of recovering ecosystem structure than the standard MCMC or L2 minimum norm (L2MN) approaches, and also outperforms an L2MN with δ15N approach. Furthermore, we find that the MCMC with δ15N approach is robust to the removal of input equations and hence is well suited to typical pelagic ecosystem studies for which the system is usually vastly under-constrained. Our approach is easily extendable for use with δ13C isotopic measurements or variable carbon:nitrogen stoichiometry.
Reconstructing ecosystem structure and trophic flows through planktonic ecosystems is crucial for understanding fisheries production, pelagic biogeochemistry, and the response of each of these to a changing climate. However, quantitative investigation of energy transfer between plankton functional groups is hampered by methodological limitations in separating ecological groups and in making rate measurements on specific trophic levels. As a result, oceanographers often rely on mass-balance ecosystem models such as EcoPath  or Linear Inverse Ecosystem Models (LIM [2, 3]) to reconstruct trophic structure from sparse measurements. However, the paucity and poor taxonomic resolution of common marine ecological measurements leaves such modeling approaches vastly under-constrained (e.g. ). Stable isotope signatures (δ15N and δ13C) offer additional information on diet and trophic position but are limited by difficulties in physically separating plankton functional groups with overlapping size [5–7]. Additionally, in contrast to terrestrial or estuarine systems where bulk stable isotopes of different primary producers are very different (e.g. depending on C3 or C4 plant metabolism), such differing bulk values have not been observed within plankton size classes , although arguably this could also be due to the size overlap of organisms with different isotope values . Approaches capable of combining isotopic data and mass-balance approaches are thus clearly needed .
We use two different forward models (NEMURO and DIAZO) and two separate sets of boundary conditions for each to create four different sets of fully constrained ecosystem flows. Because LIM models are designed to investigate the static, mass-balanced ecosystem structure, we run each model to steady-state in a simple one-box ecosystem configuration. We then determine input measurements—including ecosystem rate measurements (phytoplankton net primary production, nitrate uptake, mesozooplankton grazing, and sinking particle flux) and δ15N values of allochthonous nitrate sources, DOM, mesozooplankton, and sinking particles–that are representative of the types of measurements that can be made in the field. We use these input measurements to constrain LIMs solved using four different approaches: standard L2MN and MCMC approaches and L2MN and MCMC approaches that incorporate δ15N data. We then evaluate the ability of these LIM approaches to recover key ecosystem parameters related to nitrogen biogeochemistry (the relative proportion of phytoplankton production supported by new nitrate or nitrogen fixation) and plankton trophic dynamics (mesozooplankton mean trophic level and secondary production). We also conduct additional tests in which fewer ecosystem rate measurements are used as inputs to the inverse model. The following sections explain the forward models used and the four LIM approaches tested.
Input measurements used for all LIM approaches