Inferring Biomass, Rates, and Diversity from Biogeochemical Profiling Float Arrays

Gael Forget1, Stephanie Dutkiewicz2, Oliver Jahn1 and Michael J Follows1, (1)Massachusetts Institute of Technology, Cambridge, MA, United States, (2)Massachusetts Institute of Technology, Department of Earth, Atmospheric and Planetary Sciences, Cambridge, MA, United States
Abstract:
Autonomous platforms will provide a wealth of new biogeochemical and ecological information in coming decades. Here we assess how the planned biogeochemical-Argo array of profiling floats may perform in terms of monitoring marine ecosystems. We simulate measurement of 6 key oceanic biogeochemical variables, as planned, by global profiling float arrays from a state of the art model simulation of marine biogeochemistry and ecosystems over two decades. The model solution, from the CBIOMES project, is a global ocean state estimate that is based on Forget et al 2015 for ocean physics, constrained by core-Argo and satellite data, and on Dutkiewicz et al 2015 for marine biogeochemistry and ecosystems. The model’s detailed representation of ecosystems, with 35 phytoplankton types and 16 zooplankton types, inherent optical properties, and radiative transfer components make it ideal for simulating bio-chemical and bio-optical data. Our observing system simulation experiment first samples the ocean exactly like the core-Argo program did for temperature and salinity over the past fifteen years. We then generate ensembles of randomly selected float subsets to evaluate the skill of float arrays as a function of their density. As a first step, the simulated data sets are used to map out each observed variable, individually, in space and time via standard averaging and interpolation techniques. We then train neural networks to predict the biomass of major plankton groups, supply and biological rates, as well as biodiversity indices from observed variables. This allows us to assess float array performances with regard to biogeography over the global ocean.