Can machine learning extract the mechanisms controlling phytoplankton growth from large-scale observations? – A proof-of-concept study

A key challenge for biological oceanography is relating the physiological mechanisms controlling phytoplankton growth to the spatial distribution of those phytoplankton. Physiological mechanisms are often isolated by varying one driver of growth, such as nutrient or light, in a controlled laboratory setting producing what we call “intrinsic relationships”. We contrast these with the “apparent relationships” which emerge in the environment in climatological data. Although previous studies have found machine learning (ML) can find apparent relationships, there has yet to be a systematic study examining when and why these apparent relationships diverge from the underlying intrinsic relationships found in the lab and how and why this may depend on the method applied. Here we conduct a proof-of-concept study with three scenarios in which biomass is by construction a function of time-averaged phytoplankton growth rate. In the first scenario, the inputs and outputs of the intrinsic and apparent relationships vary over the same monthly timescales. In the second, the intrinsic relationships relate averages of drivers that vary on hourly timescales to biomass, but the apparent relationships are sought between monthly averages of these inputs and monthly-averaged output. In the third scenario we apply ML to the output of an actual Earth system model (ESM). Our results demonstrated that when intrinsic and apparent relationships operate on the same spatial and temporal timescale, neural network ensembles (NNEs) were able to extract the intrinsic relationships when only provided information about the apparent relationships, while colimitation and its inability to extrapolate resulted in random forests (RFs) diverging from the true response. When intrinsic and apparent relationships operated on different timescales (as little separation as hourly versus daily), NNEs fed with apparent relationships in time-averaged data produced responses with the right shape but underestimated the biomass. This was because when the intrinsic relationship was nonlinear, the response to a time-averaged input differed systematically from the time-averaged response. Although the limitations found by NNEs were overestimated, they were able to produce more realistic shapes of the actual relationships compared to multiple linear regression. Additionally, NNEs were able to model the interactions between predictors and their effects on biomass, allowing for a qualitative assessment of the colimitation patterns and the nutrient causing the most limitation. Future research may be able to use this type of analysis for observational datasets and other ESMs to identify apparent relationships between biogeochemical variables (rather than spatiotemporal distributions only) and identify interactions and colimitations without having to perform (or at least performing fewer) growth experiments in a lab. From our study, it appears that ML can extract useful information from ESM output and could likely do so for observational datasets as well.

Mesozooplankton grazing minimally impacts phytoplankton abundance during spring in the western North Atlantic

The impacts of grazing by meso- and microzooplankton on phytoplankton primary production (PP) was investigated in the surface layer of the western North Atlantic during spring. Shipboard experiments were performed on a latitudinal transect at three stations that differed in mixed layer depth, temperature, and mesozooplankton taxonomic composition. The mesozooplankton community was numerically dominated by Calanus finmarchicus at the northern and central station, with Calanus hyperboreus also present at the northern station. The southern station was >10 °C warmer than the other stations and had the most diverse mesozooplankton assemblage, dominated by small copepods including Paracalanus spp. Microzooplankton grazing was detected only at the northern station, where it removed 97% of PP. Estimated clearance rates by C. hyperboreus and C. finmarchicus suggested that at in-situ abundance these mesozooplankton were not likely to have a major impact on phytoplankton abundance, unless locally aggregated. Although mesozooplankton grazing impact on total phytoplankton was minimal, these grazers completely removed the numerically scarce > 10 µm particles, altering the particle-size spectrum. At the southern station, grazing by the whole mesozooplankton assemblage resulted in a removal of 14% of PP, and its effect on net phytoplankton growth rate was similar irrespective of ambient light. In contrast, reduction in light availability had an approximately 3-fold greater impact on net phytoplankton growth rate than mesozooplankton grazing pressure. The low mesozooplankton grazing impact across stations suggests limited mesozooplankton-mediated vertical export of phytoplankton production. The constraints provided here on trophic transfer, as well as quantitative estimates of the relative contribution of light and grazer controls of PP and of grazer-induced shifts in particle size spectra, illuminate food web dynamics and aid in parameterizing modeling-frameworks assessing global elemental fluxes and carbon export.

Time-series of direct primary production and phytoplankton biomass in the southeastern Bering Sea: responses to cold and warm stanzas

Sub-Arctic and Arctic regions are warming faster than nearly all other areas of the global ocean, leading to significant changes in ice quality and the duration of ice-covered periods. The impacts of this warming and sea ice variability on higher trophic levels in the Bering Sea is well documented, but the effects on lower trophic levels are less well understood. Phytoplankton biomass (as chlorophyll a [chl a]) and primary and nitrogen production measurements in the Bering Sea are presented from 2006-2016, a period that covers relatively colder (2007-2012) and warmer (2014-2016) temperature regimes. In warm spring periods, relative to cold spring periods, the frequency of subsurface chl a maxima increased, but with no significant differences in integrated chl a inventories. In contrast, cold fall periods were characterized by greater integrated chl a inventories than warm fall periods. Integrated net primary production (NPP) increased from the cold period (2007-2011) to the warm period (2014-2016). The difference in patterns in chl a and NPP resulted in higher phytoplankton growth rates during warm periods. Nitrate uptake rates increased from spring to fall during cold periods, while rates decreased from spring to fall during warm periods, suggesting changes in the balance of new versus regenerated production. While changes in phenological timing cannot be ruled out, changes in phytoplankton growth rate appear more important than changes in chl a biomass underlying increasing daily NPP. This distinction directly impacts our understanding of the linkages between warming temperatures and phytoplankton production and its implications in evaluating and understanding energy flow to higher trophic levels.

Multivariate control of heterotrophic bacterial abundance and zooplankton grazing in Labrador fjords (northeastern Canada)

This study was conducted in 4 Labrador fjords (Nachvak, Saglek, Okak, and Anaktalak) during the summers of 2007 and 2013, early fall 2010, and late fall 2009. Our results show that water temperature combined with the availability of nutrients and organic substrates are the main abiotic factors controlling the abundance of heterotrophic bacteria in Labrador fjords. Bacterivory also played a crucial role, with heterotrophic bacteria exerting a significant bottom-up control on the abundance of heterotrophic nanoflagellates (r = 0.35, p < 0.05) and ciliates (r = 0.70, p < 0.01). During summer 2013, the intrinsic phytoplankton growth rate varied between <0 and 0.64 d-1, with a mean value of 0.36 d-1. The herbivory rate was highly variable, ranging from 0.01 to 0.86 d-1, with a mean value of 0.31 d-1. Grazing mortality was 6-fold higher than phytoplankton growth rate. Mean phytoplankton growth and herbivory rates in Labrador fjords were comparable to the Barents and Bering seas. The intrinsic growth rate of total heterotrophic bacteria ranged between <0 and 0.68 d-1, with a mean value of 0.30 d-1. Bacterivory varied from 0.01 to 0.95 d-1, with a mean of 0.30 d-1. Mortality due to grazing was up to 2.3 times higher than total bacterial growth rate. This study improves our understanding of the factors influencing the dynamics of heterotrophic bacteria and indicates that herbivory and bacterivory exert substantial control on microbial communities in Labrador fjords.

The influence of temperature on seasonal and inter-annual variability of some structural and functional characteristics of the phytoplankton in coastal waters of the Black Sea (Sevastopol region)

The seasonal and inter-annual variability of the phytoplankton growth rate and biomass in the coastal waters of the Black Sea near Sevastopol was studied. The nature of the seasonal dynamics and the amplitude of these parameters almost coincide. An increase in the average annual values of water temperature causes changes in the species structure of phytoplankton, its seasonal succession, and a decrease in the average annual values of the specific growth rate and phytoplankton biomass by about 2 times.

Error assessment of biogeochemical models by lower bound methods

Biogeochemical models, capturing the major feedbacks of the pelagic ecosystem of the world ocean, are today often embedded into Earth System models which are increasingly used for decision making regarding climate policies. These models contain poorly constrained parameters (e.g., maximum phytoplankton growth rate) which are typically adjusted until the model shows a reasonable behavior. Systematic approaches determine these parameters by minimizing the misfit between the model and observational data. In most common model approaches, however, the underlying functions mimicking the biogeochemical processes are non-linear and non-convex. Thus, systematic optimization algorithms are likely to get trapped in a local minimum and might lead to non-optimal results. To judge the quality of an obtained parameter estimate, we propose to determine a preferably large lower bound for the global optimum, that is relatively easy to obtain and that will help to assess the quality of an optimum, generated by an optimization algorithm. Due to the unavoidable noise component in all observations, such a lower bound is typically larger than zero. We suggest to derive such lower bounds based on typical properties of biogeochemical models (e.g., a limited number of extremes and a bounded time-derivative). We evaluate this approach with synthetic observations and demonstrate a real-world example, consisting of phytoplankton observations in the Baltic Sea.

Modeling Of the Phytoplankton Dynamics Considering the Mechanisms of Ectocrine Regulation

In the paper set of models that take into account various mechanisms for ectokrine regulation of phytoplankton growth are proposed and studied. The considered models are following: with a nonlinear dependence of phytoplankton growth rate on the concentration of the metabolite in the environment; with a metabolite, which increases both the growth rate and mortality of the phytoplankton; with two different metabolites, one of which increases the rate of growth of phytoplankton, and the second - is reduced. These models are modifications of the classical space-time model of the dynamics of phytoplankton, linking change in its density and the concentration of nutrients in space and time, which do not consider the mechanism of ectokrine growth regulation. By the means of the model it has been shown, that the metabolites secreted by algae are able to stabilize the growth of phytoplankton in conditions of excessive nutrient availability.

An early spring bloom of large diatoms in the ice-covered Saroma-ko Lagoon, Hokkaido, Japan

Saroma-ko Lagoon, the largest body of water that has complete ice coverage during winter in Japan, was not completely covered by ice in the winter of 2009. This condition is considered to be a result of the progression of global warming. A bloom of large diatoms was observed in the ice-free area between February and April. This early spring bloom seemed to have started in the latter part of January, and lasted for about three months. The maximum chlorophyll-a (Chl a) concentration of about 10 mg m−3 was observed in March, and was similar to the level of 5–20 mg m−3 previously reported for the ordinary spring bloom in Saroma-ko Lagoon. The maximum primary production of 786 mgC m−2 day−1 and the maximum Chl a-specific primary production, an index of the phytoplankton growth rate, were also found in March. Species changes from Thalassiosira spp. to Chaetoceros spp. were observed during the bloom. This early spring bloom could extend into the ordinary spring bloom period. Its duration was obviously longer than that of the spring bloom, which is typically about one month. These results show the phytoplankton condition that could be expected during winter and spring as global warming progresses.