The evolution of light and vertical mixing across a phytoplankton ice edge bloom

During summer, phytoplankton can bloom in the Arctic Ocean, both in open water and under ice, often strongly linked to the retreating ice edge. There, the surface ocean responds to steep lateral gradients in ice melt, mixing, and light input, shaping the Arctic ecosystem in unique ways not found in other regions of the world ocean. In 2016, we sampled a high-resolution grid of 135 hydrographic stations in Baffin Bay as part of the Green Edge project to study the ice-edge bloom, including turbulent vertical mixing, the under-ice light field, concentrations of inorganic nutrients, and phytoplankton biomass. We found pronounced differences between an Atlantic sector dominated by the warm West Greenland Current and an Arctic sector with surface waters originating from the Canadian archipelago. Winter overturning and thus nutrient replenishment was hampered by strong haline stratification in the Arctic domain, whereas close to the West Greenland shelf, weak stratification permitted winter mixing with high-nitrate Atlantic-derived waters. Using a space-for-time approach, we linked upper ocean dynamics to the phytoplankton bloom trailing the retreating ice edge. In a band of 60 km (or 15 days) around the ice edge, the upper ocean was especially affected by a freshened surface layer. Light climate, as evidenced by deep 0.415 mol m–2 d–1 isolumes, and vertical mixing, as quantified by shallow mixing layer depths, should have permitted significant net phytoplankton growth more than 100 km into the pack ice at ice concentrations close to 100%. Yet, under-ice biomass was relatively low at 20 mg chlorophyll-a m–2 and depth-integrated total chlorophyll-a (0–80 m) peaked at an average value of 75 mg chlorophyll-a m–2 only around 10 days after ice retreat. This phenological peak may hence have been the delayed result of much earlier bloom initiation and demonstrates the importance of temporal dynamics for constraints of Arctic marine primary production.


Introduction
This document provides supplementary material to our study on Baffin Bay hydrography and the environmental constraints during a phytoplankton bloom at the retreating ice edge. It provides more detail on some of the methods, but mostly shows more variables that may help the interested reader in getting a better overview over the immense dataset the Green Edge campaign has produced. All data are accessible at the Green Edge database (http://www.obs-vlfr.fr/proof/php/GREENEDGE/greenedge. php).

Acknowledgments
This document was created as a Jupyter Lab notebook and exported to pdf partly using Chris Sewell's ipypublish template.

More overview maps
The following true color images are meant to provide more context for the campaign, especially with regards to floe sizes and general impressions of the ice cover that are hard to capture in sea ice concentration charts. Surface chlorophyll-a and sea temperature show the phytoplankton growth and warming trailing behind the retreating ice edge.             2 Hydrography -Some additional figures     The "open water days" (OWD) parameter is defined as follows. First, we fix a time interval that captures the full sea ice variability in the region of interest, say from a date T 0 (here, taken to be doy 120, April 29) before the melt starts, to a later date T e when the ice extent has disappeared from the study area (here, doy 210, July 28). We then define the number of "open water days since sea ice retreat" (ODS) at a given location at time T 0 < t < T e by simply summing the days where c < 15% since T 0 , and the number of "Ice days until sea ice retreat" (IDU) by summing the days where c > 15% until T e . Then OWD ≡ ODS−IDU. Because the ice edge retreats steadily (does not meander much or go back and forth), there are very few sampling stations where IDU and ODS are both non-zero ( Fig. S3.1). This was one of the reasons to conduct the experiment there in the first place and facilitates the analysis of the ice edge bloom. An instance where IDU and ODS are both non-zero will occur whenever a location has open water one day, but later on becomes ice covered again before the end of the melt season.
The steady retreat of the ice edge is demonstrated mainly by two facts: 1) ODS (Open water Days Since ice retreat) vs. IDU (Ice cover Days Until ice retreat) did not have much overlap where both are non-zero, and 2) there is a good correlation between distance from the ice edge and OWD (Open Water Days, defined as OWD ≡ ODS−IDU) The plot immediately below also demonstrates that as a consequence, an alternative definition of OWD, namely OWD=IDS when ice concentration is smaller than 15%, and OWD = −IDU when ice concentration is larger than 15%, leads to largely identical values.

Underwater light measurements using the C-OPS
The C-OPS is an oceanographic instrument composed of a surface reference downwelling irradiance sensor, E 0+ d (λ), as well as a set of two free-fall profiling radiometers (maximum depth of 125 m), namely a downwelling irradiance one, E d (z, λ), and either a upwelling irradiance one, E u (z, λ), or an upwelling radiance one, L u (z, λ, φ). Here, z denotes the depth, λ the wavelength and φ the viewing angle (nadir in our case). These sensors record their respective radiometric quantity along nineteen wavelengths, spread in the visible spectrum, from 320 nm to 875 nm. An ice floe version of this instrument, the so-called IcePRO, has been designed for deployment through an auger hole in an ice floe. (Once the IcePRO has been deployed in the hole, fresh snow is shoveled back in the hole before the profiling starts. This prevents any artificial bright spot just above the sensors.) Both versions have been used during the Green Edge project. The sensors have been factory calibrated before the cruise (January 2015 and February 2016), and a custom R code following the well recognized methodology originally published by Smith and Baker (1984) has been used to process the data. Furthermore, a PAR was computed from all the available multispectral irradiances values (in and above surface, downwelling and upwelling), following Here, h is Planck's constant and c is the speed of light in vacuum, and the summing is performed using a trapezoidal method.
When the data acquisition was performed in open water, the in-water radiometric quantities were extrapolated until z = 0 − , i.e. up to just under the surface, and when the data acquisition was performed under the ice, no extrapolation was performed, and the shallowest valid data point has been kept.
From the derived PAR d (z) and PAR 0+ quantities, a transmission profile was computed as This transmission profile includes the transmission loss due to the snow/sea ice layer, plus any water layer being between surface sensor and the profiling sensor when it is at the depth z.
For no C-OPS station was it the case that it was both 1) sampled on ice and 2) AMSR2 indicated ice concentrations below 80%, indicating a good match between in-situ sampling conditions and ice concentrations inferred by AMSR2.  The combined ice+snow transmittance where available; empty circles mean that measurements were not done from the ice. Estimated from C-OPS measurements as the ratio between downwelling PAR irradiance directly under sea ice and the simultaneously measured incoming above-surface PAR.   The fuzzy c-means clustering algorithm (Ross, 2010) (implemented in python's skfuzzy library(scikitimage team)) assigns a membership value ∈[0,1] to each data point. All data points, characterized by a number of properties, are sorted into a pre-defined number of clusters (here, two).

Above-surface PAR irradiance
Consequently, based on a few select properties, each hydrographic station is assigned a value between 0 and 1, where 0 and 1 themselves correspond to exclusive membership in either cluster, and values in between represent the statistical uncertainty in the attribution (fuzziness).
The following properties were included, centered and normalized: (1) maximum temperature in the AW layer, (2) salinity at the estimated convection depth, and (3) ANP at 20~m depth, which summarize the strength of Atlantic inflow and the resulting hydrography and nutrient composition. The clustering reproduced as expected the longitudinal gradient present in the majority of all variables. 6 Winter overturning depth and the associated pre-bloom nutrient inventory     Integrating only over the full HPLC profile will underestimate integrated chl-a whenever the last HPLC value did not cover the entire fluorescence profile. In other words, integrating the calibrated chl-a fluorescence profile until the depth where fluorescence reaches 1% of its maximum value will yield a somewhat larger value.
The following two figures demonstrate that uniformly integrating HPLC-determined total chlorophyll-a to a depth of 80 m misses little overall biomass.  The differences are small, which is why we used the simple vertical integration of HPLC total chl-a in this study. The underestimation is then given by the difference between the rosette chl-a fluorescence integrated down to the maximum HPLC sampling depth, and the rosette chl-a fluorescence integrated down to the 1% fluorescence level.