AOP derived from radiometric data


1. Data processing notes

The radiometric data are acquired with a so-called C-OPS system (“Compact - Optical Profiling System”, from Biospherical Instruments, Inc), which is a set of 3 optical radiometers, each measuring radiance or irradiance signal along 19 different wavelengths, spread in the visible part of the spectrum (between 300 and 900 nm). One atmospheric reference sensor measures above surface irradiance, while 2 profiling radiometers measure the downward irradiance and either the upward irradiance or the upward radiance (only 2 of the \(E_d(z)\), \(E_u(z)\) and \(L_u(z)\) radiometers can be fitted on the profiling frame), up to a depth of 125 m. The 3 sensors record their data synchronously, at a (user defined) frequency of 15 Hz. A dynamic range of more than 8 orders of magnitude allow the C-OPS or its ice floe version, the so-called ICEPRO, to measure either very strong optical signal close to the surface without saturating, or very weak optical signal under ice floe or in very turbid waters, at greater depths, with a very good signal to noise ratio.

The code that was used to process the Green Edge data is based on a recognized and published methodology (Smith and Baker, 1984) that is also used by the NASA to define their standards for their spatial observations vicarious calibration and validation activities (Mueller and Austin, 1992, revised in 1995, then updated by Austin in 2000, 2002 and 2003). Below, only the more relevant steps of the data processing code are then summarized.


1.1. Cleaning of the data

This preliminary step consists in cleaning the raw data, i.e. removing points were the sensors were held at the surface or at any given depth instead of profiling (initial stabilization and positioning of the profiling unit, end of profiling before stopping the acquisition, etc.), and points acquired when the instrument’s tilt was more than 5 degrees. Sometimes this 5 degrees threshold was increased by a few degrees, to avoid removing too many points. Indeed, for few profiles that were recorded under very strong tidal current conditions, a 5 degrees cutoff was discarding too many points, and not enough points were remaining for a good fit to be applied (see below for the smoothing step).


1.2. Application of a depth offset correction

A depth offset is applied to each of the sensors, as their optical detectors are not aligned with the deploying frame depth sensor diaphragm (the corrected offset is either null for the atmospheric reference, or up to 30 cm approximately for the wet sensors).


1.3. Atmospheric signal processing

The atmoshperic signal is processed as being a profiling one. This may seem odd, but one must see it as a trick to assess the sky stability (or instability) as a function of the depth at which the wet sensors were (the so-called “wet” sensors are the sensors mounted on the deploying frame). This is possible as the atmospheric light field and the water column light field signals are logged with the same timestamps, i.e. they are synchronized. For example, if a cloud decreased the atmospheric light intensity while the profiling sensors were acquiring data at 45 m depth, then a graphical view of the atmospheric signal as a function of the depth of the wet unit will show it straightaway, in an instinctive way. Then, an atmospheric correction is applied to all the profiling data, to correct them for the time variation of the atmospheric signal. This correction is performed for all the wavelengths separately, and consists of normalizing the profiling data by the time-varying atmospheric signal, as follows (see for example Antoine et al., 2013):

\[\Gamma(z, \lambda, t_0) = \Gamma(z, \lambda, t)\frac{E_d(0^+,\lambda,t_0)}{E_d(0^+,\lambda,t)} \space \space \space (eq.1)\]

where \(\Gamma\) is any one of the underwater radiometric quantities (radiance or irradiance, downward or upward), \(z\) is the depth, \(\lambda\) is the wavelength and \(t\) and \(t_0\) represents time. With this normalization, \(\Gamma(z,\lambda,t_0)\) represents the radiometric quantity as it would have been recorded at all depths \(z\) at the same time, \(t_0\) (\(t_0\) is usually taken at the start of the profile), and corrects very well slow to moderated atmospheric condition changes (in time and in amplitude).


1.4. Loess smoothing

The profiles are then smoothed using a local polynomial (low degree) method, the so-called “loess” method. The span of the regression, i.e. the extent of the data that are considered around the point of interest in the profile, is a user-defined parameter, that is manually tuned for each profile during the data processing. If the span is too high (too many points are taken into account on each side of the point of interest), the smoothed / fitted data may not describe any quick change / fine feature in the light profile. If the span is too small though, the fitted data might be noisy then spiky. The goal here is to set the span as low as possible, to resolve properly (i.e. with no noise) small scale processes as much as possible, which is oftentimes possible, due to the C-OPS design (high acquisition frequency). For very few profiles, a couple of spikes remain, but if the span would have been increased, fine features in the water column wouldn’t have been resolved. The user might want to manually remove these spikes. The smoothing is performed on a depth grid that has a increasing step toward greater depths, so that it is highly resolved in the top of the water column, and less resolved as the depth increases. It is defined in the table 1 below:

\[ \begin{array}[ht] {ccc} between~z_1... & ...and~z_2 & the~step~[m]~is \\ 0 & 1 & 0.01 \\ 1 & 2 & 0.02 \\ 2 & 5 & 0.05 \\ 5 & 10 & 0.10 \\ 10 & 20 & 0.20 \\ 20 & 50 & 0.50 \\ 50 & 100 & 1.00 \\ 100 & 200 & 2.00 \\ \end{array} \] \[\textbf{Table 1.} \space Depth \space grid \space step \space definition\]

Note 1: During the Green Edge data acquisition cruises, 2 types of profiles were recorded. The more frequent profile type is an under-ice one, where the deploying frame was lowered in the water column through a hole in the ice. The second one, essentially performed from the Amundsen or her barge in open waters or in leads in the floe, was performed by letting the frame free fall in the water column. One major difference between these 2 types of profiles processing is the smoothing. Indeed, the ice-based profiles are not extrapolated to a \(z=0^-\) depth, but rather to the closest depth from the depth grid available. For example, if the profile did start at a depth of 1.432 m, the fit will be defined from 1.44 m, and the values above this depth will be set at -999. Indeed, it would not make sense to extrapolate the light intensity from the water column into the ice, where the absorption and scattering properties are more likely very different. For open water measurements, the first available depth is very often, if not always, few cm from the surface, so that an extrapolation to just below the surface is meaningful. In other words, the fit is performed either from just below the ice surface, or extrapolated to just below the sea surface.


1.5. Corrupted data debugging

For some profiles, technical issues occured during or just before the data acquisition. The two sub-sections below give details about these issues and about the way they have been addressed.


      1.5.1. Corrupted \(E_d^{0+}(560)\)

For all the sampling days at the ice camp in 2015, the downwelling irradiance signal at 560 nm was corrupted. The optical signal between the two neighbour channels, namely 555 nm and 589 nm has been interpolated at 560 nm to replace the corrupted one. As 560 nm is very close to 555 nm, the interpolation is usually very close to the signal at 555 nm (as it is for not corrupted data set). The impact of this issue on the data quality is then minor.


      1.5.2. Asynchronous atmospherical / profile data acquisition

On 29 June 2016, on the ice camp, the deck unit of the C-OPS instrument was not working properly and prevented any data acquisition this day. Before the issue could have been resolved completely (on 12 July), only one of the two acquisition ports was working, preventing synchronous data acquisition between the atmospheric reference and the profiling radiometers. The users had to sample the atmospheric reference first, then they had to disconnect this sensor before they could record the profiling data. The data processing code has been modified to take this temporary issue into account. Namely, the atmospheric irradiances were averaged over each field day sampling period. The figure 1 below show the evolution of each channel of the atmospheric reference as a function of the minutes elapsed since the start of the C-OPS sampling effort, for the 1st, 4th, 6th and 8th July 2016. It can be seen that the atmospheric signal for the 1st and the 4th July was noisy but rather stable, that it was not noisy and stable for the 6th July and that it was not noisy but rather unstable for the 8th July. It is not ideal to take the average, but it is the only option available to process these profiles.

\[ \textbf{Figure 1.} \space Averaged \space atmospheric \space downwelling \space irradiance \]


1.6. PAR estimations

Three different PAR values are estimated from the radiometric data; They are explained below.


      1.6.1. Instantaneous PAR

From the \(E_d(z,\lambda)\) and \(E_u(z,\lambda)\) fitted profiles, the \(PAR(z)\) values are estimated by integrating spectrally a spline interpolation of the spectra at each depth of the grid:\[PAR(z_{grid})=\int_{\lambda}E_{d/u}(z_{grid},\lambda)\times d\lambda \space \space \space (eq.2)\] The spline interpolation is performed on a 1 nm mesh. Before this integral computation, the irradiance are converted from \(Watts/m^2.nm\) to \(quanta/m^2.nm.s\), using the Planck-Einstein relation, which gives the energy of one photon, \(E_{1~quanta}[W.s]=\frac{h{\times}c}{\lambda}\). The conversion is as follows:

\[E_{d/u}[quanta/m^2.nm.s]=E_{d/u}[Watt/m^2.nm]\frac{\lambda}{h \times c} \space \space \space (eq.3)\]

Finally, the PAR is converted from \(quanta/m^2.s\) to \(Einstein/m^2.s\) by dividing this last value by Avogadro’s number, \(N_A\).

As the atmospheric irradiances \(E_d^{0+}\) can be seen as a function of the wet unit depth (see section 1.3), a \(PAR_0^+\) is also estimated using the same method as above, i.e. the \(PAR_0^+\) estimations are given as a function of the profiling radiometers depth.


      1.6.2. Percentage of PAR

From these instantaneous PAR profiles, percent PAR profiles, which are light transmission profiles, are estimated as percentage of the incident one. Due to the atmospheric correction applied (see section 1.3), the first atmospheric PAR in time, i.e. \(PAR_0^+(t_0)\) at the first available depth, is taken as the normalization value used to estimate the percent of light reaching each depth, using: \[PAR_{d/u}(z,t)[\%] = \frac{PAR_{d/u}(z,t)}{PAR_0^+(t_0)} \times 100 \space \space \space (eq.4)\]

Note that the value of \(PAR_0^+(t_0)\) taken as a reference to compute this percentage of light is taken from the smoothed data, not from the measured ones.


      1.6.3. Daily PAR

Finally, a daily PAR estimate is also computed. When no other measurements, more reliable, were taken or were available (see below), a daily PAR is estimated from the C-OPS / ICEPRO data. To do so, the atmospheric PAR estimate (sections 1.6.1 and 1.6.2 above), which corresponds to a \(PAR_0^+(t_0)\) value at one precise moment in the day, is propagated all along the day, with a function of the cosine of the Sun zenith angle, \(sza\). Specifically, the punctual PAR estimate from fitted data at time \(t_0\), \(PAR_0^+(t_0)\) is normalized by the cosine of the Sun zenith angle at time \(t_0\), and multiplied by time series of cosines of Sun zenith angle along one day, by 7min 30sec minutes increments. This time series is then integrated using a trapezoidal method: \[PAR_{daily}^{0+}[Ein/m^2.day]=\sum_{t_i=0h}^{t_i = 24h}\frac{PAR_0^+(t_0)}{cos\big(sza(t_0)\big)}\times cos\big(sza(t_i)\big) \times \Delta t_i \space \space \space (eq.5)\]

The figure 2 below shows the punctual PAR estimate (red dot) at the time of the profile acquisition (red dashed line), and its daily propagation (black dots). The daily PAR is estimated as being the area below the curve.

\[ \textbf{Figure 2.} \space Daily \space PAR \space estimate \]

In the output data file, this daily PAR value is provided, as well as an estimate of the daily PAR values at depth, simply by multiplying the percentage of PAR estimated above (see section 1.6.2 above) by this daily atmospheric PAR estimate.

When other sources of daily PAR estimates were available, as they are more reliable (because they were acquiring data either continuously or all the day with a higher time resolution), they were used as a first daily PAR estimate. See below for details.

Note 2: the cosine of the Sun zenith angle function taken to propagate the instantaneous PAR value along one day supposes that the sky conditions do not change all over the course of the day, which is, naturally, not always (not often) representative of the real atmospheric conditions. Daily atmospheric PAR values have been estimated from other sources of measurements, which were acquiring data continuously (by a pyranometer for example). When available, these data were taken as a preferred source of daily PAR value (it was mostly the case for the Amundsen based stations). These so-called “in-situ” data were sometimes available. A maximum distance of 40 km has been set (40 km corresponds to 2 hours sailing at 10 knots, roughly). When the distance is less than this threshold, the in-situ daily PAR value has been kept (which is most often the case when in-situ data are available), and it was discarded when the distance was above this threshold (which only happened for 3 profiles). Note that, even when discarded by the distance threshold test, the in-situ daily PAR estimates are still available, in the log file (see below), and that the user could still replace the C-OPS one with this in-situ one if desired.


2. (Other) estimations of downwelling PAR irradiance above the sea or ice surface

Photosynthetically available radiation (PAR) is defined in this section as the integral irradiance between 400 and 700 nm (units of \({\mu}mol\space quanta.m^2.s^{-1}\)). Irradiance was measured with different instruments and types of sensors (PAR or pyranometer) during the 2015 ice camp, the 2016 ice camp and the 2016 Amundsen cruise. In addition, measurements were not available during the entire duration of the ice camps. Here we describe how a homogeneous PAR product was created by (i) treating in situ time series in a consistent way, and (ii) using satellite observations of the atmosphere (sensor MODIS Aqua) coupled to an atmospheric radiative transfer model.


2.1. In-situ PAR measurements

      2.1.1. Ice camp 2015

A CNR4 instrument (Kipp & Zonen) was used to measure downwelling shortwave irradiance (K) at 1 \(min^{-1}\) frequency. K is defined here as the total energy (\(W.m^{-2}\)) between 300 nm and 2800 nm (see document GE2015_IC_NetRadiation.doc, responsible scientist Brent Else). K was converted to PAR irradiance assuming that PAR accounts for a fraction of downwelling K that oscillates between 41 and 44% (Kirk 2011, see section 2.2.2), and applying afterwards a conversion factor (cf) to convert radiant energy to quantum units, \[PAR=\frac{fract \times K \times \ cf \times 10^6}{N_A} \space \space \space (eq.6)\] where cf = 2.77 x 1018 \(quanta.s^{-1}.W^{-1}\) (Morel & Smith 1974), \(N_A\) is the Avogadro number, and the factor \(10^6\) is used to convert from \(mol \space quanta\) to \(\mu mol \space quanta\). The PAR fraction “fract” was modulated as a function of the solar zenith angle (\(sza\)) following Kirk (2011) and references therein: fract was set to 0.41 for \(sza \gt 80 ^\circ\), to 0.44 for \(sza \lt 50 ^\circ\), and linearly interpolated between 0.41 and 0.44 as a function of \(sza\) for \(50^\circ \lt sza \lt 80^\circ\). It must be noted that the constant \(cf\) used here is only valid for the spectral distribution of irradiance above the sea surface. The PAR time series is shown in figure 3a, and the corresponding histogram in figure 4. The comparison of the PAR histograms for 2015 and 2016 (when PAR was measured with a PAR-specific sensor) suggests that the simple conversion described here is appropriate.


\[ \textbf{Figure 3.} \space PAR \space time \space series \space at \space the \space ice \space camp \space in \space 2015 \space (top) \space and \space 2016 \space (bottom) \\ comparing \space measured \space and \space modeled \space PAR. \space The \space larger \space range \space of \space the \space data \space in \space 2015 \space is \\ due \space to \space the \space higher \space measurement \space frequency \space (1 \space minute \space in \space 2015 \space and \space 1 \space hour \space in \space 2016) \]


\[ \textbf{Figure 4.} \space Histograms \space of \space the \space PAR \space time \space series \space for \space the \space ice \space camps \space and \space the \space Amundsen, \\ with \space the \space frequency \space (y \space axis) \space in \space linear \space scale \space (left) \space and \space log10 \space scale \space (right) \]


      2.1.2. Ice camp 2016

A LI-190SA instrument (Li-COR) was used to measure downwelling PAR at 1 \(h^{-1}\) frequency (see document PAR_META_INFO_GE2016-ICECAMP_cyber.doc, responsible scientist Joannie Ferland). The PAR time series is shown in Fig. 3B, and the corresponding histogram in Fig. 4.


      2.1.3. CCGS Amundsen 2016 legs 1A and 1B (Green Edge cruise)

A Kipp & Zonen PAR Lite was used to measure downwelling PAR at 1 \(min^{-1}\) frequency (see document GE2016-AMUNDSEN_TonyaBurgers_RAD.doc, responsible scientist Tonya Burgers). The corresponding histogram is shown in Fig. 4. No figure is shown for the PAR time series because, in fact, it also reflects ship movement.


2.2. PAR modeled with radiative transfer and MODIS atmosphere data

A plane-parallel atmospheric radiative transfer model, SBDART (Richiazzi et al., 1998) was used to compute a look-up table (LUT) (Simon Bélanger, UQAR; see also Laliberté et al. 2016). Then, above-surface spectral irradiance, \(E_d^{0+}(\lambda)\), was calculated by interpolating in the LUT the satellite-observed atmospheric properties at each station. The pre-computed LUT used has the following dimensions:

  • Wavelength, \(\lambda\): between 290 and 700 nm at 5-nm intervals (83 values),
  • Solar zenith angle, \(sza\): 0 to 90 degrees at 5-degrees intervals (19 values),
  • Total ozone column: 100 to 550 Dobson units (DU) at 50-DU intervals (10 values),
  • Cloud optical thickness, \(COT\): 0 to 64 in powers of 2 (\(2^n\)) intervals (8 values),
  • Surface albedo \(A\): 0.05 to 0.95 in 0.15 intervals (7 values).

For a given station and time, \(E_d^{0+}(\lambda)\) is calculated under two situations: for a cloud-free sky, \(E_{d, \space CLEAR}^{0+}(\lambda)\) (COT set to 0), and for the satellite observed COT, \(E_{d, \space CLOUD}^{0+}(\lambda)\). The final \(E_d^{0+}(\lambda)\) output is calculated as the mean of cloudy and cloud-free \(E_d^{0+}(\lambda)\) weighted by the MODIS-Aqua daily cloud fraction (CF), which expresses the proportion of the time spent under cloudy conditions:

\[E_d^{0+}(\lambda) = E_{d, \space CLOUD}^{0+}(\lambda) × CF + E_{d, \space CLOUD}^{0+}(\lambda) × (1 – CF) \space \space \space (eq.7)\]

This approach is indeed a simplification but, as shown in section 2.3, the results agree well with most observations.

The MODIS-Aqua atmospheric product used has a temporal resolution of 1 day and a spatial resolution of 1 degree. For latitudes between 67 and 70 degrees N, this corresponds to a footprint of ca. 38–43 km longitude × 111 km latitude. This is indeed very coarse compared to the stations grid of the Amundsen cruise, and cannot represent local weather features of the ice camp (discussed in section 4).

For a given day, spectral irradiance was calculated every hour (UTC time) to account for variations in \(sza\). Hourly \(E_d^{0+}(\lambda)\) corresponds to the middle of each 1-h period, e.g. 12:30 for the 12:00-13:00 period, and so on. For the 2015 and 2016 ice camps, daily \(E_d^{0+}(\lambda)\) files at hourly resolution were produced for the period between April 1st and August 15th. For the Amundsen cruise, four daily files were produced for each CTD cast, spanning between 2 days before the station to the day after. PAR was finally calculated as the spectral integral between 400 and 700 nm by the sum of trapezoids.


      2.2.1. Surface albedo effects and estimation

Here we define albedo (A) as the proportion of the incident light (integrated in the PAR range, 400 to 700 nm) that is reflected by a surface. Ice and snow have very high albedo (up to 95% in cold snow) compared to seawater (ca. 5%). Thus, surface albedo is a key player in radiative transfer in polar environments. Regarding the downwelling component of irradiance, this is particularly important under cloudy and foggy conditions, when multiple scattering can “trap” photons between the ice/snow surface and the cloudy layer. This enhances downwelling irradiance compared to an ice- and snow-free situation, where the water body would absorb most photons transmitted through the atmosphere.

To account for albedo effects here we applied an extremely simple parameterization. Total albedo (A) was calculated as the average of the ice/snow albedo (called \(A_{IS}\)) and that of open water (\(A_W\)) weighted by the sea ice concentration (SIC; obtained from the satellite sensor AMSR-2, see document READ-ME_ice_history.pdf by Philippe Massicotte):

\[A = A_{IS} × SIC + A_W × (1 – SIC) \space \space \space ( 8)\]

Open water was assigned a fixed albedo, \(A_W = 0.10\), (chosen because it optimized model predictions compared to lower albedos) whereas the ice/snow surface was assigned one of the three following albedo categories (Perovich et al. 2002 and references therein; Perovich & Polashenski 2012):

  • Cold snow: \(A_{IS} = 0.85\),
  • Melting snow: \(A_{IS} = 0.7\),
  • Melting and ponded ice: \(A_{IS} = 0.5\).

We developed a simple threshold algorithm to assign each station to one of the three \(A_{IS}\) categories. However, we used slightly different criteria in the ice camp and the Amundsen cruise in order to optimize the agreement between modeled and measured PAR.

For the ice camp:

  • Cold snow was assigned if \(SIC \ge 0.9\) \(\underline{and \space either}\) the date was earlier than June 1st \(\underline{or}\) snow depth \(\ge\) 0.1 m. This category represents prevailing winter and spring conditions. In practice, \(A_{IS}\) was set to 0.85 for day of year \(\le\) 167 (mid June).
  • Melting snow was assigned \(\underline{if}\) “cold snow = FALSE” and SIC \(\ge\) 0.75, \(\underline{and \space either}\) we were in the month of June \(\underline{or}\) snow depth \(\ge\) 0.1 m. This category represents the short period during which snow melts but melt ponds are not yet widespread. In practice, \(A_{IS}\) was set to 0.7 for day of year between 168 and 182.
  • Melting or ponded ice was assigned \(\underline{if}\) “cold snow = FALSE” \(\underline{and}\) “melting snow = FALSE”. In practice, \(A_{IS}\) was set to 0.5 for day of year \(\ge\) 183 (first of July).

The algorithm was designed after examining the time series of snow and ice depth at the ice camp, as well as the studies of Perovich et al. 2002, Perovich & Polashenski 2012 and Flocco et al. (2012). The latter study indicates that, at pan-Arctic level, melt pond formation rate is maximal in mid June and decelerates clearly in July, and melt pond area starts decreasing by mid July. Images taken from drone transects at the ice camp during the melt pond season suggest that albedo varied roughly between 0.3 and 0.7 (Simon Lambert-Girard, pers. comm.), so that the value of 0.5 for melting/ponded ice seems adequate.

For the Amundsen cruise:

  • \(A_{IS}\) = 0.7, corresponding to melting snow, was assigned to all stations. This constant albedo yielded better results than the more complex threshold criteria used for the ice camp.

Although our parameterization of ice albedo is obviously an oversimplification, and the threshold criteria and albedo values somewhat subjective, it helped improve the modeled PAR significantly. For example, for the Amundsen 2016 cruise stations with heavy ice cover (SIC >80%), downwelling PAR would be underestimated by a median 40% if we failed to account for ice albedo, whereas the median underestimation is reduced to ca. 5% when accounting for surface albedo. This slight bias is probably well within other sources of uncertainty in modeled PAR; chiefly, the spatial scale mismatch between in situ and satellite measurements. Overall, the model reproduced the measurements satisfactorily.


2.3. PAR metrics and comparison between in situ and modeled data

In situ and modeled PAR time series were processed in exactly the same way to extract light history metrics relevant to biological and photochemical processes. The metrics extracted can be grouped as follows:

With respect to the sampling time:

  • Instantaneous irradiance (closest measurement),
  • Instantaneous irradiance interpolated between the two closest measurements,
  • Mean irradiance during the previous hour,
  • Mean irradiance during the previous 3h,
  • Mean irradiance during the previous 24h,
  • Mean irradiance during the previous 48h.

With respect to the local day:

  • Irradiance during the hour centered at noontime,
  • Mean daily irradiance.

The amount of data used to calculate the means were recorded and used for quality control. Mean values were removed if they were affected by data gaps during more than 20% of the measurement period. Note also that, for time series with hourly resolution, the instantaneous and hourly mean irradiance may be the same. Yet, we retained the instantaneous irradiance in the database because it may give additional information when the measurement frequency was 1 \(min^{-1}\) (ice camp 2015 and Amundsen 2016).

In situ and modeled PAR metrics were compared during the entire measurement periods and also at ice camp stations and Amundsen CTD casts. A visual comparison for the entire measurement periods at the ice camp can be seen in figure 3. Figures 5 and 6 show scatterplots comparing modeled and measured PAR. The statistics for the modeled vs. measured PAR are compiled in Table 2. Note that in the Amundsen cruise the comparison between modeled and measured PAR degrades with increasing integration times due to ship movement. The opposite is true at the ice camp, where longer integration periods yielded more accurate estimations.


\[ \textbf{Figure 5.} \space Scatterplots \space for \space the \space ice \space camp \space and \space the \space periods \space of \space 24h \space and \space 48h \space prior \space to \space sampling. \\ Left, \space 2015; \space right, \space 2016 \]


\[ \textbf{Figure 6.} \space Scatterplot \space for \space the \space Amundsen \space cruise \space and \space a \space period \space of \space 3h \space prior \space to \space sampling \space (left), \\ model/measurement \space ratio \space sorted \space by \space sea \space ice \space concentration \space (center), \\ and \space boxplot \space of \space the \space model/measurement \space ratio \space sorted \space by \space five \space sea \space ice \space concentration \space categories \space (right) \]


\[\textbf{Table 2.} \space Model \space skill\space statistics \space used \space to \space evaluate \space the \space performance \space of \space the \space atmospheric \\ radiative \space transfer \space model \space compared \space to \space in \space situ \space PAR \space measurements. \space Pearson's \space correlation \\coefficient \space (r), \space root-mean-square \space error \space (RMSE, \space in \space \mu mol \space photons.m^{-2}.s^-1), \space mean \space bias \space (\%), \\ mean \space absolute \space percentage \space error \space (MAPE), \space and \space model \space vs. \space measurements \space slope \space using \space either \\ ordinary \space linear \space least \space squares \space regression \space (Slope) \space or \space a \space type \space I \space major \space axis \space regression \space (SlopeMA)\\ N \space refers \space to \space the \space number \space of \space comparisons, \space i.e. \space The \space number \space of \space 3h, \space 24h \space or \space 48h \\periods \space where \space in \space situ \space data \space were \space available \space for \space comparison \space with \space SBDART \space outputs\]


3. Data delivery structures and formats

3.1. Files and directories description

      3.1.1. Files contents

Each profile’s data are provided in several files, all located in a separated sub-directory. Each of these sub-directories is named after the profile’s name (see below for the naming convention). The data files are provided both as ASCII (*.txt - tab. separated values) and as BINARY (*.rds - R binary files format) files. In each profile’s subdirectory, the following files are provided:

  1. d.fit.v.01.txt (ASCII) and d.fit.v.01.rds (BINARY), giving the atmospheric and the downward data,
  2. u.fit.v.01.txt (ASCII) or u.fit.v.01.rds (BINARY), giving the atmospheric and the upward data.

Notes:

  • Downward and upward sensors are not at the exact same depth on the deployment frame (the offset could be up to 30 cm). To avoid getting several depth columns in the data, and to avoid too heavy files, the downward data and the upward data have been split in separate files, into which atmospheric data have been copied. A user who is mainly interested in the downward light data would then just upload and use the downward data files,
  • The version of the data processing is indicated in these files filenames, so that if a new version is made available in the future, this version number will be amended.

The files include the parameters listed below. Note that they are not necessarily provided in the order of the list below (because of 2 output file organizations, see below), and also note that the water column data are all given on the depth grid used for the smoothing (see section 1.4 for details on the depth grid):

  1. The profile filename, with the following convention: GE201X.CCCC_INST_YYMMDD_CAST_XXX, where:
    • X stands for the year, i.e. 5 or 6 (for 2015 or 2016, resp.),
    • CCCC stands for the cruise. The values are “ICMP” for “Ice Camp”, “AMIS” for “Amundsen - Ice Stations” and “AMOW” for “Amundsen - Open Waters (or leads)”,
    • INST stands for the instrument, i.e. “COPS” for the open water based measurements (with the “C-OPS”), or “ICEP” for the ice based measurements (with the “ICEPRO”),
    • YYMMDD stands for the date, with YY being the year, MM the month and DD the day (all on 2 digits),
    • XXX stands for the cast number in that day,
  2. A flag, called flag.quality, which is set to “TRUE” when the profile’s data were synchronised, and set to “FALSE” for the few profiles for which the instrument’s deck box was damaged, and for which the atmospheric data were not acquired at the same time than the water column ones (see section 1.5.2.),
  3. The date, either as 6 different columns for the year, month, day, hour, minute and second (ref. UTC), and also as a unique column with the date and time in a POSIX format,
  4. The latitude and the longitude (decimal degrees),
  5. The Sun zenith angle at the time of the profile acquisition (degrees),
  6. The hole type used for the sampling. This is relevant only for a part of the ice camp 2015 sampling efforts, during which two different holes in the ice were sampled. The first one was a hole drilled in a floe that was covered with a thin layer of snow (“low snow hole”), the other one was drilled in a floe that was covered with a thick layer of snow (“high snow hole”). For all the other profiles, this parameter has been set to “-999” as it was irrelevant,
  7. The loess smoothed (see section 1.4) downward irradiance above the sea surface or above the ice surface, \(E_d^{0+}(\lambda)\), in the 19 available wavelengths ( \({\mu}W/cm^2.nm\) ),
  8. The estimated instantaneous atmospheric PAR, estimated from the above \(E_{d-fit}^{0+}(\lambda)\) spectra, \(PAR_0(t)\), in \({\mu}Ein/m^2.s\),
  9. The estimated daily atmospheric PAR, \(PAR_{daily}^0\) (\(Ein/m^2.day\)). This value is estimated firstly from other PAR sensors or estimations, when available. These are data acquired by continuous PAR measurements or by continuous pyranometer measurements. As these data were collected continuously over the day, they are more accurate. When this first choice is not available, the second one is retrieving MODIS satellite observations coupled with a radiative transfer model based LUT data (see section 2.2.). These two first choices were almost always available, except for a few profiles were the C-OPS measurements were performed more than 40 km away, in which case the daily PAR value above the surface has been estimated from the above C-OPS based estimations of \(PAR_0(t)\). The First available value of \(PAR_0(t = 0)\) is taken for this estimate. This can be justified by the fact that the first values of the PAR (time wise) are the ones that are the most important to estimate the light percentage profile. Indeed, the wet unit records its data at the top of the water column, where the signal is the most important, at the beginning of the profile. Furthermore, the profile dara are corrected for the atmospheric signal variations (see section 1.3), so that if the atmospheric light conditions are not too heterogeneous, the atmospheric punctual PAR value taken as a reference should not affect the daily estimate that much,
  10. A flag, called flag.PAR, giving the source used to estimate the daily downwelling PAR above the surface (see point 9 above). This can be “IN-SITU” (continuous PAR sensor or pyranometer), “SBDART” (MODIS obs. / RT LUTs) or “C-OPS” (C-OPS based),
  11. The depth of the profiling radiometer, which is either the downward irradiance, or the upward [ir]radiance meters (m),
  12. The loess smoothed (see section 1.4) downward irradiance at depth, \(E_{d-fit}(\lambda,z)\), or the upward irradiance at depth, \(E_{u-fit}(\lambda,z)\), or the upward radiance at depth, \(L_{u-fit}(\lambda,z)\), in the 19 available wavelengths ( \({\mu}W/cm^2.nm\) for irradiances, \({\mu}W/cm^2.nm.sr\) for radiances). The files d.fit.v.01.txt and d.fit.v.01.rds give the atmospheric irradiances and the downward irradiances at depth values, while the files u.fit.v.01.txt and u.fit.v.01.rds give the atmospheric irradiances and either the upward irradiances at depth, or the upward radiances at depth, depending on which sensor was mounted of the profiling frame,
  13. The estimated instantaneous PAR at depth, estimated by the profiling radiometer’s data, \(PAR_d(z)\), or \(PAR_u(z)\) both in \({\mu}Ein/m^2.s\). Note that the PAR is estimated only from irradiance values. It means that when the upward radiance sensor is fitted on the profiling frame, no upward PAR value is estimated from it,
  14. The above estimated instantaneous PAR at depth, divided by the first atmospheric PAR value, to get the percentage of light as a function of the depth, \(PAR_{d - percent}(z) ~or~ PAR_{u - percent}(z)(\%)\),
  15. The estimated daily at depth, \(PAR_d^{daily}(z)\) or \(PAR_u^{daily}(z)\), both in \(Ein/m^2.day\), which is computed as the PAR above the surface multiplied by the previous light transmission,
  16. The estimated PAR at depth integrated during 1 hour around the solar (local) noon, \(PAR_{d/u}^{\space noon \space 1h \space LOC}(z)\),
  17. The estimated PAR at depth integrated during 1 hour before the closest rosette / CTD station (closest in time), \(PAR_{d/u}^{\space p1h}(z)\),
  18. The estimated PAR at depth integrated during 3 hours before the closest rosette / CTD station (closest in time), \(PAR_{d/u}^{\space p3h}(z)\),
  19. The estimated PAR at depth integrated during 24 hours before the closest rosette / CTD station (closest in time), \(PAR_{d/u}^{\space p24h}(z)\),
  20. The estimated PAR at depth integrated during 48 hours before the closest rosette / CTD station (closest in time), \(PAR_{d/u}^{\space p48h}(z)\).

As briefly mentioned above, the files are provided with 2 different file organizations:

  1. The first one is a so-called “Excel format”, which gives all the above parameters for every profiling frame depth. It means that there is one line per depth, giving the 19 wavelengths of atmospheric downward irradiances, the 19 wavelengths of the downward irradiance or upward [ir]radiance and the several PAR estimates, all in different columns,
  2. The second one is a so-called “normal form”, or “ODV format” (i.e. as per the “TidyR” way of organizing data for the R community), which gives one parameter per column, and one observation per line. With this output format, the irradiances are not given along 19 columns, but rather as 3 columns only: “signal.type” (which can be “ed0” or “edz” or “euz” or “luz”), then the “wavelength”, then the “signal.value”.

Note that whatever the output file organization is (Excel or normal form), both the ASCII and the BINARY formats are provided.

Below is the exhaustive list of the parameters names and description inside the provided files (order differs):

  • profile.filename: file name of the profile, see the file naming convention above,
  • flag.quality: giving the quality of the profile, see above,
  • year [UTC]: year (UTC),
  • month [UTC]: month (UTC),
  • day [UTC]: day (UTC),
  • hour [UTC]: hour (UTC),
  • minute [UTC]: minute (UTC),
  • second [UTC]: second (UTC),
  • POSIXct.date [UTC]: POSIX date and time (UTC),
  • latitude [dec.degrees]: latitude, North positive,
  • longitude [dec.degrees]: longitude, East positive,
  • sun.zenith.angle [degrees]: angle between the Sun and the zenith (i.e. a vertical line starting at the acquisition point),
  • hole.type: type of the hole into which the ICEPRO was lowered - only for the ice camp 2015 cruise. “H” stands for the high snow site, “L” for the low snow site, “-999” in any other case,
  • ed0.fit.LLL [mu.W.cm-2.nm-1]: atmospheric (i.e. above surface) downward irradiance at wavelength LLL nm. These are the smoothed values (using a Loess local regression, see section 1.4),
  • PAR.0.fit [mu.Ein.m-2.s-1]: estimate of the instantaneous atmospheric downward PAR, from the ed0.fit.LLL data,
  • PAR.0.fit.daily [Ein.m-2.day-1]: estimate of the daily atmospheric downward PAR (1st choice: from continuous in situ data, 2nd choice: from satellite observations / model, 3rd choice: from the PAR.0.fit data estimated from C-OPS data),
  • flag.PAR, giving the data source used to estimate the above surface downwelling PAR (see above),
  • depth.edz [m]: depth of the downward irradiance profiling radiometer,
  • depth.euz [m]: depth of the upward irradiance profiling radiometer,
  • depth.luz [m]: depth of the upward radiance profiling radiometer,
  • edz.fit.LLL [mu.W.cm-2.nm-1]: at depth downward irradiance at wavelength LLL nm (again, smoothed values),
  • euz.fit.LLL [mu.W.cm-2.nm-1]: at depth upward irradiance at wavelength LLL nm (again, smoothed values),
  • luz.fit.LLL [mu.W.cm-2.nm-1.sr-1]: at depth upward radiance at wavelength LLL nm (again, smoothed values),
  • PAR.d.fit [mu.Ein.m-2.s-1]: at depth estimated instantaneous downward PAR, from fitted downward irradiance at depth,
  • PAR.u.fit [mu.Ein.m-2.s-1]: at depth estimated instantaneous upward PAR, from fitted upward irradiance at depth,
  • PAR.d.fit.percent [percent]: at depth estimated downward PAR percentage, i.e. \(\frac{PAR.d.fit(z)}{PAR.0.fit}\),
  • PAR.u.fit.percent [percent]: at depth estimated upward PAR percentage, i.e. \(\frac{PAR.u.fit(z)}{PAR.0.fit}\),
  • PAR.d.fit.daily [Ein.m-2.day,1]: at depth estimated daily downward PAR, i.e. \(PAR.0.fit.daily\times\frac{PAR.d.fit(z)}{PAR.0.fit}\),
  • PAR.u.fit.daily [Ein.m-2.day,1]: at depth estimated daily downward PAR, i.e. \(PAR.0.fit.daily\times\frac{PAR.u.fit(z)}{PAR.0.fit}\),
  • PAR.d.noon1hLOC [Ein.m-2.1h-1]: at depth estimated downward PAR integrated over 1 hour around the solar (local) noon, \(PAR_d^{\space noon \space 1h \space LOC}(z)\),
  • PAR.u.noon1hLOC [Ein.m-2.1h-1]: at depth estimated downward PAR integrated over 1 hour around the solar (local) noon, \(PAR_d^{\space noon \space 1h \space LOC}(z)\),
  • PAR.d.p1h [Ein.m-2.1h-1]: at depth estimated downward PAR integrated over 1 hour before the closest rosette / CTD station (closest in time), \(PAR_d^{\space p1h}(z)\),
  • PAR.u.p1h [Ein.m-2.1h-1]: at depth estimated upward PAR integrated over 1 hour before the closest rosette / CTD station (closest in time), \(PAR_u^{\space p1h}(z)\),
  • PAR.d.p3h [Ein.m-2.3h-1]: at depth estimated downward PAR integrated over 3 hours before the closest rosette / CTD station (closest in time), \(PAR_d^{\space p3h}(z)\),
  • PAR.u.p3h [Ein.m-2.3h-1]: at depth estimated upward PAR integrated over 3 hours before the closest rosette / CTD station (closest in time), \(PAR_u^{\space p3h}(z)\),
  • PAR.d.p24h [Ein.m-2.24h-1]: at depth estimated downward PAR integrated over 24 hours before the closest rosette / CTD station (closest in time), \(PAR_d^{\space p24h}(z)\),
  • PAR.u.p24h [Ein.m-2.24h-1]: at depth estimated upward PAR integrated over 24 hours before the closest rosette / CTD station (closest in time), \(PAR_u^{\space p24h}(z)\),
  • PAR.d.p48h [Ein.m-2.48h-1]: at depth estimated downward PAR integrated over 48 hours before the closest rosette / CTD station (closest in time), \(PAR_d^{\space p48h}(z)\),
  • PAR.u.p48h [Ein.m-2.48h-1]: at depth estimated upward PAR integrated over 48 hours before the closest rosette / CTD station (closest in time), \(PAR_u^{\space p48h}(z)\),


      3.1.2. Directories contents

Both the ASCII and BINARY outputs of the Excel formatted files are provided in the directory called RES.EXCEL. Similarly, both the ASCII and BINARY formats of the ODV/normal form formatted files are provided in the RES.TIDY directory. Then, for illustration purposes, a graphical view of the data are provided as PNG plots in the directory called RES.PNG. In these graphical views, profiles of measured irradiances and/or radiances are plotted along with profiles of their respective smoothed data, as well as esimtated PAR values and profiles (again, from measured and from smoothed data). The directory called RES.PDF gives graphical view of the instantaneous atmospheric PAR propagation along the day for each profile. These last views are views that are similar to the view of the figure 2 above.
In all these directories, all the files are located in sub-directories named after the profile file name.
Finally, the directory called LOG contains the field log files where ancillary data can be found. See the note below, but the raw file names have been modified for homogeneity sake in the set of file names of all the Green Edge sub-cruises. The correspondence between the original file names and the homogenized file names can be found in the rename.raw.files.GE201x.cccc.csv files (where GE201x.cccc stands for the Green Edge sub-cruise name, “x” for the year, and “cccc” for the subcruise type, e.g. “ICMP” for Ice Camp).
The LOG directory also contains two files, called log.insitu.GE201x.cccc.tsv and log.sbdart.GE201x.cccc.tsv. They give the source of the daily PAR estimate, i.e. either the C-OPS / ICEPRO measurements, or other sources, more reliable, but only when available (see note #2 above). The priority in the daily PAR estimation has been set as 1) from the continous in-situ sensors data - 2) from the SBDART/sat. data and 3) from the punctual C-OPS measurements. In these two files, the profile.filename field gives the C-OPS / ICEPRO profiles filenames and the flag.insitu or the flag.sbdart field is set to FALSE when no other sources were available (or available but too far, see above for the threshold), and was set to TRUE whenever they were. When the flag was set to TRUE, the other fields give the following several PAR estimates, namely:

  • PAR.COPS.msr: daily PAR estimated from C-OPS measurements data,
  • PAR.COPS.fit: daily PAR estimated from C-OPS smoothed data,
  • PARp1h: 1 hour period PAR estimated by in-situ continuous acquisition (0.5 hour each side of the ship-based station closer in time to the C-OPS / ICEPRO profile),
  • PARp3h: 3 hours period PAR estimated by in-situ continuous acquisition (0.5 hour each side of the ship-based station closer in time to the C-OPS / ICEPRO profile),
  • PARp24h: 24 hours period (daily) PAR estimated by in-situ continuous acquisition (12 hours each side of the ship-based station closer in time to the C-OPS / ICEPRO profile),
  • PARp48h: 48 hours period PAR estimated by in-situ continuous acquisition (24 hours each side of the ship-based station closer in time to the C-OPS / ICEPRO profile),
  • PARdayLOC: 24 hours period (daily) PAR estimated by in-situ continuous acquisition (12 hours each side of the local noon at the station closer in time to the C-OPS / ICEPRO profile),
  • dist.cops.insitu: distance between the C-OPS / ICEPRO profile and the in-situ based measurement, in km.
    all the PAR values are given in Einstein per square meter and per either day (24 hours PAR), 48 hours period, 3 hours period or 1 hour period

Finally, the LOGdirectory contains 2 other files, PAR0.sbdart.dailyMetrics.GE201x.cccc and PAR0.insitu.dailyMetrics.GE201x.cccc, giving the full set of in-situ and SBDART metrics that are available. Data used to estimated the several PAR products have been taken from these two files.

Just for illustration purposes, below are comparison of the several daily downwelling PAR estimates above the surface. It was not expected to get a very well correlated set of data, as punctual PAR measurements won’t integrate variable cloudiness over one day for example, but the order of magnitude is rather very good.


Below is a plot showing a comparison of the daily PAR estimated by either the C-OPS / ICEPRO measurements and the in-situ sources:


and a plot showing the comparison between the C-OPS/ICEPRO estimations and the SBDART ones:



3.2. Notes

Below are few notes that are important to keep in mind when using the provided data.

      3.2.1. Filenames: UTC ref.

Each profile’s data are provided in a subdirectory, which names contain the date and the time of the profile acquisition. Be careful here, as the reference time is UTC, so a profile that has been acquired at 10pm local time will present a subdirectory name containing the next day’s date, as UTC time is local time plus 4 hours.


      3.2.2. wavelengths.

Several C-OPS systems have been used to acquire the data at the ice camp and from the Amundsen. These different C-OPS systems do not have the same set of wavelengths! They are very close sets, but a few wavelengths differ between all the systems. The wavelengths are provided in the output files.


      3.2.3. PAR estimates.

PAR values are estimated from planar irradiance sensors. The theoretical PAR computation from irradiance should be performed from scalar irradiance measurements, so that the PAR estimates are rather underestimated.


      3.2.4. Measured data.

In the data provided, only the smoothed data are given. The measured one are available upon request. The reason is that the size of these measured data files is very large.


      3.2.5. Original files renaming.

As the green edge sampling effort was performed from several sub-cruises, the original raw file names formats are very similar, but not fully consistent. This is not necessarily due to the user’s action, it is sometimes due to different versions of the acquisition software (this is not impacting the data quality!). Therefore, all the files have been renamed, to observe the file naming convention found in the section 3.1.


4. References

  1. Smith, R. C. and Baker, K. S.: The analysis of ocean optical data, in: Ocean Optics VII, edited by: Blizard, M., Proc. SPIE, 478, 119–126, 1984.

  2. Mueller, J. L. and Austin, R. W.: Ocean optics protocols for SeaWiFS validation, NASA Tech. Memo. 104566, Vol. 5, edited by: Hooker, S. B. and Firestone, E. R., NASA Goddard Space 30 Flight Center, Greenbelt, Maryland, 43 pp., 1992.

  3. Mueller, J. L. and Austin, R. W.: Ocean optics protocols for SeaWiFS validation, Revision 1, NASA Tech. Memo. 104566, Vol. 25, edited by: Hooker, S. B., Firestone, E. R., and Acker, J. G., NASA Goddard Space Flight Center, Greenbelt, Maryland, 66 pp., 1995.

  4. Mueller, J. L.: Overview of measurement and data analysis protocol, in: Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 2, NASA Tech. Memo. 2000–209966, edited by: Fargion, G. S. and Mueller, J. L., NASA Goddard Space Flight Center, Greenbelt, Maryland, 87–97, 2000.

  5. Mueller, J. L.: Overview of measurement and data analysis protocols, in: Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 3, Volume 1, NASA Tech. Memo. 2002–210004/Rev3–Vol1, edited by: Mueller, J. L. and Fargion, G. S., NASA Goddard Space Flight Center, Greenbelt, Maryland, 123– 137, 2002.

  6. Mueller, J. L.: Overview of measurement and data analysis methods, in: Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 4, Volume III: Radiometric Measurements and Data Analysis Protocols, NASA Tech. Memo. 2003–211621/Rev4 Vol.III, edited by: Mueller, J. L., Fargion, G. S., and McClain, C. R., NASA Goddard Space Flight Center, Greenbelt, Maryland, 1–20, 2003.

  7. D. Antoine, S. B. Hooker, S. Bélanger, A. Matsuoka, and M. Babin: Apparent optical properties of the Canadian Beaufort Sea – Part 1: Observational overview and water column relationships, in: Biogeosciences, 10, 4493-4509, 2013.

  8. Flocco, D., Schroeder, D., Feltham, D. L., & Hunke, E. C. (2012). Impact of melt ponds on Arctic sea ice simulations from 1990 to 2007. Journal of Geophysical Research: Oceans, 117(C9).

  9. Kirk, J. T. O. (2011). Light and photosynthesis in aquatic ecosystems, 3rd edn Cambridge University Press.

  10. Laliberté, J., Bélanger, S., & Frouin, R. (2016). Evaluation of satellite-based algorithms to estimate photosynthetically available radiation (PAR) reaching the ocean surface at high northern latitudes. Remote Sensing of Environment, 184, 199-211.

  11. Morel, A., & Smith, R. C. (1974). Relation between total quanta and total energy for aquatic photosynthesis. Limnology and Oceanography, 19(4), 591-600.

  12. Perovich, D. K., Grenfell, T. C., Light, B., & Hobbs, P. V. (2002). Seasonal evolution of the albedo of multiyear Arctic sea ice. Journal of Geophysical Research: Oceans, 107(C10).

  13. Perovich, D. K., & Polashenski, C. (2012). Albedo evolution of seasonal Arctic sea ice. Geophysical Research Letters, 39(8).