Evidence for iceberg fertilization of the NW Atlantic

. Icebergs are known to have a significant fertilizing impact on primary productivity in the Southern Ocean, 8 but this link is yet to be investigated in the Northern Hemisphere. This study combines sightings of icebergs with 9 satellite-derived ocean colour data from 1998 to 2015, to seek such a relationship in the NW Atlantic. Despite the 10 obscuring coincidence of the seasonal iceberg flux with the spring bloom of chlorophyll, it is shown that there is a 11 large-scale, one-month-lagged regional correlation between iceberg flux and chlorophyll levels. In addition, a spatial 12 time-lag analysis is consistent with the main cause for the iceberg-chlorophyll relationship being through advection of 13 the nutrients entrained in iceberg meltwater. This leads to a delayed fertilization response of 2-4 weeks. There are a 14 range of possible sources for the nutrients likely leading to this delayed response. The Northern Hemisphere impact of 15 iceberg meltwater on primary production is much less pronounced than in the Southern Ocean, but it is discernible.

with (x,y) being (longitude, latitude) of each respective square, I maps (x,y) being the number of icebergs at the grid point 140 (x,y) on the IIP maps and I sightings (x,y) deriving from the iceberg sightings database as described above. 141 A measure of the fit of the monthly map reconstruction using the sightings database is the correlation coefficient C n , 142 where C n is given by: 143 https://doi.org/10.5194/os-2021-61 Preprint. Discussion started: 9 August 2021 c Author(s) 2021. CC BY 4.0 License. .
(4) 144 The overall correlation of the method to be discussed below can then be expressed in an iceberg number weighted 145 mean, C T,n , over the 84 months from January 2009 to December 2015, where 146 .
(5) 147 The correlation measure C T,n is shown in Fig. 2. Here the dependence of C T,n on the length of time before the selected 148 day over which the Markov chain is constructed is shown. Even for the finest resolution of 1 degree squares, C T,1 149 reaches ~ 0.7 if 16 or more previous days are used, while for the bigger squares C T,n exceeds 0.95 for those cases. Later

Methods 158
Seeking an unequivocal link between iceberg fluxes and chlorophyll levels in the Labrador Sea is complicated by the 159 temporal correspondence of the peak iceberg season of spring and early summer with the marine productivity spring 160 bloom (Zhao et al., 2013), in a region close to coasts, with rivers which will have major seasonal variability, including a 161 spring peak flow due to snow melt (Vorosmarty et al., 1998), and considerable benthic sediment resuspension from 162 extensive shallow shelves. The latter can be of similar magnitude to fluxes of iceberg iron (Wadley et al., 2014;163 Raiswell et al., 2016). Remote sensing analysis in the Southern Ocean also suggests that the impact of fertilization from 164 melting icebergs can have a time lag of some weeks (Schwarz and Schodlock, 2009;Duprat et al., 2016). Two methods, 165 using different mixtures of iceberg and chlorophyll datasets, were employed to try to disentangle any signal from what 166 will be a complex and multiply-forced environment. This task is made even more complicated by the sub-grid scale size 167 of icebergs in the Labrador Sea, relative to ocean colour pixels, compared to the often much larger Southern Ocean 168 icebergs. The two basic methods are described here, with the results of the relevant analyses given in the Results 169 section. 170

Large-scale comparison 171
The first approach to seek a link between iceberg flux and chlorophyll levels uses large-scale comparisons between the 172 IIP's monthly iceberg number crossing 48 o N (I48N), and a monthly average chlorophyll over the main area of the 173 Labrador Sea through which icebergs drift, or where meltwater entrained in the North Atlantic Drift may be advected 174 over later weeks. A nearby coastal area, which will experience many of the same environmental forcings as the main 175 area, but which almost no icebergs reach, and little iceberg meltwater (see Fig. 13 of Wilton et al., 2015), was also iceberg main area, a series of correlation tests of the MODIS chlorophyll data between the control and iceberg main 182 areas were carried out. Over the almost 18 year period of July 2002 -May 2020 the monthly mean chlorophyll in the 183 control region, at 0.56±0.22 mg m -3 , is indistinguishable from that in the iceberg area (0.69±0.19 mg m -3 ), with a 184 statistically significant correlation (at better than the 1% level) of 0.47. Note that this correlation increases to 0.61 if the 185 impact of icebergs is minimized by correlating only months from the 5 years with less than 100 icebergs in total passing 186 48 o N. That icebergs have an impact on the chlorophyll is seen by correlating the monthly mean chlorophyll between the 187 control and iceberg areas for just the months during the peak iceberg season of April-June. Including all years leads to a 188 statistically insignificant correlation of 0.12, while comparing this spring bloom period for years with less than 100 189 icebergs raises the correlation to 0.52, which is statistically significant at the 5% level. Thus, using chlorophyll in the 190 control area as a means to control for the wider non-iceberg environmental factors affecting regional chlorophyll levels 191 is a valid approach. 192 https://doi.org/10.5194/os-2021-61 Preprint. Discussion started: 9 August 2021 c Author(s) 2021. CC BY 4.0 License.
To do this, Pearson correlations were first calculated between the anomalies of the two parameters I48N and 193 chlorophyll (averaged over each of the control and main areas), relative to their monthly averages over the period  2015. These were calculated over a range of lags from -2 months to 2 months, where a lag > 0 means I48N leads the 195 chlorophyll. In addition, in order to attempt to correct for changes purely due to non-iceberg effects, partial correlations 196 (Stuart et al., 2008) using the monthly anomalies from the control area as a control parameter, were also carried out 197 using the Matlab function partialcorr (https://uk.mathworks.com/help/stats/partialcorr.html). This approach, of using 198 partial correlation analysis to control for common variation in processes unrelated to the main question being 199 investigated, has been successfully employed in a number of environmental fields, with recent examples in untangling 200 large scale measure of the tropospheric circulation over the North Atlantic, strongly linked to Arctic atmospheric 225 circulation as well as being highly correlated with a range of climatic factors across the northern Atlantic (Hurrell and 226 Deser, 2009;Reintges et al., 2017). However, previous work has shown that the NAO is not linearly correlated with the 227 Greenland Ice Sheet mass balance (Hanna et al., 2011), which is one of the main variables indirectly responsible for the 228 Greenland iceberg flux (Zhao et al., 2016). There is a weak, non-linear, link between significantly lagged (> 6 months) 229 NAO and I48N (Zhao et al., 2016), nevertheless, the recent NAO Index can be used as a control parameter as it affects 230 current and recent climate, but not iceberg flux. 231 The control parameter uses the NAO lagged by 3 months compared to the month of examination, as this lag had the 232 most impact on reducing the natural variability component of the chlorophyll. Note that the use of partial correlation 233 analysis as a way to remove the impact of specific climate signals from a process study is common, with recent 234 examples in untangling rainfall/cyclone:sea temperature relationships (He et al., 2016;Hong et al., 2018;Srinivas et al., 235 2018), and the impact of the lower stratosphere on tropical cyclone intensity (Ferrara et al., 2017).  This result is consistent with there being a large-scale link between increasing iceberg numbers and fertilization of the 262 ocean a month later. It is suggestive rather than definitive, however, as the correlations are low in magnitude and the 263 increasing correlation of I48N with chlorophyll within the control area, with greater lag, is consistent with there being a 264 strong spring bloom amplification occurring after the peak iceberg flux of April/May. This result cannot shed light on 265 the question of the potential reality or cause of such a link, if real, whether it is restricted to the immediate vicinity of 266 the melting icebergs, extends to the mixing downstream of the iceberg meltwater with the ocean, or is a mixture of both 267 local and remote effects. In the next section we will explore this question further. 268 Fig. 5a shows a spatial view of the instantaneous correlation of monthly main area iceberg total number, I T, with local 270 monthly mean chlorophyll, with the significant squares shown in Fig. 5b. This shows a region of zero-lag positive 271 correlation off Newfoundland, in the control region and where iceberg density is greatest (Fig. S2), but also other 272 patches of statistically significant positive and negative correlation elsewhere across the NW Atlantic (Fig. 5b). 273

Local temporally evolving comparison 269
However, many of these significant correlation patches, particularly south of 48 o N, are linked to the variation in the 274 weather, as reflected in the NAO state 3 months previously (see Figs. 5c and d). The partial correlation, controlled for weather variability through use of the 3-month lagged NAO Index, is therefore 283 shown in Figs. 5e and 5f. The latter shows that many of the more southerly statistically significant regions seen in Fig.  284 5b disappear under this control, however, the signal in the main area, off Newfoundland remains, as do areas off 285 Greenland. These areas of statistically positive correlation that remain, however, tend to be somewhat fragmented and 286 there are also some significant negative correlations in an arc from the Labrador coast into the central North Atlantic.
Nevertheless, note the lack of any coherent area of statistically significant partial correlation within the control region. 288 These equivocal signals are consistent with the limited direct correlation between I48N and chlorophyll shown in Fig. 4. 289 That analysis suggested that it required a month before the full impact of fertilization by the melting icebergs became 290 visible. The analysis is therefore extended to consider this lag component in more detail. 291 Using the technique outlined in Sect. 2.2.2, a series of partial correlations over 1998-2015 between 5 day series of I T 292 and chlorophyll were carried out over a range of lags from 0 days to 70 days. A sub-set of the correlations showing 293 those squares with a statistical significant partial correlation at the 0.05 level is shown in Fig. 6. An animation of this 294 full correlation map sequence is available in the Supplementary Material. Note that the region over which correlations 295 are possible varies with the lag combination due to the persistent presence of cloud cover in the northern Atlantic 296 restricting the number of cloud-free pixels, particularly for the larger dataset of shorter lag. 297 This series of lagged partial correlations shows an interesting evolution, with a distinct zero lag positive correlation in 298 the vicinity of the main iceberg stream in the Labrador Current (that is, the main area of Fig. 3), changing towards a 299 more general, if lower, correlation over the NW Atlantic by a lag of 10 days, which then strengthens into a more 300 homogeneous region of positive correlation south of Greenland by day 20, peaking by day 30. This begins to decay 301 after day 30 and has essentially disappeared by day 50. There are initially some regions of relatively strong positive 302 correlation in the central east Atlantic and negative correlation in the control region and the Sargasso Sea, but these 303 decay fairly quickly. They are probably remnants of environmental chlorophyll impacts not due to the iceberg flux, but 304 not totally removed by the NAO control. A possible cause could be a light-and stratification-related spring bloom 305 signal (Behrenfeld and Boss, 2014) occurring during the peak iceberg months of April and May (Bigg et al., 2014). It is 306 also worth noting that the high chlorophyll correlation adjacent to the Greenland coast in Fig. 6 may be partially due to 307 local icebergs (not monitored by the IIP) or meltwater from Greenland glaciers (Arrigo et al., 2017). However these 308 local icebergs are closely confined to coastal waters (see charts at https://www.dmi.dk/en/groenland/hav/ice-charts/), 309 and the impact of local fjord meltwater is limited to periods after the I48N iceberg peak has decreased (early July 310 onwards) and restricted to the Greenland shelf region of the Labrador Sea (Arrigo et al., 2017). The majority of the 311 correlation signal south of 57 o N is therefore likely due to the relationship with I48N rather than Greenland meltwater. 312 icebergs and the high nutrient-low chlorophyll nature of the region allows clear and pronounced plumes of production 326 to be visibly associated with icebergs (Schwarz and Schodlok, 2009;Duprat et al., 2016;Wu and Hou, 2017). In the 327 NW Atlantic the peak of the natural light-driven spring bloom, enhanced by the inputs from land and coastal waters, as 328 well as the presence of a pronounced nutrient source through winter mixing (Fragoso et al., 2016), coincides with the 329 natural release of icebergs into the Labrador Current after being locked in winter sea-ice further north (Marko et al., 330 1994). Thus, while a correlation between icebergs and chlorophyll is discernible in Figs. 4-6, the use of partial 331 correlations, with a control for chlorophyll in a control area (Fig. 4), and inter-annual and monthly climate variability 332 through the NAO (Figs. 5-6), was required to make the signal more robust. fertilization at lags of 0 (bright green) to 1 month (pale green), where the zero lagged fertilization is likely to be 338 due to the direct impact of the icebergs melting, while the delayed response is due to the impact of meltwater 339 advection. See text for more details. Note that the boundary of the Labrador Sea is taken to be the line extending 340 from the eastern tip of Newfoundland to the southern tip of Greenland. 341 multiplies this by the matrix of probabilities generated by Algorithm 1 to obtain the probabilities of presence for the 439 next day. This needs to be repeated for every remaining day until the 15th, which is equivalent to taking the matrix to 440 the power of the number of remaining days. Note that if the last sighting happens exactly on the 15th, then that power is 441 zero and the matrix reduces to the identity matrix, which will correctly record one iceberg at the right position. The 442 main caveat of this method is that it assumes that the probability of an iceberg crossing from one square to another is 443 independent of the time it spent in the former square. However, the procedure gave satisfactory results even with this 444 approximation. 445 https://doi.org/10.5194/os-2021-61 Preprint. Discussion started: 9 August 2021 c Author(s) 2021. CC BY 4.0 License.