Comparison of new and primary production models using SeaWiFS data in contrasting hydrographic zones of the northern North Atlantic

The accuracy of two satellite models of marine primary (PP) and new production (NP) were assessed against 14C and 15N uptake measurements taken during six research cruises in the northern North Atlantic. The wavelength resolving model (WRM) was more accurate than the Vertical General Production Model (VGPM) for computation of both PP and NP. Mean monthly satellite maps of PP and NP for both models were generated from 1997 to 2010 using SeaWiFS data for the Irminger basin and North Atlantic. Intra- and inter-annual variability of the two models was compared in six hydrographic zones. Both models exhibited similar spatio-temporal patterns: PP and NP increased from April to June and decreased by August. Higher values were associated with the East Greenland Current (EGC), Iceland Basin (ICB) and the Reykjanes Ridge (RKR) and lower values occurred in the Central Irminger Current (CIC), North Irminger Current (NIC) and Southern Irminger Current (SIC). The annual PP and NP over the SeaWiFS record was 258 and 82 gC m− 2 yr− 1 respectively for the VGPM and 190 and 41 gC m− 2 yr− 1 for the WRM. Average annual cumulative sum in the anomalies of NP for the VGPM were positively correlated with the North Atlantic Oscillation (NAO) in the EGC, CIC and SIC and negatively correlated with the multivariate ENSO index (MEI) in the ICB. By contrast, cumulative sum of the anomalies of NP for the WRM were significantly correlated with NAO only in the EGC and CIC. NP from both VGPM and WRM exhibited significant negative correlations with Arctic Oscillation (AO) in all hydrographic zones. The differences in estimates of PP and NP in these hydrographic zones arise principally from the parameterisation of the euphotic depth and the SST dependence of photo-physiological term in the VGPM, which has a greater sensitivity to variations in temperature than the WRM. In waters of 0 to 5 °C PP using the VGPM was 43% higher than WRM, from 5 to 10 °C the VGPM was 29% higher and from 10 to 15 °C the VGPM was 27% higher.


Introduction
Recent warming of the Global Ocean has potentially important impacts on carbon cycling in the marine ecosystem, which is hard to quantify due to the scarcity of data over both space and time. Recent advances in Earth Observation (EO) have enabled us to fill these gaps, but the accuracy of satellite models requires careful consideration if we are to employ them to quantify the carbon fluxes in an ecosystem.
Quantitative estimation of the draw-down of carbon dioxide from the atmosphere to the ocean via primary production (PP) and the exported fraction of this fixed carbon, is important for calculating the global CO 2 flux and for modeling carbon transfer through the pelagic food web to the deep sea (Azam, 1998;Falkowski, 1988;Sarmiento, Hughes, Stouffer, & Manabe, 1998). NP is that proportion of PP driven by allocthonous nutrients that are supplied to the ecosystem through physical processes, nitrogen fixation or atmospheric input (Dugdale & Goering, 1967;Yentsch, Yentsch, Phinney, Lapointe, & Yentsch, 2004). The remaining fraction, the regenerated production, is driven by nutrients produced by re-mineralisation processes which include ammonification (Dugdale & Goering, 1967) and nitrification (Yool, Martin, Fernandez, & Clark, 2007) within the euphotic zone. It is balanced by export to deep waters or higher trophic levels, and makes a direct contribution to carbon removal from the photic zone. The measurement of NP is conventionally based on the uptake of 15 N-labelled nitrogen (NO 3 − , NH 4 + , urea), which has undoubtedly aided our understanding of the global importance of the world's oceans in exporting carbon. In many regions, few NP measurements are available resulting in sparse spatial and temporal data coverage and a vague impression of Remote Sensing of Environment 156 (2015) [473][474][475][476][477][478][479][480][481][482][483][484][485][486][487][488][489] phytoplankton NP dynamics. There has, therefore, been a concerted effort to derive accurate estimates of NP from satellite data to fill this gap (Falkowski, 1988;Laws, Falkowski, Smith, Ducklow, & McCarthy, 2000). These have been aided by the inverse relationship between temperature and nutrients in certain regions of the Atlantic and Pacific (Chavez, Service, & Buttrey, 1996;Goes et al., 2000;Morin, Wafar, & Lecorre, 1993;Sathyendranath, Gouveia, Shetye, Ravindran, & Platt, 1991), particularly in upwelling and tidally driven areas (Babin, Therriault, & Legendre, 1991;Dugdale, Morel, Bricaud, & Wilkerson, 1989;Morin et al., 1993;Waldron & Probyn, 1992), to enable the determination of large-scale estimates of NP from satellite Sea Surface Temperature (SST) data (Alvarez-Salgado et al., 2002;Dugdale, Davis, & Wilkerson, 1997;Kamykowski, Reed, & Kirkpatrick, 1992;Traganza, Nestor, & McDonald, 1980). The relationship between SST and nitrate breaks down however, when the water column becomes thermally stratified (Henson, Sanders, Allen, Robinson, & Brown, 2003). Under these scenarios, NP can be underestimated by N50% especially when PP exceeds~700 mg C m −2 d −1 (Laws, 2004). Goes, Saino, Oaku, and Jiang (1999), Goes et al. (2000) thus developed an approach to derive NP for the Pacific basin that accounts for phytoplankton consumption of nitrate using a second order polynomial of SST and SeaWiFS Chlorophyll-a (Chla). The algorithm has been used accurately to study the effects of El Niño and land mass warming on NP in the Pacific (Goes, Gomes, Limsakul, Balch, & Saino, 2001) and the Arabian sea (Goes, Gomes, Limsakul, & Saino, 2004). Similarly, Laws et al. (2000) developed a food web model that can be used to predict export production from SST and PP. In this model, the ratio between NP and PP (f-ratio) is temperature-dependent and accounts for 86% of the variability in f-ratios derived from 15 N and 14 C uptake measurements. In the Pacific upwelling region, when PP is b850 mg C m −2 d −1 this method is accurate to 10-20% (Laws, 2004), but requires further validation in the Atlantic Ocean, especially at extreme northerly latitudes where few in situ data are available.
The Irminger Basin, located to the Southwest of Iceland and to the Southeast of Greenland, is an important source of export production for the North Atlantic and North Sea (deYoung et al., 2004). From satellite-derived chlorophyll concentrations, NP in the Irminger Basin is predicted to be high~100-150 gC m − 2 yr − 1 (Falkowski, 1988;Laws et al., 2000) and therefore potentially important on a global scale. However, lower values of NP~30 to 65 gC m −2 yr −1 have been predicted from in situ nutrient uptake (Henson, Sanders, Holeton, & Allen, 2006;Sanders, Brown, Henson, & Lucas, 2005;. Recently, Henson et al. (2011) developed a model of export production based on PP and SST and their relationship with thorium ( 234 Th) derived export, which proved to be 60% lower than the e/f-ratio based estimate. Controversy therefore still exists as to the most accurate and appropriate model to estimate NP in this region, which is hampered by a lack of independent in situ observations.
Over the past two decades NASA have conducted five intercomparisons of satellite models of PP, to determine the most accurate models and to characterise sources of error (Campbell et al., 2002;Carr et al., 2006;Friedrichs et al., 2009;Saba et al., 2010Saba et al., , 2011. Few inter-comparisons of satellite derived NP have been conducted (Silio-Calzada, Bricaud, Uitz, & Gentili, 2008). In a recent error analysis in computation of net PP using the vertical generalised production model (VGPM), Milutinovic and Bertino (2011) quantified the highest uncertainty to be in the northern North Atlantic (N70%) due to errors in irradiance-depth dependent and photosynthetic rate functions.
In this paper we compare the accuracy of two satellite models, the VGPM of Behrenfeld and Falkowski (1997) and a Wavelength Resolved model (WRM) of Morel (1991) at estimating PP and NP in the northern North Atlantic. Each model is used to generate a time series of PP and NP from 1997 to 2010 using SeaWiFS data for six hydrographic zones. Interannual differences between models are assessed and in relation to climate indices. A sensitivity analysis was performed to assess which parameters contribute most to errors in the models.

Study areas and sampling regimes
Integrated PP was derived from 14 C uptake measurements at 83 stations on six field campaigns in the Atlantic basin (D261, D264, D267, FISHES, JC011, JC037) and 15 N uptake NP measurements were made at 18 stations on two of these campaigns (D261, FISHES) ( Fig. 1, Table 1). On all research cruises, vertical profiles of temperature, conductivity, fluorescence, oxygen and photosynthetically active radiation (PAR) were acquired using a Seabird 911 + CTD and Chelsea Instruments PAR sensor fitted to a rosette with either 24 × 20 dm 3 or 12 × 30 dm 3 Niskin-type sampling bottles to collect water samples for the determination of NP, PP, Chla, dissolved inorganic nutrients and photosynthetic parameters (described below). Euphotic depth (Z eu ) was determined from the CTD profiles of PAR. The field measurements were used to validate satellite models of PP and NP, which were then applied to the Irminger Sea and northern North Atlantic.
The Irminger Sea is a diverse region with influences from the subtropical thermocline via the North Atlantic Current, and from the Arctic via the dense northern overflows , which in turn may influence new production. A number of major physical zones have been described where different surface mixing and restratification processes dominate : The Central Irminger Current (CIC) has characteristically low temperatures and salinity (6.53-9.72°C, 34.73-35.08) and its surface waters are dominated by cool fresh Sub-Arctic surface water that originates in the Labrador Sea and spreads across the sub-polar gyre. The Irminger Current Zone, a branch of the North Atlantic Current positioned west of the mid-Atlantic Ridge, is the warmest and most saline (7°C, 35.00) feature of the true Irminger Sea. Two Irminger Current zones are described: The southern Irminger Current (SIC) zone extends from 54-57°N to 60°N and between 28 & 32°W. The North Irminger Current (NIC) occurs between 60 & 62°N and 28 & 32°W. The East Greenland Current Zone (EGC) dominates the western part of the Irminger Sea, including the continental shelf. The EGC is persistent along the length of the Greenland continental slope, carrying cold, fresh (b0°C, b 34.50) Arctic Polar Water and Arctic Intermediate Water (0-3°C) from the Arctic and Nordic Seas into the sub-polar gyre (Foldvik, Aagaard, & Torresen, 1988). The Reykjanes Ridge (RKR) separates the Irminger Basin from the Iceland Basin (ICB), and is characterised by warm, saline (N8.0°C, 35.10) water originating from the Iceland basin. The western boundary of the zone is clearly marked by a sharp front with high salinity and temperature gradients, and a distinctive change to weaker stratification and lower density compared to the adjacent Irminger Current.

Phytoplankton pigments
Chla was determined by high performance liquid chromatography (HPLC) on all field campaigns. Water samples were filtered through Whatman GF/F filters and stored in liquid nitrogen. Pigments were extracted with the aid of sonification in 90% acetone, clarified using centrifugation (5 min at 4000 r.p.m) and analysed in the laboratory, using reverse phase HPLC following the procedure outlined in Barlow, Cummings, and Gibb (1997). Pigments were separated using a 3 μm Hypersil MOS2 C8 column on a Thermo separations product HPLC, detected by absorbance at 440 nm and identified by retention time and on line diode array spectroscopy. Pigment absorption was measured against quantified standards: Chla standard was obtained from Sigma-Aldrich, and divinyl chlorophylls a and b from R. Bidigare and M. Ondrusek, University of Hawaii. Other pigment standards were purchased from the DHI Institute for Water and Environment, Denmark. Limits of detection were of the order of 0.001 mg m − 3 (Barlow, Aiken, Moore, Holligan, & Lavender, 2004).

Simulated in situ 14 C primary production
For field campaigns D264, D267 and FISHES, water samples were taken from 6 to 9 depths and transferred from Niskin bottles to blackout carboys to prevent shock to the photosynthetic apparatus of the phytoplankton cells. Water from each sample was sub-sampled into three 75 mL clear and one black, pre-HCl cleaned polycarbonate bottle. Each sample was inoculated with between 185 and 370 kBq (5-10 μCi) NaH 14 CO 3 according to the biomass of phytoplankton. The bottles were transferred to an on-deck simulated in situ (SIS) incubation system which uses neutral density and blue filters to simulate subsurface irradiance over depth to 97%, 55%, 33%, 20%, 14%, 7%, 3%, 1% or 0.1% of the surface value. The samples were matched to the irradiance levels in the incubator and were maintained at near-surface temperature by pumping sea water from a depth of~7 m through the system. After 24 h incubation from pre-dawn to pre-dawn of the next day, the water samples were filtered through Whatman GF/F filters. The filters were exposed to concentrated HCl fumes for 12 h and then immersed in 2.5 mL scintillation cocktail. 14 C disintegration time per minute (DPM) was measured on board using Beckman LS6000 (FISHES, D261, D264, D267), TriCarb 2900 TR (JC037) and Wallac 4040 (JC011) liquid scintillation counter (LSC) using the external standard and the channel ratio methods to correct for quenching. The quantity of 14 C added to the experimental bottles was determined by adding aliquots of the stock 14 C solution to a CO 2 absorbing scintillation cocktail, which was counted immediately on the LSC. The 14 C fixed from dawn to dawn of the following day, was integrated to 1% irradiance depth.
For research cruises D261, JC011 and JC037, photosynthesisirradiance (PE) experiments were conducted in photosynthetrons illuminated by 50 W, 12 V tungsten halogen lamps following the methods described in Tilstone, Figueiras, Lorenzo, and Arbones (2003). Each incubator houses 14 sub-samples in 0.125 L polycarbonate bottles (Arbones, Figueiras, & Varela, 2000) which were inoculated with between 185 k Bq (5 μCi) and 370 kBq (10 μCi) of 14 C labelled bicarbonate. All bottles were pre cleaned following JGOFS protocols (IOC, 1994) to reduce trace metal contamination. The samples were maintained at ambient temperature using a digital temperature controller. After 1 to 2 h of incubation, the suspended material was filtered through Whatman GF/F filters at a vacuum pressure of b 20 cm Hg and the filters were then exposed to concentrated HCl fumes for 12 h and immersed in 2.5 ml scintillation cocktail and counted as above. The broadband light-saturated Chla-specific rate of photosynthesis P m B [mg C (mg chl a) −1 h −1 ] and were estimated by fitting the data to the model of Platt, Gallegos, and Harrison (1980), as follows: where P z B is the chlorophyll-specific rate of photosynthesis [mg C (mg Chla) −1 h −1 ] and β B is the coefficient of photoinhibition [mg C (mg Chla) −1 h −1 (μmol m −2 s −1 ) −1 ] at each given depth. The PAR absorbed by phytoplankton i.e. the photosynthetically usable radiation [E PUR (μmol m − 3 s − 1 )] at each position in the incubator and for each sampling depth was estimated according to Dubinsky (1980) using measurements of a ph (λ). The maximum quantum yield of carbon fixation [ϕ m mol C fixed (mol photons absorbed) − 1 ] was determined by fitting the Chla-specific photosynthetic rates P z B [mg C (mg chl a) − 1 h −1 ] to E PUR following Figueiras, Arbones, and Estrada (1999). The daily integrated PP (mgC m −2 d −1 ) was estimated using a bio-optical model which inputs E PUR , Chla, P m B and ϕ m to integrate PP at minute by minute intervals, down to 0.1% irradiance depth following Tilstone et al. (2003) as follows: The SIS determination of PP is closer to net PP as the 24 h incubations account for the dark respiration phase of carbon fixation, whereas the PE method is closer to gross PP as respired carbon is not accounted for during the short term incubation (Marra, 2009). To account for the discrepancy between the two methods, we reduced the PE method estimates by 20% following Joint et al. (2002), as an estimation of carbon loss through the dark phase of photosynthesis.

in situ 15 N new production (NP (IS) )
Assimilation rates for NO 3 − and NH 4 + were determined on field campaigns D261 and FISHES following the incorporation of the stable isotope 15 N. Triplicate water samples from 6 depths within the euphotic zone were distributed into 640 mL clear polycarbonate bottles and 15 N-NO 3 and 15 N-NH 4 were added at a final concentration of 10% ambient NO 3 − and NH 4 + , except under oligotrophic conditions when the minimum addition was 5 nmol l −1 (Rees, Woodward, & Joint, 2006). Incubations were made in the same on-deck incubators used for the PP measurements for 24 h. Incubations were terminated by filtration (b 40 cm Hg vacuum) onto ashed Whatman GF/F filters, which were dried onboard and stored over silica gel desiccant until they were returned to the laboratory. Particulate nitrogen concentration and 15 N atom% were determined by continuous flow stable isotope mass spectrometry using techniques described by Barrie and Lemley (1989) and Owens and Rees (1989). Rates of uptake were calculated according to Dugdale and Goering (1967) and when appropriate were corrected for over addition of 15 N tracer (Rees, Joint, & Donald, 1999). Nitrification in the euphotic zone can make a variable contribution to the NO 3 − pool available to phytoplankton, which can therefore lead to an overestimate in f-ratio and subsequently estimates of new production (Yool et al., 2007). Nitrification rate measurements were not performed during these field programs and, therefore, derived new production should be considered to be at the upper range of the values. We have confidence that these are not anomalous as measurements during D261 were made during the early spring bloom period when NO 3 − is dominated by that derived from winter mixing and those during FISHES were at high latitude where the influence of nitrification on f-ratio estimates is likely to be minimal, in part due to the short growing season (Yool et al., 2007). NP was estimated from the product of net primary production and the f-ratio (Eppley & Peterson, 1979). NO 3 − is therefore considered to be "new" and NH 4 + is considered as the regenerated nitrogen pool, so that the f-ratio was determined as follows: 2.5. Remote sensing algorithms of primary (PP) and new production (NP) Two satellite models were used to estimate PP; the VGPM and WRM. Using PP from each model, NP was then computed using the model of Laws et al. (2000). The description of each model and parameterisation is given below. The VGPM was selected as it is the most cited PP algorithm currently used. It is empirical in nature and offers a simple approach to estimate PP by integrating over wavelength and depth (WIDI) in the water column using surface values as seen from satellite. This was compared against the WRM which is far more complex and computationally expensive as it resolves PP over both wavelength and depth (WRDR). Both WIDI and WRDR models have been compared in recent NASA round robin exercises (Carr et al., 2006;Friedrichs et al., 2009;Saba et al., 2011). To assess their accuracy, each model was run using both satellite and in situ data.

VGPM model of primary production (PP VGPM )
The model of Behrenfeld and Falkowski (1997) was driven by surface Chla. The euphotic depth (Z eu ) was computed from Chla following the relationships of Morel and Berthon (1989). The optimum photosynthetic rate (P opt B ) was derived from SST using the seventh order polynomial of Behrenfeld and Falkowski (1997). PP VGPM is calculated as follows: 2.5.2. WRM of primary production (PP WRM ) The wavelength resolving model (WRM) of Morel (1991) was implemented following Smyth, Tilstone, and Groom (2005). The maximum quantum yield for growth (ϕ m ) and the maximum phytoplankton Chla-specific absorption coefficient (a ⁎ max ) were parameterised using Chla following Morel, Antoine, Babin, and Dandonneau (1996). The above-water spectral light field was generated using the Gregg and Carder (1990) model run at 5 nm wavelength and 30 minute time resolution so that PP can be integrated over the day. Meteorological and ozone data to drive the model were obtained from National Centers for Environmental Prediction (NCEP) and Earth-Probe Total Ozone Mapping Spectrometer data (EPTOMS), respectively. Cloud fields were obtained from European Centre for Medium Range Weather Forecasts (ECMWF) model output and used to modify the above water light field following Reed (1977). The light field was propagated through the water column by calculating the spectral attenuation coefficient for downwelling irradiance following the methods of Morel (1988) as outlined in Tilstone, Smyth, Gowen, Martinez-Vicente, and Groom (2005). Hourly rates of PP were weighted to the water column light field and carbon fixation was integrated over the light hours for each day down to 1% irradiance depth. Integration was performed over all daylight hours, for wavelengths 400-700 nm and computed through the iterative approach of Morel and Berthon (1989). The model was run using surface Chla and temperature assuming a homogenous water column profile of Chla, a* max and ϕ m , since this is what is available from satellite. PUR irradiance was estimated using Morel (1991) and PAR derived from the Gregg and Carder (1990) model. PP WRM was calculated as follows: 2.5.3. Photosynthesis-temperature model of new production Satellite estimates of NP were calculated from the model of Laws et al. (2000) from which the f-ratio was computed as a function of temperature and net photosynthesis (mg N m −3 d −1 ). The look up table of f-ratio, temperature and net photosynthesis was downloaded from the JGOFS web site and implemented in C ++ to derive f-ratios from satellite values of Pathfinder Advanced High Resolution Radiometer (AVHRR) SST v2009 and PP from either the VGPM or WRM.

Satellite maps of primary and new production
Daily satellite images of PP and NP for match-up analysis were generated using 1 km data. Mean monthly satellite maps of PP and NP were generated using 9 km data. For each model, PP were generated from 1998 to 2010 using monthly NASA SeaWiFS OC4v6 Chla from the 2010 reprocessing (R2010) downloaded from the NASA Ocean Colour web site. AVHRR SST and PAR (Frouin & Pinker, 1995) for each province. The PAR monthly fields are average daily integrated values; downwelling irradiance values at each wavelength E d (λ) were retrieved using a look up table as described in Smyth et al. (2005). Each PP model was coupled to the NP model (Laws et al., 2000) to generate mean monthly maps of NP for the Irminger Sea from 1998 to 2010. Though we illustrate the variability in monthly mean PP and NP for 2010, these data were not used to calculate annual PP and NP due to the influence of the Icelandic volcano Eyjafjallajökull, which erupted during the summer of 2010 (Achterberg et al., 2013). SeaWiFS Chla 2010 values are likely to be erroneous for the Irminger Sea due to failure in the satellite atmospheric correction associated with volcanic particles (Porter, Kono, & Nielsen, 2005).

Statistical analyses and hydrographic zones
In situ data within ±3 h of satellite overpass were compared to the mean of a 3 × 3 pixel array around the sampling station following the procedures outlined in Bailey and Werdell (2006). Considering all six cruises, there were 14 satellite match-ups for PP and 5 for NP. The in situ and satellite data were compared using linear regression analysis and the following statistics to evaluate model performance: Mean (M) and standard deviation (S) of the log 10 -difference error between measured and satellite NP and PP and also the log 10 root-mean square (log 10 -RMS) between measured and satellite NP and PP. We also used the inverse transformed ratio between satellite and measured values M (F med ), M − S (F min ) and M + S (F max ) following Campbell et al. (2002). The relative percentage difference (RPD) was calculated to illustrate the uncertainty between measured and satellite NP and PP. The distribution of PP and NP were transformed until no significant difference was found between the expected and the observed distributions using the Kolmogrov-Smirnov with Lilliefors test to ensure homoscedasticity (Sokal & Rolf, 1997). One-way analysis of variance (ANOVA) was also used to test whether there were significant differences between satellite and in situ estimates of NP and PP. The ANOVA results are given as F 1,df = x, P = y where F is the mean square to mean square error ratio, the sub-script numbers denote the degrees of freedom and P is the ANOVA critical significance value.
We evaluated differences in mean monthly and annual PP and NP between satellite models in the six major hydrographic zones described in Section 2.1. Mean monthly and annual PP and NP were extracted for each model in each of these zones and differences between them were analysed by ANOVA. Monthly and annual PP and NP data were spatially integrated to achieve a single number per province. To assess differences between models over different temperature ranges, PP values were extracted at every 10 km along SW-NE transects from 60°N, 42.5°W to 65°N, 32.5°W in the EGC; 59°N, 40°W to 63°N, 31.5°W in the CIC; 56.5°N, 36°W to 63°N, 31°W in the SIC; 58.5°N, 33°W to 63°N, 25°W in the RKR; 55°N, 30°W to 61°N, 22.5°W in the ICB and along a SE-NW transect in the NIC from 63°N, 27.5°W to 65°N, 30°W (Fig. 1). PP VGPM was then regressed against PP WRM as a function of SST at 5°C increments from 0 to 15°C to assess the temperature dependency of each model.
For time series analyses, mean monthly primary production anomalies were calculated by subtracting from each monthly value the corresponding monthly average for the time series from 1998 to 2010. The cumulative sums method was applied to the anomalies to further decompose the signal to highlight major changes in monthly data values along the time-series (McQuatters-Gollop, Mee, Raitsos, & Shapiro, 2008). The average annual cumulative sum in anomalies of NP was compared against the cumulative sum in NAO, AO and MEI. The NAO monthly indices were taken from www.cgd.ucar.edu/cas/ which is based on the normalised sea level pressure difference between Ponta Delgada, Azores and Stykkisholmur Reykjavik. AO was downloaded from NOAA: www.cpc.ncep.noaa.gov/products/precip/ CWlink/daily_ao_index/ao_index.html. AO is the difference in pressure between the Arctic and the northern middle latitudes and is calculated by projecting the monthly mean 1000 hPa height poleward of 20°N onto its first leading EOF mode. MEI was also downloaded from NOAA: www.esrl.noaa.gov/psd/enso/mei/. MEI is calculated as the first unrotated Principal Component from six variables: sea-level pressure, zonal and meridional components of the surface wind, sea surface temperature, surface air temperature, and total cloudiness fraction of the sky.

Accuracy assessment of primary and new production satellite algorithms in the North Atlantic
The performance of the VGPM and WRM models was assessed using Chla and SST, from, firstly, in situ measurements (Fig. 2, Table 2) and, secondly, satellite estimates ( Fig. 3B, C, Table 2) to calculate PP. PP WRM had the lowest mean log-difference error in measured and satellite (M), standard deviation mean log-difference errors in measured and satellite (S), log-RMS and intercept, the highest r 2 and F med , F max and slope closest to 1 using both the in situ and satellite data (Table 2), indicating that PP WRM was the most accurate model. There was no significant difference between in situ and modeled PP estimated from both in situ measurements of Chla and SST (F 1,165 = 1.58, P = 0.211) and satellite data (F 1,27 = 0.202, P = 0.657). The relative percentage difference (RPD) of PP WRM compared to in situ PP was~30% whereas the RPD of PP VGPM was 80%. PP VGPM had the highest M indicating a significant bias with a tendency to over-estimate PP, as indicated by the high F max values. This increased the slope, intercept and log-RMS and resulted in a significant difference between in situ and PP VGPM using both in situ (F 1,165 = 15.75, P b 0.001) and satellite data (F 1,35 = 102.95, P = b0.001) to estimate PP (Table 2). Using satellite data alone, PP WRM was within 45% whereas the percentage difference between PP VGPM and in situ values was higher.
Compared with in situ 15 N uptake, NP VGPM explained 54% of the variance in in situ NP, whereas NP WRM explained 73% of the variance (Fig. 2C, D). There was no difference between in situ NP and NP VGPM (F 1,35 = 1.12, P = 0.297), which were within 72% of the in situ values. There was also no difference between in situ NP and NP WRM (F 1,35 = 0.002, P = 0.965), which were within 10% of the in situ values. For NP WRM , the slope, F med , F max and F min were closer to 1 and the log-RMS error, M and S were lower compared to NP VGPM (Table 2, Fig. 2C, D). Fig. 3A gives in situ Chla versus SeaWiFS OC4v6 Chla at the stations sampled ( Fig. 1) and corresponding validation statistics are given in Table 2. OC4v6 and in situ Chla were highly correlated having a high percentage variance explained, low bias, error, intercept, log 10 -RMS and UPD and a slope, F min , F med and F max close to 1 (Table 2). Consequently, there was no significant difference between in situ and SeaWiFS OC4v6 Chla (F 1,27 = 0.002, P = 0.964).
Using daily SeaWiFS data, there were only five in situ match-up stations for validating the NP models with satellite data (Fig. 3C), so a comprehensive validation was not possible. Similar to using in situ data, however, SeaWiFS derived NP WRM was more accurate than NP VGPM with a lower M, S and log-RMS, slope F med and F max closest to 1 (Table 2). There was no significant difference between in situ and NP WRM (F 1,27 = 0.259, P = 0.625) and the RPD was 11%. NP VGPM exhibited a negative regression caused by an over-estimation at low values, which resulted in disproportionately high F med , M and log-RMS and a significant difference between in situ and NP VGPM (F 1,27 = 15.92, P = 0.004) (Fig. 3C) and very high RPD (Table 2).

Satellite estimates of new and primary production for the northern North Atlantic
Mean monthly PP and NP images generated from the two models for April, June and August from 1998 to 2010 of the Irminger Sea are given in Figs. 4 and 5, respectively. The corresponding mean monthly time series for each hydrographic province are given in Figs. 6 and 7. Both models exhibited the same temporal pattern, with an increase in PP and NP from April to June, followed by a decrease in August. They also showed the same spatial patterns, with higher values in the EGC, ICB and RKR and lower values in the CIC, NIC and SIC. PP VGPM and NP VGPM consistently had significantly higher values than PP WRM and NP WRM (Table 3). Over all zones, PP VGPM was 1.4 times higher compared to PP WRM and NP VGPM was 2.07 times higher than NP WRM (Table 3). PP VGPM was 1.54 times higher in the spring and 1.28 times higher during summer and winter. The models showed better agreement in the autumn when PP VGPM was only 1.02 times higher than PP WRM . The differences between models varied by province and were greatest in the ICB by a factor of 1.45, and closest in EGC where there was a factor of 1.31 difference (Figs. 4, 6). The inter-annual variability also changed between models and hydrographic zones (Figs. 4, 6). For PP WRM Tables 4 and 5. Both models had the highest annual PP in the ICB and lowest in the CIC (Table 4). For NP, both models exhibited the lowest values in the CIC. The highest mean annual NP VGPM values were in the ICB, whereas NP WRM was slightly higher in the RKR (Table 5).

Climate forcing and primary production in the northern North Atlantic
The average annual cumulative sum of the anomalies in NP VGPM , and NP WRM were compared in the six hydrographic zones (Fig. 8). For these cumulative sums in NP VGPM and NP WRM , the trend was similar in all provinces and for both models, with a gradual decrease from 1998 to 2009, followed by an increase to 2010. The magnitude of the decrease was greater for NP VGPM compared to NP WRM in the ICB, NIC and CIC, which was on average 9.3, 1.3 and 1.1 times higher, respectively. In the ICB, the NP VGPM showed an initial rise in cumulative sums until 2000, whereas the NP WRM decreased until 2000. In the SIC, the decrease in cumulative sums was greater for NP WRM compared to NP VGPM . By contrast, in the EGC, there was little difference in cumulative sums between both models. The cumulative sums of the anomalies in NP for each model were then correlated against those for NAO, AO and MEI ( Table 6). The mean cumulative sum of the anomalies in NAO increased slightly from 1998 to 2000 and then gradually decreased until 2006, when there was a slight increase to 2007 followed by a decrease to 2010. NP WRM exhibited a significant positive correlation with NAO in the EGC and CIC. NP VGPM also showed a significant positive correlation with NAO in the EGC and CIC and additionally in the SIC.
Both NP VGPM and NP WRM showed significant negative correlations with the cumulative sum in AO, which was consistently higher in the EGC and CIC compared to the other hydrographic zones ( Table 6). The AO changed from negative in 1998 to positive from 1999 to 2009. The response to this positive phase and especially in EGC and CIC, was a consistent decrease in the mean cumulative sum of the anomalies in NP VGPM and NP WRM . From 2009 to 2010 there was a reduction in mean cumulative sum of the anomalies in AO, which was paralleled by a decrease in those for NP. NP VGPM exhibited a significant negative correlation with MEI in the ICB. There was an initial increase in NP VGPM In the ICB from 1998 to 2000 as MEI switched from positive to negative, followed by a decrease in NP VGPM from 2000 to 2007 as MEI switched from negative to positive (Fig. 8).

Validation of satellite algorithms of new and primary production in the northern North Atlantic
There has been much debate on the global applicability of satellite models since some areas exhibit atypical optical (Dierssen & Smith, 2000;Volpe et al., 2007) or photo-physiological characteristics (Sathyendranath, 2000). A comprehensive knowledge of the extent to which these vary, both spatially and temporally, is required so that global models can be fine-tuned to regional conditions. Even at local scales, the optical properties of the world's oceans are changing (Dierssen, 2010) and on-going validation of ocean colour algorithms is therefore necessary to continually monitor their accuracy. The northern North Atlantic is typified by low solar zenith angles during winter and long day length during summer, which can cause a high variation in phytoplankton photo-physiology (Moore, Lucas, Sanders, & Davidson, 2005). PP WRM and NP WRM were more accurate than PP VGPM and NP VGPM using both in situ and satellite data (Table 2). Campbell et al. (2002) assessed the performance of 12 satellite algorithms forced with in situ data, against 89 14 C measurements. WRMs, WIDIs such as the VGPM and time-independent models (TIM) showed a similar performance. Each model however, exhibited a regional dependency. In their study in the North Atlantic, data were available from only 20 stations and WRM and TIM types were more accurate. Carr et al. (2006) compared 24 satellite and ecological models forced with mean monthly satellite data. The WRM and DIM models proved to be closest to the mean of all 24 models, especially in the Atlantic. There was however no in situ PP data to perform a comprehensive model validation. Tilstone, Smyth, Poulton, and Hutson (2009) compared PP models in the Atlantic Ocean and found that a WRM was more accurate than VGPM in six out of nine Atlantic Ocean biogeographic provinces. Saba et al. (2010) compared 36 models at Bermuda Atlantic Time series Study (BATS) and the Hawaii Ocean Time series (HOT) using nearly two decades of in situ data. At both these sites the average bias of the biogeochemical ocean circulation models (BOGCMs) was almost twice that of the ocean colour models. Saba et al. (2011) then compared 36 models including 22 ocean colour models and 14 BOGCMs against a large in situ dataset of 14 C (N = 1156) and found that the average uncertainty across all models was 31% and WRM type models proved to be most accurate in eight out of ten regions. Similarly, we found for the North Atlantic that PP VGPM was 85% higher than in situ PP, whereas PP WRM was within 30%. This varied between hydrographic zones and was highest in the ICB and lowest in the EGC (Fig. 4, Table 2). For NP, VGPM was 72% higher than in situ data whereas for the WRM the difference was only 10% (Fig. 5). Using the models with SeaWiFS data, differences were greatest in the ICB and lowest in the EGC (Fig. 6, Table 4), however there were only a limited number of satellite match-ups (N = 5) with in situ NP. Measurements of NP are rare and there is an obvious need for more in situ measurements of NP to improve both remote sensing and ecosystem models and in turn to enhance our understanding of both food web dynamics and carbon cycling.

Causes of differences between satellite algorithms of new and primary production in the North Atlantic
To assess reasons for differences in model performance, we used sensitivity analyses to ascertain which of the input parameters produced the greatest error in PP estimates from the two models. Each variable used to compute PP was fixed at its mean and PP was then re-computed by varying the parameter through its observed maximum and minimum. The results are summarised as box plots in Fig. 9. The total mean error given below each plot, indicates the degree a variable influences the computation of PP; the higher the error, the greater the influence. For PP VGPM , C insitu and Z eu had the greatest influence while for PP WRM , C insitu and ϕ m exhibited the greatest effect. In both models, Z eu and ϕ m are derived from C insitu , which accounts for the high error in the computation of PP. That there was no significant difference between in situ and SeaWiFS Chla (Fig. 3, Table 2), suggests that the principal source of differences between models was not due to Chla. For the VGPM, plotting modeled Z eu against in situ Z eu derived from CTD profiles of PAR (Fig. 9C), illustrated a tendency for the VGPM to overestimate Z eu especially in low Chla waters (diamonds in Fig. 9C). The percentage difference between measured and modeled Z eu was 22% for the in situ stations. Behrenfeld and Falkowski (1997) concluded that their estimation of Z eu would contribute to b 15% of the error in the VGPM. The over-estimation of Z eu may partially explain why PP VGPM consistently resulted in higher values compared to PP WRM , especially at the lower range of PP (Figs. 3B, 6B, C, D). Similarly, Saba et al. (2011) found the depth of the water column to be the main parameter limiting the performance of most PP models. They observed that there was a tendency for the VGPM to under-estimate PP at bottom depths b 750 m and over-estimate PP at bottom depths N4000 m due to errors in the simulation of Z eu . Siegel et al. (2001) also found that the bias in the VGPM, when compared to a global in situ PP dataset, was more pronounced (− 38%) when Chla concentration was low, and therefore Z eu was deep. An under-estimate in Chla would lead to an over-estimate in Z eu and higher PP values at low in situ values, which we also observed (Figs. 2B, 3B). Carr et al. (2006) concluded from the PPARR3 inter-comparison that a better parameterisation of photo-physiology is required to improve model skill and accuracy. More recently, a comprehensive error analysis of the uncertainties associated with the VGPM also concluded that the largest individual contributor to the random uncertainty in PP was the input term that describes the physiological state of phytoplankton and changes in the rate of Chla-normalised photosynthesis over depth (Milutinovic & Bertino, 2011). Both the WRM and VGPM use SST to derive the photo-physiological term. We therefore assessed the temperature sensitivity of the photo-physiological term in each model by varying SST through the range in in situ values at each of the 83 stations sampled, whilst keeping Chla and PAR constant. P b opt and ϕ m were then computed at 1°C increments to estimate PP VGPM and PP WRM over the natural range in SST (Fig. 10). Both models exhibited a sigmoidal dependence to SST, with low sensitivity of PP over the temperature range 5 to 9°C, when PP VGPM was 15% higher than PP WRM . Both models exhibited a sharp increment in PP from 9 to 10°C, which was similar from 10-18°C. Over this temperature range PP VGPM was 40% higher than PP WRM . Milutinovic and Bertino (2011) found that only 9% of the variability in P b opt could be described by the polynomial function of SST, indicating that temperature alone is a poor predictor of photosynthetic rates. This is further compounded by the depth dependency of P b opt with irradiance, which is not well characterised in the VGPM. When we plot PP WRM against PP VGPM from data extracted every 10 km along S-N transects (shown in Fig. 1) in the six hydrographic zones (Fig. 11) and as a function of temperature; from 0 to 5°C, PP VGPM was 43% higher than PP WRM ; from 5 to 10°C PP VGPM was 29% higher and from 10 to 15°C PP VGPM was 27% higher. Over the entire SST range, PP VGPM was 35% higher than PP WRM , which illustrates the temperature dependency of each of the models in this region. The differences between the models in each hydrographic zone are related to the temperature range that was encountered in each zone. For example, differences were greatest in the ICB where the temperature range was between 8 and 18°C and lowest in the EGC where the temperature range was 5.5-15°C. This is reflected in the sensitivity analysis which indicated that from 5 to 9°C there will be a 15% difference between PP VGPM and PP WRM , whereas from 10 to 18°C, the difference between models was greater (Fig. 10). The dependency of P b opt is also determined by changes in irradiance and nutrients; however, these other variables often co-vary with temperature, which makes SST a convenient parameter to model changes in photosynthetic rates ( 1997), especially since SST can be sensed remotely at the same spatial and temporal scales as ocean colour. In the North Atlantic over the Scotian Shelf for example, 65% of the variation in the chlorophyllnormalised maximum photosynthetic rate (P m B ) can be described by temperature alone (Bouman et al., 2003). Below 20°C, increasing temperature results in an increase in P m B due to linear reaction kinetics (Raven & Geider, 1988). There are situations where temperature cannot account for changes in light and nutrients due to a decoupling between these parameters. For example, in warm, permanently stratified regions of the North Atlantic (typically in summer), where light is non-limiting, the P b opt dependency with SST does not account for the nutrient limitation that can occur. By contrast, in cool, well mixed environments, such as the northern North Atlantic during winter and spring when nutrients are replete, the depth dependency of P b opt with irradiance can become limiting. Under such conditions, P b opt has poor predictive capabilities of the photosynthetic rate and the uncertainty associated with the VGPM estimate of PP becomes large (Milutinovic & Bertino, 2011). Subsequently, a number of approaches to estimate changes in photo-physiology as a function of nutrient concentration or status have been proposed (Behrenfeld, Boss, Siegel, & Shea, 2005;Behrenfeld, Maranon, Siegel, & Hooker, 2002). These models are based on assumptions about ambient nutrient fields that are not available from remote sensing data. In addition, the correlation between dissolved inorganic nitrogen (DIN) and phytoplankton biomass or photosynthetic rates is often poor, because in winter when DIN is high, phytoplankton photosynthesis can be low due to light limitation. There can also be a lag response between a new supply of DIN and photosynthetic rate and once photosynthesis reaches a maximum at the peak of the bloom, DIN has been taken up by phytoplankton, which again results in a poor correlation between ambient DIN and photosynthetic rates (Maranon, Cermeno, Latasa, & Tadonleke, 2012). In view of these limitations, and since the kinetics of photosynthesis are tightly coupled to temperature especially over the range from 5 to 20°C, selection of the most accurate model that uses an SST function to describe photo-physiology is a priority for accurate estimation of both PP and NP. Some new approaches to the parameterisation of P m B as a function of SST have recently become available (Saux-Picart, Sathyendranath, Dowell, Moore, & Platt, 2014) and warrant further validation to find the most accurate PP and NP models for both global and regional seas. Although both the VGPM and WRM use SST to describe changes in photosynthetic rates, we observed that variations in SST had a greater effect on P b opt in the VGPM than they did on ϕ m in the WRM. Behrenfeld and Falkowski (1997) argued that variability in PAR and its capacity to influence the relative depth of light saturation has a small impact on the resulting PP. Milutinovic and Bertino (2011) found that the PAR function used in the VGPM can result in large uncertainty (~12%) in PP. By comparison, the WRM incorporates a spectralirradiance dependence of ϕ m , which varies over depth as irradiance becomes attenuated and over time, as irradiance decreases either side of zenith. Despite the shortcomings of SST in describing variations in photo-physiology, the parameterisation of ϕ m in the WRM formulated over two decades ago (Morel, 1991), provides more accurate estimation of PP and NP compared to the VGPM. On a global basis, Milutinovic and Bertino (2011) found that the systematic positive errors in P b opt in the VGPM contributed to a 6% bias in each pixel. The accumulated error from individual pixels for 2005, led to an overestimate of 2.5 Pg C of the 46.1 Pg C annual global PP. We found that there was a 28% difference in PP between the VGPM and WRM in 2005, which corresponded to a difference of 76 gC m − 2 yr − 1 for the northern North Atlantic alone. These differences in PP between models propagated to differences of 50% in NP estimates for the northern North Atlantic.

Intra-and inter-annual variability in new and primary production in the Irminger Basin in different hydrographic zones
NP and the f-ratio are important components of food web dynamics and carbon cycling. There are limited in situ measurements of NP available however, and especially in the North Atlantic. Based on in situ measurements, previous studies in the Greenland Sea showed that new production during a Phaeocystis sp. bloom can reach 1100 mgC m − 2 d − 1 (Smith et al., 1991). Anderson, (2000) reported NP up to 5700 mgC m −2 d −1 for the Greenland Sea. From the in situ data we collected, NP varied from 100 to 2800 mgC m −2 d −1 in the Iceland Basin and 1000 to 2600 gC m −2 d −1 in the Celtic Sea. Based on changes in in situ N, Sanders et al. (2005) reported 36 gC m −2 yr −1 in the Irminger Basin in 2002. Annual new production based on temperaturenitrate relationships and Argo float data or silicate uptake has been predicted to be 65 gC m −2 yr −1 in 2002 (Henson et al., 2003(Henson et al., , 2006. By comparison, in this study the mean NP WRM from 1998-2009 over all hydrographic zones was~41 gC m −2 yr −1 . From satellite data with an empirical algorithm, NP in the Irminger basin had previously been reported to be between 100-150 gC m −2 yr −1 (Falkowski, 1988;Laws Table 3 One Way Analysis of variance between VGPM-derived and WRM-derived differences in primary (PP) and new (NP) production. VGPM is vertical generalized production model; WRM is wavelength resolving model. F is the mean square to mean square error ratio and P is the ANOVA critical significance value. Level of significance to P = 0.05 is indicated; * is 0.05-0.01, **0.01-0.001, *** is b0.  , 2000). By comparison, using the WRM we found NP to vary between 26 gC m −2 yr −1 in the CIC in 2002 to 70 gC m −2 yr −1 in the RKR in 2008, which is the upper limit reported for the seasonally stratified NW European Shelf (Heath & Beare, 2008). By comparison, for the northern North Atlantic NP VGPM was between 45 and 166 gC m −2 yr −1 . Biological production in the Irminger Basin can often be limited by silicate (Allen et al., 2005;Sanders et al., 2005), which becomes depleted in late spring potentially leading to a limitation of diatom growth (Henson et al., 2006). These studies suggest that a nutrient proxy for photosynthesis would best describe PP and NP. The estimates of NP by Henson et al. (2006) used the VGPM, but constrained the model using Si(OH) 4 limitation in the region and their results (~65 gC m −2 yr −1 ) were lower than those of Falkowski, Barber, and Smetacek (1998) and Laws et al. (2000). More recently, Henson et al. (2011) developed a model of export production using a large database of thorium ( 234 Th) derived export measurements, which they extrapolated globally through correlations between PP and SST, and from this, estimated global integrated carbon export to be 5 GtC yr −1 lower than previous estimates, which is 60% lower than the e/f-ratio based estimate. The PP input used in the Henson et al. (2011) model was from three algorithms (Behrenfeld & Falkowski, 1997;Carr, 2002;Marra, Trees, & O'Reilly, 2007). Based on our study, if PP WRM were used as input to the Henson et al. (2011) model, the resulting export production would be 7-8 GtC yr −1 lower. Our results suggest that the PP model used as input to either new or export production model is critical in computation of these parameters.

Variability in new production as a function of climate indices
Predicted changes in temperature, stratification and nutrient input can affect the f-ratio and thus impact upper trophic levels and carbon export. Saba et al. (2010) found that between 1989 and 2007 in situ PP at BATS and HOT increased by an average of nearly 2% per year, which at the BATS site was forced by more frequent mixing events during negative NAO phases. They found that none of the satellite models tested could reproduce this change over time. Over the past 30 years, there have been major fluctuations in atmospheric forcing in the northern North Atlantic which is reflected in the NAO (Hakkinen & Rhines, 2004). Positive NAO results in a later spring bloom in North Atlantic sub-polar waters (Henson, Dunne, & Sarmiento, 2009). A large positive NAO index is usually associated with strong westerly winds and low pressure around Iceland (Hurrell & Deser, 2009). Bentsen, Drange, Furevik, and Zhou (2004) observed that convective mixing in the Labrador and Irminger Seas is linked to the NAO such that negative NAO leads to deeper vertical mixing in these regions. This may suggest that negative NAO in winter results in deeper convective mixing, which could enhance the supply of nutrients to the photic zone for the onset of the spring bloom and thus affect the magnitude of the annual new production. Herein, NP WRM and NP VGPM were positively correlated with NAO in the EGC and CIC. In addition, NP VGPM was also correlated with NAO in the SIC. During the SeaWiFS time series from 1997-2000, there was an intense reversal of the winter NAO index (Hakkinen & Rhines, 2004), which caused an initial increase in the average annual cumulative sum of the anomalies in NAO and a decrease in those of NP (Fig. 8). From 2001-2005 the NAO then fluctuated between weak positive and negative (Hakkinen & Rhines, 2009), which lead to a decrease in the average annual cumulative sum of the anomalies in NAO and NP. After 2006, the NAO switched to positive and the average annual cumulative sum of the anomalies increased, whereas those for NP continued to decrease in parallel with further decrease in NAO from 2007to 2009. From 2009 to 2010 the average annual cumulative sum of the anomaly in NP increased, though this could be due to the summer 2010 eruption of the Icelandic volcano Eyjafjallajökull (Achterberg et al., 2013). Volcanic particles are known to cause failure in satellite atmospheric correction, and in turn could cause anomalies in ocean colour which may have caused increases in annual NP values. Zhai, Platt, Tang, Sathyendranath, and Walne (2013) observed two competing mechanisms that affect the mixed-layer depth in the North Atlantic linked to the NAO. During positive NAO years, in the central North Atlantic the vertical mixing induced by strong westerly winds deepens the mixed layer. In the sub-Arctic and northern North Atlantic positive NAO years enhances the southerly transport of cold and fresh Arctic water which promotes strong stratification and the mixed layer shoals. This partially explains the decrease in average cumulative sums in the anomalies in NP in the EGC and CIC. These changes in response to NAO contrast with those in the Sargasso Sea, where negative NAO is associated Table 4 Mean annual satellite estimates of primary production (PP) gC m −2 yr −1 for different biozones in the Irminger Basin. VGPM is vertical generalized production model; WRM is wavelength resolving model. EGC-Eastern Greenland Current; NIC-North Irminger Current; SIC-South Irminger Current; CIC-Central Irminger Sea; RKR-Reykajnes Ridge; ICB-Iceland Basin, TOTAL is the spatial average of all zones.
Since 1998 the Arctic Oscillation has alternated between positive and negative, with a record negative phase in the winter of 2009-2010 (Foster, Cohen, Robinson, & Estilow, 2013). On an annual basis, the mean AO was negative in 1998, 2000-01, 2004-05 and 2009-10 and positive in 1999, 2002-03 and 2006-08. The positive phase of the AO brings ocean storms northwards, which makes the weather wetter in North Atlantic (Wang, Wang, Yang, & Lu, 2005). Pabi, van Dijken, and Arrigo (2008) showed that from 1998 to 2006 PP increased in open water areas of the Arctic and reached a maximum in 2006. They also showed that inter-annual variability in PP is tightly coupled to changes in sea ice and that variations in SST and incident irradiance contribute less to the variability in PP. When the springtime AO is strongly positive, snow melts earlier. When it is strongly negative, snow disappears later in the spring (Foster et al., 2013). Modelling studies have also suggested that during positive AO there will be an increase in PP and diatoms due the reduction in ice cover (Walsh, Dieterle, Maslowski, & Whitledge, 2004). By contrast, in the neighbouring Irminger Sea we observed a strong negative correlation between the average annual cumulative sum of the anomalies in AO and both NP VGPM  Table 6 Pearson rank correlation between mean annual cumulative sums of the anomalies in NP from the WRM and VGPM, North Atlantic Oscillation (NAO), Arctic Oscillation (AO) and multiple ENSO index (MEI) for different hydrographic zones from 1998-2010. r is Correlation coefficient, P is level of significance. Significant correlations are given in bold; * indicates significant correlation at the 5% level; ** is 0.5% level and *** is 0.05% level.

Model
Bio and NP WRM in all hydrographic zones, indicating that as AO becomes more positive, NP decreases (Fig. 8). This may be related to the fact that positive AO phase corresponds to the warm weather conditions further south (Wang et al., 2005), which would enhance stratification of the water column and thus reduce nutrient availability. This could also decrease NP (Fig. 8).
The effects of MEI on PP have been reported along the Californian Coast and as far North as British Columbia (Goes et al., 2004). More recently this has been extrapolated to the Global Ocean (Messie & Chavez, 2012). A decrease in global PP from 1998 to 2006 has been coupled to an increase in global SST as a result of the shift in MEI index (Behrenfeld et al., 2006). Goes et al. (1999Goes et al. ( , 2000 found that the primary driving force on NP in the sub-Arctic Pacific was the wintertime monsoon winds, which are stronger during El Niño. Based on remotely sensed estimates, they found that NP was highest in 1997 and 1998 corresponding with strong ENSO events and new production was lower when La Niña occurred. To further assess differences between NP models in relation to climate forcing, we compared the cumulative sums of the annual anomalies in MEI and NP VGPM and NP WRM in the six hydrographic zones (Fig. 8). The only significant correlation was between NP VGPM and MEI in the ICB, which during La Niña (negative MEI) events from 1998-2003 caused a decrease in the average cumulative sum of the anomalies, whereas El Niño years (positive MEI) from 2004-2010, caused a reduction in the decrease followed by an increase in NP VGPM . This correlation was not apparent for NP WRM . Interestingly, in the ICB, NP VGPM exhibited the greatest difference compared to NP WRM and was 2.44 times higher (Fig. 6, Table 5). Since MEI and NAO are calculated from differences in SST and due to the sensitivity of the PP VGPM to fluctuations in SST, NP VGPM exhibited significant trends with NAO and MEI in more hydrographic zones than NP WRM . It is therefore not surprising that  Behrenfeld et al. (2006) reported a strong coupling between global PP VGPM and MEI.

Conclusions
In situ 14 C and 15 N uptake measurements were used to assess the accuracy of the VGPM and WRM at estimating NP and PP in the northern North Atlantic. Both in situ and satellite estimates of Chla and SST were used as inputs to the models. Overall, the WRM was more accurate than VGPM at estimating both PP and NP which were within 30 and 10% of in situ data, respectively. The difference between models was greatest in the ICB where the temperature range was between 8 and 18°C and lowest in the EGC where the temperature range was 5.5-15°C. Both models however, had the highest annual PP in the ICB and the lowest annual PP in CIC.
The inter-and intra-variability in NP and PP calculated using SeaWiFS data from 1998 to 2010 in the northern North Atlantic with both models was also compared. PP and NP estimated using the VGPM was always higher than the WRM and the difference was greatest at low values. NP for the WRM was 20-70 gC m −2 yr −1 , whereas NP for the VGPM was 45-166 gC m −2 yr −1 . The effect of climate forcing on NP in contrasting hydrographic zones showed that the VGPM is more sensitive to variations in NAO and MEI compared to the WRM. Both models exhibited a significant negative correlation between the average annual cumulative sum in the anomalies of AO and NP.
The cause of differences in PP and NP using the WRM and VGPM were assessed. A sensitivity analysis on model input parameters illustrated that Chla has the greatest effect on both models, followed by Z eu and P b opt in PP VGPM and ϕ m in PP WRM . Since Chla was the same input for both models, errors in estimates of PP VGPM were principally from Z eu and the temperature dependency of P b opt , whereas for PP WRM these were principally from ϕ m . Over the temperature range found in the northern North Atlantic, the errors associated with estimating P b opt , resulted in PP VGPM to be 15-40% higher than PP WRM which propagated to a 50% difference between NP VGPM and NP WRM over 13 years of SeaWiFS data. The difference between the VGPM and WRM in each hydrographic zone was related to the temperature range encountered in each zone.
Based on these results, we recommend using the WRM of PP from Morel (1991) coupled with the NP model of Laws et al. (2000) to provide the most accurate satellite estimates of these parameters in the northern North Atlantic. Fig. 11. Satellite estimates of primary production (mg C m −2 d −1 ) for the VGPM and WRM extracted every 10 km from SW to NE transects through each hydrographic province (see Fig. 1 for location of transects) over the temperature range from (A.) 0-5°C, (B.) 5-10°C, (C.) 10-15°C. The colour bar is SST (°C). Dashed line is the regression line; solid line is 1:1 relationship.