Two decades of inorganic carbon dynamics along the Western Antarctic Peninsula

Propose spatial gradients seen Abstract We present 20 years of seawater inorganic carbon measurements collected along the 37 western shelf and slope of the Antarctic Peninsula. Water column observations from 38 summertime cruises and seasonal surface underway pCO 2 measurements provide unique 39 insights into the spatial, seasonal and interannual variability of this dynamic system. 40 Discrete measurements from depths >2000 m align well with World Ocean Circulation 41 Experiment observations across the time-series and underline the consistency of the data 42 set. Surface total alkalinity and dissolved inorganic carbon data showed large spatial 43 gradients, with a concomitant wide range of Ω arag (< 1 up to 3.9). This spatial variability 44 was mainly driven by increasing influence of biological productivity towards the 45 southern end of the sampling grid and melt water input along the coast towards the 46 northern end. Large inorganic carbon drawdown through biological production in 47 summer caused high near-shore Ω arag despite glacial and sea-ice melt water input. In 48 support of previous studies, we observed Redfield behavior of regional C/N nutrient 49 utilization, while the C/P (80.5 ± 2.5) and N/P (11.7 ± 0.3) molar ratios were significantly 50 lower than the Redfield elemental stoichiometric values. Seasonal salinity-based 51 predictions of Ω arag suggest that surface waters remained mostly supersaturated with 52 regard to aragonite throughout the study. However, more than 20 % of the predictions for 53 winters and springs between 1999 and 2013 resulted in Ω arag < 1.2. Such low levels of 54 Ω arag may have implications for important organisms such as pteropods. Even though we 55 did not detect any statistically significant long-term trends, the combination of ongoing 56 ocean acidification and freshwater input may soon induce more unfavorable conditions 57 than the ecosystem experiences today.

represented by a first order linear model. We then randomly divided the PAL-LTER surface measurements (depth <5 m) into 10 data subsets using the 10-fold cross validation method [Stone, 1974;Breiman, 1996]. Using 9 of the ten data sets we derived a regression model, predicted the TA with the model, and calculated the model coefficients and root mean square errors (RMSE). We repeated these steps so every data subset was left out once. The coefficients for the final model were calculated from the mean of the ten regression coefficients. We found the best fit in the following equation: TA pred (μmol kg -1 ) = 57.01 (± 0.88) x S +373.86 (± 35.26), which resulted in a linear correlation coefficient of r = 0.88 and a RMSE of 15.2 μmol kg -1 ( Figure A2). In combination with the pCO 2 measurement precision of 3 μatm, the RMSE of TA prediction resulted in a mean error in calculated Ω arag of 0.0219 units and pHT of 0.0043 [Glover et al., 2011]. Note that the calculated Ω arag and pHT estimates implicitly require that the approximately linear summertime TA-salinity relationship holds for the other seasons, a reasonable assumption if dilution and mixing substantially affect TA patterns." The change in sea ice in the region is also mentioned a number of times in the paper, but there is no description of how sea ice is changing for the region (seasonally, interannual and spatial change over 20 years) apart from one sentence in the introduction. If there is more information on sea-ice change, it seems important to include and discuss the relevance in influencing the biogeochemical properties. Also, I was also not able to find a good description of the physical oceanography of the region data are presented from. Again, there are mentions here and there in the text, but it is scattered and difficult to follow.

Response:
To address this comment we added text and several citations to the Introduction Section to better describe in more detail the regional ocean physical circulation.
Lines 98-110: "The physical oceanography of the region is strongly influenced by equatorward flow at the continental shelf/slope break associated with the eastward flowing Antarctic Circumpolar Current that abuts the continental slope along the WAP region. On the shelf, there are indications of one or more cyclonic circulation cells with poleward flow inshore [Hofmann et al., 1996;Dinniman and Klinck, 2004;Martinson et al., 2008]. Water mass properties are strongly influenced by subsurface intrusions onto the continental shelf of warm, nutrient and DIC rich Upper Circumpolar Deep Water (UCDW), that appears to be modulated by topographic depressions and canyons Dinniman et al., 2011;Martinson and McKee, 2012]. In winter, respiration processes and the entrained deep CO 2 -rich water increase the DIC concentration in surface waters to supersaturated levels of CO 2 with respect to the atmosphere [Carrillo et al., 2004; Wang et al., 2009;Tortell et al., 2014;Legge et al., 2015]." Changes of sea ice and their potential influence on biogeochemistry are discussed in the introduction and discussion. Readers interested in more details on climate and sea-ice trends are referred to the cited articles including (Ducklow et al., 2007 and2012; Stammerjohn et al., 2012).
Other comments: p 6931, lines 9-11: It is not clear how la Nina years influence the carbon cycle dynamics. Do more intense storms and a poleward displacement of the polar jet have an influence. There is a description of possible changes in carbon cycling for SAM. What does the literature indicate is happening to sea ice extent over the period?

Response:
We adjusted the text in response to these questions> Lines 137-145: "During La Niña years, storms become longer and more intense, temperatures increase and sea ice extent decreases in the WAP region as a result of a strong low-pressure system driven by the poleward displacement of the polar jet [Yuan, 2004]. Positive SAM phases are also associated with positive temperature anomalies over the Antarctic Peninsula and decreased sea-ice extent [Kwok, 2002;. Furthermore, the SAM brings the Southern Hemisphere westerly winds closer to Antarctica, which amplifies the typical features of La Niña. During these periods, nutrient and CO 2 -rich Circumpolar Deep Water intrudes more frequently on to the shelf , potentially increasing [CO 2 ] on the shelf." p 6933, section 2.1:Nutrient data are used in the paper, but I cannot find information on how these data were measured and where to access these.

Response:
We agree with the referee that more detailed information about the nutrient sampling technique is required. To address this concern we added the following text: Line 192-216 " "Water for inorganic nutrient analysis was subsampled from Niskin bottles into acid washed 50 mL Falcon tubes and frozen at -70 °C.. The samples were analyzed using a Lachat Quickchem 8000 autoanalyzer at the University of California at Santa Barbara Marine Science Institute Analytical Lab from 1993 2007 and later at the Marine Biological Laboratory (Woods Hole MA, 2008 -2012). Inorganic nutrient data reach a precision of ± 1%. All PAL-LTER data and a detailed description of the sampling methodology are publicly available at http://pal.lternet.edu/ (dissolved inorganic nutrients, PAL LTER dataset 27)." p 6935, line 7: how many outlier (per cent) were excluded out of the total number of samples? The text suggests there may be an analytical problem. I suspect this isn't the case, but the sentence beginning "These outliers included..." indicates there were many more than described in the section.
Response: There were only 4 outliers. We agree that this sentence was confusing and adjusted it (Lines 263-265): "Four PAL-LTER pCO 2 outliers that underestimate/overestimate pCO 2 relative to the underway observations by more than 150 μatm were removed." p 6935, lines 10-19: I am not sure what the point of this regression analysis broken down into different years is. I first thought the intercept might be meaningful, but it seems more like the authors are trying to check the internal consistency of their measurements. Why not consider the residuals? Is the need to split the years used to compare pCO2 measured and pCO2 calculated an indication that the quality of the measurements has issues some years? If so, please state what years and why?
Response: In this paragraph we are discussing potential reasons why the correlation between the calculated PAL-LTER pCO2 and the measured LDEO pCO2 is not >0.82. Among matching issues and the resulting large pCO2 set-offs, we believe that years with a smaller range of pCO2 variability may be responsible for a large error of the intercept parameter, leading to a lower correlation. We agree that this may be confusing and not useful and therefore adjusted text and figure accordingly.
Lines 265-273: "Analysis of the corrected data set with a Linear Regression Type II model suggests a correlation of r = 0.82 ( Figure A1, Table 1). Some of the observed discrepancies may be attributed to errors in matching the times of bottle samples with those of underway pCO 2 measurements. Seawater inorganic carbon chemistry is highly variable along the WAP due to the influence of productivity, respiration, freshwater and upwelling of CO 2 -rich subsurface water [Carrillo et al., 2004]. Small matching errors may therefore introduce small DIC and TA offsets, which would translate into larger fractional differences in pCO 2 due to the large Revelle Factor (∂ ln pCO 2 / ∂ ln DIC) common in the region [Sarmiento and Gruber, 2006]" P 6936, section 2.4: Why are nutrient concentrations ignored in the TA vs salinity relationship given what appears to be a large range in pCO2 and presumably nutrient concentrations? Nutrient data are used with TA on page 6939. I am also unclear on the relevance of this salinity vs TA relationship. Most of the following sections in the paper do not seem to use the relationship as there are TA, DIC and nutrient data used to calculate the carbonate system parameters, or is this incorrect? Section 3.4 does use the relationship and it would be helpful to state in section 2.4 that it is used later with data pCO2 data to calculate the saturation state in fall, winter and spring seasons when bottle data are not available.

Response:
We agree with the reviewer that the use of the salinity vs TA relationship was not clear. We therefore merged section 2.4 with section 3.4 and reworded the text. This should clarify that the TA-salinity relationship, in combination with surface pCO2 measurements, was used to calculate Ω arag for fall, winter and spring when DIC and TA bottle data are not available. Because nutrient data are also not available for these seasons, we did not use nutrients to predict salinity, even though we agree with the reviewer that it would probably improve the TA prediction. These are the first two paragraphs of the newly merged section 3.4: Lines: 536-557 "3.4 Seasonal variability To get insights into the carbon dynamics during winter, spring, and fall, when direct measurements of DIC, TA and nutrients are either scarce or not available, we developed a regional TA algorithm (based on PAL-LTER summertime data). In combination with seasonal LDEO pCO 2 , salinity and temperature data, we calculated Ω arag for the missing seasons. Due to the weak correlation between PAL-LTER temperature and TA (r = 0.50), we based the TA algorithm on salinity only ( Figure A2, r = 0.88). Applying the Akaike information criterion [Burnham and Anderson, 2002], we determined that TA along the WAP will be best represented by a first order linear model. We then randomly divided the PAL-LTER surface measurements (depth <5 m) into 10 data subsets using the 10-fold cross validation method [Stone, 1974;Breiman, 1996]. Using 9 of the ten data sets we derived a regression model, predicted the TA with the model, and calculated the model coefficients and root mean square errors (RMSE). We repeated these steps so every data subset was left out once. The coefficients for the final model were calculated from the mean of the ten regression coefficients. We found the best fit in the following equation: TA pred (μmol kg -1 ) = 57.01 (± 0.88) x S +373.86 (± 35.26), which resulted in a linear correlation coefficient of r = 0.88 and a RMSE of 15.2 μmol kg -1 ( Figure A2). In combination with the pCO 2 measurement precision of 3 μatm, the RMSE of TA prediction resulted in a mean error in calculated Ω arag of 0.0219 units and pHT of 0.0043 [Glover et al., 2011]. Note that the calculated Ω arag and pHT estimates implicitly require that the approximately linear summertime TA-salinity relationship holds for the other seasons, a reasonable assumption if dilution and mixing substantially affect TA patterns." p 6937, section 3.1: This is OK, but it averages data from the Summer, when there is large variability. The point that there are large and persistent decreases inshore relative to offshore is well defined. However, the section does not indicate the range of values used in the averaging. For example, what range of sDIC and salinity values occurs inshore compared to offshore for the averaged data points. It would be good to get some idea of the variability.

Response:
We think that the Anderson paper is an appropriate citation here as it presents a nice dataset of alkalinity measurements from glacial streams. Both citations refer to the fact that DIC and TA are much lower in sea ice (Yamato-Kawai) and glacial Anderson) meltwater. The part of the sentence that mixing of seawater with meltwater leads to dilution of [CO32-] is an explanation for the reader of the effect of lower TA and does not need a citation. p 6944, lines 11-12: These refer to DIC drawdown in the WW layer as biological, which seems reasonable as an ultimate cause of drawdown. I suppose this drawdown will occur in the summer season? Is this correct and why can't the DIC decrease in Figure 5 be due to mixing of surface water into the WW layer or mixing of lower DIC WW water from other regions.

Response:
We agree with the reviewer that in addition to biological DIC drawdown in the WW layer, other physical mechanisms may have an influence. For example, vertical mixing could play a role either by mixing in low DIC surface waters during the summer or during the prior winters when low DIC surface water is contributing to the formation of winter water, which then carries a signal of past surface productivity.
Lines 758-762: "The observed DIC drawdown in the winter water ( Figure 5 and A3) may be a result of biological productivity, which is supported by previous observations of Chl a maxima in the euphotic part of the winter water, likely due to increased iron concentrations there [Garibotti et al., 2003;Garibotti, 2005]. However, it is also likely that lateral advection or vertical mixing of low DIC water into the winter water have caused this signal. p 6944, lines 14-18: Is the text here referring to Figure 5? This is the only figure I could locate that shows anything that might relate to the text.

Response:
This text does not refer to a specific figure. The information was taken out of a numerical analysis and discusses the low levels measured off-shore (see section 3.1). We now mention that there is no figure for this statement. Lines 764-766: "Low Ω arag values (< 1.35) observed offshore coincided with surface waters supersaturated with regard to atmospheric CO 2 , salinities >33.5, and temperatures between 1.3 -1.5 °C (not shown)." p 6945, lines 10-25: Why would not accounting for the drivers of TA influence the TA vs salinity relationship? If TA+nutrients are used, it may help the relationship with salinity, but the authers have not done this. Invoking ikaite is unlikely to explain the differences. The occurence of ikaite in sea ice is limited and it is not clear how changes in a 1-2 m sea ice layer spread over a 50m mixed layer could have much effect (ie any effect would be diluted in the 50m thick mixed layer). This section is not much more than a statement that TA variability could be explained by just about any process. One other possible explanation is the TA measurements have a large amount of error although the methods section states the measurements are high accuracy.

Response:
We constructed the TA relationship to estimate seasonal TA and Ω arag . Because there are no nutrients available for the other seasons, there was no use in constructing a TA relationship based on nutrients. This is now more clearly described in the following lines: Lines 536-540: "To get insights into the carbon dynamics during winter, spring, and fall, when direct measurements of DIC, TA and nutrients are either scarce or not available, we developed a regional TA algorithm (based on PAL-LTER summertime data), and in combination with seasonal LDEO pCO 2 , salinity and temperature data, calculated Ω arag for the missing seasons" We don't see why we can't discuss potential reasons for the large TA variability and therefore did not change the text.
p 6946 line 16-20: Why have two high values been singled out to consider the decadal rates of change in the central sub-region? The fall and spring are when rapid change might occur and it is not clear from Table 3 or the text if this is a persistent pattern each year or due to limited data. The more interesting data may be for winter when biological effects are small compared to Spring. Here, the decadal trend is small in the central region and similar to the atmospheric increase in the north region. Do these changes agree with Takahashi's previous estimates and why the differences? The same applies to the fall and spring rates of change (ie why the regional differences?).

Response:
We redid the trend analysis based on the corrected version of the LDEO pCO2 data set that was recently published (Takahashi et al., 2015). Furthermore, we restricted our analysis to the central sub-region, which corresponds with the LTER sampling region. As a result, none of the trends are significant anymore, which shows how difficult it is to distinguish between real secular trends and natural variability. It also corresponds with analysis done by Munroe et al., [in press]. This correction led to a variety of adjustments in the text and table 3: Lines 55-58: "Even though we did not detect any statistically significant long-term trends, the combination of ongoing ocean acidification and freshwater input may soon induce more unfavorable conditions than the ecosystem experiences today." Lines 630-637: "3.5 Temporal trends Trend analysis of the PAL-LTER data showed no statistically significant annual trends (at the 95% confidence level) in the measured carbon parameters, temperature or salinity in surface waters in summer between 1993 and 2012 ( Table 2). As a comparison, we conducted a trend analysis for the LDEO surface underway pCO 2 data set (1999 -2013) in the same region. LDEO observations show an increasing, but not statistically significant trend in surface pCO 2 , supporting our results above ( Table 3). The largest increasing trend was found in fall, (1.9 ± 0.95 μatm yr -1 ), but this trend was also slightly outside the confidence interval and therefore statistically not significant." Lines 790-804: "The large uncertainties in our estimated temporal trends are caused inherently by the large spatial and temporal variability of our data. Nevertheless, our mean rates of 1.45 ± 2.97 for summer and 0.43± 0.77 μatm yr -1 for winter suggest that the surface water pCO 2 has been increasing at a slower rate than the atmospheric pCO 2 rate of about 1.9 μatm yr -1 , and that the air-to-sea CO 2 driving potential has been increasing. Our results may be compared with the recent analysis of the 2002-2015 time-series data obtained across the Drake Passage by Munro et al. [in press]. In the waters south of the Polar Front (their Zone 4, closest to the LTER area), they observed that the surface water pCO 2 increased at a rate of 1.30 ± 0.85 μatm yr -1 in summer and 0.67 ± 0.39 μatm yr -1 in winter, which are comparable with ours along the WAP. We observed the strongest but still insignificant increase in surface pCO 2 in fall (1.9 μatm year -1 , p = 0.0685). This increase corresponds with the mean atmospheric pCO 2 increase of 1.9 μatm per year, which causes a pHT decrease of about 0.02 per decade [Takahashi et al., 2014]. Interestingly, Stammerjohn et al., [2008aStammerjohn et al., [ , 2008b found that sea ice extent and wind are also changing most rapidly in spring and fall, which may enhance sea-air gas exchange and therefore facilitate positive pCO 2 trends. Furthermore, it is likely that the strong counter effect of biological productivity successfully masks the pCO 2 trend in summer, and decreased gas exchange due to sea ice weakens the trend in winter. However, the WAP climate and oceanography are regulated by large-scale atmospheric patterns, such as El Niño Southern Oscillation and Southern Annular Model [Stammerjohn et al., 2008a], which may also influence the region's inorganic carbon chemistry on an interannual scale. A longer measurement period may be needed in order to be able to distinguish with certainty between natural variability and secular trends [Henson et al., 2010]." Response to anonymous referee #2: This manuscript represents a very important evaluation of one of the highlights of longtem monitoring of the carbonate system in the Southern Ocean -the PAL-LTER program. The wider utilization of the summertime data to enable extrapolation to annual scales in conjunction with the more prolific surface pCO2 data illuminates the changing nature of the carbonate system and thus ocean acidification. The structure of the work done is very logical and well laid out. The carbonate system reporting and data analysis is generally performed and well described in accordance with common practice. However, data normalization to deep water values between cruises is not performed. Certain broad assumptions are made regarding the development of carbonate system proxies, nutrient utilization and the physical setting that weaken the scientific merit of the paper and subsequent interpretation of the results.The language of this manuscript would benefit from a general sharpening of the text. The sentences are often long and statements and descriptions of scenarios are repeated. Overall, this manuscript is a valuable contribution to the scientific field and after I suggest that this manuscript be accepted for publication after successfully addressing or challenging the comments laid out below.
General comment: Regular mispelling of ueq -replace with _meq. Use pHT throughout to clearly denote the scale.

Response:
We changed pH to pHT. We measured/reported TA in ueq/kg and it is therefore not a misspelling. Since ueq/kg TA is equivalent to umol/kg and because this is becoming the more commonly used unit, we changed it throughout the document. Abstract P6930 L5 "this" dynamic system Done L6 change "The discrete" to "Discrete" Done L8 remove "Analysis shows". Propose "Large spatial gradients were seen in: : :" Done L8 total alkalinity Done L9 remove "from values" and bracket (<1 to 3.9) Done L17 These were not "predictions". They were calculated values but even this is not necessary here. Just use aragonite saturation.
The seasonal values of aragonite saturation state were salinity-based Introduction P6931 L5 use general "change" Done L9 higher trophic organisms. Krill and fish are not species. Done L11 oceanographic Done P6932 L1 not sure what you mean by "timing of sampling". Time of year? Added Lines 132-133: "…, but possibly also the timing of sampling in relation to the timing of sea ice retreat and phytoplankton blooms" L1-2 remove "dark" and "months". Done L8 remove "a" and change timescale to timescales Done L22. Replace "has" with "have". Done P6933 L11 remove "of each transit" Done L26 "variables" not "parameters" Done P6934 L20 replace "calculations" with "procedure" or "program" Done L24 remove "well" Done P6935 L18 remove "of" Done L24. They are "offsets" in CT and AT and not "errors" Done L25 "differences" not "errors" Done P6936 L6. There is no direct AT v T plot and no correlation information in Figure A2. Response: The correlation information is in the text. We now point to the figure in relation to the salinity-TA algorithm: Lines 540-541: "Due to the weak correlation between PAL-LTER temperature and TA (r = 0.50), we based the TA algorithm on salinity only ( Figure A2, r = 0.88)." L20. An evaluation of the error in calculated pH would be useful here too. Done Lines 552-554: "In combination with the pCO 2 measurement precision of 3 uatm, the RMSE of TA prediction resulted in a mean error in calculated Ω arag of 0.0219 units pHT of 0.0043 [Glover et al., 2011]." P6937 L10 variables Done P6938 L3 Remove "above-presented" Done L10 replace with "can decrease (increase) Done L16 How robust is this assumption considering the high ammonium stock in the WAP region (e.g. Nutrients in the Southern Ocean GLOBEC region: variations, water circulation, and cycling Serebrennikova and Fanning, 2004) Response: We agree with the reviewer that some data show surprisingly high ammonium stocks in this region. But typical levels are around ~1 uM/liter over much of the study area, which are much lower than NO3 (Serebrennikova and Fanning, 2004). Furthermore, net community production, which is therefore likely based on NO3 uptake, is responsible for the DIC drawdown. This is clarified here: Lines 421 -425: "Since nitrate is more abundant than ammonium in WAP surface waters [Serebrennikova and Fanning, 2004], nitrate was assumed as the nitrogen source. With a Redfield stoichiometry of 6.6 mol C/mol N then TA should increase by 1/6.6 = +0.15 mol TA per umol DIC consumed. Precipitation of biological CaCO 3 material reduces both DIC and TA with the effect on TA twice as large as that on DIC (2 umol /umol)." P6939 L1 How can you interpret this from Figure 5? The reader should not have to evaluate this from interpreting depth from the salinity. Response: As indicated in the figure, all dots that don't have a black or grey frame are upper-ocean data. L2 Where is this "excess" AT coming from relative to the end members? Highlight this leading to the discussion on P6945.
Response: This and all other findings are discussed in the discussion (lines 801-815). We don't think that it is necessary to put more emphasis on this finding than on others in the results. L25. Why have you used a constant PCO2 of 390? Why not use the relevant annual (or even better, seasonal) values over the measurement period? Response: We think that constant atmospheric pCO2 is good enough for this back-of-the-envelop calculation, especially given the fact that we only have seawater pCO2 measurements from January or February, which are extrapolated to rest of the summer months. L26 Is this globally averaged transfer rate representative of the Southern Ocean? Response: The gas transfer rate used for this calculation is the estimated gas transfer rate for the Southern Ocean and is not as previously stated a global mean. We thank the reviewer for paying close attention and are glad that we caught this mistake. Lines 480-484: "To account for DIC concentration changes due to gas exchange with the atmosphere, we assumed a constant atmospheric concentration of 390 μatm between 1993 and 2012, and a gas transfer rate (k) of 5 (±1) milli-mol CO 2 m -2 μatm -1 month -1 , which is the estimated mean rate for the Southern Ocean area south of 62 °S [Takahashi et al. 2009]." P6940 L3 As the MLD can be easily calculated from the CTD profiles, why choose a "d" of 50m. The episodic nature of wind-stress and a rapidly evolving MLD require that a much more locally informed, at a minimum a monthly climatological value should be used. Response: There are no CTD profiles for the months November and December. We therefore chose to use the published value of summer average mixed layer depth.

P6943
L7 sDIC We don't understand this comment.

P6944
L14 replace "overlapped" with "coincided" Done P6946 L18 replace "what was" with "that" Done L25 Calculate not predicted pH replaced with "estimated" P6947 L2 Replace "Additional decades" with e.g. " A longer measurement period" Done L4 replace "predicted" with "calculated" replaced with "estimated" L16 "to" be able to Done Table 1. Legend: Remove "statistics for" Done Why were only selected years chosen for Figure A1? Figure now shows all years. Table 2. According to your criterion, none of the trends are statistically significant. This needs to be stated more clearly. Why are the regional trends not shown? These are much more important than the dataset mean.
Response: "3.5 Temporal Trends" states this clearly: Lines 630-637: "Trend analysis of the PAL-LTER data showed no statistically significant annual trends (at the 95% confidence level) in the measured carbon parameters, temperature or salinity in surface waters in summer between 1993 and 2012 (Table 2)." Trends from the north are not shown because PAL-LTER data is only available from the central sub-region. Figure 2. These plots clearly show the offset between cruises in the deep water. Why were the data not corrected according to the practice adopted for CARINA, for example? Or can you show that the offset are due to spatial differences? Response: Since 1980s, DIC measurements were calibrated using the CRM (produced by A. Dickson, SIO), which was, in turn, based upon C. D. Keeling's manometric CO 2 determinations. The CRM used are reported to be accurate to ± 1 umol/kg. The WOCE/CLIVAR section and Palmer time-series ocean DIC data presented in Fig.  2 are all based upon the CRM, and the precision of the shipboard DIC measurements has been estimated to be about ± 2 umol/kg. Although measurements are also subjected to expedition-to-expedition variability, differences in DIC values exceeding ± 3 umol/kg may be attributed to time-space variability of the ocean. We will add a brief mention of underlying commonality of the CRM calibration to the text (Line 241).
As noted by the reviewer some inorganic carbon synthesis projects such as GLODAP and Carina have used deep-water cross-over analysis and related techniques to generate suggested corrections for DIC. In the regional deep-water DIC data shown in Figure 2, we did not feel that there were sufficient offsets between the Palmer DIC data and the WOCE/CLIVAR DIC to warrant any offset. Figure 3. This is not a very clear figure. The data density is too great and the colour coding is too similar for many of the years. Please simplify or remove. Response: We agree with the reviewer. We changed the color-coding to all black (now Figure 3) and just pointed out a few special data points that were mentioned in the text. Figure 5. This figure does not, contrary to its legend, depict the physical and biological controls on inorganic carbon chemistry We removed the title. Figure 6. Similarly, the legend is misleading. Not all the processes leading to the movement in TA/DIC space are of biological nature. The grey dots and lines should have a slightly darker shading. We removed the title and adjusted the figure. Figure 7. Please explain better the plot in the legend. "Nutrient consumption" is incomplete and incorrect regarding the lower plot. "Nutrient consumption" is removed Figure 8. There are no "dynamics" shown in this plot. Replaced with "system" Figure 9. Here is stated that after "clear outliers were removed". In both plots there are differences between the two approaches of 150ppm. What criterion was used to define that these were also not clear outliers? Response: As stated in the text, we removed all outliers that showed a bigger difference than 150 ppm. To make this more clear, we added the following sentence to the figure caption: Lines 1159-1160: PAL-LTER pCO2 outliers that underestimate/overestimate pCO2 relative to the underway observations by more than 150 uatm were removed Please also see the figure below, which shows that the difference between the two datasets is < 150 uatm after removing the outliers. We present 20 years of seawater inorganic carbon measurements collected along the 37 western shelf and slope of the Antarctic Peninsula. Water column observations from 38 summertime cruises and seasonal surface underway pCO 2 measurements provide unique 39 insights into the spatial, seasonal and interannual variability of this dynamic system. 40 Discrete measurements from depths >2000 m align well with World Ocean Circulation 41 Experiment observations across the time-series and underline the consistency of the data 42 set. Surface total alkalinity and dissolved inorganic carbon data showed large spatial 43 gradients, with a concomitant wide range of Ω arag (< 1 up to 3.9). This spatial variability 44 was mainly driven by increasing influence of biological productivity towards the 45 southern end of the sampling grid and melt water input along the coast towards the 46 northern end. Large inorganic carbon drawdown through biological production in 47 summer caused high near-shore Ω arag despite glacial and sea-ice melt water input. In 48 support of previous studies, we observed Redfield behavior of regional C/N nutrient 49 utilization, while the C/P (80.5 ± 2.5) and N/P (11.7 ± 0.3) molar ratios were significantly 50 lower than the Redfield elemental stoichiometric values. Seasonal salinity-based 51 predictions of Ω arag suggest that surface waters remained mostly supersaturated with 52 regard to aragonite throughout the study. However, more than 20 % of the predictions for 53 winters and springs between 1999 and 2013 resulted in Ω arag < 1.2. Such low levels of 54 Ω arag may have implications for important organisms such as pteropods. Even though we 55 did not detect any statistically significant long-term trends, the combination of ongoing 56 ocean acidification and freshwater input may soon induce more unfavorable conditions 57 than the ecosystem experiences today.

Comparison with deep-water WOCE/CLIVAR inorganic carbon system data 233
We checked the consistency of the PAL-LTER DIC and TA data by comparing PAL-234 Ocean -Variability, Predictability, and Change (CLIVAR) cruises along parts of sections 237 A21 and S4P that were overlapping with the PAL-LTER grid (data available at 238 http://www.nodc.noaa.gov/woce/wdiu/). The WOCE and CLIVAR shipboard 239 measurements were calibrated using seawater certified reference materials (prepared by 240 A. G. Dickson, Scripps Institute of Oceanography), leading to an estimated precision of 241 ±2 µmol kg −1 . DIC was measured on all cruises. When necessary, TA was calculated 242 from DIC and either fCO 2 or pCO 2 following the same procedure as described in Section 243 7 2.1. Figure 2a shows the stations along the WAP where deep-water samples were taken 250 during PAL-LTER and WOCE cruises. PAL-LTER DIC and TA measurements were 251 within the range of sampled/calculated DIC and TA from the WOCE and CLIVAR 252 cruises, (Figures 2b and c). After removing five outliers, mean deep-water DIC (DIC mean 253 = 2260.6 ± 3.8 µmol kg -1 ) and TA (TA mean = 2365.4 ± 7.0 µmol kg -1 ) from PAL-LTER 254 cruises corresponded well with the data measured/calculated from WOCE cruises 255 (DIC mean = 2261.8 ± 3.0 µmol kg -1 ; TA mean = 2365.9 ± 9.3 µmol kg -1 ). 256 257

Comparison with underway-surface pCO 2 data 258
We also undertook a quality check of the PAL-LTER discrete surface DIC and TA data 259 Here, we examine the observed spatial summer patterns of DIC, TA, pHT and Ω arag along 276 the WAP and explore the underlying biological and physical drivers. We then discuss 277 regional carbon -nutrient drawdown ratios and present our seasonal Ω arag predictions that 278 give initial insights into the chemical environment in the more poorly sampled spring, fall 279 8 and winter months. Finally, using the LTER and LDEO data sets, we investigate temporal 320 trends over the past two decades.  (Figures 3b and c), 344 decreasing along the coast towards the north to pHT ~8.2 and Ω arag ~1.9, and reaching 345 the lowest levels in northern offshore waters (pHT min = 8.1; Ω arag min = 1.7). TA also 346 exhibited north-south and onshore -offshore gradients, with values as low as 2185 µmol 347 kg -1 in the northern near-shore regions and as high as > 2300 µmol kg -1 offshore. The low 348 TA values along the northern part of the coast coincided with the lowest salinity values of 349 31.8, suggesting dilution of TA due to freshwater input (Figures 3d and e). Higher TA 350 values offshore were also reflected in increased DIC and salinity concentrations, with 366 temperatures between 1.3 -1.5 °C. DIC also exhibited an onshore-offshore gradient with 367 values about 80 to 100 µmol kg -1 lower in the near shore region compared to offshore, but 368 there was no significant north-south gradient despite the presence of freshwater in the 369 north (Figure 4f). Salinity normalized DIC (sDIC, normalized with UCDW salinity = 370 34.7) was lowest in the southern region, thereby indicating that biological processes 371 likely counteracted the expected north-south DIC gradient due to the pronounced 372 freshwater influence on DIC in the north (Figure 4g). 373 374

Physical and biological drivers of the inorganic carbon system 375
In this section we examine the physical and biological mechanisms that control the 376 observed variability in DIC and TA. DIC can decrease (increase) through dilution with 377 freshwater (evaporation), organic matter production (remineralization), CO 2 outgassing to 378 the atmosphere (CO 2 uptake) and/or precipitation of CaCO 3 (dissolution). While positive 379 net community production decreases DIC, the biological effect of organic matter 380 production on TA depends on the source of nitrogen, where nitrate consumption 381 increases TA and ammonium consumption decreases TA [Goldman and Brewer, 1980]. kg -1 is visible in the winter water (grey diamonds), which increased to more than 200 422 µmol kg -1 in the mixed layer, leading to Ω arag as low as 1.5 and as high as 3.9. 423 The DIC drawdown relative to the salinity mixing-dilution line is most likely due 424 to biological production of organic matter. Figure 6 shows sDIC as a function of salinity-425 ΔpCO 2 (pCO 2 atm -pCO 2 ML ) was between -143 µatm and 312 µatm, as pCO 2 ML ranged 462 from 533 µatm to 78 µatm, indicating that there was potential for both oceanic CO 2 463 uptake and outgassing. Assuming that d = 50 m [Ducklow et al., 2013], we estimate that 464 the monthly ΔDIC due to air-to-sea CO 2 gas exchange was in the range of -14 to 31 µmol 465 kg -1 month -1 . Since the first large phytoplankton blooms generally occur after the sea-ice 466 retreats in November (Δt ~3 months), we assume that by the time of sampling at the end 467 of January, ΔDIC would fall in the range -43 to 94 µmol kg -1 . The DIC corrected for gas 468 exchange is illustrated as grey dots in Figure 6. While applying the gas exchange proportions C/N/P = 106:16:1 [Redfield, 1958]. Our findings suggest that the carbon-478 nutrient cycles along the WAP depart from the standard Redfield values (Figure 7). In a 479 few samples, the standing stock of PO 4 3became depleted before NO 3 -, and overall the 480 regression indicates a low N:P ratio of 9.8 ± 0.4 in the mixed layer (Figure 7a, black) and 481 N:P = 11.7 ± 0.3 for all data (dark grey) relative to the standard Redfield value of 16 482 molN/mol P. The mole/mole C:P ratio was also considerably smaller than the Redfield 483 ratio (Figure 7b). C:P yielded 43.1 ± 2.3 in the mixed layer and 55.0 ± 1.7 for all data. 484 However, after applying the gas exchange correction on DIC (see section 3.2), the C:P 485 ratio shifted closer to the Redfield Ratio and resulted in a value of 80.5 ± 2.5 (light grey 486 dots and lines). Correcting the DIC for gas exchange shifted the molar ratio from 4.5 ± 487 0.2 (mixed layer depth) and 4.7 ± 0.1 (all data) to 6.7 ± 0.2 and resulted in a Redfield-like 488 C:N ratio. 489 490

Seasonal variability 491
To get insights into the carbon dynamics during winter, spring, and fall, when direct 498 measurements of DIC, TA and nutrients are either scarce or not available, we developed a 499 regional TA algorithm (based on PAL-LTER summertime data). In combination with 500 seasonal LDEO pCO 2 , salinity and temperature data, we calculated Ω arag for the missing 501 seasons. Due to the weak correlation between PAL-LTER temperature and TA (r = 0.50), 502 we based the TA algorithm on salinity only ( Figure A2, r = 0.88). Applying the Akaike 503 information criterion [Burnham and Anderson, 2002], we determined that TA along the 504 WAP will be best represented by a first order linear model. We then randomly divided 505 the PAL-LTER surface measurements (depth <5 m) into 10 data subsets using the 10-fold 506 cross validation method [Stone, 1974;Breiman, 1996]. Using 9 of the ten data sets we 507 derived a regression model, predicted the TA with the model, and calculated the model 508 coefficients and root mean square errors (RMSE). We repeated these steps so every data 509 subset was left out once. The coefficients for the final model were calculated from the 510 mean of the ten regression coefficients. We found the best fit in the following equation:  in winter and spring (Figure 8b). Some summertime TA was predicted to be as low as 578 2056 µmol kg -1 . 579 Prediction of seasonal Ω arag revealed that surface waters of the WAP were 580 supersaturated with regard to aragonite throughout the years (Figure 8c). The most 581 frequent occurrence of low Ω arag was in winter and spring, when most of the predicted 582 values resulted in Ω arag < 1.4. 20 % of spring and winter values were Ω arag < 1.2, with the 583 lowest predicted surface Ω arag reaching near aragonite undersaturation in winter. Similar 584 to the LTER observations, predicted summertime Ω arag displayed a large range, spanning 585 from 1.1 to 4.1, with the majority of predictions between 1.3 and 1.8. Biological 586 production in summer is sufficiently intense to prevent low Ω arag values during the active 587 growing season when its effects might be most pronounced. 588 589

Temporal trends 590
Trend analysis of the PAL-LTER data showed no statistically significant annual trends 591 (at the 95% confidence level) in the measured carbon parameters, temperature or salinity 592 in surface waters in summer between 1993 and 2012 ( Table 2). As a comparison, we 593 conducted a trend analysis for the LDEO surface underway pCO 2 data set (1999 -2013) 594 in the same region. LDEO observations show an increasing, but not statistically 595 significant trend in surface pCO 2 , supporting our results above (Table 3). The largest 596 increasing trend was found in fall, (1.9 ± 0.95 µatm yr -1 ), but this trend was also slightly 597 outside the confidence interval and therefore statistically not significant. The 20 year-long PAL-LTER seawater inorganic carbon time-series showed a distinct 601 upper-ocean spatial pattern of onshore-offshore and north -south gradients and suggests 602 that the summertime carbon dynamics are primarily controlled by biological productivity 603 and freshwater input in near-shore areas. 604 Surface Ω arag was distributed across a wide range (<1 to values > 3) in freshwater-605 influenced areas with salinities S < 32 ( Figure 5). To better understand how such a wide 606 range of Ω arag at relatively low salinities was possible, we quantified the effect of 607 freshwater and biological production. Mixing of seawater with sea-ice or glacial 608 meltwater leads to a 'dilution' of CO 3 2ions and a decrease in Ω arag because TA and DIC 679 in glacial and sea-ice meltwater are much lower than in seawater [Anderson et al., 2000; productivity, which is supported by previous observations of Chl a maxima in the 718 euphotic part of the winter water, likely due to increased iron concentrations there 719 [Garibotti et al., 2003;Garibotti, 2005]. However, it is also likely that lateral advection 720 or vertical mixing of low DIC water into the winter water have caused this signal. 721 Low Ω arag values (< 1.35) observed offshore coincided with surface waters 722 supersaturated with regard to atmospheric CO 2 , salinities >33.5, and temperatures 723 between 1.3 -1.5 °C (not shown). These physical properties are associated with modified 724 UCDW, a mixture between UCDW and Antarctic Surface Water [Smith et al., 1999] and 725 indicate that upwelling of DIC and TA rich water into the mixed layer may lead to lower 726 Ω arag conditions offshore [Carrillo et al., 2004]. 727 The PAL-LTER data indicate N:P uptake ratios lower than the Redfield ratio of Subantarctic South Pacific [Hales and Takahashi, 2012]. Consistent with the low N/P 736 ratio, the observed C:P ratio (80.5 ± 2.5 ,corrected for gas exchange) was also lower than 737 the classic Redfield ratio. This indicates that the regional phosphate cycle shows non-738 Redfield behavior, which is in agreement with the observed C:P ratio of 91.4 ± 7.9 in the 739 mixed layer south of the Polar Front [Rubin et al., 1998]. A2) relative to other studies that either had additional parameters at hand (i.e. O 2 or 757 nutrients) to derive inorganic carbon system parameters in coastal environments [Juranek 758 et al., 2009;Kim et al., 2010;Evans et al., 2013] or that used salinity algorithms to 759 predict TA in open-ocean regions [Takahashi et al., 2014]. Furthermore, TA varied by 760 more than 70 µmol kg -1 at salinities >33.7, which led to an unbalanced distribution of 761 residuals ( Figure A2c springtime Ω arag were near Ω arag = 1 and 20 % were between 1.0 and 1.2 (Figure 8). The large uncertainties in our estimated temporal trends are caused inherently by 790 the large spatial and temporal variability of our data. Nevertheless, our mean rates of 1.45 791 ± 2.97 for summer and 0.43± 0.77 µatm yr -1 for winter suggest that the surface water 792 pCO 2 has been increasing at a slower rate than the atmospheric pCO 2 rate of about 1.9 793 µatm yr -1 , and that the air-to-sea CO 2 driving potential has been increasing. Our results Stammerjohn et al., [2008a, 2008b] found that sea ice extent and wind are also changing 803 most rapidly in spring and fall, which may enhance sea-air gas exchange and therefore 804 facilitate positive pCO 2 trends. Furthermore, it is likely that the strong counter effect of 805 biological productivity successfully masks the pCO 2 trend in summer, and decreased gas 806 exchange due to sea ice weakens the trend in winter. However, the WAP climate and 807 oceanography are regulated by large-scale atmospheric patterns, such as El Niño 808 Southern Oscillation and Southern Annular Model [Stammerjohn et al., 2008a], which 809 may also influence the region's inorganic carbon chemistry on an interannual scale. A 810 longer measurement period may be needed in order to be able to distinguish with 811 certainty between natural variability and secular trends [Henson et al., 2010]. 812 813

Conclusions 814
This study gives new insights into the spatial and temporal variability of the WAP 895 inorganic carbon system and its main physical and biological drivers. In particular, we 896 found that large inorganic carbon drawdown through biological production in summer 897 caused high near-shore Ω arag , despite glacial and sea-ice melt water input. Furthermore, 898 the data do not show a significant long-term trend in any of the inorganic carbon 899 chemistry variables measured. Continuation and expansion of the inorganic carbon 900 chemistry timeseries across other seasons is necessary to distinguish between natural 901 variability and secular trends and to better understand synergistic effects of ocean 902 acidification and climate change. Due to the region's physical complexity of circulation 903 and forcing, and strong dynamic response to climate variability, we recommend 904 development of a highly resolved biogeochemical model to complement our 905 observational work. Implementation of modeling studies will improve our mechanistic 906 understanding of how interannual variability and anthropogenic climate change impact 907 the inorganic carbon chemistry along the WAP, which is imperative to predict the 908 potential impact on the unique WAP ecosystem. 909

910
Author Contributions 911 Designed research: HD and TT. Field sampling and analytical measurements: TT, HD 912 and ME. Data analysis and interpretation: CH with help from all co-authors. Wrote the 913 paper: CH with help from SD, TT, and HD.

915
Acknowledgements 916 We thank past and present members of the Palmer LTER program as well as the captains 917 and crew of the U.S. Antarctic research vessels. We are especially grateful to Richard 918 Iannuzzi and James Conners for their support with data management, and to Tim 919 Newberger for underway pCO 2 measurements. We gladly acknowledge support from the