Long-term trends in pH in Japanese coastal seawater

In recent decades, acidification of the open ocean has shown a consistent increase. However, analysis of longterm data in coastal seawater shows that the pH is highly variable because of coastal processes and anthropogenic carbon inputs. It is therefore important to understand how anthropogenic carbon inputs and other natural or anthropogenic factors influence the temporal trends in pH in coastal seawater. Using water quality data collected at 289 monitoring sites as part of the Water Pollution Control Program, we evaluated the long-term trends of the pHinsitu in Japanese coastal seawater at ambient temperature from 1978 to 2009. We found that the annual maximum pHinsitu, which generally represents the pH of surface waters in winter, had decreased at 75 % of the sites but had increased at the remaining sites. The temporal trend in the annual minimum pHinsitu, which generally represents the pH of subsurface water in summer, also showed a similar distribution, although it was relatively difficult to interpret the trends of annual minimum pHinsitu because the sampling depths differed between the stations. The annual maximum pHinsitu decreased at an average rate of −0.0024 yr−1, with relatively large deviations (0.0042 yr−1) from the average value. Detailed analysis suggested that the decrease in pH was caused partly by warming of winter surface waters in Japanese coastal seawater. The pH, when normalized to 25 C, however, showed decreasing trends, suggesting that dissolved inorganic carbon from anthropogenic sources is increasing in Japanese coastal seawater.

than 2000 monitoring sites that cover most parts of the coastline (Fig. 1), so the dataset provides the 100 opportunity to estimate the overall trend in pH in Japanese coastal areas and the regional variability in 101 the trends from data of known precision. Suitable analytical methods could make up for these 102 shortcomings of the WPCL dataset. In this study, we focused on the general characteristics of the 103 overall pH trends at the all monitoring sites rather than examining the trend in pH at each site in detail, 104 after carefully considering the accuracy of the dataset. 105 In the present study, we examined the pHinsitu trends in surface coastal seawater from data measured 106 as part of WPCL monitoring programs. We then examined the trends at specific locations. The 107 remainder of this manuscript is organized as follows: the data and methods are described in Section 2, 108 and trends in pHinsitu are presented in Section 3, the results are discussed in Section 4, and the 109 concluding remarks are provided in Section 5. (www.nies.go.jp/igreen; http://www.nies.go.jp/igreen/md_down.html). We downloaded pHinsitu data 116 from 1978 to 2009 for the trend analysis. We also downloaded temperature (T) and total nitrogen (TN) 117 data that were measured at the same sites as the pHinsitu data for the same time period (data for T and 118 to ±0.05. The pH was measured immediately after the water samples were collected, at the ambient 139 water temperature. The repeatability permitted in each measurement was ±0.07. The pH data were assumed that the minimum and maximum pH data coincided with the highest and lowest temperatures, 153 respectively (Fig. 2), and we used these data to calculate the pH25 in Section 4.2. 154 The monitoring operations were carried out by licensed operators as outlined in the annual plan of 155 the Regional Development Bureau of each prefecture. These specific licensed operators were retained for the duration of the measurement period, which means that the same laboratories were always in 2.2 Quality control procedures and assessing the consistency of the WPCL monitoring data 161 We selected all the data for fixed sites in coastal seawater that had continuous time series from sites covered almost all the coastline in Japan (Fig. 1). 165 As explained in more detail later in this section, we applied a three-step quality control procedure. 166 We excluded 1) discontinuous time sequences, 2) time sequences that had extreme outliers in each year, 167 and 3) time sequences that included significant random errors, and which were only weakly correlated 168 with time sequences at adjacent sites.

169
When we excluded the sites that had discontinuous pHinsitu time sequences from 1978 to 2009, 1481 170 sites remained (Fig. 1). We then excluded time sequences with outliers, defined as sites with data points 171 that were more than three standard deviations from the average of minimum and maximum pHinsitu 172 values for each year. After this step, 1127 sites remained (not shown). We calculated the trends in the 173 unbroken continuous time sequences of the minimum and maximum pHinsitu data at each site with 174 linear regression (Fig. 3), and the slopes of the linear regression were taken as the minimum and maximum pHinsitu trends (e.g. Fig. 3). The linear regression trends might have been influenced by 176 random errors or variations at different temporal scales in the data for each site. To eliminate the 177 influence of these errors and variations as far as possible, we removed the data that had significant were correlated at adjacent monitoring sites in the same prefecture ( Fig. 4). At most of the stations, the 183 correlations between the temperatures at the site pairs were relatively strong, which indicates that the 184 temperature followed similar patterns over time at adjacent sites ( Fig. 4a−b). The correlations tended 185 to be strong when the sites were close together, but gradually weakened with increasing distance 186 between sites. The pHinsitu correlations followed a similar pattern (Fig. 4), which indicates that the 187 pHinsitu and temperature data at adjacent monitoring sites varied in the same way. In other words, the 188 relative ratios of the measurement errors in pHinsitu and the natural spatio-temporal variations at these 189 monitoring sites were similar to those for temperature. The absolute values of the correlation 190 coefficients for the pHinsitu were slightly lower than those for temperature for each corresponding pair 191 of sites (Figs. 4 and 5), and might reflect the fact that pHinsitu, but not the water temperature, is subjected 192 to strong forcing by coastal biological processes and other severe physical processes in summer, which 193 causes the pHinsitu to vary on the short-term. The correlations between the minimum pHinsitu data (Fig.  4c) were weaker than those for the maximum pHinsitu data (Fig. 4d) because the degree of biological 195 forcing varied by season and was stronger in summer when the pHinsitu was at a minimum and weaker 196 in the winter when the pHinsitu was at a maximum. Despite the influence of biological processes on the 197 pHinsitu, the correlation coefficients remained high and were significant (r=0.367, p<0.05) at most of 198 the monitoring sites, especially at sites that were less than 5 km apart within the same prefecture, where 199 the pHinsitu followed similar patterns. In the final step of the quality check procedure (step 3), we 200 removed all the time sequences with weak and insignificant correlations for temperature and pHinsitu 201 (Fig. 5), because we considered that the monitoring sites having both significant correlations for water 202 temperature and pHinsitu were reliable. After this final step, 289 sites remained. As shown in Table 2, 203 the correlations between temperature and pHinsitu at sites within 15 km of each other strengthened after 204 steps 2 and 3, which suggests that the reliability of the dataset improved at each step of the quality 205 control.

206
The monitoring in each prefecture is carried out by different licensed operators, decided by the

227
The number of sites decreased at each step of the quality control, but the shapes of the histograms were 228 generally similar for both the minimum and maximum pH trends. The total trends showed overall 229 normal distributions with a negative shift at all levels of quality control. 230 We detected both positive (basification) and negative (acidification) trends, which contrasts with  The negative trends were relatively weak for the minimum pHinsitu data and relatively strong for 239 the maximum pHinsitu data, but there was an overall tendency towards acidification. At the 289 sites, were acidification and basification trends at 70% and 30% of the monitoring sites, respectively, and at 243 75% and 25% for the maximum pHinsitu data, respectively. 3.2 Local patterns in acidification and basification 246 We examined the pHinsitu trends for the 289 sites for local patterns in acidification and basification 247 (Section 2.2) and found that the trends seemed to be randomly distributed. For example, the values 248 were different at sites that were less than 50 km apart (Fig. 8). There are many monitoring sites in the for the maximum pHinsitu were larger than those for the minimum pHinsitu in these prefectures. 262 We found more acidification trends for the minimum pHinsitu in the southwestern prefectures of  and 0.0020 yr −1 , respectively (Fig. 7). The average trend in the maximum pHinsitu, however, shifted 287 from zero in a negative direction at a rate of more than 0.0020 yr −1 for all three scenarios (Fig. 7b, d,

288
f). This result implies that, averaged over the whole country, the Japanese coast was acidified in winter to a degree that could be detected from the historical WPCL pH data, even with an uncertainty of ±0.1.

290
The observed standard deviation for the maximum pHinsitu was also larger than the expected value of 291 0.0020 yr −1 because of local variations in the pH trends. The average shift in the minimum pHinsitu data 292 was smaller than 0.0020 yr −1 , but all three scenarios showed negative shifts in the average minimum 293 pHinsitu value (Fig. 7a, c, e). 294 We used Welch's t test to assess the direction of the average minimum and maximum pHinsitu trends.  Welch's t test confirmed that the average trends for both the minimum and maximum pHinsitu data were 301 negative.

302
We also applied a paired t test to determine whether the two trends calculated from the averaged 303 minimum and maximum pHinsitu data were significantly different. The population mean and the sample 304 size were 0.0 and 289, respectively. The t value of 4.64 (with 288 degrees of freedom) shows that the 305 null hypothesis was rejected, with the paired t test thus indicating that the two trends calculated from 306 the averaged minimum and maximum pHinsitu data were significantly different.

308
The WPCL dataset did not discriminate between surface (0.5-2 m) and subsurface (10 m Usually the pH is lower in subsurface water than in surface water, as primary production decreases 317 and increases the DIC concentrations in surface and subsurface water, respectively, because of 318 decomposition when Particulate Organic Carbon (POC) is produced by primary producers. We 319 therefore speculate that the annual maximum pH includes very little data from a depth of 10 m, and so 320 this value does represent the winter pH of surface waters. In contrast, the annual minimum pH was 321 somewhat difficult to interpret, as it may have contained data from 10 m at some monitoring sites but 322 only surface data at other sites shallower than 10 m.

323
Results of statistical analysis (Section 4.1) confirm that the trends in minimum and maximum pHinsitu 324 data tended to be negative in the seawater around Japan. The negative tendency of the annual maximum 325 pHinsitu trends may imply a trend of overall acidification in winter in surface waters around the Japanese 326 coasts, but the pattern in the annual minimum pHinsitu trends was difficult to interpret. Nevertheless, the annual minimum pHinsitu trends were, as for the annual maximum pHinsitu, also negative (Section

374
The wide variations in DIC (B (T, N)) and Alk (S) between regions might have caused the regional 375 differences in pHinsitu trends among stations, contributing to relatively large standard deviations in both 376 the minimum and maximum pHinsitu trends (Fig. 7). The three-step quality control procedures 377 effectively removed the sites with high variability due to analytical errors, and this process may also 378 have removed the effect of large local processes (e.g. heavy phytoplankton bloom, or freshwater 379 discharge change). Nevertheless, we still are able to detect regional scale difference in distribution of 380 positive/negative trends (e.g. Fig.8). Therefore, we discuss the effects of global processes on the 381 overall average pH trends and of regional effects, separately, in later sections (Sections 4.3.1 and 4.3.2).
where a1 was set to −0.015 and T was the observed temperature.

398
The distributions of the trends in pH25 after applying equation 3 are shown in Fig. 10. The minimum 399 and maximum pH25 data were normally distributed, meaning that the distributions of the pHinsitu trends 400 were maintained after applying equation 3 (Fig. 7e, f). The averages (± standard deviations) of the 401 minimum and maximum pH25 trends were −0.0010±0.0032 and −0.0014±0.0041 yr −1 , respectively. We found regional differences in the pHinsitu values (e.g. Fig. 6) and pHinsitu trends (Figs. 8−9). The 437 negative pHinsitu trends (acidification) were more significant in southwestern Japan than in northeastern 438 Japan, especially for the minimum pHinsitu data ( Fig. 9 and Section 3.

455
We assumed that the TN was mainly dissolved inorganic nitrogen and determined the correlations 456 between TN and the minimum and maximum pHinsitu trends (Fig. 14). There were statistically suggests that the coastal environment might not be completely devastated by acidification. If we can manage the coastal environment effectively (e.g., control nutrient loadings and 518 autotropic/heterotrophic conditions), we might be able to limit, or even reverse, acidification in coastal 519 areas.    Monitoring sites that met the strictest criterion (n = 302).

719
The value of pH25 was estimated using the method of Lui and Chen (2017).    Table 2 Average mutual correlation coefficients among water temperature and pHinsitu measurements at 741 adjacent monitoring sites in the same prefecture. The averages were calculated from the data for the 742 highest and lowest temperature, and minimum and maximum pHinsitu within 15 km for the three 743 criteria. We refined the sites using three quality control steps, yielding 1481 (step 1), 1127 (step 2), 744 and 302 (step 3) sites. The two columns on the right show the significance level of 5% and the degrees 745 of freedom for the correlation coefficients of each quality check procedure. 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12    Scatter plots of correlation coefficients for water temperature and pH insitu at adjacent monitoring sites in the same prefecture. Fig. 5a is for the highest temperature and the minimum pH inisitu data and Fig. 5b for the lowest temperature and the maximum pH insitu data, respectively.        Table 2 Average mutual correlation coefficients among water temperature and pH insitu measurements at adjacent monitoring sites in the same prefecture. The averages were calculated from the data for the highest and lowest temperature, and minimum and maximum pH insitu within 15 km for the three criteria. We refined the sites using three quality control steps, yielding 1481 (step 1), 1127 (step 2), and 302 (step 3) sites. Two right columns represent a significant level of 5% and a degree of freedom for the correlation coefficients of each quality check procedure.