---

GO BACK TO OCEAN HYDROGRAPHY

---

Introduction

For the last fifteen years, the most complete and consistent source of "observed" data for ocean modelers working on both regional and global analyses has been the Levitus [1982, 1994] atlas. Recently, Lozier, Owens and Curry [1995] presented a new climatology for the North Atlantic (Hydrobase). To allow better evaluation of the variability in the basin they paid particular attention to the determination of mean properties. Here we present the Hydrobase climatology for the North Pacific.

Hydrobase is a set of:

• Quality controlled hydrographic observations [Curry, 1996]
• Executables which allow the observed data to be isopycnally averaged and gridded on to user specified surfaces.

Advantages of Hydrobase:

• It is extremely flexible to the ever increasing resolution needs of modelers because it can be gridded by the individual user.
• Isopycnal averaging avoids the creation of unrealistic water properties which are produced when averaging on depth or pressure surfaces.

The Data Source

The initial data for this climatology were the bottle data of World Ocean Atlas 1994 [Levitus, 1994] Marsden squares designated as North Pacific lying between 70N and the equator. Every observation was required to include both a temperature and a salinity measurement. Stations which had failed all three (monthly, seasonal and annual) of the WOA94 statistical checks [Levitus and Boyer, 1994] were excluded from the start. Stations with bottom depths less then 200 m although converted to Hydrobase format, were excluded from the quality control as a different set of statistical criteria would be be needed for the highly variable near shore observations. Due to the shift in the oxygen values resulting from changes in titration standards in the mid 1950s, all the oxygen data taken prior to 1960 were removed. The resulting raw data set(Fig.1) was comprised of some 268,000 stations.

The Quality Control Procedure

Designed to follow the work of Curry [1996], the quality control (QC) procedure is a statistical analysis performed in isopycnal bins. We use the same three basic steps in the quality control to identify outliers: a range check to remove obviously bad data points which could scew the calculation of the mean, a statistical check to eliminate observations lying outside a defined range of the local -S and -O2 relationships and a visual scan of a number of horizontal property and property variance plots to allow for a broader check, especially in those regions which had been divided very finely geographically in the statistical check. Two additional steps were included prior to the statistical check: Removal of unreliable data and a removal of duplicate observations.

To perform the statistical check the data were first divided into geographical subregions in which the local -S and O2-S relationships were well defined. The idea being to obtain as a tight a relationship as possible while simultaneously retaining sufficient observations to obtain meaningful statistics (Fig. 2).

Having the divided the basin horizontally each local region was then divided in the vertical into a set of isopycnal bins. Within each bin a mean property relationship and slope was defined. These linear segments define the property curve within the bins (Fig. 3). Once the isopycnal bins were chosen, all observations which fell outside 2.3 (a 97.9% confidence radius delimited by the thick lines in Fig. 3) were eliminated along with any observation not falling in a bin, all values in bins with less than 4 observations and all observations in stations in which greater than 25% of the observations were discarded for whatever reason. The statistics were then recalculated and the elimination procedure was repeated.

!Click on the thumbnail to see a bigger version of the figure!
Figure 1. The station positions of the source data are taken from the observed data provided by the World Ocean Atlas 1994.	Figure 2.Illustration of the size ( in square degrees) of the horizontal geographical binning used in the statistical check	Figure 3.An example showing the segmented linear fit to the -S curve. The dots represent the position of the mean with the isopycnal bin. The lines represents the slope of the curve at the mean and at +/- 2.3 standard deviations from the mean.

The check of the -O2 (or S-O2) values proceeded in a similar fashion except that a 3 limit was used and rather then removing entire observations the bad oxygen values were simply flagged. All discarded -S observations have been kept so that users may access them if they so desire. A record has been kept of the discarded oxygen data.

Approximately 3/4 (200,000) of the raw stations (Fig. 1) lie with a 30 radius of Japan, not surprisingly within the path of the Kuroshio. Much of these data were taken by the Japanese navy and fisheries prior to 1945. Unfortunately, these data are suspect due to insufficient calibration in the salinity measurements. The number of such bad observations was so large that meaningful statistics could not calculated without their removal prior to the statistical checking procedure. Fig 4. is an example illustrating the quality control results with and without the removal of these suspicious salinity data.

Distribution of the quality controlled data

The quality controlled dataset contains 114,822 stations. 4990 stations (4.2%) were eliminated completely on statistical considerations. Some 120,000 (6.9%) individual observations were discarded (4% in the first pass). 45,000 (6.5%) oxygen values were flagged (again 4% in the first pass). These statistics can be compared to the North Atlantic in which 132,608 stations survived the quality control procedure. 9,552 (6.7%) stations were removed completely. 6.6% of the observations were discarded in the first pass.

The Horizontal Data Distribution (Fig. 5):

The horizontal coverage provided by the final quality controlled dataset is illustrated in Fig. 5. The horizontal distribution of the data was not changed by the quality control. Well measured regions lie in the southwest tropics, to the northeast near the coast of North America and the California current and around the Hawaii. The worst data coverage is in the tropics to the southeast of Hawaii. It is in this region that statistical analysis on scales less then 10 became impossible.

Comparing Fig 5. to Fig. 1, it can be seen that the very densely observed regions are the same. What is not obvious is the disproportionately large number of observations removed from Japanese waters due to the questionable salinity values discussed in the previous section. It is hoped that many of these observations can be recovered in the future.

The Vertical Data Distribution (Fig. 6):

The QC procedure has not changed the vertical distribution of the data. Fig. 6 highlights the low sampling of the deep waters and the removal of the early oxygen values. The first column of values is the number of QC stations with observations in the depth bin. The second column is the number of raw stations. Fig. 6C describes the change in the vertical distribution of the data around Japan when the suspect navy and fishery data were removed. It shows that although the character of the vertical distribution of these suspect observations is similar to that of the data which was kept, their vertical coverage is not as good. Nevertheless, they are a rich source of data near the surface and represent a good fraction of the available data even at some of the deeper levels. Such a large amount of data will be invaluable in filling in the time history of the regional hydrography.

The Temporal Distribution (Figs. 7 and 8):

!Click on the thumbnail to see a bigger version of the figure!

Figure 7.

The temporal distribution of the quality controlled data set is shown in Fig. 7 along with the distribution of WOCE and PREWOCE ctd data and the raw data. Fig 7a shows that most of the data in the region of Japan affected by the salinity calibration problems were made between 1965 and 1975, with somewhat fewer taken between 1975 and 1985. These two decades account for 75% of the data. Over the rest of the North Pacific (Figs. 7a and 7b) there was a general build up in the number of observations made beginning in 1950, peaking between 1965 and 1975 and falling off sharply after 1975. 70% of the observations were made over the two and half decades spanning 1950 to 1975. Fig. 7c indicates that about 20% of all the data was taken in the 5 years between 1965 and 1969 and about 15% in 1970 to 1975.

The distribution of the data over the different seasons is also fairly irregular. Figs. 8a and 8b compare the distribution of August (with the maximum number of stations) data to December (with the minimum number of stations) data. The monthly distribution varies enormously over most of the basin. Integrating over seasons helps, but does not eliminate the problem as there are twice as many summer stations available as winter ones. (20,606 stations in the period December through February, 28,708 in March through May, 40,610 in June through August and 24,898 in September through November).

!Click on the thumbnail to see a bigger version of the figure!
Figure 8. Station locations for the months with the maximum (8a) and minimum (8b) coverage.

Property Maps (Figs 9a and 9b)

!Click on the thumbnail to see a bigger version of the figure!

Figure 9a.

Figure 9b.

Maps representing the annual average pressure, potential temperature, salinity, and variance of these three properties for two isopycnal surfaces ( = 26.80, 2 = 36.90) are shown Figs. 9a and 9b. The maps were created from the QC version of the North Pacific Hydrobase climatology. The data were averaged into 1 degree bins - attempting to maintain at least 10 observations per bin by allowing a search radius of up to 3 degrees. This technique results in lesser smoothing in the well sampled regions ( around Hawaii and near the east coast) and greater smoothing in the more poorly sampled mid-ocean gyre. Where no observations were available within the 3 radius the maps show a blank gray 1 square. Here, we use the 26.80 isopycnal to illustrate the distribution of North Pacific Intermediate Water NPIW as seen through Hydrobase (Fig. 9a) The climatological 26.80 isopycnal does not outcrop within the North Pacific Basin. Talley (1993) suggests that the salinity minimum is formed in the northwest region of the Pacific between the Kuroshio extension and the Oyashio front, the so called mixed water region. It is this here, that the 26.80 isopycnal displays its shallowest(200m), freshest(33.5) and coldest(1.5 -2 C) extremes.

The deep waters of the North Pacific are not formed locally, but are made up Lower Circumpolar Waters, which are themselves modified Antarctic Bottom Water and North Atlantic Deep Water. The LCPW are transported northward and across the equator via western boundary currents. The 36.90 2 isopycnal (Fig. 9b) which represents the top of the North Pacific deep water layer, lies at about 2100 db over most of the basin, the only exceptions being in the adjacent seas in the west. Considered to be the most uniform of all the deep oceans, the North Pacific contains little in the way of property gradients at these depths. Salinities decrease northward from about 34.65 at the equator to 34.6 at 50 Temperatures range from about 2 C at the equator to 1.7 C north of the Aleutian Island chain. The strongest gradients seen by the North Pacific Hydrobase lie in the South China Sea, but there are few observations in the region.

Compared to Levitus (Fig. 10)

A large number of regional and global numerical models either initialize or relax to the climatological values of the gridded WOA94 (Levitus, 1994). It is therefore necessary to make some comparison between the Hydrobase climatology presented here and the gridded WOA94 product. A number of versions of the World Ocean Atlas (commonly known as the Levitus Atlas) now exist. Here we will make our comparison with the World Ocean Atlas 1994 (Levitus, 1994) as we began with same observed data set as this version.

Fig. 10 illustrates the pattern and magnitude of differences found between the two climatologies in the North Pacific. The 1 gridded WOA94 data is used to create the difference maps. Data from both data sets have been projected onto the same isopycnals for the purpose of comparison.

On the 26.80 NPIW isopycnal the pressure differences appear in zonal bands. The WOA94 surface is somewhat deeper (~20 db) near the equator, shallower in the subtropics (~20 db) and deeper from about 35 N to 45 N by as much as 200 db. In salinity and potential temperature the greatest differences occur in this 35 to 45 N zonal band emanating from the western boundary current region. WOA94 values are as much as 1.3 C warmer and 0.2 saltier at 37 N, 145 E. A comparison to observed data in / S, / depth and S / depth space in this region reveals an offset in WOA94 gridded values which is most likely due to isobaric averaging of steep gradients. Large differences can also be seen in the the Sea of Japan and the Okhokst Sea. Lesser differences spread in a broad band between about 10 and 20 N, spanning the entire basin. Here the WOA94 values are 0.3 to 0.5 C warmer and 0.05 to 0.075 saltier than the Hydrobase values. It should be also noted that the removal of Japanese navy and fishery data from the Hydrobase climatology must also be contributing to differences found within and around the western boundary.

The deep waters of the North Pacific are more homogeneous than anywhere else within the World Ocean resulting in relatively flat isopycnals over most of the basin and therefore, lesser differences between Hydrobase and the WOA94 here than higher in the water column. Most of the differences which do exist between the two climatologies at depth appear to be due to the manner of handling the sparse observation density at depth.

Public Access to the North Pacific Hydrobase

The North Pacific Hydrobase climatology contains some 115,000 stations from the historical archive. It is supplemented by another 3000 CTD stations from the WOCE and PRE-WOCE cruises.

The North Pacific Hydrobase climatology quality controlled data is available in both observed and gridded format through the anonymous ftp site ftp.whoi.edu under pub/whoinet/amacdonald/hydrobase under the subdirectories /qc (the quality controlled data) and /grid (the 1 quality controlled gridded data set). Note, please move directly to the hydrobase directory or you will find yourself in an un-named empty directory. The data are divided geographically into files which are labeled according to the 10 ~Marsden square into which they fall (Fig. 11).

Some 200,000 coastal stations and another 140,000 stations with dubious salinities have not been included in the current incarnation of the quality controlled North Pacific Hydrobase climatology. For those users who would like access to the North Pacific hydrobase format raw data set, it is also available in the /hydrobase subdirectory /raw.

All files have been gzip'd for easier transfer. The raw and qc files are in ascii Hydrobase format. The gridded data are in netcdf format which is also read by the hydrobase software.

Note there is now a global Hydrobase which has updated software, uses a new data format and which includes the North Pacific climatology described here. For information about this latest version contact Ruth Curry rcurry@whoi.edu or see http://www.whoi.edu/science/PO/hydrobase/.

More details on the North Pacific Hydrobase are available in Macdonald et al. 2001.

---

GO BACK TO OCEAN HYDROGRAPHY

---

Any Questions? e-mail me.