2. Stages in the Creation of Winter PHC 2.1
- Preprocessing of WOA, AOA, and BIO data
- World Ocean Atlas (WOA)
- Arctic Ocean Atlas (AOA)
- Bedford Institute of Oceanography (BIO)
- About the BIO Database
- BIO vs WOD
- The data queried from BIO for PHC 2.1
- Preparing the BIO data for Optimal Interpolation
- Padding Western Baffin Bay
- Summary of Data Inputs into the Optimal Interpolation Routine
- Optimal Interpolation (OI) Details
- Post OI Processing
3. Monthly Fields: some additional details
It's advisable to first read the PHC 2.1 Overview for an introduction to PHC, as well as About the Upgrade from PHC 2.0 for a summary of what it provides relative to the World Ocean Atlas (WOA, Levitus), and PHC 2.0. All of the basic details of what PHC 2.1 is are presented there. The following report will present the finer details involved in creating the new PHC 2.1 fields.
|2. STAGES IN THE CREATION OF PHC 2.1|
The three sources of input data used in the creation of PHC 2.1 were AOA, WOA, and BIO. These data were prepared based on our goal of providing a well represented Arctic Ocean, while maintaining the high quality world atlas provided by NODC (Levitus) over the rest of the globe. This was accomplished by decimating the WOA fields in our "Arctic Region" where the fields were based on fewer than two actual observations. We replenished this region with the high quality Arctic Ocean Atlas (AOA) fields prior to our optimal interpolation (OI) procedure. The decimation of the WOA fields, however, left the Canadian Archipelago region and surrounding bays underepresented, a deficiency we compensated for with the BIO data.
Each data source is described in turn below, followed by a more detailed discussion of the steps taken to prepare each for the optimal interpolation routine.
PHC 2.1 uses the 1998 World Ocean Atlas (sometimes refered to as the "Levitus data set"), provided by NODC. The details of this product are summarized below:
The input data used to create WOA 98 are compiled in the World Ocean Database (WOD) 1998 (version 1) which primarily covers the time period up through 1994, although some newer data were also used.
0 150 800 1750 10 200 900 2000 20 250 1000 2500 30 300 1100 3000 50 400 1200 3500 75 500 1300 4000 100 600 1400 4500 125 700 1500 5000 5500
Obtaining PHC defined winter fields from WOA
The WOA winter fields are defined by January, Febuarary, and March, whereas for PHC, winter is defined as March, April, and May in order to match the analyzed fields provided by EWG (see below). To generate PHC defined winter fields from WOA, we averaged the WOA monthly fields of March, April, and May. The surface salinity fields for WOA's March, April, and May are shown below, followed by the average of these three months, which is the field used in calculating our Winter PHC 2.1 fields. We refer to this field as the MAM WOA field, where MAM stands for March, April, and May. To see the surface Temperature equivalent of these plots, run the mouse over the temperature link.
|Choose Fields To View:|
|Temperature (degrees C)||Salinity (parts/1000)|
|WOA surface analyzed monthly fields||Average of Mar,Apr,May|
Obtaining data for input into the OI routine from the prepared WOA fields
To prepare the WOA data for interpolation, we first discarded all field values where fewer than two observations existed inside our defined Arctic Region. This procedure was performed on the Winter (MAM) and Summer WOA fields at all levels to create the final input fields into the OI routine. Here's an example of this decimation procedure for winter surface temperature:
|MAM Surface Temperature (WOA)|
|Number of Observations (center plot)
|Temperature (right and left plots)
|Surface Temperature||Number of observations||Decimated MAM Field|
|This is the MAM averaged field from WOA. Our Arctic Region is outlined in black.||This plot shows the total number of observations that NODC used to create March, April, and May fields. It is the sum of Mar+May+Apr observations.|
PHC 2.1 uses the Russian 1997 (winter) and 1998 (summer) Arctic Ocean Atlas (AOA), created by the Environmental Working Group (EWG). The details of this product are summarized below:
- Spatial Coverage:Latitudes 65N - 90N
- Temporal Coverage:1950-1989
The AOA atlases contain gridded, interpolated mean fields for 1950-1989, and also for the decades 1950-59,1960-69, 1970-79, and 1980-89.
- Horizontal Resolution:50 km Cartesian grid (Lambert projection)
- Depth Levels (m):
0 150 1500 5 200 2000 10 250 2500 15 300 3000 25 400 3500 50 500 4000 75 750 4400 100 1000
- Climatological smoothed, analyzed fields:
- Seasonal: Winter=MAM, Summer=JAS
- Data:Potential Temperature, and Salinity
- Orginal Profile Data Available?Some
There were only two modifications necessary to prepare the AOA for the OI routine. First, the winter temperature field had to be converted to in situ to be consistent with PHC. Second, we linearly interpolated the AOA depths onto our PHC depths (which are coincident with WOA depths). The AOA surface input fields for winter temperature and salinity are shown below.
AOA Winter Surface Input Fields Temperature
In addition to the AOA and WOA data used in PHC 2.0 and PHC 2.1, we also incorporated data from the Bedford Institute of Oceanography into PHC 2.1. We acquired these data via their website:
Here's a brief description of the BIO database as it appears on their site:
"The Climate Database
The Ocean Science hydrographic database is a collection of temperature and salinity data for the area roughly defined by 35° - 80° N and 42° - 100° W. The data comes from a variety of sources including hydrographic bottles, CTD casts (either up or down casts), spatialy and temporally avereged Batfish tows, and expendable, digital or mechanical bathythermographs. Near real-time data in the form of IGOSS Bathy or Tesac messages are also included. The database currently consists of approximately 525,000 profiles and 15 million individual observations from 1910 to the present. Updates are made monthly."
There are a great deal of data available in this database in the Canadian Archipelago region, Hudson Bay, Baffin Bay, and the Labrador Sea. BIO does a great job of making this data accessible based on the needs of the user. By way of a query form, the user is able to make selections such as specific subregions, time frames, and bin averaging. The query results are emailed back to the user lickity split, usually under 24hrs. More information about this great resource and how to access it is available on their site at
Both BIO and the World Ocean Database (WOD), out of which WOA was created, used some common sources of data. Before beginning our project of acquiring additional data from sources outside of NODC to pad the winter coverage of the Archipelago and surrounded bays, we first combed through the WOD carefully. Our goal was to give any raw profiles we might find addition weight and scope to stretch their influence through the region. We did discover a few data profiles in the WOD in this region, but found that they were coincident with BIO values. Since the BIO database contained more profiles than the WOD, and included data found in the WOD, we used the BIO data alone for our patching purposes.
These are the specifications of the data we used from BIO:
Region: 100W to 50W between the latitudes of 50N and 80N.
Time Frame: From 01/01/1900 to 05/03/2002, March, April, and May data only.
Bin Averaging: Averaged over the individual months, but not over depth.
The data were first interpolated onto PHC depths, and then bin averaged into 1X1 degree bins. March, April and May were averaged together into one "Winter" field each for temperature and salinity. Analyzing the winter temperature profiles, we discovered that some of the data values were unrealistically warm. We applied a filter to be rid of them. This filter eliminated profiles where data in the top 50m was greater than .5C above freezing. Corresponding salinity profiles were also eliminated. A few slightly below freezing values were also discovered. We simply set them back to freezing.
We were only interested in supplementing the region that was underepresented by the WOA and AOA data sets. Since the Labrador Sea has good coverage in the WOA, and a lot of the data from BIO is in fact the same stuff that was used to create the WOA, we did not include these points. Thus, we discarded all BIO points south of 65.5N, and east of 61.5W.
For plots illustrating some of this process: Illustrative Plots
If you click on the "Illustrative Plots" link above, and look at the final input fields, you'll see that there are no data points in the western portion of Baffin Bay in winter. This becomes a problem in the optimal interpolation routine, because it allows the points in the eastern portion of the bay to significantly influence western Baffin Bay. This causes all of Baffin Bay to be characterized by the West Greenland Current, which is not physically accurate. To produce a more realistic Baffin Bay, we found it necessary to "pad" western Baffin Bay with an engineered physically realistic profile of our own. In all, we added 11 identical profiles to western Baffin Bay, in the locations listed below. These points are shown in the "Summary of Data Inputs" plots below. The engineered salinity and temperature profiles are shown here, along with a description of how they were made: Western Baffin Profiles
Western Baffin Bay
Engineered Profile Locations
Latitude Longitude 72.5 -67.5 70.5 -63.5 68.5 -62.5 69.5 -62.5 71.5 -64.5 69.5 -63.5 70.5 -62.5 71.5 -64.5 72.5 -65.5 67.5 -60.5 67.5 -59.5
|Example of Data Coverage: Winter Surface Temperature Inputs to the OI routine|
Our techniques for optimally interpolating the input data for PHC 2.1 were largely developed in the making of PHC 1.0 and 2.0. It may be helpful for a new user to review the general aspects involved before reading the specifics associated with PHC 2.1. Below is a table of references to assist you in this endeavor.
Table of General Optimal Interpolation Aspects LINKS DESCRIPTION Optimal Interpolation This page describes how optimal interpolation is implemented. You'll find a qualitative and quantive description discussing the parameters involved. Some sample IDL code is also provided. PHC 2.0 Paper:
This section of the PHC 2.0 paper reviews the WOA and AOA data sets available for input (which has not changed for PHC 2.1). It also describes how the background field was created from this data. The same technique was used for PHC 2.1, aside from some modifications in the Canadian Archipelago region and surrounding bays, which will be discussed below. Also discussed is the meaning of the error assignments.
For PHC 2.1, the Archipelago region was divided into several subregions, as shown below. It was necessary to subdivide the region to create an accurate background field in all parts of the region. The background field in each region was created from data only in that region. The plot below also illustrates the correlation length scales used in each region, represented by different shades. The table below outlines how relative error values were assigned to each data set in each main region. Low error (say .0001) implies that we trust the data a lot, while high error (say .5) means the opposite. There were not much data available in region 1, even with the addition of the Canadian Data. To compensate for this, we attempted to create high quality background fields in each subregion. At the same time, we stretched the correlation length scale to extend the influence of each data point, while assigning moderate error so that each calculated value would not drift too far from the background field.
The large length scale in region 1 was necessary to stretch the influence of the sparse data across the region. Some of the subregions have unique characteristics, though, that should not be shared. These subregions were isolated from the others so that they could retain their identity and not influence the rest of the region. This "isolation" was accomplished by increasing the error of all data outside an "isolation subregion" while the oi routine was calculating a value inside the isolation subregion. The data inside the isolation subregion retained all of the relative error values described in the table. When the oi routine moved outside of the isolation subregion, then all of the data inside the isolation subregion was assigned high error, while the data outside returned to the normal error distribution.
PHC 2.1 Archipelago Region
Correlation Length Scales (km):
The Background Field: Each lettered subregion had a unique background field calculated from values only in that region. For regions 2 and 3, the background was a zonal average of the raw input data, as described in the PHC 2.0 Paper (see the link in Table of General OI Aspects above).
PHC 2.1: Error Table REGION Region Definition Error Description AOA error WOA error BED error 1 The combination of all the
.0001 .5 .1 2 The Arctic Region,
excluding Region 1
.0001 .5 no data 3 The non-Arctic Region. .5 .0001 no data
There are two subregions that required isolation. They are James Bay (subregions D and E), and the West Greenland Current (subregion C). The reason James Bay needed to be islolated is because it has large rivers flowing into it, making its characteristics unique compared to Hudson Bay and the rest of region 1. Subregion C needed to be isolated because subregion B had very little data, while C had comparatively reasonable coverage. Before isolation, this caused all of Baffin Bay (north subregion B, and subregion C), to look like subregion C, in effect stretching the West Greenland Current westward across the whole Bay.
The tables below detail how the errors were assigned based on the above description.
Error Definitions Normal Errors: High Errors: AOA: .0001 10 WOA: .5 10 BIO: .1 10
The Isolation Subregions Subregion Name Subregion label James Bay D and E combined West Greenland Current C
THEN FOR A GIVEN ISOLATION SUBREGION:
Value to be calculated... Using data
INSIDE the isolation
OUTSIDE the isolation
INSIDE the isolation
Normal Errors High Errors OUTSIDE the isolation
High Errors Normal Errors
One last modification to these error rules was necessary because of the strong front that exists at the boundary between regions 1 and 3. Because the length scale is so high in region 1, warm salty water from the Laborador Sea (region 3) bleeds into region 1. This happens because near this boundary on the region 1 side, there isn't much data, while just on the otherside of the boundary (in Region 3) there's plenty. If the oi routine attempts to calculate a value on the region 1 side, most of the data it grabs is on the region 3 side, and they're given a fair amount of weight. To keep this from happening, region 3 was isolated from region 1, but region 1 was not isolated from region 3. To see this demonstrated in a few example plots, check out: EXAMPLES.
After optimally interpolating the raw data fields, some additional processing was required. The following is an outline of the steps taken to create our final PHC 2.1 products.
- We optimally interpolated the raw winter data for salinity and temperature, at each of our PHC 2.1 standardized depths.
- All of these fields were then smoothed with a gaussian smoother, with a correlation length scale = 200km.
- The temperature fields were corrected for below freezing points (i.e., set back to freezing).
- The PHC 2.0 summer fields were smoothed in Region 1 (the Canadian Archipelago and surrounding bays). This smoothing consisted of a 9 point median smoother, followed by a gaussion smoother with a correlation length scale of 100km.This is the only modification that was made between the summer fields of PHC 2.0 and PHC 2.1. There's plenty of summer data in Canadian Archipelago and surrounding bays, so our PHC 2.0 summer fields are realistic. However, because the region is so complicated with islands and small-passages, the optimal interpolation routine had some difficulty creating smooth fields in the region, and so were improved by this procedure.
- The winter PHC 2.1 fields were patched in Region 1 with summer PHC 2.1 data at depth.This patching was done to ensure that there is not much difference between the summer and winter fields at depth in PHC 2.1. This procedure had to be done carefully, though, to ensure that discontinuities were not introduced into the profiles. This was our strategy:
For temperature, beginning at the surface, we searched downward looking for the first instance where the winter temperature was greater than the summer temperature. We patched the winter profile with the summer profile beginning at this level down to the bottom (2000m for Baffin Bay). For Salinity, the same was done, search for instances when winter salinity was less than the summer salinity, and patching downward from there. In both cases I only searched down to 200m. If the winter and summer profiles didn't cross before 200m as described above, then the winter field was patched from 200m on down to the bottom.
- The salinity field was corrected for inversions from the top down.
- A final smoothing was performed on this winter field in Region 1, indentical to the smoothing that was done on the summer PHC 2.0 fields to create summer PHC 2.1. A 9-point median smoother was applied, followed by a gaussian smoother with correlation length = 100km.
3. Monthly Fields: Some Additional Details
PHC 2.1 monthly fields were created by fitting a cycloid-sinusoid function through the winter and summer fields. The same function was used to generate monthly PHC 2.0 fields, though it was applied differently. For a quantitative description of the function, review the PHC 2.0 paper: Creating Monthly AOA Fields. How this function was applied to create PHC 2.1 monthly fields was discussed in the "Upgrade from PHC 2.0" section of this site: The function used to create monthly fields from the seasonal fields. It describes what we did, and why our PHC 2.1 monthly fields are an improvement over the PHC 2.0 fields. However, the conclusion of this section brings to light some occasionally anomalous findings in the relative values between our summer and winter fields. At depth, winter temperature was sometimes found to be greater than the summer temperature, and similarly, winter salinity was sometimes found to be fresher than summer salinity. The purpose of this section is to discuss these "anomalies", and describe how our function handled them.
The location of the anomalies
These anomalies were introduced into our PHC Arctic Region from the EWG data. Generally, the surface layers do not exhibit this reversal of what is expected between summer and winter values. The anomalies seem to creep in below the mixed layer, through the halocline, and into the Atlantic Water Layer and deeper. The biggest anomalies appear between the depths of 25m to 200m in the EWG winter and summer fields. To see plots demonstrating the extent of these EWG anomalous values, check out: EWG anomalies.
Investigating the anomaly
Hakkinen Model Results
In a paper by Hakkinen and Mellor, "One Hundred Years of Ice Cover Variations as Simulated by a One-Dimensional, Ice-Ocean Model" (JGR, Vol. 95 No. C9, Sept. 15,1990), a seasonal equilibrium profile from their model is plotted for the central Arctic (shown below). We see that the model results allow for winter salinity to be fresher than summer salinity. This is because the fresh summer ice melt signal takes months to penetrate into the halocline.
Hakkinen Central Arctic Model Result
Summer and Winter Raw Data Profile Comparisons
To see whether this modeled result could also be seen in real data, two SCICEX cruises were compared. The first, SCICEX 98, was a summer cruise from June to August. The next, SCICEX 99, occured the following Arctic winter, from March to April. A region in the central Arctic where the cruises overlapped was selected, and their profiles compared: Raw SCICEX data
These profiles illustrate the fluctuant nature of the central Arctic. The profiles for winter and summer cross each other frequently in both the temperature and salinity profiles. While this brief investigation provides far from conclusive evidence, it does suggest the possibility that the EWG anomalies maybe realistic. Of course, they are also likely a result of interpolation onto a grid from sparse data.
The Monthly values generated at anomalous points
Our function for generating monthly fields relies on fitting a function between peak high and low values, where we assumed that these would occur in August and April. We assigned our seasonal summer values to August, and our winter values to April (this is a simplification of our actual procedure). Typically for temperature, then, the August (summer) value is high and the April (winter) value is low, and for salinity, the August value is low and the April value high. For the cases where this was reversed, we simply reversed the function, i.e., the seasonal curve is just flipped over its horizontal axis of symmetry.
Return to Table of Contents