This discontinuity is most evident in the PHC 2.1 summer temperature fields, though it can also be seen in the winter temperature fields. At its worst, it is on the order of .1 degree C in the surface fields of PHC 2.1 summer temperature. The source of this discontinuity was traced back to an anomalous region of similar magnitude in the AOA temperature fields. This region, as illustrated below, approximately follows the 83 N latitude line, along the arc between longitudes 315 W to 100 E. In PHC 2.1, these anomalous values where spread around the entire 83.5 N latitude circle by our optimal interpolation routine. This happened because of the way we calculated the Background Field, an input to the optimal interpolation routine (review "Optimal Interpolation Information and Software" for details about the role of the background field in the optimal interpolation). In PHC 2.1, we used a zonally averaged field of the raw input data to create the Background Field. In the Arctic Ocean, this "raw data" is primarily AOA data. Zonal averaging of the AOA data acted to spread the anomaly uniformly around the latitude circle in the background field, which then propagated into the final PHC 2.1 product.
Temperature: AOA Summer Temperature
This is our primary input field in the Arctic Ocean. Anomalous values are seen below at ~83N, between longitudes 315 W to 100 E.
PHC 2.1 Summer Temperature
The anomalous values from the AOA field at left propagated into our PHC 2.1 field via our zonally averaged background field.
Great thanks goes to Ralf Doscher, of the Swedish Meteorological and Hydrological Institute (SMHI), for letting us know about this problem.
Our PHC products primarily combine the Russian Arctic Ocean Atlas (AOA) with the NODC World Ocean Atlas (WOA). Since the WOA is currently widely used, we used the exact same gridding style for PHC, with land and bathymetry flags in the same locations. The AOA data, however, is gridded completely differently, and uses a different bathymetric mask. This caused some problems in PHC 2.1 at depth in the Arctic, where the AOA data often stops short of the bottom as defined in WOA, and hence PHC. When this happened, the optimal interpolation routine was forced to rely on whatever WOA data was in the region. This WOA data, especially at depth, was often inconsistent with the AOA data. This led to areas of instability at depth in PHC 2.1. The profiles below show the worst case of this problem, which occured in the Amundsen Basin.
PHC 2.1 Eurasian Basin profiles Contour plots at 4000m and 4500m for temperature (top) and salinity(bottom) This problem was brought to our attention by Jinlun Zhang here at the Polar Science Center. The deep density anomaly shown above caused an unrealistic clockwise gyre over the Amundsen Basin in his model.
To address the issue of the discontinuity in the temperature fields at 83 N, we re-engineered the background field. Instead of using a zonally averaged background field (which spread the AOA anomalous region around the latitude band), we applied a gaussian smoother with a correlation length scale of 200m to the raw data to create the background fields. Below is a table illustrating the input data that was used to create the background fields in each of our defined regions: the Arctic Region, the Non-Arctic Region, and the Canadian Archipelago and surrounding bays.
|Raw data inputs for Temperature and Salinity
gaussian smoothed Background Fields used for PHC 3.0
|SEASON||Arctic Region||Canadian Archipelago |
and surrounding bays
|Summer||AOA data only||WOA data only||WOA data only|
|Winter||AOA data only||WOA data and BIO data,|
with no gaussian smoothing.
The same background field that was
painstakingly developed for this region
in PHC 2.1 was used again for PHC 3.0.
|WOA data only|
|Region Definitions for PHC 3.0 are the same
as those for PHC 2.1. The basic division is
the "Arctic Region" vs. the "Non-Arctic Region".
||This plot shows the Canadian Archipelago
and surrounding bays, identified with the #1 below. It is the
combination of all the lettered
sub-regions. "2" is the remaining Arctic Region, and "3"
is the Non-Arctic Region.|
This solution addresses the problem of bad values in PHC 2.1 at depth. To understand what was done to correct this problem and why, let's first review the cause of the problem. The box below illustrates that our PHC climatology has deeper values in many places than the primary input field in the Arctic Region: AOA. When the oi routine attempted to calculate a value at a point and depth where there was no AOA representation, it found only sparcely scattered WOA data. If there was a WOA point close enough, the calculated value would approximate the WOA value. Otherwise, the value would default to the background field. This caused problems in either case, because the deep WOA values are unrealistic, and the background field at these depths were weighted toward these WOA values.
The fix to this problem was accomplished in a couple of steps. First, below 1000m, wherever there was summer AOA data that didn't exist in the winter AOA fields, we patched the winter blanks with the summer data from the same location. (There is not much difference between summer and winter fields at this depth). Next, missing AOA values were filled in at each depth by interpolating from surrounding AOA data at that same depth. In cases where extrapolation would have been necessary to obtain an AOA grid point value, the nearest AOA neighbor was used instead. This was successful at all levels except 4500m, where there is no AOA data at all, and the WOA data is wrong. Since salinity and temperature does not change much between 3000 and 4500m, AOA values for the 4500m level were taken from the deepest available AOA profile value at each 4500m location.
|Correction of the 83.5 N Discontinuity|
|Summer Surface Temperature (Celcius)
Mouseover the colorbar to magnify the scale
|Density Instability Correction at 88.5 N, 4.5 E|
|PHC 2.1||PHC 3.0|
Return to Table of Contents