Labrador Sea Drifters Research Topics and Issues
LabSea Publication
Figures and Tables for
Mesoscale Correlation Length Scales from NSCAT and Minimet Surface Wind Retrievals in the Labrador Sea (December 2001).
MINIMET Data
There were 21 drifters in the Labrador Sea that were grouped into two deployments. The lengths of data records are shown below. For each drifter the wind direction record is on top, the wind speed record is in the center, and the position record is on the bottom.
All drifter tracks are shown here. The open circles mark the beginning of the track, and the closed circles mark the end of the observations.
As an example of an individual track, all the recorded positions of drifter 16895 are shown here. The red line shows the half-day smoothed track.
The MINIMET drifters measure wind direction for 160sec at 1Hz. Raw compass data are binned and the bin with the largest number of observations is recorded as the wind direction.
The bin size varies between 3° and 8°. All recorded wind directions from drifter 16895 are shown here. Every wind direction data is marked by an "+" below the horizontal axis. Also indicated are the values of the bin centers and the differences between them. The number of data in each bin ("N data") is depicted by the vertical lines. Note that the bins are more closely spaced at intermediate wind directions (100° to 270°) than at small and large wind directions.
MINIMET Comparisons with NSCAT-KU2000
Drifter data can be compared with satellite scatterometer measurements of wind direction and wind speed. We have chosen the KU-2000 NSCAT product from Remote Sensing System. For a particular drifter measurement there is only a successful co-location if the NSCAT data was collected within 50km and 60min of the drifter data. The following figures show examples of drifter and co-located NSCAT data for drifter 16895. The drifter wind direction data in these figures has already been corrected (see explaination below). The figure below shows the entire drifter record for air pressure (a), wind speed (b), and wind direction (c). Also shown in figures b and c are co-located NSCAT data (open circles).
The wind direction difference vs. time, the distance between the drifter and the co-located NSCAT data vs. time, and the time difference of the co-located data vs. time are shown below.
Finally, scatterplots of wind speed, and of the wind direction difference vs. distance, vs. time difference, and vs. wind speed are presented.
MINIMET Wind Direction Corrections
Because of magnetic contamination of the drifter instrument package prior to launch, it is necessary to calibrate/correct the drifter wind direction data. For this purpose it would be ideal to only consider co-located data that are taken at exactly the same time and at exactly the same location. This is, obviously, not possible because of the difference in sampling frequency in space and time of the two measuring systems. In order to derive a correction algorithm only a subset of the co-located data is used (``Nfit''). This subset is supposed to represent the highest quality and most reliable co-located data, excluding data when either the drifter or NSCAT might have been reporting erroneous measurements.
The criteria to eliminate possibly erroneous data include the magnitude of the wind direction difference between the two instruments, the distance, the time difference, and also the wind speed at which the wind direction was measured. If these criteria are too restrictive, most of the data would be eliminated and the resulting correction coefficients would not be very reliable. On the other hand, if the restrictions are too relaxed, erroneous data could be included in the derivation of the correction coefficients.
The resolution of the NSCAT data is 25 km, and the above scatterplot of wind direction differences vs. distances indicates that most of the co-located data come from within a satellite swath, i.e. distance < 25 km. So Nfit, the number of data used for the correction algorithm, is restricted to instances when the distance between a drifter buoy and an NSCAT WVC is less than or equal to 20 km. The scatterplot of time differences indicates that the wind direction difference does not increase with time difference, at least not for the range of spatial and temporal differences considered here. So all data within 60min of each other are included in Nfit. An alternative scheme was also used by restricting Nfit to time differences within 30min. As the scatterplot indicates, this scheme eliminates a lot of data. The resulting correction coefficients were not found to be significantly different from the ones here presented. The 60 min time difference limit was deemed statistically more reliable because of the greater number of data under consideration.
The NSCAT instrumentation is most accurate for wind speeds of 3 to 30 m/s (see NSCAT User's Manual). Also, at very low wind speeds, the drifter buoy's wind vane can be much more influenced by pitching and rolling due
to surface waves. The scatterplot of speed does indicate a few anomalously high differences between drifter and NSCAT data at small wind speeds, so a minimum wind speed of 5 m/s is used to further limit Nfit. Lastly, when the wind direction difference between drifter and NSCAT exceeds 90°, data was eliminated from Nfit. This is assumed to be mostly due to erroneous NSCAT data due to the scatterometer's possible upwind-downwind ambiguity. After these 'a priori' limits for Nfit (distance <= 20 km, time difference <= 60 min, and abs(wind direction difference) < 90°), the average direction difference (avDiff) and the standard deviation (StDev) of the differences was computed. As the final step to pick Nfit data, all co-located data for which abs(wind direction difference - avDiff) are greater than 2xStDev are eliminated.
The co-located NSCAT data are used as "truth" to correct drifter observations by fitting an offset, 4 sine, and 4 cosine functions to the difference of the wind direction data:
(NSCAT-Drifter) direction vs. Drifter direction = C0 + C1 x sin(direction) + C2 x sin(2 x direction) + C3 x cos(direction) + C4(2 x direction)
For drifter 16895, there were 112 co-located data pairs. 84 pairs were used for the correction (="Nfit"). For this subset, the average distance between drifter and NSCAT data was 10km, and the average time difference was 20min. The coefficients are:
C0 = 9.7, C1 = 9.4, C2 = 6.5, C3 = 11.4, and C4 = -3.5.
Uncertainty for RMS differences between drifters and NSCAT
Here are two figures to understand the distribution of wind direction
differences that were used used to compute the rms values in figure 9 (Figures
for LabSea Paper).
The first shows the number of data (expresssed as percentage of all data
at a given lag) vs. wind direction difference. This was done for all
lags: 0-20km, 20-40km, .., and 180-200km. The direction differences are
binned into 10° intervals.
The next figure shows the distribution just for the 100-120km lag.
All moments for this distribution are printed on the figure. Also plotted
are the analytical function for a gaussian distribution with standard deviation
of 43.475 and a maximum of 12.2%. Other exponential fits are plotted as well.
The actual number of wind direction data that resulted in this distribution are
printed below the x-axis.
Extracting high-quality barometric data for drifter 16895
For the purpose of plotting the drifter observed air pressure, all values of "800.0" and "800.2" were treated as missing values. This still leaves many spikes in the plot that are unreliable data. Below is a figure of just the pressure of drifter 16895. In this plot, all pressure values that represent a pressure decrease of more than 20mbar (i.e. new value - old value < - 20) are colored in red.
Even this leaves several points that might represent unrealistic pressure changes. Those are marked with circles below:
In summary: for drifter 16895 all "bad" values were eliminated, i.e. "800.0" mbar, "802.0" mbar, and when the pressure fell by more than 20 mbar.
Those data are plotted in black below. Also plotted are data for 3 other drifters: 16886 (red), 16890 (green), and 16896 (yellow). These data sets agree very well with each other.
last modified on April 15, 2002
Go
Back
)
)
)
)
)
)
)
)
)
)
)
)