Drifters in Hurricanes Frances and Jeanne, September 2004

Detailed analysis of drifter data close to the hurricane, and the momentum equation for radial and polar wind speeds are presented in Drifter data in Hurricane Momentum Equation .

For the following figures, click on the images for enlarged versions of plots.

1. High-Resolution Surface Windspeeds from Drifters and QSCAT in Hurricane Jeanne

The subtitles in b figures indicate the time difference between drifter and satellite observations (drifter-QSCAT) in hours; only the last two digits of the drifter IDs are printed. Drifter dots with no arrows indicate observed speeds of less than 5 knots. Drifter wind arrows with no barbs indicate a valid direction but no speed.

2. Drifter Tracks during Hurricanes Frances and Jeanne

2.1 Accuracy of GTS Position Data

3. Drifter Data Sampling Frequency

In the following figures, "Time of observation", the drifter observations are plotted as day vs. hour of day: blck dots indicate position data only (possibly SST and pressure), blue indicates direction data available, and red means direction and speed available. Note, that QSCAT observations in the area of the drifters occur around 10:30 and 22:30.

4. Drifter observed wind speed, SST, and air pressure

4.1 Metocean (9 drifters with speed, SST, and air pressure)

4.2 ADOS (8 drifters with speed, SST, and air pressure)

4.3 Minimet (14 drifters with SST and air pressure)

4.4 SVP-15m (3 drifters with SST and air pressure)

The three SVP-15m drifters stayed very close together after the deployment. Their SST observations are not the same, however, but differ by offsets on the order of 0.3-1.2°C. The next figures show the comparissons between the drifters. Plotted in each figure are time series of two drifters, the SST difference of the two, the separation distance (in km), and the 2D plot of the two drifter tracks. North and East directions are indicated, the length of each (50km) provide the scale for the tracks.

4.5 SVP-100m (12 drifters with SST only)

4.6 SST (17 drifters with SST only)

4.7 SST&pres (3 drifters with SST and air pressure)

5. Corrections for air pressure and SST, and more frequent "Pacific Gyre" data

5.1 Corrected Metocean (9 drifters with speed, SST, and air pressure)

5.2 Corrected ADOS (8 drifters with speed, SST, and air pressure)

5.3 Corrected Minimet (14 drifters with SST and air pressure)

5.4 Corrected SVP-15m (3 drifters with SST and air pressure)

5.5 Corrected SVP-100m (12 drifters with SST only)

5.6 Corrected SST (17 drifters with SST only)

5.7 Corrected SST&pres (3 drifters with SST and air pressure)

6. Drifter observed wind speed and direction

6.1 Accuracy of GTS Wind Speed Data

6.2 Sound Pressure Data from ADOS drifter 41921

Comparisons with collocated QSCAT data are presented below. Because of the sampling frequency of the drifters and the resulting time differences with local overflights of the sactterometer (see fig.3.1 & 3.2), it was necessary to use a maximum time difference of 4hr for collocations. The maximum distance allowed was 100km. The resulting average time difference of the collocations is ca. 140min (2:20hr); and the average distance is 17km (ranging from 1 to 95km).
Presented are timeseries of wind speed: drifter speeds derived from the 6-10kHz frequency and standard QSCAT. Also shown are the wind speed differences and the time differences of the two collocated measurements. Note that the time period after day 281 (when the relative magnitudes of the three sound channels change) does not exhibit anomolous differences with respect to QSCAT.

Scatterplots of wind speeds from drifter and QSCAT are presented below. Rain-flagged data are marked in red.

The few rain-flagged data (Nrain=4) do not exhibit very large wind speed differences with respect to QSCAT, and, therefore, will not be excluded from the comparisons.
About a third of the drifter measurements tend to overestimate wind speed. Breaking surface waves produce large acoustic noise that result in larger measured sound levels than only due to wind speed at the surface of the ocean. That contamination of the acoustic signal would bias the results towards higher wind speeds.

Also plotted are scatterplots of wind speed differences vs. time differences and separation distances.

In the time difference plot, the average speed differences for 4 time difference (Tdif) intervals are drawn with blue lines. While the speed differences for Tdif = 0-60min are the smallest (average=0.7m/s), the averages for longer Tdif do not change very much for time differences from 60 to 230min (average speed differences vary between 1.7 and 2.1m/s). Therefore, it seems reasonable to include the relatively large time differences (up to 230min = 3:50hr) in the collocation data sets.
Distances between collocated data tend to be within 20km. The standard QSCAT product is provided on a 25x25km grid. Larger distances only occur in patches of missing QSCAT wind vector cells (WVC),or when the drifters were at the edge of the scatterometer swath.

The next figure presents the functional fit between sound pressure level (spl) and wind speed. Black dots indicate the wind speeds derived from Equation I (a=20., b=104.5, c=53.91). When compared with collocated QSCAT, the average speed difference is 1.2m/s (Stdev=2.1m/s). Using the collocated QSCAT wind speeds, one can derive an estimate of the conversion coefficients based on a least-square-fit: a=39.21, b=1.1, c=2.84. This conversion function and the resulting wind speed estimates are plotted in red.

The data are limited to wind speeds of less than 15m/s. For small wind speeds (< 7m/s), the best-fit function agrees very well with the function from Equation I. For stronger winds, the best-fit function would result in lower estimates of wind speeds.

Hourly estimates of drifter derived wind speeds are used in the following figures with collocated QSCAT. The hourly estimates are computed by averaging only the two lowest sound pressure levels in every hour. The wind speed satterplots show that the correlation betwen drifter and QSCAT speeds gets better: the average speed difference drops from 1.2 to 0.6m/s, and stdev shrinks from 2.1 to 1.6m/s. Also the best-fit conversion function moves closer to the theoretical function.

From the two figures above it is easy to see that most of the speed differences have been reduced, when comparing the hourly with the 15min data. In particular, it is striking how speed differences within 20km separation distances have greatly been reduced. In the 15min data, 27% (17 out of 63) of the data within 20km had speed differences of more than 2.2m/s (max=7.7m/s). In the hourly data, this percentage has been reduced to 8% (5 out of 59, max=5.5m/s).
There are seven outliers in the hourly data set with relatively large speed differences (2.2-5.0 m/s). These outliers are either associated with large time differences (5 with Tdif>90min) or large separation distances (1 with dist>80km). In the next section, a sub-sample of the hourly is investigated, from which these seven outliers have been removed.

A subsample of hourly estimates is presented below. For the subsample large speed differences with either large time differences (speed diff>2.2m/s & Tdiff>90min) or large separation distances (speed diff>2.2m/s, and dist>80km) were eliminated (7 out of 67 samples were removed).
For the resulting 60 collocated pairs of drifter derived wind speeds and QSCAT measurements, the average speed difference is 0.4m/s, with Stdev = 1.0m/s.

The best-fit conversion function, based on this subsample, requires coefficients of (20.3, 98.0, 53.2), for which the theoretical values are (20.0, 104.5, 53.9). The best-fit average speed difference is 0.0m/s, with Stdev = 1.0m/s.

6.3 Comparison of wind direction records from GTS and Pacific Gyre files

In this section, GTS data are compared with Pac data. For this comparison, the time stamps of Pac data are not adjusted by the "age" value, which indicates the time difference between measurement time and reception by ARGOS. Only the most recent measurements from each 1-hour sampling cycle is compared, since GTS files do not include any other 15-min measurements. The un-adjusted Pac time stamps differ from GTS, generally, by not not more than 4 minutes. Furthermore, in order to compare the same data, Pac data records of bin number (1-72) is converted to wind direction by multiplying by 5° (but not subtracting 2.5°, which should be done to be exact). The resulting values are rounded to nearest 10degrees, which reflects the accuracy of the GTS data. Also, none of the directions are corrected for battery effect (which is required for ADOS data, and can be as large as 34°), or for magnetic declination (9-13°).

In the following figures, Pac wind directions are plotted with black dots, GTS are in red. GTS records with time stamps that match Pac time stamps (generally, within 4min), but which have different direction values, are marked in blue and are counted as "Ndiff" data in the subtitle. GTS records with time stamps that do not match any Pac time stamps, are marked in green ("Nextra"). Any black Pac values with no red mark, indicate Pac values with no matching GTS time stamps ("PAC: Nextra").

7. Wind direction scatterplots

8. Wind speed scatterplots

9. SST in CBLAST region, after Frances

The first 11 days after deployment, and the next 11 days are shown here.

The following figures show maps of SST during the first 3 weeks after Frances passed over the drifters in the CBLAST region. SST is plotted as the difference of observed SST minus the pre-Frances value at each drifter. The IDL "Min_Curve_Surf" function was used to interpolate the irregularly gridded set of drifter points onto a surface with "minimum curvature", at a resolution of 0.25° x 0.25°. This function does an acceptable job of finding a reasonable SST surface. The resulting 3-hourly maps are shown for 12:00 UTC on September 1 (just before Frances arrived), 2, 3, 4, 5, 10, 15, and 20. Printed on the plots are the drifter SST data that went into generating each map. Each drifter location includes the last two digits of the drifter and a letter for the drifter type: M=Minimet, m=Metocean, A=ADOS, S=SVP-100m, s=SVP-15m, p=SST&pres, t=SST. Also shown is the track of Frances, with 3-hourly marks from 9/1 6:00hr to 9/2/04 15:00hr. On September 20, 12:00 Jeanne had already passed over the drifters and was at 26.4°N, 288.3°W, just north of the plotted region. Its track is shown on the last map, too.

This interpolation scheme applies the same surface curvature in the longitudinal and latitudinal directions. Since the SST cooling happens roughly parallel to the track of Frances, a more sophisticated mapping routine should incorporate this preferred orientation in order to generate maps with features more parallel to the track. For example, on Sep 2 the two regions of relatively cold SST at 288.5°W, 23°N (coldest drifter SST=-1.8°) and 291°W, 22°N (coldest drifter SST=-1.7°) are not connected parallel along the track of Frances (in a region with no drifter data), but instead the mapping routine generates an area with temperatures of only -1.0°, due to drifter data of -0.8° further north of the track (from drifter 19A-0.80 at 290°W, 23°N) and data of -0.6° south of the track (drifter 17A-0.60). Similarly, on Sep 20 the region parallel and east of Jeanne's track, with drifter data as low as -1.8° in two regions, are not connected along the track (through an area with no drifter observations). Instead, the mapping function creates a small isolated area with slightly warmer SST separating the two cold regions.

10. Presentation at CBLAST Hurricane Conference, Miami, April 2005

The ppt file of the presentation, "Air-deployed Drifting Buoy Observations during Fabian, Isabel, Frances, and Jeanne" , can be viewed here .