TEPPS Sonde Data

1.  General Description

Two hundred and fourty nine rawinsondes were successfully launched
during the 49 days of the PACS TEPPS 1997 cruise on the NOAA Ship
Ronald H. Brown.  All of the soundings were made with Vaisala RS80
sondes.  Of these 45 used Omega tracking to derive the winds (model
15N.15), like those used for the Integrated Sounding Systems (ISS)
sites in TOGA COARE.  The majority of the sondes (204) used GPS
tracking to derive the winds (model 15G).  The sonde data were
recorded using Vaisala software running on a DigiCora sonde system,
which was set up to store the data at 2-s intervals.  Vertical
resolution was determined by the rate of data transmission and by the
filtering applied to the data within the DigiCora before the data are
stored.  The Omega sonde wind data are transmitted on 10 s intervals
and held in a 4 min buffer, from which they are then filtered and
stored with a final vertical resolution of approximately 300 m.  The
GPS sonde wind data are transmitted at 2-s intervals and are also
filtered, in this case to eliminate the pendulum motion of the sensor
detectable by this more sensitive system.  The GPS sonde data are
filtered over 60-s intervals, yielding a vertical resolution of
approximately 30 m.  Both systems transmit thermodynamic data every
1-2 s, which are then median filtered over 11-s intervals.  Both wind
and thermodynamic variables are interpolated to 2-s intervals before
being written to disk.  Omega sonde launches were distributed randomly
throughout the cruise, except for the 00 UTC and 12 UTC sonde launches
which were always GPS sondes.  The order of the particular sondes
launched was randomized relative to their shipping cases in order to
minimize the effect of biases associated with a particular sonde
batch.

Independent surface measurements of temperature, relative humidity,
pressure, wind speed and wind direction were provided by the Woods
Hole Oceanographic Institution integrated suite of improved
meteorological (IMET) instruments and other ship's sensors.  The 1-min
average data at sonde launch time from the surface meteorology time
series are used as the first sounding data point.  The IMET
temperature and relative humidity instruments were located at 13 m
above sea level, and the wind instruments were at 14 m above sea
level.  The ship's barometer was at 15.5 m above sea level.

The procedure for preparing and launching the sonde was designed to
minimize heating or cooling of the sonde sensor prior to launch.
Ground checks of the temperature, relative humidity and pressure from
the sonde were compared to reference instruments.  The sondes were
released via a launch tube contained in a non-airconditioned van
located approximately 10 m from the sea surface.  The van door was
usually left open to equilibrate the inside of the van as closely as
possible to the ambient outside environment.  A second set of
independent surface points for air temperature and relative humidity
was collected at the time of launch using a hand-held sling
psychrometer.

The soundings were taken at regular intervals throughout the cruise.
The launch schedule was as follows:

                                       Sondes/   Total  Total 
          Location      Dates          day       GPS    Omega
          ------------  -------------  -------   -----  -----

          Panama to     970728-970808   6        34     33
          8N, 125W

          on station,   970808-970823   6        93      1
          (8N, 125W)

          8N, 125W      970824-970825  12        22      0
          to San Diego

          San Diego to  970826-970829   2         9      0
          subtropics

          subtropics    970830-970906   8        49     11


The sonde data were quality controlled at the Joint Office for Science
Support (JOSS) at NCAR, using methods similar to those used to quality
control the TOGA COARE sounding data (Loehrer et al., 1996; their
sections 3a-e).  The only significant departure was in the automated
quality control check for large inversions as described in Section 3.2
below.).


2.  Errors in surface measurements

Post-cruise corrections to air temperature and relative humidity have
been applied to the WHOI IMET data based on a recalibration of the
instruments at Woods Hole immediately following the TEPPS cruise.  The
correction increased the humidity by a few percent, giving better
overall agreement with the hand-held sling psychrometer measurements.
Figure 1 shows the sling psychrometer measurements centered about the
one-to-one line with the corrected IMET data.  The uncorrected IMET
data (solid blue circles) lie below the one-to-one line, indicating a
dry bias.  Few sonde measurements are above the one-to-one line in
Figure 1, implying fewer instances where the relative humidity above
the surface layer is greater than that measured at the surface (see
sections 3.1 and 3.4 for details).  Air temperature corrections were
small, increasing air temperatures by an average of 0.2 deg C, or
equivalent to less than a 1% decrease in relative humidity (Figure 2).

The errors for the IMET surface meteorological instruments, used for
the first sounding data point, are expected to be similar or greater
on a ship than those on a WHOI IMET buoy.  The estimated accuracies of
IMET buoy measurements (Weller and Anderson, 1996) are given in the
table below.


          Parameter            Instantaneous accuracy  Units
          -------------------  ----------------------  -----

          wind speed           5                       %
          wind direction       10                      deg
          barometric pressure  0.5                     hPa
          air temperature      0.2                     deg C
          relative humidity    4                       %


3.  Error analysis

3.1  Omega sonde biases

A sounding is considered to have a high or low moisture bias if the
offset between the IMET surface measurement and the sonde mixed layer
point is greater than that predicted from surface similarity theory
for the surface layer moisture gradient by an amount exceeding the
combined measurement errors of the IMET and sonde sensors.  To
investigate the occurance of biases, a mixed layer moisture value was
estimated from the 960 mb level.  The difference between this value
and the surface IMET measurement is compared to surface similarity
theory, which predicts that for the conditions encountered during
TEPPS, a moisture gradient of about 1 g/kg is expected across the
surface layer.

The sonde relative humidity sensor has a measurement error on the
order of 4% - 5%, and thus increases with increasing humidity.  This
error also applies to the calibration of the sonde which implies that
even if the sonde measurements are well within the measurement
accuracy of the instrument, the calibration error could offset these
precise measurements by 4% - 5%.  The IMET relative humidity sensor
has an average accuracy of 4%, with higher errors when the instrument
is heated in conditions of direct sun and low winds (Anderson and
Baumgartner, 1998).  A 4% to 5% error in relative humidity is
equivalent to 0.7 g/kg to 1 g/kg in mixing ratio, given the typical
range of observed surface relative humidity values in TEPPS of 70% to
80%.  Therefore, in order to consider a sounding moist bias, the
difference between the mixed layer and surface moisture values must
exceed the predicted value of 1 g/kg by an amount equivalent to the
sum of the errors of both the IMET and sonde instruments, or 2 g/kg.
Thus soundings with a difference greater than 3 g/kg are considered to
have a moist bias.  Likewise, soundings with a difference between the
mixed layer and surface moisture values of less than -1 g/kg are
considered to have a dry bias.

Most of the Omega sondes launched had been on the shelf for several
years and were donated to TEPPS for use before the Omega system was
turned off on 15 September 1997.  These older Omega sondes account for
all but 1 occurance of mixed layer relative humidity values higher
than those of their corresponding 10-m IMET measurements (Figure 3),
with 26 out of the 45 Omega sondes recording these high mixed layer
values.  Of these 26 Omega sondes, only 6 have differences less than
-1 g/kg (moist bias), while only 1 out of 204 GPS sondes exhibit a
moist bias.  Drift in the original calibration due to the age of the
sensor could be a cause of this problem for the Omega sondes.  Also
observed in Figure 3 are soundings for which the difference between
the surface and mixed layer moisture values is greater than 3 g/kg,
indicating a dry bias.  There are 8 soundings which meet this
criterion, of which all are Omega sondes.  The Omega sonde batches,
determined from their serial numbers, are seen to be well correlated
with the occurance of a moist or dry bias in the sounding.  Different
batches are indicated in Figure 3 by different colored open squares
about the Omega points (red diamonds).  The magenta and yellow batches
tend to be dry biased, while the cyan, blue and green batches tend to
be moist biased.  This result favors the interpretation that the
biases result from a calibration error of the particular batch.  As we
have no method to correct these data for moist or dry biases, they
stand uncorrected.


3.2  Inversions

Part of the NCAR/JOSS quality control procedure is to flag the
temperature profiles for large inversions.  During COARE, a questionable
data flag was used for inversions greater than 5 C/km and a bad data
flag was used for inversions greater than 30 C/km.  Given the
different environmenal conditions in the eastern Pacific, the flagging
procedure is slightly modified for the TEPPS tropical soundings
(970726-970828).  A questionable data flag is used for inversions
greater than 30 C/km and no bad data flag is used.  For the
subtropical soundings (970829-970906), a questionable data flag was 50
C/km above 700 mb and 100 C/km below 700 mb.  The tolerance was
increased since strong inversions at the top of the boundary layer are
typical for the subtropics, and to avoid flagging dry layers observed
periodically above the boundary layer which are believed to be real.


3.3  Superadiabatic lapse rates

A subset of the temperature data exhibited superadiabatic lapse rates
in both the tropical and subtropical regions.  NCAR/JOSS examined the
data on 6-s intervals and flagged lapse rates greater than 15 deg C/km
as questionable, and lapse rates greater than 30 deg C/km as bad data.
In addition, following their quality control procedure, where the
temperature data are flagged for superadiabatic lapse rates, the
humidity data are also flagged with a lesser warning flag.  Out of
751028 data points, the NCAR/JOSS analysis flagged 9661 as
superadiabatic, or about 1.3% of the data.  However, nearly all fo the
data points flagged as superadiabatic were questionable (9456, or
~98%) and only 205 (~2%) were flagged bad.  From our own analysis of
these data, 80% of the superadiabatic lapse rates in the surface layer
are accompanied by a dry surface layer which may be erroneous (Figure
4).

3.4  Low-level temperature and humidity errors

The low-level temperature and humidity data exhibited questionable
behavior for the first few data points for about half of the soundings
collected in each region.  Dry surface layers are observed with and
without superadiabatic temperature gradients and occur with varying
recovery depths and shapes in the moisture profile (Figure 5).  These
dry, and at times warm, surface layers are similar to those observed
at some of the ISS sites in COARE.  Following COARE, this problem was
investigated (Cole, H., 1993) and corrections were developed.  Further
investigation into the dry surface layers has also been done by the
University of Washington on the TEPPS data.  In general it is found
that while a dry layer appears in a fair number of TEPPS soundings,
the magnitude is small, on average 1.2 g/kg, and relaxes fairly
quickly, in about 8 s, to the ambient value.  A more detailed
description of the problem and a discussion of the results are given
in the following sections.

3.4.1  Description of the moisture problem

The anomalous moisture data are characterized by a decrease in
humidity from the independently measured surface point at
approximately 10 m to the first sonde point, which is typically
measured anywhere between 20 and 40 m.  This decrease is then followed
by an increase in humidity with height, as the sonde relaxes to the
ambient mixed layer value, estimated from the value of the mixing
ratio at 960 mb (Figure 5b).  Such dry layers are found in 62% of the
soundings in the ITCZ and 81% of the soundings in the stratocumulus
region, or in 72% of the soundings overall.  For our best soundings,
those which are not exhibiting a dry or moist bias (section 3.1), and
do not contain a dry layer structure, the average difference between
the first sonde measurement and the mixed layer value, in terms of the
mixing ratio, is 0.3 g/kg.  For these same soundings the average
difference between the surface and mixed layer values is 1.0 g/kg, in
agreement with surface similarity theory, giving confidence in our
choice of this subset of soundings for determining an acceptable
difference between the first sonde and mixed layer moisture values of
0.3 g/kg.  For the subset of soundings that exhibit a dry layer, this
difference is in the range of -3 g/kg to -0.01 g/kg, with an average
difference of -0.6 g/kg.  While these dry layers appear to be a real
and erroneous pattern in the data, a value of -0.6 g/kg is within the
4% - 5% measurement error of the sonde humidity sensor. Including
measurement error the range of measured values for this difference is
then -1.7 g/kg to 2.3 g/kg, based on an expected value of 0.3 g/kg.
Only 4 soundings have dry layers which exceed a -1.7 g/kg difference
between the first sonde measurement and the mixed layer value.  No
soundings exceed a 2.3 g/kg difference, indicating no moist layers are
observed.

3.4.2  Description of the temperature problem

At times a dry layer is accompanied by a superadiabatic lapse rate at
the surface, determined from the first three sonde measurements and
excluding the independent surface point.  The method of least squares
was used to determine the lapse rates, for which those greater than 10
deg C/km are considered superadiabatic for the purposes of this
discussion.  These superadiabatic lapse rates have been determined as
part of the analysis done at the University of Washington and are
independent from the NCAR/JOSS calculations that determine which data
are flagged in the final released data set (section 3.3).  Overall,
superadiabatic temperature gradients occured at the surface 46% of the
time in the ITCZ, and 56% of the time in the stratocumulus region.
Using the 10 deg C/km criterion, anomalous dry layers are seen 83% of
the time in conjunction with superadiabatic lapse rates at the
surface.  Dry layers for this overlap statistic are defined as
moisture structures for which the first sonde moisture value is less
than the mixed layer value, this difference not necessarily being less
than -1.7 g/kg.

3.4.3  TOGA COARE error analysis

Many TOGA COARE soundings had a similar dry surface layer structure in
their temperature and humidity profiles at low levels (see Cole, H.,
1993, Figures 6 and 10).  After much post-experiment analysis, it was
concluded that the sonde sensors were heated by the sun, resulting in
incorrectly low relative humidity measurements.  This results because
the relative humidity on the sonde is determined relative to the
sensor arm.  If this arm is heated, the higher arm temperature
produces the erroneous humidity values.  Relative humidity is very
sensitive to small changes in temperature.  Thus, this error may not
necessarily be as evident in the temperature measurements, explaining
why superadiabatic temperature gradients are not always measured for
all instances of dry layers.  Once the problem was determined, the
sondes were run through experiments in a wind tunnel to determine the
heat transfer function of the sensor arm.  The time constant for the
sonde to adjust to ambient temperatures given wind speeds typically
seen on ascent is 20-25 seconds.  Within these 20-25 seconds, the
relative humidities exponentially relax to ambient.  If the initial
temperature of the humidity sensor are known, the correct humidity can
be backed out of the erroneous data from this relationship.  However,
for TEPPS we do not have the original temperature data recorded by the
sonde at launch time, so it is not possible to extract the correct
values of the humidity in this way.


3.4.4  TEPPS error analysis

We suspect that the sonde sensor arm was heated relative to ambient by
a small temperature difference between the launching chamber and the
outside air.  It is fairly certain that the sensors were not heated by
solar radiation, first, because the sondes were not exposed to the sun
until after they were launched, and second, because of the lack of
correlation between the near surface temperature and humidity
gradients and incoming solar radiation (Figure 6). The correlation
coefficient between the incoming solar radiation at the surface and
the near surface temperature gradient is 0.46, while that between the
humidity gradients and the incoming solar radiation is -0.32.  Even
lower correlation coefficients are found between the air temperature
and the near surface temperature gradient.  These correlations do not
indicate a relationship between the behavior of the measurements at
the surface and the incoming solar radiation.  Our analysis indicates
that the likely source of the heating of the sensor arm above ambient
prior to launch was related to the launch chamber temperature, and
that the chamber temperature is only loosely correlated to the
incoming solar radiation.  A study is currently underway to measure
the temperature difference between the launch chamber and the van and
outside temperatures, the results of which will hopefully clarify this
issue.


The recovery rate of the affected TEPPS soundings has also been
estimated by finding the time and height at which the difference
decreased by 1/e from its maximum value.  The recovery is similar for
both regions, with values of 53 m and 8 seconds for the ITCZ and 51 m
and 7 seconds for the stratocumulus region.  Thus, the TEPPS sonde low
relative humidity values recovered more quickly than the COARE sondes,
likely because of a smaller sensor arm temperature perturbation from
ambient in comparison to COARE.


4. Summary of the sonde data quality analysis

Overall we have good confidence in the TEPPS sonde data above 100 m,
with the exception of the subset of Omega soundings (14) and one GPS
sounding which exhibit a significant dry or moist bias (Figure 3).
Below 100 m, anomalous temperature and moisture structure are
observed, and although these anomalies are generally within
measurement error, they appear to be systematic and erroneous.
Therefore we recommend viewing each sounding before any analysis is
undertaken involving the surface layer.  Anomalous temperature
structure, defined as surface layers with superadiabatic temperature
gradients, were observed for roughly 50% of the soundings (Figure 4a).
Superadiabatic lapse rates greater than 30 deg C/km, the bad data
criterion, occurred for 4% of the soundings (Figure 6a).  Anomalous
moisture structure in the form of dry layers, defined as soundings
with a higher mixed layer moisture value than the corresponding first
sonde measurement, were observed for roughly 70% of the soundings
(Figure 5b).  Observed moisture differences varied between just less
than zero to -3 g/kg, with an average value of -0.6 g/kg.  Of these
cases only 4 have moisture differences outside of the error range of
the sonde moisture sensor (Figure 6b).  The moisture deficit recovers
to the mixed layer value within an average of 60 m.  The overlap
between the occurance of a superadiabatic layer and a dry layer is
high, with roughly 80% of the anomalous temperature structures also
having an anomalous moisture structure at the surface.  As these
particular statistics were not calculated for the COARE sounding data,
it is difficult to determine how the TEPPS anomalous surface layers
compare.  We can state that the higher vertical resolution of the GPS
sondes provides a more accurate estimate of the the depth of the
anomalous layer in TEPPS.


Contacts and data access

For information on how to obtain these data see our web page at
http://www.atmos.washington.edu/home/disk/rain2/asap/qc/index.html.
For more information regarding quality control issues or general
questions on the sonde data email Yolande L. Serra at
yserra@u.washington.edu.


References

Anderson, S.P. and M.F. Baumgartner, 1998:  "Radiative heating errors
in naturally ventilated air temperature measurements made from buoys",
JAM, in press. 

Cole, H., 1993: "The TOGA COARE ISS Radiosonde Temperature and
humidity sensor errors".  Technical Report, Surface and Sounding
Systems Facility (SSSF), National Center for Atmospheric Research
(NCAR), Boulder, 26 p.  

Loehrer, S. M., T. A. Edmands and J. A. Moore, 1996:  "TOGA COARE
upper-air sounding data archive:  development and quality control
procedures", BAMS, 77, 2651-2671.

Lucas, C. and E.J. Zipser, 1996: "The Variability of Vertical Profiles
of Wind, Temperature and Moisture During TOGA COARE".  Seventh
Conference on Mesoscale Processes, September 9-13, 1996, Reading, UK. 
American Meteorological Society, Boston, 125-127.  

Weller, R. A. and S. P. Anderson, 1996: "Surface meteorology and
air-sea fluxes in the western equatorial Pacific warm pool during the
TOGA Coupled Ocean Atmosphere Response Experiment", J. Climate, 9,
1959-1990.


Figure captions

Figure 1: This figure shows 3 sets of points; the sling psychrometer
relative humidity values verses the IMET corrected relative humidity,
the first sonde relative humidity measurement (the second point in the
sounding; n=2) verses the IMET corrected relative humidity and the
uncorrected IMET relative humidity verses the corrected IMET
relative humidity.  Sling measurements were collected on all ship
decks as well as on the RHIB (rigid hulled inflatable boat) which went
out upwind of the ship once a day for measurements of ocean
temperature, air temperature and humidity.

Figure 2: Same as Figure 1 only for air temperature.

Figure 3: The difference in mixing ratio (r) between the corrected
IMET sensor at the surface and the mixed layer value (estimated at 960
mb) is plotted verses time.  Data from the Omega sondes constitute all
but one of measurements where this difference is negative.  A negative
difference indicates a possible moist bias in the sonde data, with
differences less than -1 g/kg being significant.  A difference greater
than 1 g/kg indicates a possible dry bias in the sonde data, with
differences greater than 2 g/kg being significant for these biases.
The occurance of either a dry or moist bias is seen to be associated
with particular Omega sonde batches, indicated by different colored
open squares around the Omega points.  

Figure 4: Examples of anomalous low level temperature and moisture
structure.  Temperature (a) and mixing ratio (b) profiles for 970817
2229 UTC.  The lower portion of the temperature profile is
superadiabatic, as seen when compared to the dry adiabat shown.
This superadiabatic structure is coincident with a dry surface layer
seen in the moisture profile.  Such anomalous dry layers often accompany
superadiabatic lapse rates in the TEPPS soundings.

Figure 5: Two examples of temperature profiles (a) and their
corresponding anomalous dry surface layers (b) found in the TEPPS
soundings.  The strongest dry layer observed, which occured on 970816
2238 UTC (green) has a -3 g/kg difference between the first sonde
measurement and the mixed layer.  On average, this difference was much
lower, or approximately -0.6 g/kg.  The example from 970820 1840 UTC
(blue), represents a more typical dry layer, with a recovery to mixed
layer humidity values by 60 m.  

Figure 6: The lapse rate (a) and difference in mixing ratio between
the first sonde measurement and the mean mixed layer value (b) versus
incoming shortwave radiation is shown.  There is no observed
correlation between surface gradients of temperature and moisture and
the incoming solar radiation, indicating that heating of the sonde
sensor during TEPPS was not related to the external environment, as
was the case in COARE.  
