Figure 2. Simulated retrieval accuracy for four differently weighted sets of retrieval coefficients including soundings with low level inversions.
Figure 2 shows the results when the two test radiosondes with inversions are included. As expected, the retrievals deteriorate significantly. However, since the raw MTP measurements clearly show brightness temperatures expected for inversions, it was expected that with proper retrieval coefficients -- that is, those calculated from radiosondes with inversions present -- these errors would drop significantly.
In the process of examining the various sets of retrieval coefficients
(RCs) with different weights on the downward-looking observables, it was
realized that there was a "catch-22" with how the weights were used. They
were used to determine which set of retrieval coefficients best matchs the
measurements, but they were already used in calculating the RCs as well.
Until the analysis of the TexAQS 2000 data, the a priori measurement
errors were fixed at each elevation angle that the MTP views. Because of
the possible contamination of the downward-looking measurements due to excess
ground emission, a scheme was developed that would weight-down the contaminated
measurements so that they would not impact the retrieval. As mentioned above,
this was done by requiring 5 optical depths between the MTP line of sight
and the ground. However, this neglected to take into account the fact that
this process would result in different observable errors at each flight altitude.
This only affects the retrieval coefficient selection, and not the actual
retrievals. What it does mean however is that the work above on evaluating
the accuracy of each set of RCs was incorrect. This evaluation would have
to be redone by including a two dimensional array of observable errors (one
dimension for each elevation angle observed, and one dimension for each altitude
at which retrieval coefficients are calculated).
The analysis programs were modified to take this into account. But before
re-evaluating the effectiveness of the different RCs, the simulation program
was used to better understand how much ground contamination there actually
was. Code was written to allow this excess emission to be included in the
brightness temperatures calculated using radiosondes (for the forward radiative
transfer calculation). It turned out that although there is a large effect
on the brightness temperatures themselves, the effect on the retrievals was
only second order. It was also noted that if a negative excess was used the
observables looked much like what would be seen if there was a ground inversion.
This led to an idea. Rather than weight down the downward-looking observables,
why not add excess ground emission when calculating the RCs? Then, in principal,
both the temperature profile and the surface temperature could be retrieved,
and there would no longer be arbitrary weighting of the measurements. This
approach would use all the data instead of throwing some of it away.
The next question to answer was: "How much warmer than the lowest radionsonde
level is the surface?" This had previously been examined
using the data from the two Heimann IR temperature probes
on the Electra. It was determined that over land the coldest IR temperatures
were about equal to the surface air temperature, that the warmest temperatures
were about 20 K warmer, and that these temperatures fluctuated rapidly
between these limits. Therefore it was decided to randomly add 0-20 K of
excess surface temperature to evaluate how well this worked.
Another question was: "What radiosondes should be used to calculate new retrieval coefficients?" There were two issues here: the NCAR Electra normally flew between 15-23 UT so that 0000UT and 1200UT soundings from nearby launch sites (BRO, CRP, LCH, SIL, SHV, FWD) would not represent the atmospheric conditions very well, and related to this, it was necessary to calculate RCs using sondes which included low-level inversions. It was decided to use the MM5 model output to extract "fake" soundings at eleven locations over ocean, coastal areas, and inland. These soundings were extracted by TAMU personnel from two model runs at five times (15, 17, 19, 21 and 23 UT) for 6 days that the Electra was flying between August 23 and September 1, for a total of 660 soundings. To these soundings were added 209 Airsondes launched from downtown Houston (HOU) and LaMarque (HSE).
Before calculating RCs, an attempt was made to bin these 869 sondes into
five or six different sets to minimize the formal retrieval errors. (If
all the sondes were put in one group, there would be too wide a spread in
temperature/shape that would lead to larger errors.) The first thought was
to separate the sondes into those with and without inversions and then to
bin them according to time of day. Examination of the MM5 fake soundings
and the HOU/HSE soundings without inversions by time of day showed that there
was a ~17 K spread in temperature near the surface, 15 K at top of the PBL,
and 6 K at 10000 feet. The average surface temperature (Ts) was 305 K from
19-23 UT, and 303 K from 15-19 UT. The temperature spread has little to do
with time of day. The average Ts for 572 soundings was 304 K with Ts ranging
from 295-315 K.
Since these temperature spreads were considered too large for accurate
retrievals, binning by location was considered with the following results.
The Ocean sites (GP01, GP02, and GP09) have an average Ts = 302 K and a
6 K spread, the Inland sites (GP08, GP10 and GP11) have an average Ts =
307 K and a 20 K spread, and the Houston sites (GP03, GP04, GP05, GP06 and
GP07) have an average Ts = 304 with 17 K spread. Most of the spread in the
latter is diurnal.
This led to the following conclusions; calculate five set of RCs as
follows:
1) Treat the Ocean sites separately (average Ts = 302 for 129 soundings)
with 0 K excess skin temperature.
2) Combine the Houston and Inland sites (average Ts=305 K for 443 soundings)
and calculate two sets of RCs based on whether a given sonde is warmer
or colder than the average temperature at flight level. For these two sets
of RCs a random excess skin temperature of 20 K peak amplitude was included.
3) Separate the soundings with inversions into two groups: one with ground
inversions and one with inversions above 500 feet. No excess surface temperature
was included on the first go around so as not to decrease the effect of
the inversion.
Figure 3. Simulated retrieval accuracy using five sets of revised retrieval coefficients based on MM5 fake soundings and HOU/HSE soundings .
As before, these five sets of RCs were used to perform simulated retrievals on the seven test case soundings. These were done for flight at 2000, 3000 and 4000 feet. The average difference and the standard deviation between the actual sounding and the retrieved profile for flight at 2000 and 4000 feet is shown in the Figure 3. Clearly, for flight at 2000 feet (FL020) the retrievals are very good: the average error (Avg) is close to zero (dark blue trace) and its standard deviation (SD) is <0.5 K 1000 feet below the aircraft (dashed red trace). However, at 4000 feet (FL040) the MTP is unable to "see" the low level inversions and the standard deviation on the retrievals increases above 1 K near 1000 feet (3000 feet below the aircraft). Therefore, since capturing the inversions is very important, it is recommended that only flight data near 2000 feet be used as input to the MM5 model. Table 1 shows the seven test case soundings and the corresponding simulated MTP retrievals. Of particular interest are test cases (3) and (4) which have low level inversions and yet very good retrieved profiles. This would not have been possible if the downward-looking observable had been down weighted.
|
Simulated Retrievals for The Seven Test Cases (click each image to enlarge it) White trace is the sounding; Yellow trace
is the retrieval. Vertical axis is pressure altitude (left in km; right in
kft) Bottom axis is temperature in kelvin.
|
(1) HSE 8/31/2000 2100UT |
(2) HOU 8/29/2000 2301UT |
(3) HOU 8/30/2000 1400UT |
(4) HOU 8/30/2000 1700UT |
(5) HOU 8/30/2000 2000UT |
(6) HOU 8/30/2000 2259UT |
(7) HSE 8/28/2000 2033UT |
Table 1. The seven test case soundings (white) and the corresponding simulated MTP retrievals (yellow). Click each figure to enlarge.
It should be noted that radiometric noise was not included in these retrievals . This noise near flight level is <0.5 K. Since the temperature at flight level is ~300 K, this would affect the simulated brightness temperatures by <0.2% and would not have a significant effect on the retrievals. Since the MM5 model will be run with horizontal resolution of 4 km, it is suggested that the MTP data be averaged for 3 scans to minimize the effect of radiometric/retrieval noise. This will correspond to ~4km along the flight track.
It was mentioned above that the raw MTP brightness temperatures for August 30, 2000, clearly show low level inversions while flying near downtown Houston. Although the simulated retrievals also "see" this inversion, it is important to check that the inversions are present in the actual retrievals that are going to be used in the MM5 modelling.
Figure 4. The NCAR Electra flight track on August 30, 2000 (yellow).
The blue dots and numerals represent the Universal Time in thousands of seconds
(ks).
Figure 4 shows the Electra flight track for August 30, 2000. The red portion of the flight track (from 57.8 to 59.7 ks) shows the period of time when the MTP retrievals show low level inversions. During this time the aircraft altitude varied from 2000 to 1000 feet.
Figure 5. A comparison of MTP retrievals (white and yellow) with HOU soundings (red, magenta, and green).
Figure 5 compares MTP retrievals at 58.4 ks (yellow) and 58.5 ks (white) to three HOU soundings at 1400 UT = 50.4 ks (red), 1700 UT = 61.2 ks (magenta) and 2000 UT = 72.0 ks (green) UT. The aircraft altitude at the time of these MTP retrievals was 2000 feet, and it would be expected that the MTP retrievals should have an accuracy of <0.5 K (see below). The Electra was two miles north of HOU at 58.4 ks, and the 1700 UT sounding at HOU was made 2.8 ks later. The temperature difference between the MTP and the HOU sounding at 2000 feet is ~2 K. This is much larger than the expected accuracy of the MTP measurement, so it must be concluded that this difference is because two measurements were made at different times and at different locations. More importantly, it is clear that the actual (i.e., not simulated) MTP retrievals are able to "see" the low level inversion.
Figure 6. A comparison of NCAR Electra outside air temperature measurements
(OAT or Tnav) with radiosonde temperatures at flight level (Traob).
Since the accuracy of the MTP retrievals is important, a few words will be said temperature calibration. The MTP measurements are calibrated using radionsondes launched near the Electra flight track. In practice we generally compare the outside air temperature (OAT or Tnav) measurement made aboard the aircraft to radiosondes, and then transfer the calibration to the MTP data using a corrected OAT measurement. Figure 6 shows the results of this comparison. The formal result is: Tnav – Traob = +0.22 +/- 0.47 K, so that within the errors the agreement is excellent and no correction was applied to the OAT measurements. This is the first time that we have ever encountered such good agreement, and we attribute it to the fact that because the Electra generally flies at low altitudes, it can be accurately calibrated using tower fly byes. This is why we feel that the MTP measurement at flight level has an accuracy of <0.5 K, because it is very nearly equal to the OAT measurement.
MTP Home Page: http://mtp.jpl.nasa.gov