MTP Documentation

A Summary of MTP Results for PREDICT


MJ Mahoney/JPL and Julie Haggerty/NCAR

Initial Page:     May 31, 2011

Last Revision:  June 20, 2011

We summarize on this web page, the results of the analysis of the Microwave Temperature Profiler (MTP) data obtained on the NSF/NCAR GV (NGV) during the PREDICT field campaign. Its purpose is two-fold: to present the final MTP data with comments on each flight, and to discuss the excellent temperature calibration that was achieved.


In our archived 'MP-files' we provide in the header record a comment urging users of these files to come to this web page for specfic comments on the quality of the MTP data for each of the PREDICT research flights. For this reason we put the comments first in this summary. Following these comments, we provide information on how the temperature was calibrated (very successfully) for the PREDICT campaign. It can be safely said that the PREDICT MTP data is the highest quality MTP data ever measured, and the temperature calibration is the best ever achieved. The only hiccup in the MTP data set is that on several flights the MTP frequency synthesizer, which normally tunes to three frequencies, would get stuck on a single frequency. In these cases, single frequency retrievals were performed and interspersed with the three frequency retrievals in the final archived data so that no data would be lost.

Comments on the PREDICT MTP Final Data

The following table provides a link to a color-coded temperature curtain (CTC) for each of the PREDICT test and research flights. More importantly, the table also provides a 'Comments' field where specific comments will be provided for each of the flights. Clicking on each thumbnail image will show the full-sized CTC image in a popup window, so that the image can be viewed while reading the comments.

On each of the following CTC plots the x-axis is the Universal Time (UT) in kilo-seconds (ks), the left y-axis is the pressure altitude in kilometers (km), and the right y-axis is the pressure altitude in thousands of feet (kft). On the right is the color-coded temperature scale, which ranges from 170- 320 K. Also shown on each plot is the GV's altitude (black trace), the tropopause altitude (white trace), and a quality metric (gray trace at the bottom). The quality metric, which we call the MRI for historical reasons, ranges from 0 to 2 on the left pressure altitude scale. If the MRI is <1, we consider the retrieval to be reliable; if it is >1 the retrieval is less reliable, and users should contact us as to whether it can be used or not. All the MTP final data have been editted to include retrievals with the MRI<1.

For most of the flights the CTC plots (which in fact are plotted using the archived data) are restricted to ±8 km from flight level. CTCs are not provided for flights for which avionics data was not available, such as the test and ferry flights. They could be processed if anyone is in need of the data. In the table below we use TFxx for test flights, RFxx for research flights, FF for ferry flights and GTxx for ground tests.


CTC Comments
TF01 - 20100803 Local test flight from RMMA
TF02 - 20100804 Local test flight from RMMA
TF03 - 20100806 Local test flight from RMMA

FF01 - 20100812

Ferry flight from RMMA to St. Croix

RF01 - 20100815

St Croix local flight

RF02 - 20100817

St Croix local flight

Frequency collapse to Ch2 at 40.0 ks. Did single channel retrieval for entire flight.

RF03 - 20100818

St Croix local flight

RF04 - 20100821

St Croix local flight

Frequency collapse to Ch3 at start of flight. Did single channel retrieval for entire flight.

RF05 - 20100823

St Croix local flight

Frequency collapse to Ch2 at start of flight. Did single channel retrieval for entire flight.
FF02 - 20100829 Ferry flight to Barbados

RF06 - 20100830

Barbados local flight

RF07 - 20100831

Barbados local flight

RF08 - 20100901

Barbados to St Croix

Frequency collapse to Ch3 from 54.7 - 57.3 ks. Did single channel retrieval during this period.

RF09 - 20100902

St Croix local flight

RF10 - 20100903

St Croix local flight
GT01 - 20100904 St Croix ground test

RF11 - 20100905

St Croix local flight

RF12 - 20100906

St Croix local flight
RF13 - 20100907 Ferry flight to Barbados (MTP not on GV)
FF03 - 20100908 Ferry flight Barbados to St Croix (also MTP ground test)

RF14 - 20100910

St Croix local flight

RF15 - 20100910

St Croix local flight

RF16 - 20100911

St Croix local flight

RF17 - 20100912

St Croix local flight

MTP thought to be overheating at ~43 ks and was turned off for an hour

RF18 - 20100913

St Croix local flight

Frequency collapse to Ch2 from 41.9 - 46.8 ks. Did single channel retrieval during this period.

RF19 - 20100914

St Croix local flight

-Frequency collapse to Ch1 from 54.8 - 56.7 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch3 from 59.3 - 61.5 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch3 from 63.2 - 64.8 ks. Did single channel retrieval during this period.

RF20 - 20100920

St Croix local flight

RF21 - 20100921

St Croix local flight

RF22 - 20100922

St Croix local flight

RF23 - 20100924

St Croix local flight

Frequency collapse to Ch1 from 54.3 - 55.9 ks. Did single channel retrieval during this period.

Data from ~55-58 ks is suspect because of interference but the tropopause retrieval appears to be ok

RF24 - 20100927

St Croix local flight

-Frequency collapse to Ch2 from 46.2 - 47.4 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch1 from 51.5 - 52.3 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch1 from 55.7 - 56.8 ks. Did single channel retrieval during this period.

RF25 - 20100928

St Croix local flight

-Frequency collapse to Ch1 from 46.9 - 47.3 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch3 from 49.9 - 50.5 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch3 from 50.9 - 51.5 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch1 from 57.4 - 58.5 ks. Did single channel retrieval during this period.

-Frequency collapse to Ch1 from 62.3 - 62.9 ks. Did single channel retrieval during this period.

-As can be seen from the MRI, the retrievals from 51-53 ks are suspect. Appears to be interference.

RF26 - 20100930

St Croix local flight

Frequency collapse to Ch3 from 52.1 - 53.0 ks. Did single channel retrieval during this period.
FF04 - 20101002 Ferry flight from St Croix to RMMA

PREDICT Temperature Calibration

1) Some Background

For nearly two decades the MTP team has been refining techniques for calibrating in situ temperature measurements made aboard research aircraft using radiosondes launched near the aircraft's flight track. Initially this was done by hand, and could involve as much as a day for a single comparison because of the tedious quality control procedures that had to be implemented (such as limiting pressure altitude excursions during the comparisons, restricting allowable pitch and roll changes, and checking for radiosonde temporal and spatial variability). About a decade ago these procedures were largely automated, but the comparisons were made for the entire MTP-retrieved temperature profile at that time, not just at flight level.

Even though the MTP did not participate in the T-Rex campaign, we were asked if the MTP temperature calibration techniques could be applied to the the research and avionics temperatures measured during this campaign so that differences in these temperatures could be resolved. During T-Rex the GV flew from RMMA to near Independence, CA, where it spent most of its flight time. In addition to the NWS soundings on transit, Leeds University frequently launched radiosondes from Independence, CA (INCA), so we had a wealth of soundings with which to do comparisons. All of the radiosondes used had an accuracy of ±0.3 K. As described on another web page, we found that  both the research temperature Tres (ATRL) and the avionics temperature Tavi (AT_A) had substantial warm biases with respect to radiosondes launched near the GV flight track (Tavi - Traob = 1.21 ± 0.12, and Tres - Traob = 2.37 ± 0.12, respectively). While Tres has the largest warm bias, we also found that the Tavi warm bias was very significantly pressure altitude dependent.

This work to understand the T-Rex in situ temperature measurements aboard the GV opened the door to a new approach for doing the MTP temperature calibration. As mentioned above we had previously compared the entire retrieved temperature profile to radiosondes, not just the flight level temperature. This often required several retrieval iterations through all the flights to achieve acceptable results. Eventually it was realized that if the flight level temperature was calibrated independently of the MTP data, less work would be needed. (This is the case because previously we applied a correction to the in situ temperature measurement called OATnavCOR. Therefore, everytime OATnavCOR changed we would have to recalculate the instrument gain. If the flight level temperature is accurately calibrated from the start, then OATnavCOR is always 0.0 K, and the instrument gains do not have to be recalculated. This saves a lot of effort.)

We have continued to refine the temperature calibration techniques that we developed for T-Rex on subsequent GV campaigns. There are other web pages that describe this procedure for START-08, HIPPO-1, HIPPO-2 and HIPPO-3. In fact, much of this discussion is a repeat of what is on the HIPPO-3 web page since the PREDICT data was processed after the HIPPO-3 data.

Before discussing the calibration procedure for the PREDICT field campaign, we will first provide a little background. During the HIPPO field campaigns the GV was for the most part continuously profiling the troposphere (and sometimes the lower stratosphere). This was a significant concern for a number of reasons:

  • First, in order to obtain good temperature profile retrievals, the MTP requires that the pressure altitude of the aircraft to be relatively constant during the course of a ~20 second scan. This was blatantly not the case when the GV is behaving like an atmospheric yoyo.
  • Second, related to this is the fact that we have typically averaged 3 -7 scans to beat down noise introduced by mesoscale temperature variations. Such averaging would be impossible during rapid descents and ascents.
  • Third, in the past we have flatly refused to do radiosonde comparisons in the troposphere because of the high lapse rate, and therefore sensitivity to altitude excursions.
  • Fourth, in order to do radiosondes comparisons, you need radiosondes. Since most of the HIPPO flights were in radiosonde sparse regions (the Arctic, Antarctic and Pacific Ocean), obtaining enough comparsions to achieve good statistics would be difficult.
  • Fifth, careful consideration needs to be given to the dependence of the temperature recovery factor on Mach Number. There is no way that a constant temperature recovery factor can be used when an aircraft (and its in situ temperature probes) are profiling the atmosphere.

For these reasons our hand was forced during the HIPPO campaigns. Normally when we do radiosonde comparisons, we do them at the time of great-circle closest approach to the radiosonde launch site. We are also careful to make sure that no one radiosonde comparsion overly weights the statistics. For example, suppose that the GV was taking off or landing at an airfield where radiosondes were launched. The "closest approach" algorithm might produce multiple times of closest approach during frequent turns near the airport on take-off and landing. We would edit out these additional comparisons to avoid overly weighting the statistics to this site. Given the sparsity of oceanic and polar radiosondes, and the desire to have good statistics, we decided to try a new approach for the HIPPO campaigns (and other campaigns where atmospheric profiling is common). Instead of using the great-circle time of closest approach to make the comparison, we decided to do comparisons every 2 km in altitude from 2 km on up with the closest radiosonde launch site that was available. (If the closest radiosonde launch site was very distant, we had a filter that would exclude soundings beyond a specified distance threshold.) This approach would increase the number of potential comparisons by nearly an order of magnitude. But equally as important, it would allow us to assess whether any of the in situ temperature measurements had a pressure altitude dependence, which, as we remarked above, was the case for the avionics temperature during T-Rex. In addition to allowing tropospheric radiosonde comparisons, we would also be forced to abandon averaging of scans to beat down the mesoscale temperature noise, since (when profiling) the temperature change due to altitude change completely dominates any change due to mesoscale temperature variations.

2) What We Did For PREDICT

Compared to the HIPPO field campaigns, PREDICT was benign because there was no atmospheric profiling. However, since we got more potential comparisons when doing them every 2 km in altitude (rather than only the time of closest approach) we chose the same approach as for the HIPPO campaigns.

For PREDICT there were five in situ temperature measurements that were available for radiosondes comparisons: AT_A, ATHL1, ATHL2, ATHR1, and ATHR2. (The fast response temperature ATFR was not used -- apparently because it ices up.)  When we analyzed the data for HIPPO-1, we initially attempted to do a correction to the in situ temperatures as a function of pressure altitude. However, since the correction from total temperature (Tt) to static temperature (Ts) involves Mach Number squared (see this web page for more information, this equation ignores the recovery factor):

and the MTP data processing software did not use Mach Number, we needed to create a proxy for Mach Number as a function of pressure altitude Zp. We did this for a segment of a HIPPO-1 flight on 20090109, and obtained:

          M2(Zp) = 0.02089 + 0.06555 Zp - 0.00122 Zp2

This approach has some inherent error however, since there is not a one-to-one correspondence between pressure altitude and Mach Number. Therefore, when dealing with the PREDICT (and the previously completed HIPPO-2 and HIPPO-3) temperature calibrations, we decided to update the data analysis software and spreadsheets to use Mach Number squared (M2). Using Tx to represent one of the five in situ temperature measurements available for PREDICT, we plotted Tx-Traob versus M2, and found for the corrected temperatures (sub-script 'c'):

    AT_Ac  = AT_A  * (1 - 0.0074 * M2) + 0.56

    ATHL1c = ATHL1 * (1 + 0.0198 * M2) - 2.50

    ATHL2c = ATHL2 * (1 + 0.0146 * M2) - 1.16

    ATHR1c = ATHR1 * (1 + 0.0130 * M2) - 1.18

    ATHR2c = ATHR2 * (1 + 0.0121 * M2) - 0.95

The Mach Number corrections for these corrected temperature measurements are important. For example the maximum value of Mach Number squared when we made radiosonde comparisons was 0.65. If we assume a nominal temperature of 200 K at this Mach Number then the corrections in these five equations are -0.40 K, 0.07 K, 0.73 K, 0.51 K, and 0.62 K, respectively. These are significiant temperature corrections, having a range of  1.13 K. Notice that Mach Number correction (the coefficient of M2) is larger for the four research temperatures, compared to the avionics temperature (AT_A), which has the opposite sign. Although ATHL1 had the smallest temperature correction (0.07 K) for M2= 0.65, AT_A had the smallest standard error (0.08 K compared to 0.34, 0.33, 0.09 and 0.28 K for the research temperatures). For these and other reasons, we adopted the corrected AT_A (that is, AT_Ac) as the in situ temperature to be used for PREDICT to calibrate the MTP gain.

Figure 1a. One hundred and seven radiosonde comparisons WITHOUT a Mach Number correction.


Figure 1b. One hundred and seven radiosonde comparisons WITH a Mach Number correction. The green cells show the bias and slope of the Mach Number correction.


Figure 1c. One hundred and seven radiosonde comparisons with the Mach Number correction correction applied in the analysis code to verify the correction. Note that the bias and slope of the Mach Number correction is zero (green cells), as it should be in this verification plot.


Because we did radiosonde comparisons every 2 km from 2 km on up whenever the GV made a descent and ascent, we ended up with 219 potential comparisons within a range of 260 km of the aircraft. After these were editted for the criteria discussed above, including non-redundancy, the total number of comparisons was reduced 107. Figure 1a shows these 107 comparisons without a Mach Number correction, Figure 1b shows the same comparisons with a bias correction of +0.56 K and a slope correction of -0.0074, and Figure 1c just verifies that when the corrections were applied in the data analysis software the bias and slope corrections go to zero. Note the 'slope correction' is really a pressure altitude correction, but it is more closely tied to Mach Number Squared than it is to pressure altitude.

With the corrected avionics temperature (AT_Ac) in hand, we could calculate the MTP instrument gains for each observing frequency as:

       G = [Counts (Horizon) - Counts (Target)] / [AT_Ac - Ttarget]  in Counts/Kelvin.

where Counts (Horizon) and Counts (Target) are just the output of MTP when looking at the horizon (i.e., an in situ measurement in front of the GV because the MTP integrates along the line of sight) and the reference target. (The gain calculation is actually not this simple, but we'll spare you the details!) With the gains in hand, we could now do retrievals. After the first pass through all the flights, we calculate what we call a Window Correction Table (WCT).  These are small temperature corrections that are applied to the measured brightness temperatures to correct for scan mirror side lobes. By design the WCT is always 0.0 K when the scan mirror elevation angle is zero, so this does not affect the flight level temperature calibration. Another retrieval pass is now made through all the flights with the WCT applied.

Figure 2a. Assessement of MTP performance relative to radiosondes BEFORE  RAF correction.


Figure 2b.  Assessement of MTP performance relative to radiosondes AFTER RAF correction.


At this point we assess the accuracy of the MTP retrievals at all retrieval altitudes, not just flight level. This is done in Figure 2. In Figure 2a we show the MTP accuracy with respect to flight level for the 107 radiosonde comparisons. The retrieved MTP temperatures below the aircraft are relatively well behaved; however, starting at about 3 km above the aircraft a warm bias starts to develop, and it increases with altitude. Given the season and location of the PREDICT flights, the MTP overestimates the temperature above the aircraft when flying below the tropopause (which it always did) because the degraded vertical resolution is not able to resolve any sharp thermal structure which occurs in the sub-tropics. Since we understand the cause of this bias, we apply what we call the RAF-correction to the data (this stands for REF-file After Fix). It is simply a sixth-order polynomial fit to determine the correction that gives the smallest overall bias with respect to radiosondes. This is shown in Figure 2b. This is without question the very best MTP performance that has been achieved in 25 years of MTP research.