MSU TEMPERATURES AND RAINFALL DATASET VERSION 2

The data is here.

A TOVS Pathfinder Path C Product

Daily 2.5 Degree Gridpoint Layer Temperatures and Oceanic Rainfall for 1979-93

produced by
Roy W. Spencer
Vanessa L. Griffin
ES43 NASA Marshall Space Flight Center
June 1994
Read this documentation to ensure that the Version 2 dataset error characteristics are compatible with your research objectives.
Documentation date: 29 June 1994.
Table of Contents
								           
1.  Introduction							
	1.1 Dataset Overview						
	1.2 Our Guiding Philosophy					
2.  The MSU Instrument						
3.  What the MSU Measures						
4.  Stability of the MSU Calibration					
5.  Dataset Characteristics
	5.1 Deep Layer Temperatures					
		5.1.1  Averaging kernels
		5.1.2  Drift in channel 3 calibration
		5.1.3  Limb corrections and satellite 
		          intercalibration	
		5.1.4  Limb correction errors		
		5.1.5  Gridpoint error estimates
	5.2  Oceanic Rainfall	
		5.2.1  Algorithm description
		5.2.2  Rainfall errors				          
6.  References							          
7.  Figures  (NOTE: The figures and appendix are available only in 
hardcopy)
 	Figure 1.  MSU weighting functions and averaging kernels.	
        Figure 2.  Production procedures for the daily, 2.5 deg gridpoint 
                   channel 2/3 dataset.
	Figure 3.  Single satellite gridpoint standard error of 
                   measurement (SEM) for channels "2/3" (a); "3/4" (b); 
                   and channel 4 (c) computed during  the 1982 data 
                   from NOAA-6 and NOAA-7, and the 1990 data from 
                   NOAA-10 and NOAA-11.
        Appendix.  Daily channel "2/3" coverage of the Earth during 
                   1979-93 by seven TIROS-N series satellites. 

1. Introduction

1.1 Dataset Overview

Version 2 of this dataset includes daily 2.5 deg gridpoint fields of deep-layer average temperatures derived from the Microwave Sounding Units for:
1
the lower troposphere (channel "2/3");
2
the (tropical) upper troposphere (channel "3/4");
3
the lower stratosphere (channel 4); and
4
oceanic rainfall estimates.
The dataset period of record is 1979-93 for the temperatures, and 1979- May 1994 for oceanic rainfall. The NOAA satellites contributing to this dataset are, in order of their launch, TIROS-N, NOAA-6, -7, -9,- 10,-11, and NOAA-12. NOAA-8 data were not used due to poor data quality. Each of the datasets have their own error characteristics, discussed below. Version 2 of the dataset will be replaced in approximately one year by Version 3, which will include a 1994 update to the time series, and will likely include various algorithm improvements, the most important of which will be in the limb correction procedure..

1.2 Our Guiding Philosophy

No satellite dataset is perfect, and the MSU datasets are no exception to this rule. Issues such as satellite intercalibration, limb correction, and algorithm improvement each involve a long list of decisions which must be made, and each of these decisions could be improved upon. While we know we can improve the datasets beyond their present state, we also recognize the need to make them available to researchers whose research objects are already attainable with the datasets in their present form.

We also find that these decisions often depend upon what purpose the datasets would be used for, so that in general we believe that there is no "best" version of a dataset of our parameters. Instead, datasets are constructed for specific purposes and the many decisions which must be made in the construction of the dataset are made with those purposes in mind.

We have decided to provide the present version of this dataset to researchers so that they can be exploited for purposes which the dataset error characteristics will allow, and with the hope that feedback from the users will help to make Version 3 (and later versions) still more useful for a wider range of research efforts.

The authors would appreciate a preprint of manuscripts submitted for publication which include research based upon these datasets. Users should always obtain the latest available dataset Version and documentation to ensure best results. Updates can be obtained from:

 	MSFC DAAC User Services Office
	Building 4492
	NASA/George C. Marshall Space Flight Center
	Huntsville, Alabama, 35812
	phone: 205-544-6329
	fax: 205-544-5147
	e-mail: msfcuser@microwave.msfc.nasa.gov.

2. The MSU Instrument

The Microwave Sounding Units were built by the Jet Propulsion Laboratory for NOAA to fly as part of the TIROS Operational Vertical Sounder (TOVS) instrument complement aboard the TIROS-N series of sun-synchronous polar orbiting satellites. The MSU is a four-channel Dicke-type scanning microwave radiometer observing passive microwave radiation at 50.3, 53.74, 54.96, and 57.95 GHz, which are frequencies in the 50 to 60 GHz oxygen absorption complex. As the instrument scans, it measures microwave radiation at eleven beam positions over a swath about 2000 km wide. The scan period is 25.6 seconds, which combined with the satellite orbital speed, results in scan- to-scan separation of about 150 km. The spatial resolution of the measurements, defined by the half-power (3 dB) beamwidth of 7.5 deg and the range from the satellite to the observed point on the earth, varies from about 110 km at the nadir (#6) beam position to near 200 km at the extreme (#1 and #11) beam positions.

Once every earth scan, the instrument makes calibration measurements, viewing deep space (2.7K) and high emissivity warm targets. There is one target for the two lower frequencies (channels 1 and 2) and another for the two highest frequencies (channels 3 and 4). The temperature of each target is monitored with redundant platinum resistance thermometers (PRT's). Calibration of the instrument digital counts into brightness temperatures (Tb) is simply a linear interpolation of the Earth-viewing measurements between the space and warm target measurements (see Spencer et al., 1990). Please refer to Smith et al. (1979) for further details of the MSU.

3. What the MSU Measures

The MSU was designed to be used together with the High Resolution Infrared Sounder (HIRS) and Stratospheric Sounding Unit (SSU) to obtain vertical atmospheric temperature profiles from space. Compared to the HIRS channel weighting functions, the MSU has poorer vertical resolution in the troposphere and better vertical resolution in the stratosphere. It has considerably poorer spatial resolution than the HIRS, but this gives the advantage of a much lower data rate and thus a more manageable data volume for analyses of the fifteen year data archive. The MSU is essentially insensitive to non- precipitating cirriform clouds, and so should provide a more robust air temperature signal than the HIRS. It is considerably less sensitive to liquid phase clouds than the HIRS. Neither instrument can measure air temperatures within precipitation.

MSU channel 1 (50.3) GHz (Fig. 1) has only weak oxygen absorption and so is sensitive to air temperature in only the lowest few kilometers of the atmosphere. However, this temperature information is heavily contaminated by other influences such as surface temperature and emissivity, as well as water vapor and liquid and precipitation-size ice hydrometeors in the troposphere. This limits the utility of channel 1 for monitoring lower tropospheric temperatures. MSU channel 2 (53.74 GHz) is sensitive to deep layer average tropospheric temperatures with a weighting function peaking near 500 hPa. It is only very slightly affected by variations in tropospheric humidity (see Spencer et al., 1990), but is contaminated by precipitation-size ice in deep convective clouds, which can cause Tb depressions of up to 15 deg C in midlatitude squall lines. High elevation terrain protruding into the MSU channel 2 weighting function results in proportionally less of its measured radiation coming from thermal emission by the air and more coming from the surface. The MSU channel 3 (54.96 GHz) weighting function peaks near 250 hPa and so often straddles the extratropical tropopause. MSU channel 4 (57.95 GHz) has its peak weighting at 70 hPa, and provides a good measure of lower stratospheric deep-layer temperatures.

Because the four MSU weighting functions overlap, they can be combined to retrieve information over thinner layers than the individual weighting functions represent (Conrath, 1972). This is the fundamental basis of satellite temperature retrieval schemes. For instance, a fraction of channel 3 can be subtracted from channel 2 to eliminate the lower stratospheric influence on channel 2 for middle and lower tropospheric temperature monitoring (Fig. 1). Similarly, a fraction of channel 4 can be subtracted from channel 3 for monitoring of upper tropospheric temperatures in the tropics, where the tropopause is near 100 hPa. In addition, because the MSU scans across the satellite subtrack at eleven different beam positions (six different Earth incidence angles symmetric about the center footprint), each channel actually has six slightly different weighting functions due to the variations of the view angle through the atmosphere. These different view angles can also be combined into a new weighting function, although this is done at the expense of any information about temperature gradients across the swath and, if the combinations are symmetric about the nadir measurement, the resulting retrieval represents an average temperature for the entire swath (Spencer et al., 1992b). This technique is more useful for gridpoint temperature monitoring over long time scales, or zonal averages over short time scales, where the errors due to horizontal gradients are averaged out.

The lower tropospheric air temperature influence on channel 1 is small compared to other influences, such as land emissivity and oceanic air mass humidity and liquid water path. In particular, we have used channel 1 to retrieve oceanic rainfall since its variability over the ocean is dominated by cloud and rain water activity.

4. Stability of the MSU Calibration

The MSU is externally calibrated, with the warm target and cold (deep space) radiation traveling through the same instrumental paths as the earth-view radiation. Therefore, changes in instrument temperature and gain are canceled out in the calibration procedure. We have quantitatively documented the high levels of stability for MSU channels 2 (Spencer and Christy, 1992a) and 4 (Spencer and Christy, 1993). We perform our own calibration of the instrument, the procedures for which were described by Spencer et al. (1990). There is evidence of drift in MSU channel 3 of up to 1 deg C (or more) in a couple of the satellites (NOAA-6 and NOAA-9), and corrections for this are discussed below. While drift in MSU channel 1 has not yet been rigorously addressed, the rainfall retrievals we have made using channel 1 data show no obvious evidence of drift in MSU channel 1, but further work needs to be done to quantify this.

We have found no evidence of calibration drift in any of the channels on time scales that would significantly affect variations within individual days, or the computations of day-to-day temperature changes.

5. Dataset Characteristics

5.1 Deep Layer Temperatures

MSU channels 2, 3, and 4 respond primarily to air temperature through their sensitivity to thermal emission by molecular oxygen, a well mixed and temporally stable atmospheric constituent. The weighting functions for these channels (Fig. 1) are quite broad in their response to temperature fluctuations at different altitudes. By linearly combining two or more channels, we can achieve sensitivity to shallower layers than the raw MSU channels represent. For the datasets described here, linear "averaging kernels" (Conrath, 1972) are produced for the lower troposphere; (tropical) upper troposphere; and the lower stratospheric measurements are taken from MSU channel 4 alone.

5.1.1 Averaging kernels

MSU channels 2 and 3 are linearly combined to form an averaging kernel (Tb23 = 1.6Tb2 - 0.6Tb3) which peaks near 500 hPa and has most of its radiant energy originating below about 300 hPa (Fig. 1). We loosely call this our "lower tropospheric" temperature measurement. In regions of high terrain, especially over Greenland, the Andes, Himalayas, and portions of Antarctica, most of the energy comes from the surface and so can not be interpreted as an air temperature measurement. The Tb23 data will have little utility in these regions.

MSU channels 3 and 4 are combined to form an averaging kernel (Tb34=1.35Tb3 - 0.35Tb4) which peaks near 250 hPa and receives most of its energy from the 500 - 100 hPa layer. We call this retrieval the "upper tropospheric" temperature, however, we only calculate it for latitudes 30 S to 30 N because the retrieval is only applicable in the tropics where the tropopause generally lies above 100 hPa. Tb34 is only slightly affected by high terrain.

Our "lower stratospheric" measurements come from channel 4 alone.

5.1.2 Drift in channel 3 calibration

MSU channels 2 and 4 have been documented to have no statistically significant drift in their calibration. All overlaps between satellite coverage have revealed no trend in the differences between satellites for these two channels at the level of a few hundredths of a degree Celsius over several years (see Spencer and Christy, 1992a for statistics on channel 2; Spencer and Christy, 1994 for channel 4).

Unfortunately, a couple of the MSU's (NOAA-6 and NOAA-9) had considerable drift in MSU channel 3. This drift seems to have only a low frequency component, occurring on a time scale of months to years, rather than from day to day (although an abrupt change was noticed on November 1, 1986 for NOAA-9). Since the averaging kernels for channels 2/3 and 3/4 both depend on channel 3, their fields needed adjustment for this drift. To achieve this adjustment for channel 2/3, we have forced the monthly, 30 deg zonally averaged anomalies to match those produced by channel 2R (see Spencer and Christy, 1992b). Channel 2R is a lower tropospheric retrieval which depends upon channel 2 alone and which uses the different view angles of channel 2 to produce a lower tropospheric averaging kernel. The channel 2R averaging kernel peaks somewhat lower in the troposphere than the channel "2/3" kernel, but we assume that their differences in monthly, zonally averaged anomalies are probably small. The 2R trend-adjusted channel 2/3 data can be used for long-term trend analyses on global, zonal, or gridpoint scales. The channel 2R procedures used here for satellite intercalibration (discussed below) differ somewhat from those in Spencer and Christy (1992b), and so can be expected to produce small differences in the resulting anomalies and trends.

The drift in channel 3 inferred from comparisons to channel 2R was then used to adjust the channel 3/4 measurements as well, but these adjustments have been made based upon only the tropical drifts. We do not have as much confidence in the long-term stability of the channel 3/4 fields, and they are provided here as an experimental product which is still undergoing evaluation and refinement. The interannual variability in the tropics, however, should be quite stable. The primary reason why we have released this field for distribution is that it has interesting day- to-day and interannual variability in the deep tropics.

The channel 4 fields have the best long term stability of the three temperature products provided. There has been no evidence of drift during any of the overlaps in satellite coverage. The only calibration problem we have noticed is for the NOAA-12 MSU, which showed an annual cycle in polar temperatures that was significantly different from all of the other satellites. It was found that this could be corrected with a simple linear scaling of the data which can be interpreted as a calibration slope error of about 2%.

5.1.3 Limb correction and satellite intercalibration

Because the MSU is a through-nadir scanner, the eleven footprints across the MSU swath width come from six different earth incidence angles (nadir, and two each at 11, 22, 33, 44, and 55 deg earth incidence angles on either side of nadir). Due to the longer atmospheric path lengths that the non-nadir views have compared to the nadir view, the corresponding weighting functions are lifted to higher levels. Thus, some sort of "limb correction" must be performed for all three deep layer measurements (channels 2/3, 3/4, and 4) if a spatially uniform field of temperatures is to be produced, i.e. temperatures on a constant pressure "surface". (Some numerical weather prediction modeling efforts are now focusing on the assimilation of raw radiances at their original view angles so that the data are not corrupted by assumptions inherent to limb corrections).

The Version 1 limb correction procedure (see Fig. 2) relies on the compilation of statistics over many years of the average relationships between a nadir measurement and a non-nadir measurement at a specific location (2.5 deg gridpoint); month; type of satellite (7:30 local time or 2:30 local time); and ascending or descending node of the satellite (ascending is p.m.; descending is a.m.). Third order polynomial regression equations then provide the Tb adjustments necessary to "correct" the non-nadir measurements to nadir. Thus, for each of the three atmospheric layers, there are approximately 480,000 limb correction equations (=10,000 gridpoints x 2 satellite times x 2 nodal crossing times x 12 months). These equations account for variations in both atmospheric lapse rate and surface temperature which vary seasonally and geographically. In the case of channel 2/3 and 3/4, drift in the calibration of channel 3 is inferred from the difference in channel 2/3 and 2R for monthly, 30 deg zonally averaged anomalies for the entire period of record (see Fig. 2).

Because channel 4 has not exhibited any evidence of calibration drift (Spencer and Christy, 1993), the intercalibration between successive satellites was accomplished by first limb correcting and gridding the daily channel 4 data by the same procedure represented on the left side of Fig. 2, then adjusting for the differences between satellites during operational overlap periods with those adjustments being averaged for each of 12 months in 10 deg latitude bands.

5.1.4 Limb correction errors

All three layer temperature products (channels 2/3; 3/4; and 4) have varying amounts of residual effects resulting from imperfect limb correction procedures. These effects can sometimes be seen by animating daily imagery and looking for east-west patterns in the daily grid fields that have a spatial separation coinciding with the orbit sampling of the earth, and which move eastward on successive days. When these patterns are present, a spectrum analysis of the time series at these individual gridpoints will reveal spectral peaks near 4.5 days and 8 days. These correspond to the usual rate at which the satellite orbits progress eastward across the sky (a function of the orbital altitude and inclination of a sun-synchronous satellite causing the number of orbits per day to not be an exact integer multiple).

The current channel 3/4 limb correction seems to work much better in the deep tropics than in the extra-tropics, and work is progressing on a different limb adjustment procedure suitable for the channel 3/4 characteristics. Detailed daily analyses of channel 3/4 patterns in the middle latitudes should await limb correction procedure improvements in a later version.

The channel 4 limb corrections seem to perform very well for most of the 15 year dataset.

5.1.5 Gridpoint error estimates

The precision with which daily gridpoint Tb can be measured depends mostly upon the quality of the limb correction scheme, and to a lesser extent upon the sampling errors inherent in estimating a daily average with 1-4 "snapshots" provided by one or two satellites, and the limiting radiometric resolution of a single MSU measurement (0.2 - 0.3 deg C). There are probably other, smaller sources of error such as slight variations in different MSUs' weighting functions for the same channels. It is important to note that all statistics reviewed below are for a single satellite, and the corresponding errors are smaller during two- satellite coverage (which includes about 75% of the 15 year record, see the Appendix).

Estimates of the single-satellite standard error of measurement (SEM) for the daily deep layer temperatures are shown in Fig. 3. The statistics are computed by measuring the relative levels of disagreement between two satellites' variations in daily Tb at individual gridpoints. The (single satellite) standard error of measurement (SEM) is calculated as

	  	     SEM =  (2**0.5/2) sd(T(sat1)  -  T(sat2))	      (1)
where sd is the standard deviation, T is the Tb for any layer temperature product, and the single satellite factor (2**0.5/2) assumes that each satellite contributes equally to the total error. The sd in (1) is actually an average of the standard deviations for the individual years of 1982 (NOAA-6 and NOAA-7) and 1990 (NOAA-10 and NOAA-11).

The channel 2/3 SEM (Fig. 3a) show that the daily gridpoint errors in the deep tropics generally range from 0.3 to 0.4 deg C, except over land areas where they usually range from 0.4 to 0.6 deg C range. Most mid-latitude areas have from 0.5 to 1.0 deg C SEM's. Many high altitude regions, especially portions of Antarctica, Greenland, and the Andes Mountains, show high standard errors of measurement (over 1 deg C), and the highest errors occur in strongly sloping terrain (coastal Antarctica and Greenland).

The daily gridpoint noise estimates for channel 3/4 (Fig. 3b) are generally less than 0.3 deg C in the deep tropics, increasing to 0.5 or 0.6 deg C in the middle latitudes. The errors improve again at the high latitudes. Because we know that the current limb correction scheme for channel 3/4 does not perform well in the middle latitudes, we expect a new scheme to improve the mid-latitude noise estimates in Version 2 of the dataset.

The daily gridpoint noise estimates for channel 4 (Fig. 3c) are usually below 0.2 deg C in the deep tropics, increasing to 0.3 to 0.4 deg C in the Northern Hemisphere middle latitudes, but reaching 0.5 to 0.6 deg C in the Southern Hemisphere middle latitudes. It is believed that the large noise figures are caused by the variable conditions near the boundary of the wintertime polar vortex occurring in each hemisphere.

5.2 Oceanic Rainfall

5.2.1 Algorithm description

Daily gridpoint oceanic rainfall estimates equatorward of 60 deg latitude are made through a procedure similar to that described by Spencer (1993), and numerous details such as algorithm justification and detailed rain gauge comparisons can be found there.

Rainfall is diagnosed when a channel 1 Tb threshold is exceeded, the threshold being a function of the air mass temperature deduced from channels 2/3 . For each 1 deg increment of channel 2/3 Tb, a 15% cumulative frequency distribution was calculated for the base year of 1982 (NOAA-6 and NOAA-7). The 15% thresholds in channel 1 vs. channel 2/3 Tb space was approximated by a line. There are six of these linear relationships, corresponding to the six view angles of the MSU, which are used for all satellites. Tb1 warming above the appropriate threshold is assumed to be linearly related to a footprint-averaged rain amount. The conversion into rainfall units, however, was performed after compilation of many years of average Tb1 warming above the threshold from multiple satellites. To accomplish this, careful intercalibration between satellites during overlapping periods of operation was performed in two steps. First, we calculate a Tb1 "offset" of the second satellite which produces the same frequency of rainfall diagnoses (typically near 15%) as that from the first satellite during the overlap period. Second, the average DTb1 above that threshold was then forced to equal that from the first satellite through a "magnification factor" which is a function of beam position and satellite. These two adjustments force the rain amounts diagnosed during the overlap period to be equal between the two satellites. The overlaps in satellite coverage which were used for intercalibration throughout the 15 year period ranged from three months to 1 year (see the Appendix). [The additional step of an air mass temperature correction described in Spencer (1993) was not used in Version 2 of the rainfall dataset due to insufficient rain gauge data at high latitudes where a large annual cycle in temperature exists. However, cursor comparisons to the few high latitude gauges suggest that the resulting MSU rainfall diagnoses might be biased low during the cold season. Thus, studies of the annual cycle in extratropical rainfall with this dataset might be compromised.]

Monthly gridpoint averages of this warming were compared to 7- 10 years of monthly rain gauge totals at 123 island and coastal locations throughout the world, and a single scale factor was derived which scaled the DTb1 into rain units. The locations of these gauges (part of a dataset described by Eischeid et al. (1991)) was shown by Spencer (1993). The subsequent production of daily rain grids simply used the monthly calibration factor divided by the average number of days in a month (30.4).

The channel 1 signal is actually sensitive to cloud water as well as rain water, an ambiguity common to all passive microwave emission schemes for measuring rainfall, but which is particularly true at a frequency as high as channel 1 (50.3 GHz). The response of 50.3 GHz radiation to liquid hydrometeors is so strong that it is not, in general, sensitive to vertically integrated water contents exceeding (effective) rain rates of only a few mmh-1, at which point the liquid water path becomes essentially opaque to the transfer of the radiometrically cold radiation emitted by the ocean. Thus, the variability in the MSU daily rainfall fields is primarily related to the coverage of the grid squares by cloud and rain activity, not to the variations in rain intensity per se. Therefore, much of the accuracy of the method will depend upon a significant positive climatological correlation between the true footprint- averaged rain amount and the areal coverage of that footprint by cloud and rain water.

The Special Sensor Microwave/Imager (SSM/I) has a better range of microwave frequencies for radiometric sensitivity to local rain intensity, up to rain rates of about 15 mm/hr. Thus, the SSM/I will allow more accurate rain retrieval algorithms to be developed than are possible with the MSU. However, even a perfect instrument would not be useful if it did not provide sufficient sampling of the earth. The MSU provides two advantages in this regard: 50% greater spatial coverage than the SSM/I due to its wider swath, and its 15 year period of record is the longest of any passive microwave instrument.

5.2.2 Rainfall errors

5.2.2.1 Sea ice
Even though regions of multi-year sea ice have been largely screened out by restricting the oceanic rainfall dataset to 60 N-60 S, there remain a few regions in the dataset where seasonal sea ice occurs and produces false rain signatures. These include areas near Antarctica, the Bering Sea, Labrador Sea, and Hudson Bay. Users of the data in these regions should be aware of these false rain signatures and screen them out during the cold season. We might attempt to remove these areas in Version 3 of the dataset.
5.2.2.2 Climatology errors
Small changes to the rainfall algorithm (particularly the assumed slope of the channel 1 vs. channel 2/3 15% cumulative frequency distribution line) have been observed to cause significant changes in the annual cycle in diagnosed rainfall at high latitudes. Thus, there is inherent uncertainty in the MSU rainfall climatology, especially at high latitudes. Therefore, research which examines the high-latitude annual cycle in MSU rainfall could be compromised until the algorithm is refined to correct for

As discussed by Spencer (1993), rainfall in the extratropical storm tracks appears to be too heavy, although there is surprisingly little rain gauge information with which to validate this assumption. Nevertheless, because of the inherent ambiguity between cloud water and rain water signatures, it is certainly possible that the storm track regions have a larger cloud/rain water ratio than other regions.

6. References

Conrath, B.J., 1972: Vertical resolution of temperature profiles obtained from remote sensing measurements. J. Atmos. Sci., 29, 1262- 1271.
Eischeid J.K., H.F. Diaz, R.S. Bradley, and P.D. Jones, 1991: A Comprehensive Precipitation Dataset for Global Land Areas. Department of Energy Carbon Dioxide Research Program Report TR051. Oak Ridge National Laboratory, Oak Ridge, Tennessee
Smith, W.L., H.M. Woolf, C.M. Hayden, D.Q. Wark, and L.M. McMillin, 1979: The TIROS-N operational vertical sounder. Bull. Amer. Meteor. Soc., 60, 1177-1187.
Spencer, R.W. and J.R. Christy, 1990: Precise monitoring of global temperature trends from satellites. Science, 247, 1558-1562.
Spencer, R.W., J.R. Christy, and N.C. Grody, 1990: Global atmospheric temperature monitoring with satellite microwave measurements: Methods and results 1979-84. J. Climate, 3, 1111-1128.
Spencer, R.W. and J.R. Christy, 1992a: Precision and radiosonde validation of satellite gridpoint temperature anomalies, Part I: MSU channel 2. J. Climate, 5, 847-857.
Spencer, R.W. and J.R. Christy, 1992b: Precision and radiosonde validation of satellite gridpoint temperature anomalies, Part II: A tropospheric retrieval and trends 1979-90. J. Climate, 5, 858-866.
Spencer, R.W. and J.R. Christy, 1993: Precision lower stratospheric temperature monitoring with the MSU: Technique, validation, and results 1979-91. J. Climate, 6, 1194-1204.
Spencer, R.W., 1993: Global oceanic precipitation from the MSU during 1979-91 and comparisons to other climatologies. J. Climate, 6, 1301-1326.