A NEW GLOBAL WATER VAPOR DATASET

David L. Randel*, Thomas H. Vonder Haar+, Mark A. Ringerud#, Graeme L. Stephens+, Thomas J. Greenwald*, Cynthia L. Combs#


* Cooperative Institute for Research in the Atmosphere, Colorado State University, Ft. Collins, Colorado
+ Department of Atmospheric Science, Colorado State University, Ft. Collins, Colorado
# METSAT, Science and Technology Corporation, Ft. Collins, Colorado

Corresponding author address: Dr. David L. Randel, Cooperative Institute
for Research in the Atmosphere, Colorado State University, Ft. Collins,
Colorado, 80523

Bulletin of the AMS (BAMS) - June, 1996 Vol 77, No 6

ABSTRACT

A comprehensive and accurate global water vapor dataset is critical to the adequate understanding of water vapor's role in the earth's climate system. To begin to satisfy this need, the authors have produced a blended dataset made up of global, five-year (1988-1992), 1 x 1 degree spatial resolution, atmospheric water vapor (WV) and liquid water path products. These new products consist of both the daily total column integrated composites and a multilayered WV product at three layers (1000-700, 700-500, 500-300 mb). The analyses combine WV retrievals from the Television and Infrared Operational Satellite (TIROS) Operational Vertical Sounder (TOVS), the Special Sensor Microwave/Imager (SSM/I), and radiosonde observations. The global vertical layered water vapor dataset was developed by slicing the blended total column water vapor using layer information from TOVS and radiosonde. Also produced was a companion, over-oceans only, liquid water path dataset. Satellite observations of liquid water path are growing in importance since many of the global climate models are now either incorporating or contain liquid water as an explicit variable. The complete dataset (all three products) has been named NVAP, an acronym for National Aeronautics and Space Administration Water Vapor Project.
This paper provides examples of the new dataset as well as scientific analysis of the observed annual cycle and the interannual variability of water vapor at global, hemispheric, and regional scales. A distinct global annual cycle is shown to be dominated by the Northern Hemisphere observations. Planetary scale variations are found to relate well to recent independent estimates of tropospheric temperature variations. Maps of regional interannual variability in the 5-year period show the effect of the 1992 ENSO and other features.

1. INTRODUCTION

There is an urgent need for a comprehensive and accurate global water vapor dataset to assist many important scientific studies in the atmospheric sciences. During the next decade, many World Climate Research Programme (1990) experiments will use present-day and future datasets to improve our understanding of the role of moisture in climate and its interaction with other variables such as clouds and radiation. Included in these experiments are the Global Energy and Water Cycle Experiment (GEWEX) and the GEWEX Continental-Scale International Project (GCIP). Many aspects of climate research are dependent on accurate water budget data. These include, but are not limited to, poleward energy transports, general circulation model (GCM) verification, regional climate studies, and global change baseline measurements (Vonder Haar 1994). The new dataset described here is the first of several new "pilot" datasets to address these needs (e.g., Chahine and Susskind 1991; Chedin et al. 1994). It does so with a combination of radiosonde and satellite infrared and satellite microwave retrievals. This dataset will help build the foundation from which investigators of future GEWEX-related and Earth Observing System (EOS) related work can learn and build upon.
Currently, atmospheric water vapor measurements are made from a variety of sources including radiosondes, aircraft and surface observations, and in more recent years, by various satellite instruments. Since these individual sources of data have certain limitations, it follows that a truly global moisture dataset should be derived from a combination of these measurement systems (Schubert et al. 1993). For many years, large-scale studies of atmospheric water vapor have relied wholly upon the analysis of radiosonde data (Bannon and Steele 1960; Oort 1983). Recently, there have been efforts to develop algorithms in order to retrieve the global water vapor (WV) climatology from either infrared or microwave space-based observations (Prabhakara et al. 1982; Trenberth and Guillemot 1995). Satellite-based observations are critical to this climatology effort because significant horizontal gradients in total column water vapor can exist between ground-based stations (e.g., Chesters et al. 1983). Analyses using only radiosonde data tend to smooth out these mesoscale gradients, which are important to the cloudiness, precipitation, and radiation balance fields. These fields are under special study by many scientists because of their central role in scale interactions in the climate system.
Large radiosonde data gaps over the oceans, and even over some land areas (e.g., Africa), limit the ability to define the global water vapor distribution. The newer data sources, such as those from infrared and microwave satellite sensors, can greatly enhance the global coverage on a daily basis. Examples of presently available large-scale WV datasets include satellite microwave retrievals from the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) data over ocean (Jackson and Stephens 1995), TOVS infrared retrievals over land and ocean (Rossow et al. 1991; Wittmeyer and Vonder Haar 1994), upper-tropospheric relative humidity from geostationary satellites (Schmetz et al. 1995), and a number of datasets using special radiosonde measurements for research purposes on limited time and space scales. Also available are model-analyzed 4-D data assimilated (in many cases operational analysis) global humidity fields from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP, formerly the National Meteorological Center) (Trenberth and Olson 1988).
As noted previously, each of the above datasets has significant limitations. Radiosonde measurements are made primarily over land with limited spatial and temporal coverage, infrared satellite techniques are only applicable in the absence of significant cloud cover, and microwave retrievals are presently feasible only over oceans. In addition, the satellite datasets have unobserved geographic areas each day. Figure 1, which shows the single-day plots of available data for July 10, 1989, illustrates these limitations. A comprehensive, blended WV dataset, accurate in all meteorological and geographical scenarios, must work within the limitations of these observing systems while using the advantages of each. The result is a synergistic effort far better than any single input dataset, yielding significant improvements in the daily representation of the moisture fields. Use of a blended WV dataset in climate model studies will greatly improve our understanding of many of the most difficult phenomena to characterize; those relying heavily on an accurate description of the moisture field.
The development of a global cloud liquid water product is also important to the understanding of the earth's climate system. There is a growing number of scientific studies aimed at characterizing the observed relationship between physical cloud properties and the Earth Radiation Budget (ERB). For example, Stephens and Greenwald (1991) investigated the connection between cloud albedo and cloud liquid water path (LWP), while Greenwald et al. (1995) examined the climatic characteristics of cloud LWP and also the relationship between the ERB and cloud LWP using collocated SSM/I and Earth Radiation Budget Experiment (ERBE) scanner measurements. Also, Zuidema and Hartmann (1995) investigated the cloud microphysical properties of marine stratus using a combination of ERBE albedo and SSM/I measurements.

2. INPUT DATA

To address the need for a comprehensive global water vapor dataset, including cloud liquid water over the oceans, we have produced a blended global analysis consisting of five years (1988-92), at a 1 degree lat x 1 degree long spatial resolution, with daily, pentad, and monthly temporal resolutions. This consists of both daily total column integrated composites and a multi-layered Water Vapor (WV) product at three layers (1000-700, 700-500, 500-300 mb). Radiosonde, SSM/I, and TOVS retrievals have been combined to produce these global products. The following sections describe the individual datasets used as input, the products produced from the individual input datasets, and the quality control required to produce the blended WV dataset.

a) Radiosonde

Historically, upper-air balloon soundings have been the basis for water vapor statistics used throughout the scientific literature (Oort 1983). These data have a long history and are still the only global ground truth available. For this project, radiosonde datasets were examined from a variety of sources. Differences between datasets are caused primarily by the diverse quality control applied to the observations, ranging from uncorrected raw observations to the statistically analyzed, quality controlled set described in Ross and Elliott (1996). From Ross and Elliott a five-layer WV for all stations were obtained and used for all five years of the National Aeronautics and Space Administration Water Vapor Project (NVAP) analysis. The processing involved using the five layered WV and producing three layers (surface-700, 700-500, 500-300 mb) and the total column WV. If the moisture sounding did not reach 700 mb, the sounding was discarded These observations were then inserted into 1 degree lat x 1 degree long global bins. A complete description of Elliott's data processing and quality control is given in Ross and Elliott (1996), but the following summarizes the process.
The radiosonde data was transmitted via the Global Telecommunications System, was decoded at the National Center for Atmospheric Research, and supplied to the National Oceanic and Atmospheric Administration (NOAA). Over 1800 stations have reported since 1973; however, the quality control procedure depends on stations with a long time series, thus limiting the number of stations to approximately 900. For the mandatory levels, the quality control involved computing a climatological monthly mean temperature at each station, and then requiring the observed temperature to fall within +/- 4 standard deviations of the mean. If a temperature failed a check at a mandatory level at or below 700 mb, the entire sounding was discarded. Above 700 mb, the sounding was kept, but data above the last valid level was discarded. This allows the lower levels, containing the bulk of the water vapor, to be retained.
A climatological, statistical check on the moisture soundings is more difficult since synoptic variability can cause both extremely dry and saturated soundings. Thus, some erroneous moisture soundings would fail to be detected during the temperature quality control check. For each station both the 00 UTC and 12 UTC soundings were used and the soundings were terminated at the 300 mb level since humidity measurements are considered unreliable above that point.

b) Special Sensor Microwave Imager (SSM/I)

The first of the satellite datasets used in the blended product was derived from the SSM/I, which was flown on the DMSP F8 and F11 satellites. The SSM/I is a passive dual-polarization instrument (channels at 19.35, 22.235, 37.0, and 85.5 GHz) designed to provide measurements of rainfall, water vapor, liquid water, and near-surface wind speed over ocean surfaces and the surface characteristics of sea ice and land. The DMSP F8 satellite (launched June 19, 1987) was the source of data through 1991 and measurements from the DMSP F11 (launched November 28, 1991) extended the dataset through the end of 1992. The F8 and F11 satellites have sun-synchronous equatorial crossing times of 0615 and 1704 (local time) respectively.
A retrieval scheme based on the physical method employed by Greenwald et al. (1993), was used to simultaneously retrieve over the oceans, both total column water vapor and integrated cloud liquid water. This scheme was an extension of the method of Tjemkes et al. (1991), and is based on measurements at 19.35 and 37 GHz.
The retrieval model requires several input parameters, the first of which is sea surface temperature (SST). NVAP used monthly mean SST's produced by NCEP on a 2 degree lat x 2 degree long global grid (Reynolds 1988). The next input parameter is the near surface wind speed, which is derived from the SSM/I brightness temperatures using the Goodberlet et al. (1989) algorithm. Since the Goodberlet method is less reliable in regions of high water vapor, the Bates (1991) method is used instead for SST greater than 300 degrees K. A mean cloud temperature is also required in the model and was specified as the SST minus 6 K, following Greenwald et al. (1993).
One important aspect of the processing involved intercalibrating measurements from the F8 and F11 satellites in order to provide a consistent dataset from one satellite to another. Near-coincident (i.e., within about 18 min) and collocated data from the two satellites were compared in December 1991 and the calibration factors in the model were adjusted accordingly.
There are three significant sources of contamination in the SSM/I retrievals. The first is land contamination which can result in an overestimation in the retrieved WV and LWP quantities. A 0.5 degrees lat x 0.5 degrees long land mask was used to prevent retrievals over land regions. However, land contamination was evident in certain clusters of small islands that were not included in the land mask. The final quality control procedure included manually searching and eliminating gridded data in these specific areas.
The second contamination problem is due to sea ice. Sea ice contamination was found to be more problematic than land contamination since the surface characteristics of sea ice are highly variable and since the areal extent of the ice changes seasonally. Sea ice was detected at the pixel level using a simple version of the AES/YORK algorithm developed for the SSM/I Calibration/Validation effort (Hollinger et al. 1991). As with land contamination, an additional screening of the datasets was done to eliminate erroneous data.
The final problem was precipitation contamination. For precipitating clouds, microwave retrievals of water vapor are likely to be overestimated resulting is a less reliable cloud liquid water retrieval. In this version of the dataset, quality control measures were implemented to detect and eliminate precipitation contaminated retrievals. However, caution must still be observed when using these datasets in areas of heavily precipitating cloud systems because the method could yield retrieval errors that are undetected (Greenwald et al. 1993).

c) TIROS Operational Vertical Sounder (TOVS)

Operational satellite-based WV retrievals have been made since 1978 by the NOAA/National Environmental Satellite Data and Information Service (NESDIS) (Werbowtzki 1981), using raw data collected from the NOAA series of operational polar-orbiting satellites. Data from NOAA-9, -10, -11, and -12 satellites were used during the NVAP data period. These satellites have a near-polar sun-synchronous orbit with a 102-min period and have local equatorial ascending node crossing times of 1420, 1930, 1340, and 1930 respectively. The TOVS instrument package - for retrieval of atmospheric temperature, ozone, and humidity - is carried aboard these satellite platforms and is made up of the second-generation High Resolution Infrared Radiation Sounder (HIRS/2), the Microwave Sounding Unit (MSU), and the Stratospheric Sounding Unit. These instruments are cross-track scanners with different resolutions (ranging from 18.5 - 148 km) and swath widths (1234 - 2230 km) (Kidder and Vonder Haar 1995).
Measurements from all three instruments are used for the retrieval of vertical temperature and moisture profiles. A different combination of channels are required depending on clear or clouly sky conditions. The retrieval scheme is based on the radiance variance approach proposed by Smith and Woolf (1984), and later modified by Fleming et al. (1986), and Reale et al. (1989). The three HIRS/2 channels most sensitive to water vapor are not used in the moisture retrieval. Instead, radiance data from HIRS/2 and MSU channels, primarily centered on the CO2 and O2 absorption bands are used to simultaneously generate temperature soundings and layered moisture in a single solution matrix (Reale et al. 1989). The selected channels, though not centered on the primary moisture absorbing frequencies, still contain considerable absorption by water vapor. A statistical eigenvector regression method was used before 16 September 1988 and was succeeded thereafter by a physical scheme (used for the majority of the NVAP dataset).
For the NVAP project, the operational TOVS sounding products produced by NESDIS were used. This quality controlled dataset is available from other members of the water vapor science community (e.g. John Bates at NOAA / Cooperative Institute for Research in Environmental Science). The data included total and three-layered WV for approximately 25 000 retrievals per day with geographical spacing of approximately 2 degrees. Our local TOVS processing consisted of gridding the WV retrievals into 1 degree lat x 1 degree long bins and applying only minimal quality control during the blending process. Erroneous TOVS data points, identified by comparison with the other two input datasets, were generally found in desert or coastal areas and were removed in the quality control screening, as discussed in section 3.
There are two problems inherent in all infrared moisture retrievals that tend to limit the dynamic range of the TOVS data. First, the inability to perform retrievals in areas of thick clouds can cause a "dry bias" (Wu et al. 1993). Second, limitations in infrared radiative transfer theory can cause significant overestimation of WV in regions of large-scale subsidence (Stephens et al. 1994). For these reasons, SSM/I data are given a higher total column WV confidence level than TOVS data.

3. DATA QUALITY CONTROL

Quality control is essential in order to produce a quality global data product. An important aspect of the NVAP dataset was direct human intervention, therefore, every daily grid was examined by a team of meteorologists. With each of the three input datasets, there are certain conditions for which the WV estimates may be inaccurate. Knowledge of these conditions enabled the NVAP quality control team to focus on potential problem areas and recognize possible inaccurate values. Other factors such as climate, terrain, and time consistency were also taken into account.
Use of Elliott's radiosonde dataset greatly reduced the required amount of manual quality control required. Most of the unusual or erroneous values had previously been eliminated; however, there were still a few problem stations. These would appear as stations that reported consistently higher WV than surrounding radiosonde or satellite retrievals. One possible reason for such a bias is the inherent limitations of various radiosonde instrumentation (Elliott and Gaffen 1991). There are approximately a dozen different suppliers of radiosondes worldwide, with wide differences between humidity sensors. The Finnish Vaisala, with a thin film capacitive humidity sensor and the U.S. VIZ, with a carbon hygristor, are considered to have good response times. These instruments are used predominately by the United States and European countries and their associates. However, other models have a much slower response time in humidity measurements, especially at upper levels. This lag can cause a moist bias in the readings. India uses a lithium chloride element in their humidity sensors, which is known to have a much slower response time. Another humidity device called "a goldbeater's skin hydrometer", used by many Chinese stations, is also known to have a slower response time. The slower response of these instruments provides a reasonable explanation for most of the higher values removed in these areas.
As mentioned previously, two primary quality control problems with SSM/I measurements are land and sea ice contamination, with sea ice being the most common. For both of these effects, the moisture retrievals are usually excessively high and this makes the detection of erroneous values fairly easy. However, high latitude waters such as those adjacent to Antarctica, Greenland, Siberia, and Japan can have occasional ice contamination not removed by the ice detection algorithm. Land contamination is a less significant problem than sea ice since these can be eliminated with a geographical mask. However, areas containing many small islands or rock outcroppings can contaminate the retrievals such as the area around South Georgia and the South Sandwich Islands, southeast of Argentina. Another occasional difficulty with the SSM/I measurements were erroneous data periods when the data are mislocated. In the first three years of this project (1988-90), these periods were documented (e.g., Wentz 1991) and subsequently removed during processing.
There were times when TOVS data points needed to be removed after comparisons with surrounding radiosonde and SSM/I data. Of the few points removed, the majority tended to be in dry or desert regions. These areas included central Australia, Namibia, Western Sahara, the coast of Peru, Kazakhstan, and the Middle East. Such areas are documented to have a "moist bias" (Stephens et al. 1994).
The procedure for manual quality control involved several steps. For the SSM/I data, each individual daily grid was visually inspected with emphasis on checking the known problem areas, as mentioned above. Other suspicious values, identified either visually, climatologically, or in a time series (day before or day after), were examined in the individual input product data grids. If a value was suspiciously higher than the surrounding values, it was removed. Any values that were questionable were marked for later examination in the blended product. After the three input sets were blended, as described in the next section, each daily product was visually inspected under the same guidelines as before. Suspect values were noted, then traced to one of the three original datasets. Comparisons were made between the suspicious value and the surrounding values in all three input sets. If that value was well above any of the surrounding values in all three sets, the value was removed from the appropriate input dataset. Once complete, the modified input files were again blended into the final product and visually inspected again. A similar process was followed for the layered products.

4. Blended Data


a) Total column water vapor

To create the blended WV product, the three input datasets were individually gridded into three separate daily 1 degree lat x 1 degree long global grid maps. The SSM/I gridded analyses were then checked for missing data over the oceans and spatially filled using linear interpolation. The total column WV blended products were created by using a weighting scheme that considered the radiosonde retrievals to be the most accurate, and the SSM/I retrievals more accurate than those from TOVS.
The blending process started by assuming the radiosonde points to be the truth and weighting these values at 100%. Next, the SSM/I and TOVS grids were combined together using a selected weighting of 10% TOVS and 90% SSM/I for coincident points. This blending information was recorded in a data source code (DSC) map, numerically ordered by the estimated WV retrieval error. The DSC map describes the origin of each point in the blended product with confidence values from 0 to 8. From the highest to lowest confidence the blended data points are: radiosonde (level 8), combined TOVS and SSM/I (level 7), SSM/I only (level 6), SSM/I interpolated combined with TOVS (level 5), SSM/I interpolated (level 4), and TOVS only (level 3). By referencing the DSC maps, which are produced for each day of the 5-year dataset, a NVAP data user has several basic analysis and interpolation options.
Finally, the total blended WV product was checked for missing data and missing regions up to 10 degree lat x 10 degree long were spatially interpolated. This type of grid point value was given a confidence level of 2. The remaining areas of missing data were filled using a temporal three-day running average. These results were given the lowest confidence (level 1) except for missing data (level 0), which occasionally is found in the south polar latitudes. With the individual data points identified by the DSC grid map, interpolated points could be easily removed by a user if desired.
The NVAP dataset contains daily global WV and liquid water path (LWP) fields. Figure 2 demonstrates the results of the total column WV blending process on April 25, 1988, for the North Pacific. Here we clearly see an intrusion of moisture from the Tropics into the midlatitudes. These moisture bursts are most commonly caused by the southwesterly flow in the warm sector of a mid-latitude cyclonic system, but may also be observed during times when the subtropical jet has a strong northerly component.
Figure 3 shows the global total column WV distribution for July 10,1989. The July, 1989 monthly average is also shown. The WV maximum in the Tropics and particularly in the west Pacific is clearly visible, as well as the equator-to-pole moisture gradients. The largest gradient, in a regional area, is across the Himalayas from the very moist Indian monsoon region to the dry elevated Tibetan plateau.

b) Layered water vapor

The layered WV is an important part of the NVAP dataset since the vertical distribution of WV is important to moisture transport and radiation studies. Two of the three input datasets, radiosonde and TOVS, contain layered information and were used to create the layered WV dataset. The TOVS data was ordered into in three layers: surface - 700, 700-500, and 500-300 mb and placed into a 1 degree lat x 1 degree long grid. The radiosonde retrievals were processed into identical matching layers. This layered information was then used to "slice" the blended total column WV product.
In the "slicing" technique, it was necessary to incorporate two basic assumptions with respect to the global WV distribution. The first assumes the TOVS total column WV is not as accurate as the SSM/I value, but that the fraction of total column WV in each layer is relatively correct. Second, while the total column WV may change rapidly in space and time, the fraction of the total WV in each layer changes more slowly. Layered WV from radiosondes and TOVS is divided by the total column WV to derive the global distribution of the percent-of-total (POT) WV in each layer. The variability in the POT is a strong function of latitude and season and does not vary spatially as quickly as the total column WV. These POT grids were then spatially interpolated to fill in small areas and temporally interpolated using a 5-day running average to fill in larger gaps. A 5-day average can be used (versus a 3-day average in the blended total column WV) because of the slower time variability of the POT. To create the layered WV products, the blended total WV grids produced earlier are multiplied by the POT grids. This gives the layered product the advantage of including SSM/I information along with TOVS and radiosonde data.
A Data Source Code (DSC) map is also provided for each of the daily layered WV grids (5 to 0, with 5 being the highest confidence and 0 being the lowest confidence). In order of highest to lowest confidence: radiosonde only is the highest (level 5); coincident TOVS and radiosonde points are combined together using a weighted 10% TOVS and 90% radiosonde (level 4); TOVS only points (level 3); spatially interpolated (level 2); and temporally filled data (level 1). Remaining missing data is given the lowest confidence level (level 0). There are increasing areas of missing data with the upper layers, especially in the polar regions due to persistent cloudiness that affects the TOVS retrievals.
The NVAP WV for each of the three layers for July 10,1989 are shown in Figs. 4, 5, and 6 . The monthly averaged layers for July, 1989 is also available. Since the majority of atmospheric water vapor is in the lower troposphere, it is no surprise the Surface-700 mb geographical distribution is similar to the total column WV. As we move up in the atmosphere, Figs. 4 and 5 show the decrease in WV with height. The large isolated maximum in the middle and upper troposphere in the area of the Indian monsoon is of special interest. In this area, due to the strong vertical moisture transport, over 8 mm are present above 500 mb.
In general, results from the layered WV products show that in oceanic areas, roughly 75-85% of the total WV is in the lowest layer. Depending on the surface elevation, elevated interior regions have only around 50% in this layer. In some locations, the surface may be above 700 mb, such as over the Tibetan highlands, in which case the POT for this layer is zero.
The WV annual cycles of the global and hemispheric daily averages for 1992 (Fig. 7) clearly show the global cycle is dominated by the Northern Hemisphere (NH) (Wittmeyer and Vonder Haar 1994). It is seen that the time series of global WV averages are sinusoidal in shape and have a maximum during June-July-August (JJA). Both hemispheres show a maximum during the summer months and a minimum in the winter. The differences between the NH and the Southern Hemisphere (SH) cycles are significant. The annual range of the NH cycle is twice that of the SH, and the NH summer maximum is much greater. These differences are related to the following geographical dissimilarities between the hemispheres. First, the amount of WV the atmosphere can hold before saturation depends on the temperature (as represented by the Clausius - Clapeyron relationship). The NH has a greater seasonal temperature change and the majority of the land surface area, thus allowing greater variability in the atmospheric WV. Second, the NH has a strong summer convective cloud maximum, with the intertropical convergence zone being primarily in the NH. Other factors include the strong summer monsoon season in India (NH) and the lower water vapor concentrations in the SH contributed by the cold and elevated Antarctic continent.
We also find the annual cycles are apparent in all atmospheric layers. Upon examination of the annual cycles of the global and hemispheric averages (for all five years of the NVAP data), there is some evidence the summer maximum lags with increasing height. This would suggest a time delay of the moisture transport from the surface to upper layers is discernible in the dataset.

c. Liquid water

We include, as part of the NVAP dataset, two atmospheric liquid water products. These are derived from the SSM/I and are available over oceans grid points only. Liquid water is currently of special importance since many GCMs are beginning to include it as an explicit variable. The two available products are: the LWP, and cloud liquid water (CLW). The LWP product is the liquid water in any region, averaged during all-sky conditions. We also include the monthly averages of cloud LWP, expressed as CLW, which is the liquid water in cloudy-only regions using a specified threshold of the liquid water retrievals. Extensive discussion and analyses of the LWP and CLW data are included in several papers noted in the reference section (e.g., Greenwald et al. 1993, 1995).

5. DATASET ANALYSIS

Table 1 lists the estimates of global and hemispheric averaged WV from this and previous studies. The present NVAP study includes data for 1988-92, and the results show slightly higher values than the TOVS observations but significantly lower than those using ECMWF. It is generally thought the ECMWF WV results are too high (e.g., Trenberth et al. 1987; Wittmeyer and Vonder Haar 1994), but these results may change soon since the ECMWF dataset, along with other operational analyses are being reanalyzed. The best agreement comparing NVAP with previously published results, occurs in the NH with the in situ dataset of Rosen et al. (1979).
Table 2 lists the global and hemispheric averages for the NVAP data from the layered and total column WV (for the time period 1988-92). The table shows the hemispheric differences in the total and layered WV, most striking in the total WV where the NH is 10% higher (2.4 mm) than the SH.


        Investigator                    Data            Time Period      NH       SH      GL
---------------------------------------------------------------------------------------------------------
Present Study                           3 sources       1988-1992       25.7    23.3    24.5
Wittmeyer and Vonder Haar (1994)        TOVS            1983-89         24.3    22.5    23.4
Wittmeyer (1990)                        ECMWF           1983-88         28.7    26.1    27.4
Rosen et al. (1979)                     MIT             1958-63, 68     25.7
Starr et al. (1969)                     IGY             1957-58                         26.0
Trenberth (1991)                        6 sources       1957-78                         25.3
Trenberth (1987)                        ECMWF           1978-85                         28.6

Table 1. Historic Estimates of Global and Hemispheric Averaged WV (mm). Modified from Wittmeyer and Vonder Haar (1994).


NVAP 1988-1992  N. Hemis.       S. Hemis.       Global
---------------------------------------------------------------------------
500 mb-300 mb   1.5             1.4             1.5
700 mb-500 mb   5.0             4.2             4.6
Surface-700 mb  19.4            18.4            18.9
Total           25.7            23.3            24.5

Table 2. The NVAP Layered and Total Column WV, Global and Hemispheric Averages (mm) for 1988-92.


Stability of any dataset is necessary for interannual variability studies and the NVAP processing has been undertaken with this goal in mind. The Elliott radiosonde dataset identified stations with a long-time series. The SSM/I retrievals included data from both DMSP F8 and F11 satellites but these systems were carefully inter-calibrated during a common month to eliminate any satellite bias. As was previously mentioned, the operational TOVS retrieval scheme did change in September 1988, but this was found to have a negligible impact on the global and hemispheric averages.
The total column WV climatology for the entire five years of the NVAP dataset is shown in Fig 8. The highly averaged nature of the global distribution is evident in the smooth WV distribution and agrees well with that published in Pexioto and Oort ( 1991), which used only radiosondes. Figure 9 presents the annual cycle of the NVAP WV global and hemispheric averages for the entire five years of the dataset. The interannual variability of the global averages is less than 2 mm, but this is hemispheric dependent. The NH's summer and winter months have the greatest variability near 3 mm with the SH generally less than 2 mm. The first nine months of 1988 are higher than other years, especially when compared to 1992, which tends to provide the lower bound for the global and NH cycles. While the largest WV values in the SH also occurs in 1988, 1992 does not provide the lower bound and is near average.
The global climate of 1988 was indeed much different than 1992. In 1988, the tropical Pacific had La Niña conditions, while 1992 was a moderate El Niño year (Trenberth and Hoar 1995). The difference in the WV distribution is shown in Fig. 10 - a plot of 1988 minus 1992 . Most of the world had greater WV in 1988 with the following exceptions: the central Pacific in the area of warmer than usual sea surface temperatures, off the west coast of the Americas, off the west coast of Australia, and in SH oceans south of 40 degrees. In general, the water vapor distribution during the El Niño year showed drying in the subtropics and midlatitudes, with higher atmospheric moisture only in areas near the equator.
The interannual variability of the WV, after removing annual cycle, is shown in Fig. 11. The values are expressed as a mean monthly standard deviation. The most striking high variability areas during the five years, are due to changes in the tropical circulation patterns caused by the ENSO events. Other variations over Africa warrant study. The low variability areas are easily seen over the persistent marine stratus areas and in the U.S. Pacific Northwest.
Christy et al. (1995) have recently reported on the lower troposphere temperature (LTT) anomalies during the last 15 years from the MSU on board the NOAA series of operational satellites. The global and hemispheric temperature anomalies for 1988 - 92 are shown with the concurrent NVAP WV anomalies in Fig. 12. For the global averages, the two anomaly time series are well correlated with a correlation coefficient of 0.75. The first implication of these results, mainly that total atmospheric WV and temperature apparently vary together as constrained by the Clausius-Clapeyron relationship (e.g., Stephens 1990; Gutowski et al. 1995), implies the climate system follows a constant relative humidity rather than a constant absolute humidity. The high correlation also reinforces the statement that WV is the principle greenhouse gas. An increase in global WV, therefore, points to an increase in the greenhouse effect and thus implies a positive feedback. Of course this is an oversimplification of the greenhouse situation since WV feedback may also depend of the vertical distribution of WV. In addition, other greenhouse gases and certain cloud microphysical properties affect the upper troposphere and stratosphere without being directly tied to atmospheric temperature.
Examining Fig. 12 shows us that in 1988 the global LTT was considerably warmer than 1992. The sharp drop in the global temperature, starting in mid-1991, was apparently caused by the June eruption of Mt. Pinatubo in the Philippines and was accompanied by a concurrent drop in global total column water vapor. The temperature and WV anomalies agree well, with the exception of early 1991 when the SH WV anomaly was strongly negative with no matching negative temperature anomaly. For March 1991, the NVAP dataset (Fig. 13) indicates that the SH mid-Pacific, Australia, and all regions from Australia to Indonesia were considerably drier than normal. In these areas, large-scale subsidence provides the drying mechanism, which also has a warming effect on the atmosphere. Therefore, while on a global scale, the atmospheric WV generally correspond to LTT changes, other physical phenomena and regional dynamic effects can impact significantly on the global relationship between water vapor and temperature.

6. SUMMARY

An extensive global dataset of water vapor (WV) has been produced from combining three independent data sources. The dataset includes total column integrated values, and values for three atmospheric layers for 1988 - 92. Each of the individual input datasets has significant limitations: microwave retrievals are presently feasible only over oceans; infrared satellite techniques only work in the absence of significant cloud cover; and radiosonde measurements are made primarily over land and are widely spaced, not showing small-scale WV variations. A comprehensive global dataset should draw upon the strengths of each of these methods and use the advantages of each for all meteorological and geographical scenarios. The NASA Water Vapor Project (NVAP) result is a combined column WV product far better than any single input dataset. In addition, a single method has been derived using the layered WV from radiosondes and TOVS retrievals to slice the total column WV and create the three-layer global WV dataset.
The global-averaged WV was found to be considerably higher during the 1988 La Niña event when compared to the El Niño year of 1992. The five-year NVAP dataset anomalies match well with the tropospheric temperature anomalies, and in general confirm the physical principle that a warmer atmosphere contains more WV than a cooler one. However, these correlations can be affected by large scale anomalous subsidence when a warmer atmosphere also becomes a drier one. To fully examine the question of whether the atmosphere maintains a constant relative or specific humidity, one would need a more detailed analysis of temperature anomalies at various levels. This will be pursued in further studies.
There are three primary NVAP products: total column WV, LWP derived from the SSM/I and available only over the oceans, and layered WV. In addition, supplemental products such as SSM/I WV, radiosonde WV, TOVS WV, cloud liquid water, and DSC for total column and layered WV are provided. These products are available in four possible temporal averaging periods: day, month, pentad, and annual. The complete five-year dataset along with documentation and software are available from NASA's Distributed Active Archive Center - Marshall Space Flight Center . The monthly averages are available electronically while the daily global grids are be supplied via mail. To order NVAP data, see the WWW link at the end of this article.

ACKNOWLEDGEMENTS

We wish to thank Mr. Donald Reinke for his assistance with the project organization and processing, and Mr. Ian Wittmeyer for special contributions to the TOVS processing. We gratefully acknowledge the support and helpful discussions provided during the course of this work by Dr. James C. Dodge, NASA Technical Monitor and his colleagues at NASA Headquarters. Special thanks are extended to Bill Elliott and Becky Ross at the NOAA Air Resources Laboratory, and to John Bates, NOAA ERL, for discussions and access to their datasets. Thanks also to the reviewers, who provided many useful comments. This work has been performed under NASA Contract NASW-4715 to Science and Technology Corporation and under NASA Contract NAGW-2700 to Colorado State University.

REFERENCES

Bannon, J.K. and L.P. Steele, 1960:  Average water vapor content
of the air.  Geophysical Memoir, 102, British Meteor. Office, London.

Bates, J.J., 1991:  High-frequency variability of Special Sensor
Microwave/Imager derived wind speed and moisture during an
intraseasonal oscillation.  J. Geophys. Res., 96, 3411-3423.

Chahine, M.T.,  and J. Susskind, 1991: Derivation of long-term
climate data sets from NOAA's HIRS2/MSU.  Global and planetary
change,  4, pg 121.

Chedin, A., N.A. Scott, and C. Claud, 1994: Global scale
observation of the earth for climate studies.  Advances in space
research, 14, pg 155.

Chesters D., L.W. Uccellini, and W.D. Robinson, 1983:  Low-level
water vapor fields from the VISSR Atmospheric Sounder (VAS)
"split window" channels.  J. Appl. Meteor., 22, 725-743.

Christy, J.R., R.W. Spencer, and R.T. McNider, 1995:  Reducing
noise in the MSU daily lower-tropospheric global temperature dataset.
J. Climate, 8, 888-896.

Elliott, W.P. and D.J. Gaffen, 1991:  On the utility of radiosonde
humidity archives for climate studies.  Bull. Amer. Meteor. Soc., 72,
1507-1520.

Fleming, H.E., M.D. Goldberg, D.S. Crosby, 1986: Minimum variance
simultaneous retrieval of temperature and water vapor from satellite
radiance measurements. 2nd conference on Satellite Meteorology /
Remote Sensing and Application, Williamsburg VA, 20-23.

Goodberlet, M.A., C.T. Swift, and J.C. Wilkerson, 1989:  Remote
sensing of ocean surface winds with the Special Sensor Microwave /
Imager.  J. Geophys. Res., 94, 14547-14555.

Greenwald, T.J., G.L. Stephens, T.H. Vonder Haar, and D.L. Jackson,
1993: A physical retrieval of cloud liquid water over the global oceans
using SSM/I observations.  J. Geophys. Res., 98, 18471-18488.

-------,-------, S. Christopher, and T.H. Vonder Haar, 1995:
Observations of the global characteristics and regional radiative effects
of marine cloud liquid water.  J. Climate, 8, 2928 - 2946.

Hollinger, J.P. and DMSP Cal-Val Team, 1991:  DMSP Special Sensor Microwave /
Imager calibration/ validation.  Naval Research Laboratory, Washington D.C.
Final Report, Volume 2.

Gutowski, W.L., E.A. Lindemulder, and K. Jovaag, 1995:  Temperature-
dependent daily variability of precipitable water in special sensor
microwave/imager observations.  J. Geophys. Res, 100, 22971-22980.

Jackson, D.L. and G.L. Stephens, 1995: A study of SSM/I -derived
columnar water vapor over the global oceans,  J. Climate, 8, 2025-2038.

Kidder, S.Q., and T.H. Vonder Haar, 1995: Satellite Meteorology:
An Introduction.  Academic Press, 456 pp.

Oort, A.H., 1983:  Global Atmospheric Circulation Statistics,
1958-1973, NOAA Professional Paper No 14, Goverenment printing office,
Washington D.C., 180 pp. + microfiche.

Peixoto, J.P, and A.H. Oort, 1991: Water Cycle. Physics of
Climate, American Institute of Physics, New York, NY,  270 - 307.

Prabhakara, C., H.D. Chang, A.T.C. Chang, 1982:  Remote sensing
of precipitable water over the oceans from Nimbus-7 microwave
measurements.  J. Appl. Meteor., 21, 59-68.

Reale, A.L., M.D. Goldberg, and J.M. Daniels, 1989:  Operational
TOVS  soundings using a physical approach.  Proceedings of the
IGARRS '89 12th Canadian Symposium on Remote Sensing,
Vancouver, B.C., 2653-2657.

Reynolds, R.W., 1988:  A Real-time global sea surface temperature
analysis.  J. Climate, 1, 75-86.

Rosen, R.D., D.A. Salstein, and J.P. Peixoto, 1979: Variability in
the annual fields of large scale atmospheric water vapor transport.
Mon. Wea. Rev.,  107, 26-37.

Ross,  R. J. and W.P. Elliott, 1996: Tropospheric water vapor
trends over North America, 1973 - 1993. J. Climate (submitted).

Rossow, W.B., L.C. Garder, P.J. Lu, and A.W. Walker, 1991:  International
Satellite Cloud Climatology Project (ISCCP).  Documentation of Cloud
Data.  WMO/TD-No. 266, World Meteorological Organization, 76 pp
plus appendices.

Schmetz, J. W.P. Menzel, C. Velden, X. Wu. L. van de Berg, S.
Nieman, C. Hayden, K. Holmlund, and C. Geijo, 1995:  Monthly
mean large-scale analyses of upper-troposphere humidity and wind
field divergence derived from three geostationary satellites, Bull.
Amer. Meteor. Soc., 76, 1578-1584.

Schubert, S.D., R.B. Rood, and J. Pfaendtner, 1993:  An assimilated
dataset for Earth science applications,  Bull. Amer. Meteor. Soc., 74,
2331 - 2342.

Smith, W.L., and H.M. Woolf, 1984: Improved vertical sounding
from amalgamation of polar and geostationary radiance observations.
Proceeding of the Conference on Satellite Meteorology/ Remote
Sensing and Applications,  Clearwater, Florida, 45 - 48.

Starr, V.P., J.P. Peixoto, and R.G. McKean, 1969: Pole-to-pole
moisture conditions for the IGY.  Pure Appl. Geophys. Res., 96,
15311-15324.

Stephens, G.L., 1990:  On the relationship between water vapor over
the oceans and sea surface temperatures.  J. Climate, 3, 634-645.

-------,  and T.J. Greenwald, 1991:  The Earth's radiation budget and
its relation to atmospheric hydrology.  Part II:  Observations of cloud
effects.  J. Geophys. Res., 96, 15325-15340.

-------, D.L. Jackson, and J.J. Bates, 1994:  A comparison of SSM/I
and TOVS column water vapor data over the global oceans.  Meteor.
Atmos. Phys., 54, 183-201.

Tjemkes, S.A., and G.L. Stephens, D.L. Jackson,1991: Space borne
observations of precipitable water: Part I:  SSM/I observations and
algorithm.  J. Geophys. Res., 96, 10941-10954.

Trenberth, K.E., 1981: Seasonal variations in the global sea level
pressure and the total mass of the atmosphere.  J. Geophys. Res., 86,
5238-5246.

-------, J.R.Christy, and J.G. Olson, 1987: Global atmospheric
mass, surface pressure,  and water vapor variations. J.  Geophy. Res.,
92, 14815-14826.

-------, and J.G. Olson, 1988:  Intercomparison of NMC and ECMWF
global analyses:  1980-1986.  NCAR Tech. Note 301, 81 pp.

-------, and C.J. Guillemot, 1995: Evaluation of the global atmospheric
moisture budget as seen from analyses.  J. Climate, 8, 2255-2272.

-------, and T.J. Hoar, 1995: The 1990 - 1995 El Nino Southern
Oscillation Event:  Longest on Record.  Geophy. Res. Letters, 23,
no 8, pg 57.

Vonder Haar, T.H., 1994:  The global energy budget and satellite
observations,  Advances in Space Research, 14, no. 1 131 - 144.

Wentz, F.J., 1991:  User's Manual SSM/I Antenna Temperature
Tapes, Revision 1.  Remote Sensing Systems (RSS) Tech. Report
120191, Santa Rosa, CA, 70 pp.

Werbowtzki, A., 1981:  Atmospheric Sounding Users Guide.  NOAA
Technical Report, NESS 83, U.S. Department of Commerce,
Washington, D.C., 80 pp.

Wittmeyer, I.L., 1990: Satellite based estimates of global precipitable
water vapor distribution and poleward latent heat flux.  Atmospheric
Science paper no. 473, Colorado State University, Ft. Collins,
Colorado, 76 pp.

-------, and T.H. Vonder Haar, 1994:  Analysis of the global ISCCP
TOVS water vapor climatology.  J. Climate, 7, 325-333.

World Climate Research Program, 1990:  Scientific Plan for the Global
Energy and Water Cycle Experiment. WMO/TD-No. 376, WMO,
Geneva, Switzerland, 84 pp.

Wu, X., J. J. Bates, and S.J.S. Khalsa, 1993:  A climatology of the water
vapor band brightness temperatures from NOAA operational satellites.
J. Climate, 6, 1282-1300.

Zuidema, P. and D.L. Hartmann, 1995:  Satellite determination of
stratus cloud microphysical properties.  J. Climate, 8, 1638-1657.


To order the NVAP data-set start up the WWW ordering system.


For more information about the NVAP data-sets, contact:

      Dr. David L. Randel (randel@cira.colostate.edu)
      CIRA
      Colorado State University
      Ft. Collins, CO  80523
      Tel: (970) 491-8219; Fax: (970) 491-8241

Back to the CIRA Home Page