• Bedka, S., R. Knuteson, H. Revercomb, D. Tobin, and D. Turner, 2010: An assessment of the absolute accuracy of the Atmospheric Infrared Sounder v5 precipitable water vapor product at tropical, midlatitude, and arctic ground-truth sites: September 2002 through August 2008. J. Geophys. Res., 115, D17310, doi:10.1029/2009JD013139.

    • Search Google Scholar
    • Export Citation
  • Bevis, M., S. Businger, T. A. Herring, C. Rocken, R. A. Anthes, and R. H. Ware, 1992: GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system. J. Geophys. Res., 97 (D14), 15 78715 801.

    • Search Google Scholar
    • Export Citation
  • Bevis, M., S. Businger, S. Chiswell, T. A. Herring, R. A. Anthes, C. Rocken, and R. H. Ware, 1994: GPS meteorology: Mapping zenith wet delays onto precipitable water. J. Appl. Meteor., 33, 379386.

    • Search Google Scholar
    • Export Citation
  • Collier, J. C., and G. J. Zhang, 2006: Simulation of the North American monsoon by the NCAR CCM3 and its sensitivity to convection parameterization. J. Climate, 19, 28512866.

    • Search Google Scholar
    • Export Citation
  • Collins, W. D., and Coauthors, 2006: The Community Climate System Model version 3 (CCSM3). J. Climate, 19, 21222143.

  • Cook, K. H., E. K. Vizy, Z. S. Launer, and C. M. Patricola, 2008: Springtime intensification of the Great Plains low-level jet and Midwest precipitation in GCM simulations of the twenty-first century. J. Climate, 21, 63216340.

    • Search Google Scholar
    • Export Citation
  • Data Distribution Centre, cited 2010: SRA2 scenario. [Available online at www.ipcc-data.org/ar4/scenario-SRA2.html.]

  • Dow, J. M., R. E. Neilan, and C. Rizos, 2009: The International GNSS Service in a changing landscape of Global Navigation Satellite Systems. J. Geod., 83, 191198, doi:10.1007/s00190-008-0300-3.

    • Search Google Scholar
    • Export Citation
  • Gutman, S. I., S. R. Sahm, S. G. Benjamin, B. E. Schwartz, K. L. Holub, J. Q. Stewart, and T. L. Smith, 2004: Rapid retrieval and assimilation of ground based GPS precipitable water observations at the NOAA Forecast Systems Laboratory: Impact on weather forecasts. J. Meteor. Soc. Japan, 82, 351360.

    • Search Google Scholar
    • Export Citation
  • Hack, J. M., S. G. Yeager, K. W. Oleson, M. M. Holland, J. E. Truesdale, and P. J. Rasch, 2006: Simulation of the global hydrological cycle in the CCSM Community Atmosphere Model version 3 (CAM3): Mean features. J. Climate, 19, 21992221.

    • Search Google Scholar
    • Export Citation
  • Karl, T. R., and K. E. Trenberth, 2003: Modern global climate change. Science, 302, 17191723.

  • Kiehl, J. T., and K. E. Trenberth, 1997: Earth’s annual global mean energy budget. Bull. Amer. Meteor. Soc., 78, 197208.

  • King, R. W., and Y. Bock, 1996: Documentation of the GAMIT GPS analysis software, version 9.4. Massachusetts Institute of Technology and Scripps Institution of Oceanography Rep., 192 pp.

  • Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor, 2007: The WCRP CMIP3 multimodel dataset: A new era in climate change research. Bull. Amer. Meteor. Soc., 88, 13831394.

    • Search Google Scholar
    • Export Citation
  • Revercomb, H. E., and Coauthors, 2003: The ARM program’s water vapor intensive observation periods: Overview, initial accomplishments, and future challenges. Bull. Amer. Meteor. Soc., 84, 217236.

    • Search Google Scholar
    • Export Citation
  • Rothacher, M., 1992: Orbits of satellite systems in space geodesy. Ph.D. dissertation, University of Bern, 243 pp.

  • Schmidt, G. A., and Coauthors, 2006: Present-day atmospheric simulations using GISS ModelE: Comparison to in situ, satellite, and reanalysis data. J. Climate, 19, 153192.

    • Search Google Scholar
    • Export Citation
  • Soden, B. J., and I. M. Held, 2006: An assessment of climate feedbacks in coupled ocean–atmosphere models. J. Climate, 19, 33543360.

    • Search Google Scholar
    • Export Citation
  • Soden, B. J., D. L. Jackson, V. Ramaswamy, D. Schwarzkopf, and X. Huang, 2005: The radiative signature of upper tropospheric moistening. Science, 310, 841844.

    • Search Google Scholar
    • Export Citation
  • Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M. B. Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate Change 2007: The Physical Science Basis. Cambridge University Press, 996 pp.

  • Trenberth, K. E., J. Fasullo, and L. Smith, 2005: Trends and variability in column-integrated atmospheric water vapor. Climate Dyn., 24, 741758.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., L. Smith, T. Qian, A. Dai, and J. Fasullo, 2007: Estimates of the global water budget and its annual cycle using observational and model data. J. Hydrometeor., 8, 758769.

    • Search Google Scholar
    • Export Citation
  • Tripoli, G. J., and W. R. Cotton, 1980: A numerical investigation of several factors contributing to the observed variable intensity of deep convection over south Florida. J. Appl. Meteor., 19, 10371063.

    • Search Google Scholar
    • Export Citation
  • Turner, D. D., R. A. Ferrare, L. A. Brasseur, W. F. Feltz, and T. P. Tooman, 2002: Automated retrievals of water vapor and aerosol profiles from an operational Raman lidar. J. Atmos. Oceanic Technol., 19, 3750.

    • Search Google Scholar
    • Export Citation
  • Ware, R. H., and Coauthors, 2000: SuomiNet: A real-time national GPS network for atmospheric research and education. Bull. Amer. Meteor. Soc., 81, 677694.

    • Search Google Scholar
    • Export Citation
  • Washington, W. M., and Coauthors, 2000: Parallel climate model (PCM) control and transient simulations. Climate Dyn., 16, 755774.

  • Weatherhead, E. C., and Coauthors, 1998: Factors affecting the detection of trends: Statistical considerations and applications to environmental data. J. Geophys. Res., 103 (D14), 17 14917 161.

    • Search Google Scholar
    • Export Citation
  • Wolfe, D. E., and S. I. Gutman, 2000: Developing an operational, surface-based, GPS, water vapor observing system for NOAA: Network design and results. J. Atmos. Oceanic Technol., 17, 426440.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The GCM latitude dependence of PWV for four models (GISS E20 = red, CCSM3 = blue, PCM1 = cyan, and CGCM3.1 = green) in the AR4 SRES A2 scenario is shown for (top) DJF and (bottom) JJA for the 5-yr mean 2005–09. The zonal mean model differences in the (left) 25°–50°N latitude range are enhanced for the longitude passing through the (right) U.S. Great Plains. This result suggests large model differences in PWV occur over the North American continent, where GPS networks can be used for assessment.

  • View in gallery

    Comparison of the coincident 30-min average PWV observations of two GPS receivers at the DOE ARM SGP central facility near Lamont, OK (LMNO and SG01 from the SuomiNet network), for the calendar year 2009 showing a high relative accuracy with an agreement of >99.9% in slope and a small offset (0.16 mm).

  • View in gallery

    GPS station location maps showing the (left) UCAR SuomiNet network and (right) NOAA NPN used in this study and the boundaries of the 10 regions defined in this study over the U.S. Great Plains and Midwest.

  • View in gallery

    Elevation contour levels at 100-m intervals overlaid with the region boundaries defined in this study.

  • View in gallery

    Quadratic fit of monthly-mean GPS PWV to station elevation for (bottom curve) January and (top curve) July, including all of the SuomiNet stations shown in Table A1 for the period 2002–09.

  • View in gallery

    (top) Monthly-mean PWV and anomaly time series and trend fits in the Oklahoma–Kansas region for SuomiNet (plus symbol) and NOAA NPN (circles) GPS stations without correction for station elevation. (bottom) Same data as (top) after correcting the PWV from each station for the elevation difference between each site and the regional topographic mean elevation. The trends computed from each GPS network agree after elevation correction.

  • View in gallery

    (top) GPS PWV regional climatology of the Oklahoma–Kansas region for the SuomiNet network (2002–09) and NPN (2000–09). (middle and bottom) Interannual variability in millimeters and as a percent where the symbols indicate SuomiNet (plus symbol) and NOAA NPN (circles).

  • View in gallery

    (a) Comparison of (top) observed and modeled autocorrelation and (bottom) monthly standard deviation in the Oklahoma–Kansas region for the period 2000–09. (b) Comparison of observed and modeled PWV trends in (top) millimeter per year and (bottom) percent per year in the Oklahoma–Kansas region for the period 2000–09. The GPS and model trends are consistent with zero within the 95% confidence error bars.

  • View in gallery

    Time series of GPS PWV at the DOE SGP ARM central facility site compared to the selected global climate models for the period 2000–10. The GPS SuomiNet and NOAA NPN observations are in agreement with each other, while only GISS E20 agrees with observations at this site in the summer season. All models are in reasonable agreement with GPS observations in the winter season. Symbols: Suomi—○, NPN—×, CCSM3—+, CGCM3.1—*, GISS—□, PCM1—▹.

  • View in gallery

    Comparison of the GPS and GCM mean PWV (2005–09) for the Oklahoma–Kansas region with error bars representing (top) the uncertainty in the mean, (middle) the difference between the GCM and GPS, and (bottom) the fractional error. GISS E20 best captures the total column water vapor in the summer season, while the CCSM3 is 30% too dry.

  • View in gallery

    The seasonal biases in GCMs relative to the GPS networks (Suomi and NPN) over the period 2005–09 are shown for each of the 10 regions illustrated in Fig. 3; (left) the U.S. Great Plains from North Dakota to southern Texas, and (right) the Midwest from Wisconsin to Louisiana.

  • View in gallery

    (top) Winter and (bottom) summer (left) latitude cross-section comparison of GCM PWV at the longitude of the DOE ARM SGP site (97.5 W) for the period 2005–09 with the observed SuomiNet GPS PWV regional means in the U.S. Great Plains (shown as square symbols). (right) Longitude cross-section comparison, at the latitude of 37°N (between 34° and 40°), with the Great Plains and Midwest regional GPS mean values overlaid. All models are in good agreement with observations in the winter season, but only GISS E20 captures the PWV magnitude in the summer season.

  • View in gallery

    Maps of the GCM PWV compared to the NARR for the months of (left) January and (right) August 2006. Note that the GISS model reproduces the observations in the Gulf coast region of the southern United States and propagates moisture into the U.S. Great Plains and Midwest regions better than the other GCMs.

  • View in gallery

    Mean GCM and ARM SGP radiosondes profiles for (left) winter and (right) summer over the period 2005–09 for (top) temperature (°C), (middle) relative humidity (%), and (bottom) mixing ratio (g kg−1). All models show good agreement with radiosonde mixing ratio profiles in the winter, but only the GISS model approximates the observations in the lower troposphere in the summer season. This is consistent with the GISS agreement with GPS PWV observations.

  • View in gallery

    GCM profile errors (model − sonde) by season relative to radiosondes for (a) temperature, (b) water vapor relative humidity, and (c) water vapor specific humidity at the ARM SGP site for the layer averages (left) 0–5 and (right) 5–12 km.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 111 81 11
PDF Downloads 93 47 3

Assessment of Regional Global Climate Model Water Vapor Bias and Trends Using Precipitable Water Vapor (PWV) Observations from a Network of Global Positioning Satellite (GPS) Receivers in the U.S. Great Plains and Midwest

View More View Less
  • 1 Cooperative Institute for Meteorological Satellite Studies, Space Science and Engineering Center, University of Wisconsin—Madison, Madison, Wisconsin
© Get Permissions
Full access

Abstract

Precipitable water vapor (PWV) observations from the National Center of Atmospheric Research (NCAR) SuomiNet networks of ground-based global positioning system (GPS) receivers and the National Oceanic and Atmospheric Administration (NOAA) Profiler Network (NPN) are used in the regional assessment of global climate models. Study regions in the U.S. Great Plains and Midwest highlight the differences among global climate model output from the Fourth Assessment Report (AR4) Special Report on Emissions Scenarios (SRES) A2 scenario in their seasonal representation of column water vapor and the vertical distribution of moisture. In particular, the Community Climate System model, version 3 (CCSM3) is shown to exhibit a dry bias of over 30% in the summertime water vapor column, while the Goddard Institute for Space Studies Model E20 (GISS E20) agrees well with PWV observations. A detailed assessment of vertical profiles of temperature, relative humidity, and specific humidity confirm that only GISS E20 was able to represent the summertime specific humidity profile in the atmospheric boundary layer (<3%) and thus the correct total column water vapor. All models show good agreement in the winter season for the region. Regional trends using station-elevation-corrected GPS PWV data from two complimentary networks are found to be consistent with null trends predicted in the AR4 A2 scenario model output for the period 2000–09. The time to detect (TTD) a 0.05 mm yr−1 PWV trend, as predicted in the A2 scenario for the period 2000–2100, is shown to be 25–30 yr with 95% confidence in the Oklahoma–Kansas region.

Corresponding author address: Jacola Roman, University of Wisconsin—Madison, 1225 W. Dayton St. Madison, WI 53706. E-mail: roman2@wisc.edu

Abstract

Precipitable water vapor (PWV) observations from the National Center of Atmospheric Research (NCAR) SuomiNet networks of ground-based global positioning system (GPS) receivers and the National Oceanic and Atmospheric Administration (NOAA) Profiler Network (NPN) are used in the regional assessment of global climate models. Study regions in the U.S. Great Plains and Midwest highlight the differences among global climate model output from the Fourth Assessment Report (AR4) Special Report on Emissions Scenarios (SRES) A2 scenario in their seasonal representation of column water vapor and the vertical distribution of moisture. In particular, the Community Climate System model, version 3 (CCSM3) is shown to exhibit a dry bias of over 30% in the summertime water vapor column, while the Goddard Institute for Space Studies Model E20 (GISS E20) agrees well with PWV observations. A detailed assessment of vertical profiles of temperature, relative humidity, and specific humidity confirm that only GISS E20 was able to represent the summertime specific humidity profile in the atmospheric boundary layer (<3%) and thus the correct total column water vapor. All models show good agreement in the winter season for the region. Regional trends using station-elevation-corrected GPS PWV data from two complimentary networks are found to be consistent with null trends predicted in the AR4 A2 scenario model output for the period 2000–09. The time to detect (TTD) a 0.05 mm yr−1 PWV trend, as predicted in the A2 scenario for the period 2000–2100, is shown to be 25–30 yr with 95% confidence in the Oklahoma–Kansas region.

Corresponding author address: Jacola Roman, University of Wisconsin—Madison, 1225 W. Dayton St. Madison, WI 53706. E-mail: roman2@wisc.edu

1. Introduction

The Fourth Assessment Report (AR4) of the Intergovernmental Panel on Climate Change (IPCC; Solomon et al. 2007) states that warming of the climate system is “unequivocal.” Water vapor is a major greenhouse gas and is responsible for the dominant feedback in the climate system (Kiehl and Trenberth 1997; Karl and Trenberth 2003; Trenberth et al. 2007). In a comparison of global climate models (GCMs) included in the IPCC AR4 dataset, Soden and Held (2006, p. 3359) state that “water vapor provides the largest positive feedback and that the strength of this feedback can be estimated assuming constant relative humidity in all models.” As global temperatures in the troposphere increase along with the saturation vapor pressure, there is evidence that water vapor amounts also increase (Trenberth et al. 2005). A strong correlation between sea surface temperature (SST) and precipitable water vapor (PWV) has been established using satellite microwave observations over ice-free ocean (Soden et al. 2005). Hack et al. (2006) showed that the Community Atmospheric Model, version 3 (CAM3) had a summertime dry bias relative to global satellite data in the zonal average of PWV for a latitude range (10°–40°N) dominated by continental landmasses, but that it showed reasonably good agreement in winter [December–February (DJF)]. Cook et al. (2008) compared 18 coupled atmosphere–ocean GCM simulations from the IPCC AR4 dataset, including the Community Climate System Model, version 3 (CCSM3), in their analysis of the Great Plains low-level jet and Midwest precipitation. That study showed considerable variation among models in the timing and intensity of the low-level jet, which may be related to the moisture transport from the Gulf of Mexico to the U.S. Great Plains and Midwest in the summer season [June–August (JJA)]. Similarly, Collier and Zhang (2006) examined the Community Climate Model, version 3 (CCM3) simulation of the North American Monsoon and its sensitivity to convection parameterization. Tripoli and Cotton (1980) show the importance of low-level convergence in the modeling of convection and precipitation. This study uses PWV as a measure of the low-level moisture content in the U.S. Great Plains and Midwest. A selection of models was made from the World Climate Research Programme’s (WCRP’s) Coupled Model Intercomparison Project phase 3 (CMIP3) multimodel dataset for the IPCC AR4 A2 scenario (Meehl et al. 2007). Figure 1 illustrates the latitude dependence of PWV from four GCMs for the period 2005–09. The left-hand panels show a zonal average PWV similar to the result of Hack et al. (2006) for winter and summer but in this case for the CCSM3, Canadian Centre for Climate Modelling and Analysis Coupled General Circulation Model, version 3.1 (CGCM3.1), Goddard Institute for Space Studies Model E20 (GISS E20), and Parallel Climate Model, version 1 (PCM1). Note that the CAM3 is used in the CCSM3 and PCM1 coupled climate models. GISS E20 is seen to exhibit higher summertime PWV values in the 25°–50°N latitude range, while the CCSM3 shows the smallest PWV value. Collier and Zhang (2006) examined the CCSM3 simulation of the North American monsoon and its sensitivity to convection parameterization, showing that there were biases as to the location of the monsoon and its diurnal cycle, which may explain this lack of PWV seen in this region during summertime. Another possible explanation for this lack of moisture in CCSM3 is due to the poor modeling of the low-level jet in the Great Plains, which transports moisture from the Gulf of Mexico up to the Great Plains and Midwest (Cook et al. 2008). The right-hand panels of Fig. 1 are the same latitude cross sections except restricted to a longitudinal average between 99.5° and 94.0°W that includes the U.S. southern Great Plains. Figure 1 suggests that large seasonal biases exist between GCMs in the continental land regions of North America. In our analysis, we focus on the region between the Gulf of Mexico and the Canadian border. This region has accurate moisture observations and is an area with strong continental and maritime air masses. We will evaluate regional GCM performance in the 25°–50°N latitude range through comparison of GCM water vapor product output from the IPCC AR4 A2 scenario analysis against column and profile assessment data collected in the U.S. Great Plains and Midwest during the period 2000–10.

Fig. 1.
Fig. 1.

The GCM latitude dependence of PWV for four models (GISS E20 = red, CCSM3 = blue, PCM1 = cyan, and CGCM3.1 = green) in the AR4 SRES A2 scenario is shown for (top) DJF and (bottom) JJA for the 5-yr mean 2005–09. The zonal mean model differences in the (left) 25°–50°N latitude range are enhanced for the longitude passing through the (right) U.S. Great Plains. This result suggests large model differences in PWV occur over the North American continent, where GPS networks can be used for assessment.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

The assessment data consist of observations of PWV from global positioning system (GPS) receivers at fixed ground sites (Bevis et al. 1992, 1994). Networks of ground-based GPS measurements of the water vapor column provide the time and space continuity and absolute accuracy needed to test GCMs, at least with respect to low-level moisture over the continents. Previous lack of regional sampling from ground-based PWV measurements has been addressed with growing networks of GPS receivers, including those of SuomiNet, the National Oceanic and Atmospheric Administration (NOAA) Profiler Network (NPN), and the International Global Navigation Satellite Systems (GNSS) Service (IGS) (Ware et al. 2000; Wolfe and Gutman 2000; Dow et al. 2009). We will present GCM comparisons to observations of monthly-mean PWV for a decade in the U.S. Great Plains and Midwest. To test the realism of GCM performance on regional spatial scales and seasonal to interannual time scales, results are presented as mean biases, variability, and trends in PWV over 10 regions. The spatial assessment of the results is confirmed through a comparison with the National Centers for Environmental Prediction (NCEP)’s North American Regional Reanalysis (NARR), which assimilates surface meteorological and conventional radiosonde measurements using a numerical weather prediction (NWP) model. The vertical distribution of water vapor specific humidity in the models is further examined using the research radiosondes launched from the Department of Energy’s Atmospheric Radiation Measurement Program’s Southern Great Plains (DOE ARM SGP) site near Lamont (station LMNO), Oklahoma.

2. Characteristics of ground-based GPS PWV measurements

This study expands on Bedka et al. (2010) in the use of PWV measurements derived from ground-based microwave radiometers (MWR) to assess retrieved water vapor profiles from satellite infrared sounders. In contrast to the calibrated sky brightness temperatures observed by the MWR, a ground-based GPS receiver measures the wet signal delay in the GPS signal to derive deviations in the refractivity profile from that of a dry atmosphere, which can be interpreted as due to changes in the water vapor column (Bevis et al. 1992). Accuracy of the GPS integrated water vapor retrieval method has been estimated to be less than 3 mm RMS with a long-term bias of less than 2 mm (Bevis et al. 1994; Ware et al. 2000). The paper by Bedka et al. (2010) includes a comparison of coincident MWR PWV and GPS integrated water vapor [station identification (ID) SG01] measurements at the DOE ARM SGP central facility that indicates relative agreement between these two ground-based instruments within 4% over a range of column water amounts between 10 and 50 mm over the 6-yr period 2002–08. The Bedka et al. (2010) result for the period 2002–08 is consistent with detailed water vapor intercomparison analyses performed at the ARM SGP site during the period 1996–2000 (Revercomb et al. 2003), which found the GPS PWV about 3%–4% drier than the ARM MWR. A detailed analysis of the GPS data collected at the SGP site is given in Wolfe and Gutman (2000). Therefore, the uncertainty of the GPS integrated water vapor at the ARM Southern Great Plains site is estimated to be better than 5% for water amounts in the range of 10–50 mm. Since we use many individual GPS stations in our regional analyses, we also need to assess the relative accuracy among GPS receivers. To quantify the relative accuracy, we compared the integrated water vapor from two GPS units installed 9 km horizontally and 12 m vertically near the DOE ARM SGP central facility site for the year 2009 from the SuomiNet, processed using Bernese software (Rothacher 1992). Figure 2 shows a scatterplot of the individual matchups of 30-min sky observations between the GPS stations SG01 and LMNO. A linear regression line fit of the LMNO-minus-SG01 integrated water vapor values indicates a near-zero dependence on PWV amount (0.00078 mm mm−1 ± 0.00072) with a bias of 0.16 mm ± 0.02. The sign and magnitude of this bias is consistent with the 12-m height difference between the two stations. This high relative accuracy (>99.9%) between two collocated GPS stations combined with the previously stated 5% absolute accuracy for the SG01 station suggests that ground-based GPS PWV measurements are a good reference for the assessment of studies. Moreover, the absolute calibration of the GPS refractivity measurement is traceable to an international time standard, which indicates that the entire GPS ground-based network will have some common systematic errors. Known uncertainties in the ground-based GPS PWV measurement are related to the use of surface pressure and temperature to estimate a mean atmospheric temperature profile (Bevis et al. 1994). As improved processing methods are developed, they can be applied after the fact to the entire GPS network data record. With an absolute calibration reference of this type, we can use the data from each individual sensor without the need to create overlapping data records or perform complex intercalibrations. In particular, trends are relatively immune to gaps in the data record, which allow the results from multiple sensors to be combined without leading to additional systematic errors. The GPS delay measurements are also largely immune to cloud cover and precipitation, and thus the continuous time sampling (30-min intervals) of PWV provides unbiased sampling of the diurnal cycle.

Fig. 2.
Fig. 2.

Comparison of the coincident 30-min average PWV observations of two GPS receivers at the DOE ARM SGP central facility near Lamont, OK (LMNO and SG01 from the SuomiNet network), for the calendar year 2009 showing a high relative accuracy with an agreement of >99.9% in slope and a small offset (0.16 mm).

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

In contrast to other instrumentation of similar accuracy, the GPS ground-based receivers are relatively inexpensive to deploy while providing reliable unattended operation (Wolfe and Gutman 2000). This advantage has led to a rapid expansion of the network of ground-based GPS receivers primarily in support of real-time weather applications (Gutman et al. 2004). The SuomiNet is a project of the University Corporation for Atmospheric Research (UCAR), a nonprofit consortium of research universities, on behalf of the National Science Foundation and the university community, which processes a collection of GPS sensors deployed by educational and government institutions (Ware et al. 2000). A complimentary network in the United States is provided by the NOAA Profiler Network (NPN), which includes a GPS receiver as part of the instrument suite at each site (Wolfe and Gutman 2000). The NOAA GPS Meteorology (GPS-Met) networks provide valuable time continuity of measurements dating back to at least 1996. A useful overlap of these ground-based GPS networks provides significant, but nonuniform, regional coverage in the United States. We take advantage of the regional coverage of the GPS networks to extend our analysis of PWV from the ARM SGP site in north-central Oklahoma to the regions of the U.S. Great Plains and Midwest defined by the boundaries shown in Fig. 3. With this spatial coverage, we are able to estimate the monthly-mean PWV for each region and compare these regional means to the same quantities computed using climate model output grids. This provides a valuable latitudinal sampling of the model fields over a domain for which there is considerable disagreement among climate models, as shown in Fig. 1. The U.S. Great Plains is also a region of significant “return flow” of moisture from the Gulf of Mexico, which provides much of the moisture for precipitation throughout the eastern half of the United States in the summertime (Trenberth et al. 2007). The continuous temporal sampling, the expanding regional coverage, the good absolute accuracy and precision, and the traceability of absolute calibration to an international standard reference are all advantages that make the use of ground-based GPS PWV a good choice for climate model assessment. Limitations in the use of ground-based GPS PWV in climate model assessment include limited spatial sampling, inconsistent periods of operation, and ongoing research into optimal processing methods for the extraction of PWV from the radio delay signal.

Fig. 3.
Fig. 3.

GPS station location maps showing the (left) UCAR SuomiNet network and (right) NOAA NPN used in this study and the boundaries of the 10 regions defined in this study over the U.S. Great Plains and Midwest.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

3. Observations and model output

This paper makes use of GPS PWV observations from the SuomiNet network and the NOAA GPS-Met. The GPS data used in this study were obtained through the ARM Climate Research Facility data archive (http://www.arm.gov/data/vaps/suomigps). Two types of files containing GPS station data were used: sgp30suomigpsX1 and 30wpdngps.

The file sgp30suomigps provides data from the conterminous United States (CONUS) sites starting 7 June 2001 until the present. This file provides data from the SuomiNet network and has been consistently processed with the updated Bernese GPS Software, version 5 (B5.0; Rothacher 1992), including the NOAA stations found in the NPN. These files contain a day’s worth of data, and each day of data contains measurements made every 30 min for each station. The list of SuomiNet stations used in this study is shown in the appendix in Table A1 along with the respective data start and end dates. Each file contains the station names, network ID, station latitude and longitude, and station elevation above sea level. The data for each station include the duration of validity of measurement, precipitable water vapor, precipitable water vapor error, surface atmospheric pressure, surface temperature, and surface relative humidity.

The file 30wpdngps contains data from over 500 sites, predominately from the CONUS area. This study used all the stations in the U.S. Great Plains and Midwest regions. The data are available starting from 1 January 1996 until the present. NOAA’s Earth System Research Laboratory, formally the Forecast Systems Laboratory, processed the data using GAMIT software (King and Bock 1996). The NPN files contain a day’s worth of data, and each day of data contains measurements made every 30 min at each station. A list of NPN stations used in this study is shown in the appendix in Table A2 along with the data start and end dates. Each file contains station names, station numbers, station latitude and longitude, and station elevation. For each station, the data provides measurements for PWV, surface pressure, temperature, and relative humidity (available starting 9 December 2001).

Research grade radiosonde data were also obtained through the DOE ARM data archive. For this study, the file 10rlprofmr1turn was used, which provides data from 2 October 2004 through the present at the ARM SGP central facility site, near Lamont, Oklahoma. Each daily file includes profiles of pressure, temperature, mixing ratio, and relative humidity, which are linearly interpolated from 4 times a day radiosonde launches to 10-min intervals to match the ARM Raman lidar measurements of mixing ratio vertical profiles (Turner et al. 2002). Since these profile data are continuous in time (through interpolation), they are easily matched to the continuous 30-min sampling of the GPS PWV measurements. The radiosonde water vapor mixing ratio profiles in this data product are the original measurements and have not been scaled (http://www.arm.gov/instruments/rl).

The NARR data used in this study were obtained through NOAA’s National Operational Model Archive and Distribution System (NOMADS), and was obtained for January and August 2006. This was prior to the assimilation of GPS PWV observations over North America. The NARR data used assimilated observational data from rawinsondes, dropsondes, pibals, aircraft, surface observations, and geostationary satellites. Other datasets include precipitation datasets from various sources, Television and Infrared Observation Satellite Operational Vertical Sounder (TOVS), hourly and 3-hourly surface stations, ship and buoy data, Air Force snow data, SST, sea ice data from a satellite, and Canadian lake ice from the Canadian Ice Center. The file contains the analysis and desired variables from the 0- to 3-h forecast. Variables available include surface pressure, relative humidity, surface temperature, and specific humidity as well as the total column water vapor (http://nomads.ncdc.noaa.gov/data.php?name=access#narr_datasets).

This paper uses GCM output retrieved through the WCRP CMIP3 multimodel database, which is the primary portal for the Earth System Grid (ESG). Each model output was used in the IPCC Fourth Assessment Report and is available starting January 2000 through December 2100. For this study, the Special Report on Emissions Scenarios (SRES) A2 run 1 experiment was selected, which is described by the Data Distribution Centre’s (2010) website as “a very heterogeneous world with continuously increasing global population and regionally oriented economic growth that is more fragmented and slower than in other storylines.” This study used the following models: CCSM3, CGCM3.1 Model T47, GISS E20/Russell, and PCM1 (Collins et al. 2006; Schmidt et al. 2006; Washington and Coauthors 2000). We selected the models that had the PWV variable, provided by the modelers themselves, at the time frequency we wanted for the A2 scenario. Each file contains latitudes, longitudes, time, and a specific variable. The global grid dimensions vary with each model: CCSM3—128 × 256 × 17, CGCM3.1—43 × 96 × 17, PCM1—64 × 128 × 17, and GISS E20—46 × 72 × 17. For profiles, each file would also contain some sort of level (pressure, height, hybrid-sigma level), which corresponds to the desired variable. Each model data type spans the 100-yr period, 2000–2100, although the models have various starting dates, and contain the monthly averages of PWV and vertical profiles of temperature and water vapor (https://esgcet.llnl.gov:8443/home/publicHomePage.do).

4. Methodology

We compare the PWV from GPS ground-based observations to the total column water vapor obtained from vertical integration of climate model water vapor fields. The methods used in the data and model comparison are given in the following subsections.

a. GPS PWV analysis approach

The first step was to extract the measured data from the archival GPS files. To extract the data, a loop over the daily files was used and an output file for each year was created. The output file contained the extracted measurements every 30 min of the desired variables, such as surface temperature, PWV, and surface pressure, for the selected stations. Table A3 lists the criteria used for quality control on the GPS data in the monthly statistics. Monthly averages for individual stations were then calculated. To compute these statistics, a loop over all of the output files for each year was implemented and the mean, median, standard deviation, and mean error of each desired variable (PWV, temperature, relative humidity, etc.) for each month was calculated. Along with this, the monthly-mean climatology and the interannual variability for individual stations were computed. The monthly time series anomaly for each station was also computed.

Prior to computing regional averages from the GPS station data, an elevation correction was applied to each station time series to adjust the PWV values for the elevation difference between the site and the geographic mean elevation of the defined region. The region boundaries and mean elevation are given in Table A1. This correction accounts for the topographic variation within each study region illustrated in Fig. 4. The relation between PWV and surface elevation was determined by fitting a quadratic function (y=ax2 + bx + c) to the monthly-mean PWV values of all sites in the U.S. Great Plains and Midwest between 2002 and 2009. This is illustrated in Fig. 5 for the months of January and July. Table 1 contains the PWV elevation correction coefficients determined for each month. These coefficients are only valid over the period of the fit. The offset coefficient c may be expected to change as the mean temperature responds over a region in a changing climate scenario.

Fig. 4.
Fig. 4.

Elevation contour levels at 100-m intervals overlaid with the region boundaries defined in this study.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Fig. 5.
Fig. 5.

Quadratic fit of monthly-mean GPS PWV to station elevation for (bottom curve) January and (top curve) July, including all of the SuomiNet stations shown in Table A1 for the period 2002–09.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Table 1.

Quadratic equation coefficients used to fit GPS PWV vs station elevation.

Table 1.

Next, the regional averages for each month were determined for the region boundaries shown in Fig. 3. The stations within each region were found, and the monthly statistics for each individual station in the region were used to calculate a mean, median, standard deviation, and mean error for that particular region. Also, the monthly-mean climatology and the interannual variability, by month, for the specific region were computed. From this, a time series anomaly was also calculated as the measured time series minus the monthly-mean climatology. The NPN and SuomiNet networks of GPS sensors were analyzed separately and compared. Figure 6 shows the monthly-mean and anomaly time series of the NPN and SuomiNet networks’ PWV over the Oklahoma–Kansas region for the period 2000–09 before and after adjusting for the difference of each station’s elevation from the regional mean topographic elevation. Note that the NPN and SuomiNet regional time series are in much better agreement after the site elevation correction is applied, where the difference in the two slopes relative to the combined uncertainty is 0.5% before correction and 0.1% after correction.

Fig. 6.
Fig. 6.

(top) Monthly-mean PWV and anomaly time series and trend fits in the Oklahoma–Kansas region for SuomiNet (plus symbol) and NOAA NPN (circles) GPS stations without correction for station elevation. (bottom) Same data as (top) after correcting the PWV from each station for the elevation difference between each site and the regional topographic mean elevation. The trends computed from each GPS network agree after elevation correction.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Regional trends were computed from the monthly anomaly time series for each region. A least squares fit for the trend was computed with equal weighting for each estimated monthly-mean regional PWV anomaly value. Equal weighting was chosen because, unlike the individual station monthly estimates, we do not have enough information to estimate the error in the regional monthly estimates due to spatial sampling. However, the regional estimated trends do have an advantage over individual station trends, by having the longest possible time record with the fewest number of temporal data gaps, which should lead to more robust trend estimates. For each region, an uncertainty in the trend was calculated using equations from Weatherhead et al. (1998). The equation for calculating the trend uncertainty (k = 1) is as follows:
e1
where σN is the monthly standard deviation and Φ is the monthly autocorrelation. To calculate the number of years of monthly data needed to detect a trend at a 95% confidence level with probability 0.90, the following equation was used:
e2
where σN/ω0 is the ratio of the monthly standard deviation to the annual trend and Φ is the autocorrelation. The observed SuomiNet (NPN) trend, shown in Fig. 6, is 0.01 mm yr−1 (−0.04 mm yr−1) with an uncertainty of 0.11 mm yr−1 (0.07 mm yr−1) at the 95% confidence level (k = 2) over the Oklahoma–Kansas region from 2000 until 2009, where the uncertainty is computed using Eq. (1). Note that without the site elevation correction, an artificial trend is introduced in both the NPN and SuomiNet PWV time series. After elevation correction of each site to the regional topographic mean elevation, the measured trend from NPN and SuomiNet networks is statistically consistent with each other and with zero trends in PWV over the period 2000–09 for the Oklahoma–Kansas region. A similar process was used to calculate the trends for individual stations by computing the monthly averages for the individual stations and calculating the trends over the period of the available station data. Table 2 and Fig. 6 illustrate that the independent processing methods used in computing PWV for the SuomiNet and the NPN networks show good consistency in trend determination.
Table 2.

GPS and GCM regional trends and bias statistics.

Table 2.

b. Global climate model intercomparison approach

GCMs provide a way to predict the increase in the atmosphere’s total water vapor content. However, the GCMs are largely lacking assessment of the accuracy of the water vapor predictions. Through the following intercomparison approach, we provide an assessment for the PWV over land regions, and a profile assessment for temperature, relative humidity, and mixing ratio at the ARM SGP central facility site.

For regional statistics, GCM grid points were selected that lay within each regional boundary. Each region had latitude and longitude limits that represented the regional boundary shown in Fig. 3. Grid points with latitude and longitude that lay within these regional latitude and longitude limits were extracted. The number of grid points ranged from 1 for each region for GISS E20 to more than 10 for CCSM3.

To begin, surface temperature, surface pressure, surface relative humidity, and PWV that covered the same time and spatial span as the GPS data were extracted. The files for the GCMs already contained monthly averages. From these extracted monthly averages, regional statistics, regional trends, and station trends were computed using the exact same methods that were applied to the GPS data. These trends were then directly compared to those from the GPS data.

Next, the monthly statistics from both the GCMs and the GPS data were used to compare the two datasets and determine the differences between them. One approach used was to simply subtract the predicted PWV time series of the GCMs by the observed PWV time series of the GPS at the SGP central facility site. This study also compared the two datasets through analysis of a seasonal 5 yr (2005–09) mean and interannual variability. To implement this comparison approach, the four seasons were defined: winter (DJF), spring [March–May (MAM)], summer (JJA), and fall [September–November (SON)]. The regional statistics were then averaged over each season for each year. These seasonal averages were then averaged over the 5 yr, producing a seasonal mean. This method was applied to the PWV measurements, and was used to compute the 5-yr seasonal interannual variability of the PWV, the seasonal mean PWV difference (GCM − GPS), and the seasonal mean fractional error of the PWV, for each region.

Profiles of mixing ratio, relative humidity RH, pressure P, and temperature T of the GCMs were extracted covering the same time span available for the Vaisala radiosonde data at the SGP central facility site. Since each model has a different grid size, a latitude–longitude range was determined for each GCM, which would locate the closest point to the SGP central facility site, ranging from 1° to 4° away from the exact location of the SGP central facility site. The radiosonde P, T, RH, and mixing ratio data were provided as interpolated 10-min sky mean values, which were then averaged daily and monthly, creating monthly statistics at a set of vertical height levels. Again, the GCM’s were provided already averaged monthly for each level in the profile. The datasets contained different height scales, and therefore the GCM’s were interpolated to match the finer height scale used for the radiosonde profiles. From these monthly statistics, the first approach used to assess the profiles was to compute the difference at each level (GCM− sonde) of each variable. Next, 5-yr seasonal averages were computed, for each measurement, using the same method applied to the surface data, creating a 5-yr temperature–relative humidity–mixing ratio seasonal mean of the profile. Another comparison approach was to calculate the average from 0 to 5 km in the profile for each season and each variable. Then, the difference between the sonde and the GCM was computed. This approach was used for the top of the profile as well (5–12 km).

5. Results

We present results for the estimation of PWV regional climatology using ground-based GPS receivers, an assessment of the PWV bias and trends in GCMs in the period 2000–09, and a detailed evaluation of the vertical distribution of GCM water vapor from a profiling site within the study region.

a. Regional PWV climatology using GPS

Regional statistics for the 10 study regions in the U.S. Great Plains and Midwest shown in Fig. 3 were computed separately for the SuomiNet and NPN networks for the periods 2002–09 and 2000–09, respectively. As shown in Tables A1 and A2 in the appendix, the number of sites per region has increased during this period, with the oldest sites beginning in the mid-1990s and the newest sites starting as recently as June 2008. Table 2 summarizes the monthly standard deviation and autocorrelation for each region. The autocorrelation ranges from 0 to 0.6 for the various regions. The monthly anomaly standard deviation is between 1 and 4 mm. Figure 7 shows the estimated monthly-mean climatology and interannual variability of the PWV for the Oklahoma–Kansas region from 2000 through 2009 for the two GPS networks. Despite the large standard deviation of the monthly-mean PWV anomaly, we find the interannual variability of the PWV climatology is relatively small (<1 mm). Note that the NPN shows significantly higher variability than SuomiNet in the drier months. This may be explained by differences in the data processing constraints applied to the zenith total delay time series.

Fig. 7.
Fig. 7.

(top) GPS PWV regional climatology of the Oklahoma–Kansas region for the SuomiNet network (2002–09) and NPN (2000–09). (middle and bottom) Interannual variability in millimeters and as a percent where the symbols indicate SuomiNet (plus symbol) and NOAA NPN (circles).

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

The number of years to detect a 0.05 mm yr−1 trend is shown in Table 3 using Eq. (2) and the regional autocorrelation and monthly standard deviation in Table 2. The time to detect (TTD) ranges from 25 to 30 yr across the 10 study regions. Also shown is the 95% confidence interval for the TTD using the formulation of Weatherhead et al. (1998). These TTD estimates are important because they are based on measurements of the natural variability derived from actual observations rather than from climate model simulations, which may not include all the small time and space scales found in “weather noise.”

Table 3.

Time to detect a 0.05 mm yr−1 trend.

Table 3.

To investigate the spatial and temporal sampling of the GPS network, we looked at trends from individual stations compared to the regional trends shown in Table 2. The density of station coverage per region varies from 3 for the GP_MN_WI region to 28 for the GP_OK_KS region. Many GPS stations were added to the operational network during the 1990s. The number of months of data collection varies between 3 and 103 at individual stations. We can make the general observation that stations with shorter time records have a larger variation in the fitted trend, but these trends are not statistically significant if the monthly variability and autocorrelation are taken into account. As more stations are added to a region with time, the estimate of the regional average is improved. This makes the regional time series a more robust result than the time series for any individual station within that region that may be subject to missing data or short data records. For this reason, the regional trends are believed to be more accurate than trends at individual stations and the regional estimates are used in the assessment of the GCM output.

b. GCM PWV assessment

We present the results of the GCM PWV comparison to observed PWV from GPS networks and the North American Regional Reanalysis. The GCM data were analyzed using the same methodology as that used for the GPS regional analysis described previously, except that no elevation correction was applied since the GCM grids completely cover each region. GCM and GPS observed standard deviation and autocorrelation are similar for each region, as shown in Table 2, and illustrated in Fig. 8a.

Fig. 8.
Fig. 8.

(a) Comparison of (top) observed and modeled autocorrelation and (bottom) monthly standard deviation in the Oklahoma–Kansas region for the period 2000–09. (b) Comparison of observed and modeled PWV trends in (top) millimeter per year and (bottom) percent per year in the Oklahoma–Kansas region for the period 2000–09. The GPS and model trends are consistent with zero within the 95% confidence error bars.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Table 3 shows the computed TTD for a 0.05 mm yr−1 trend for both GPS observations and model predictions. The models’ TTDs are similar in magnitude to the GPS observations. Figure 8b summarizes the trend comparison of GPS and GCM PWV for the Oklahoma–Kansas region for the period 2000–09. Figure 8b indicates that all four GCMs predict a zero trend in this decade, within statistical uncertainty, which is confirmed by both GPS networks for this region. This is in contrast to the long-term predicted trend for the AR4 SRES A2 scenario over the 100-yr period 2000–2100, which was computed, using the same methodology, to be 0.050 mm yr−1 (0.054 mm yr−1) with an uncertainty of 0.008 mm yr−1 (0.009 mm yr−1) at the 95% confidence level (k = 2) for the CCSM3 (GISS E20). Note the close agreement in the 100-yr PWV trend between the CCSM3 and GISS E20 despite large seasonal differences in the two models.

A time series comparison of PWV between GPS and global climate models is shown in Fig. 9 at the location of the SG01 GPS station (DOE ARM SGP central facility). While there is reasonable agreement in the winter season, there is a large range in the model estimates for the summer season at this location. A bar plot of the seasonal mean and standard deviation over the period (2005–09) is shown in Fig. 10 for the Oklahoma–Kansas region, where GCM-minus-SuomiNet GPS PWV bias error is shown in millimeters (middle) and as a percent (bottom). The CCSM3 has the largest summertime bias, a difference of −10.45 mm, while GISS E20 shows the best agreement, a difference of 3.701 mm. The model biases are roughly consistent across all of the 10 study regions as shown in Fig. 11.

Fig. 9.
Fig. 9.

Time series of GPS PWV at the DOE SGP ARM central facility site compared to the selected global climate models for the period 2000–10. The GPS SuomiNet and NOAA NPN observations are in agreement with each other, while only GISS E20 agrees with observations at this site in the summer season. All models are in reasonable agreement with GPS observations in the winter season. Symbols: Suomi—○, NPN—×, CCSM3—+, CGCM3.1—*, GISS—□, PCM1—▹.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Fig. 10.
Fig. 10.

Comparison of the GPS and GCM mean PWV (2005–09) for the Oklahoma–Kansas region with error bars representing (top) the uncertainty in the mean, (middle) the difference between the GCM and GPS, and (bottom) the fractional error. GISS E20 best captures the total column water vapor in the summer season, while the CCSM3 is 30% too dry.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Fig. 11.
Fig. 11.

The seasonal biases in GCMs relative to the GPS networks (Suomi and NPN) over the period 2005–09 are shown for each of the 10 regions illustrated in Fig. 3; (left) the U.S. Great Plains from North Dakota to southern Texas, and (right) the Midwest from Wisconsin to Louisiana.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Assessment of the latitude dependence of the GCM PWV in the U.S. Great Plains is shown in Fig. 12 for the period 2005–09 using the SuomiNet GPS network for comparison. The GISS model shows remarkably close agreement with the GPS regional estimates for each season, particularly in the Northern Hemisphere summer (JJA). The CCSM3, PCM1, and CGCM3.1 models also agree with GPS PWV in the winter season (DJF) but contain a significant dry bias in the summer season (~5–15 mm). This dry bias in CCSM3 and PCM1 in summer is consistent with the study of Hack et al. (2006), since both of these models use CAM3 as the atmospheric component. The right-hand panels of Fig. 12 show the cross section along a constant latitude value between 34° and 40°N passing through northern Oklahoma and southern Kansas and illustrate the large PWV variations at this latitude due to the influence of the Pacific and Atlantic Oceans, the North American, African, and Asian continents, as well as the surface elevation changes of the Rocky Mountains and the Tibetan Plateau. Note that the GISS model shows larger PWV values than the other GCMs in both North America and North Africa.

Fig. 12.
Fig. 12.

(top) Winter and (bottom) summer (left) latitude cross-section comparison of GCM PWV at the longitude of the DOE ARM SGP site (97.5 W) for the period 2005–09 with the observed SuomiNet GPS PWV regional means in the U.S. Great Plains (shown as square symbols). (right) Longitude cross-section comparison, at the latitude of 37°N (between 34° and 40°), with the Great Plains and Midwest regional GPS mean values overlaid. All models are in good agreement with observations in the winter season, but only GISS E20 captures the PWV magnitude in the summer season.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

To confirm the spatial distribution of the GCM seasonal biases seen for the study regions, we extracted the NARR PWV values for January and August 2006. Figure 13 illustrates the spatial distribution of water vapor for each model compared to the observations contained within the NARR. Using the NARR as truth, we can clearly see from this example that the GISS model in the AR4 A2 scenario output provides the most realistic representation of the northward flow of moisture from the Gulf of Mexico into the U.S. Great Plains and Midwest during the summer season. During the winter each model is reasonably consistent with the NARR monthly-mean value of PWV.

Fig. 13.
Fig. 13.

Maps of the GCM PWV compared to the NARR for the months of (left) January and (right) August 2006. Note that the GISS model reproduces the observations in the Gulf coast region of the southern United States and propagates moisture into the U.S. Great Plains and Midwest regions better than the other GCMs.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

c. GCM profile assessment

To further investigate the nature of this PWV bias between GCMs, we evaluated the height dependence of the specific humidity produced by each GCM along with the corresponding air temperature and relative humidity. For each model we compared the vertical profile for the grid cell containing the DOE ARM SGP central facility to measured vertical profiles from research radiosondes routinely launched from that site. Fig. 14 contains mean profiles for temperature, relative humidity, and water vapor mixing ratio. The large water vapor deficit in PWV in the CCSM3 is due to a dry boundary layer in the summer season. We note that above 5 km, the CCSM3 is actually moist relative to sonde measurements, which is consistent with previously published results for CAM3 (Hack et al. 2006). A summary of the model-minus-sonde error in the vertical moisture distribution is shown in Fig. 15 for each model by season. During the summer season (JJA), for the mean below 5 km, GISS E20 is within 0.5°C of the assessment data in temperature, 6% in relative humidity, and 3% in specific humidity. In contrast, the CCSM3 has large temperature errors (>2°C), large relative humidity errors (>15%), and correspondingly large mean specific humidity errors (>25%) for the same period. Figure 15 shows large model-to-model variations in the results by season for both temperature and relative humidity in the lowest 5 km for the ARM SGP site, suggesting that this is a good location for testing the realism of GCMs for regional climate prediction. Above 5 km all GCMs show similar biases relative to radiosondes, indicating a systematic error in the treatment of upper-tropospheric thermodynamic structure.

Fig. 14.
Fig. 14.

Mean GCM and ARM SGP radiosondes profiles for (left) winter and (right) summer over the period 2005–09 for (top) temperature (°C), (middle) relative humidity (%), and (bottom) mixing ratio (g kg−1). All models show good agreement with radiosonde mixing ratio profiles in the winter, but only the GISS model approximates the observations in the lower troposphere in the summer season. This is consistent with the GISS agreement with GPS PWV observations.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

Fig. 15.
Fig. 15.

GCM profile errors (model − sonde) by season relative to radiosondes for (a) temperature, (b) water vapor relative humidity, and (c) water vapor specific humidity at the ARM SGP site for the layer averages (left) 0–5 and (right) 5–12 km.

Citation: Journal of Climate 25, 16; 10.1175/JCLI-D-11-00570.1

6. Conclusions

This study uses PWV observations from the NOAA NPN and NCAR SuomiNet networks of ground-based GPS receivers to conduct a regional assessment of global climate models. Study regions in the U.S. Great Plains and Midwest highlight the differences among GCM output from the AR4 SRES A2 scenario in their seasonal representation of column water vapor and in the vertical distribution of moisture. In particular, the CCSM3 is shown to exhibit a dry bias of more than 30% in the summertime water vapor column, while the GISS model agrees well with GPS PWV observations. All models show good agreement in the winter season for the study regions. Comparison of GCM output to the NARR suggests that the northward moisture flux from the Gulf of Mexico represents the largest variation from model to model, which may explain the model differences in the U.S. Great Plains and Midwest regions studied. One possible hypothesis is that the GCMs do not simulate the Great Plains summertime low-level jet (LLJ) correctly (Cook et al. 2008), and therefore they are not representing northward moisture transport. A detailed assessment of vertical profiles of temperature, relative humidity, and specific humidity confirm that of the GCMs evaluated, only the GISS model was able to accurately represent the summertime specific humidity profile in the atmospheric boundary layer and thus the correct total column water vapor. The importance of performing a station elevation correction in the estimation of regional trends from GPS networks was demonstrated to avoid introducing artificial trends in the observations. Regional trends using station-elevation-corrected GPS PWV data from two complimentary networks are found to be consistent with null trends predicted in the AR4 SRES A2 scenario GCM output for the period 2000–09. The time to detect (TTD) a 0.05 mm yr−1 PWV trend, as predicted in the A2 scenario for the period 2000–2100, is shown to be about 25–30 yr at the 95% confidence level. Future work in this area will include the evaluation of the Fifth Assessment Report (AR5) model output for the same assessment data used here. The methodology of this study should be applied on a continuous basis in the future to provide an objective means to assess climate prediction scenarios.

Acknowledgments

We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP’s Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multimodel dataset. The Office of Science, U.S. Department of Energy, provides support for this dataset. The authors acknowledge the University Corporation for Atmospheric Research (UCAR) for the use of SuomiNet data. The water vapor data used in this analysis were obtained from the Atmospheric Radiation Measurement Program (ARM), sponsored by the U.S. Department of Energy, Office of Science, Biological and Environmental Research, Climate and Environmental Sciences Division. The authors acknowledge Dr. David Turner for his helpful suggestions in the formulation of this manuscript.

APPENDIX

GPS Station Locations

Table A1.

List of SuomiNet GPS stations used in this study.

Table A1.
Table A2.

List of NPN GPS stations used in this study.

Table A2.
Table A3.

Quality control criteria used to compute monthly-mean PWV from GPS station data. Regional statistics required a 50% uptime for each month.

Table A3.

REFERENCES

  • Bedka, S., R. Knuteson, H. Revercomb, D. Tobin, and D. Turner, 2010: An assessment of the absolute accuracy of the Atmospheric Infrared Sounder v5 precipitable water vapor product at tropical, midlatitude, and arctic ground-truth sites: September 2002 through August 2008. J. Geophys. Res., 115, D17310, doi:10.1029/2009JD013139.

    • Search Google Scholar
    • Export Citation
  • Bevis, M., S. Businger, T. A. Herring, C. Rocken, R. A. Anthes, and R. H. Ware, 1992: GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system. J. Geophys. Res., 97 (D14), 15 78715 801.

    • Search Google Scholar
    • Export Citation
  • Bevis, M., S. Businger, S. Chiswell, T. A. Herring, R. A. Anthes, C. Rocken, and R. H. Ware, 1994: GPS meteorology: Mapping zenith wet delays onto precipitable water. J. Appl. Meteor., 33, 379386.

    • Search Google Scholar
    • Export Citation
  • Collier, J. C., and G. J. Zhang, 2006: Simulation of the North American monsoon by the NCAR CCM3 and its sensitivity to convection parameterization. J. Climate, 19, 28512866.

    • Search Google Scholar
    • Export Citation
  • Collins, W. D., and Coauthors, 2006: The Community Climate System Model version 3 (CCSM3). J. Climate, 19, 21222143.

  • Cook, K. H., E. K. Vizy, Z. S. Launer, and C. M. Patricola, 2008: Springtime intensification of the Great Plains low-level jet and Midwest precipitation in GCM simulations of the twenty-first century. J. Climate, 21, 63216340.

    • Search Google Scholar
    • Export Citation
  • Data Distribution Centre, cited 2010: SRA2 scenario. [Available online at www.ipcc-data.org/ar4/scenario-SRA2.html.]

  • Dow, J. M., R. E. Neilan, and C. Rizos, 2009: The International GNSS Service in a changing landscape of Global Navigation Satellite Systems. J. Geod., 83, 191198, doi:10.1007/s00190-008-0300-3.

    • Search Google Scholar
    • Export Citation
  • Gutman, S. I., S. R. Sahm, S. G. Benjamin, B. E. Schwartz, K. L. Holub, J. Q. Stewart, and T. L. Smith, 2004: Rapid retrieval and assimilation of ground based GPS precipitable water observations at the NOAA Forecast Systems Laboratory: Impact on weather forecasts. J. Meteor. Soc. Japan, 82, 351360.

    • Search Google Scholar
    • Export Citation
  • Hack, J. M., S. G. Yeager, K. W. Oleson, M. M. Holland, J. E. Truesdale, and P. J. Rasch, 2006: Simulation of the global hydrological cycle in the CCSM Community Atmosphere Model version 3 (CAM3): Mean features. J. Climate, 19, 21992221.

    • Search Google Scholar
    • Export Citation
  • Karl, T. R., and K. E. Trenberth, 2003: Modern global climate change. Science, 302, 17191723.

  • Kiehl, J. T., and K. E. Trenberth, 1997: Earth’s annual global mean energy budget. Bull. Amer. Meteor. Soc., 78, 197208.

  • King, R. W., and Y. Bock, 1996: Documentation of the GAMIT GPS analysis software, version 9.4. Massachusetts Institute of Technology and Scripps Institution of Oceanography Rep., 192 pp.

  • Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor, 2007: The WCRP CMIP3 multimodel dataset: A new era in climate change research. Bull. Amer. Meteor. Soc., 88, 13831394.

    • Search Google Scholar
    • Export Citation
  • Revercomb, H. E., and Coauthors, 2003: The ARM program’s water vapor intensive observation periods: Overview, initial accomplishments, and future challenges. Bull. Amer. Meteor. Soc., 84, 217236.

    • Search Google Scholar
    • Export Citation
  • Rothacher, M., 1992: Orbits of satellite systems in space geodesy. Ph.D. dissertation, University of Bern, 243 pp.

  • Schmidt, G. A., and Coauthors, 2006: Present-day atmospheric simulations using GISS ModelE: Comparison to in situ, satellite, and reanalysis data. J. Climate, 19, 153192.

    • Search Google Scholar
    • Export Citation
  • Soden, B. J., and I. M. Held, 2006: An assessment of climate feedbacks in coupled ocean–atmosphere models. J. Climate, 19, 33543360.

    • Search Google Scholar
    • Export Citation
  • Soden, B. J., D. L. Jackson, V. Ramaswamy, D. Schwarzkopf, and X. Huang, 2005: The radiative signature of upper tropospheric moistening. Science, 310, 841844.

    • Search Google Scholar
    • Export Citation
  • Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M. B. Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate Change 2007: The Physical Science Basis. Cambridge University Press, 996 pp.

  • Trenberth, K. E., J. Fasullo, and L. Smith, 2005: Trends and variability in column-integrated atmospheric water vapor. Climate Dyn., 24, 741758.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., L. Smith, T. Qian, A. Dai, and J. Fasullo, 2007: Estimates of the global water budget and its annual cycle using observational and model data. J. Hydrometeor., 8, 758769.

    • Search Google Scholar
    • Export Citation
  • Tripoli, G. J., and W. R. Cotton, 1980: A numerical investigation of several factors contributing to the observed variable intensity of deep convection over south Florida. J. Appl. Meteor., 19, 10371063.

    • Search Google Scholar
    • Export Citation
  • Turner, D. D., R. A. Ferrare, L. A. Brasseur, W. F. Feltz, and T. P. Tooman, 2002: Automated retrievals of water vapor and aerosol profiles from an operational Raman lidar. J. Atmos. Oceanic Technol., 19, 3750.

    • Search Google Scholar
    • Export Citation
  • Ware, R. H., and Coauthors, 2000: SuomiNet: A real-time national GPS network for atmospheric research and education. Bull. Amer. Meteor. Soc., 81, 677694.

    • Search Google Scholar
    • Export Citation
  • Washington, W. M., and Coauthors, 2000: Parallel climate model (PCM) control and transient simulations. Climate Dyn., 16, 755774.

  • Weatherhead, E. C., and Coauthors, 1998: Factors affecting the detection of trends: Statistical considerations and applications to environmental data. J. Geophys. Res., 103 (D14), 17 14917 161.

    • Search Google Scholar
    • Export Citation
  • Wolfe, D. E., and S. I. Gutman, 2000: Developing an operational, surface-based, GPS, water vapor observing system for NOAA: Network design and results. J. Atmos. Oceanic Technol., 17, 426440.

    • Search Google Scholar
    • Export Citation
Save