Abstract

High-temporal-resolution total-column precipitable water vapor (PWV) was measured using a Radiometrics Corporation WVR-1100 Atmospheric Microwave Radiometer (AMR). The AMR was deployed at the University of Manitoba in Winnipeg, Canada, during the 2003 and 2006 growing seasons (mid-May–end of August). PWV data were examined 1) to document the diurnal cycle of PWV and to provide insight into the various processes controlling this cycle and 2) to assess the accuracy of the Canadian regional Global Environmental Multiscale (GEM) model analysis and forecasts (out to 36 h) of PWV. The mean daily PWV was 22.6 mm in 2003 and 23.8 mm in 2006, with distinct diurnal amplitudes of 1.5 and 1.8 mm, respectively. It was determined that the diurnal cycle of PWV about the daily mean value was controlled by evapotranspiration (ET) and the occurrence/timing of deep convection. The PWV in both years reached its hourly maximum later in the afternoon as opposed to at solar noon. This suggested that the surface and atmosphere were well coupled, with ET primarily being controlled by the vapor pressure deficit between the vegetation/surface and atmosphere. The decrease in PWV during the evening and overnight periods of both years was likely the result of deep convection, with or without precipitation, which drew water vapor out of the atmosphere, as well as the nocturnal decline in ET. The results did not change for days on which low-level winds were light (i.e., maximum winds from the surface to 850 hPa were below 20 km h−1), which supports the notion that the diurnal PWV pattern was associated with the daily cycles of local ET and convection/precipitation and was not due to advection. Comparison of AMR PWV with the Canadian GEM model for the growing seasons of 2003 and 2006 indicated that the model error was 3 mm (13%) or more even in the first 12 h, with mean absolute errors ranging from 2 to 3.5 mm and root-mean-square errors from 3 to 4.5 mm over the full 36-h forecast period. It was also found that the 3–9-h forecast period of GEM had better error scores in 2006 than in 2003.

1. Introduction

Atmospheric water vapor is an important element in weather forecasting. Although it only composes a small fraction of the total pressure exerted by atmospheric gases, its spatial and temporal distributions have critical implications for cloud and precipitation production and for their forecasts. The high three-dimensional variability of atmospheric moisture also makes it difficult to depict this parameter accurately in a model’s analysis using current instrumentation, especially with ever-increasing model horizontal resolution. The accuracy of any numerical forecast of precipitation (occurrence and amount) is highly dependent on the correct initialization of atmospheric moisture.

It has been shown that including more detailed atmospheric moisture data, such as observed total-column precipitable water vapor [hereinafter referred to as precipitable water vapor (PWV)], in numerical model assimilation schemes generally improves model forecast skill (e.g., Naito et al. 1998; Gutman and Benjamin 2001; Vedel and Huang 2004). The operational use of surface-based atmospheric microwave radiometers (e.g., Lesht and Liljegren 1997; Knupp et al. 2008) and GPS receivers (e.g., Deblonde et al. 2005; Wang and Zhang 2009) now makes it possible to verify a forecast model’s atmospheric water vapor field over diurnal temporal scales. Radiometric techniques are very accurate in determining PWV amounts [e.g., root-mean-square errors (RMSE) <1 mm; Niell et al. 2001], and they have been used in several studies focusing on atmospheric moisture (e.g., Guldner and Spankuch 2001; Han and Westwater 1995; Westwater et al. 2001), including its diurnal variations (e.g., Wang et al. 2002).

Canadian Global Environmental Multiscale (GEM) model validations of PWV have recently been performed over various locations in Canada by several studies using GPS networks and microwave radiometers (Deblonde et al. 2005; Fisico 2008; Smith et al. 2008). GEM is the primary weather prediction model used in Canada. It is currently run globally (0.3° lat × 0.48° long resolution) and regionally (15-km resolution) over North America (e.g., Cote et al. 1998). Using GPS-derived PWV, Deblonde et al. (2005) and Smith et al. (2008) obtained contrasting results, suggesting that GEM consistently underestimated and overestimated, respectively, PWV over their forecast period of interest. Deblonde et al. (2005) found a bias (GEM − GPS) for GEM analysis (0-h analysis) from −0.7 to −1.2 mm and from −0.5 to −1.0 mm for the 6-h forecast, depending on the region and season in question. Smith et al. (2008) reported a summer mean bias (GEM − GPS) of +0.5 mm (for the analysis through the 9-h forecasts combined) in Alberta, Canada. They also showed that the analysis period had the smallest bias (+0.2 mm) but that the bias increased thereafter up to +0.9 mm by the 9-h forecast. Both Smith et al. and Deblonde et al. showed that GEM’s PWV wet bias became larger as prognosis time increased. Differences in the studies may have arisen from 1) greater data availability (and more seasons used) in Deblonde et al. (2005), 2) differences in GEM versions that were verified (Deblonde used the global GEM whereas Smith used the regional GEM), and 3) the fact that Deblonde used many GPS stations throughout Canada while Smith focused on southern Alberta. In addition, Fisico (2008) compared microwave radiometer PWV with GEM analysis (1200 and 0000 UTC) in the Beaufort Sea region of the Canadian Arctic between October 2003 and June 2004, with an overall coefficient of determination (r2) = 0.93, GEM RMSE = 1.5 mm, and GEM mean bias of −1.0 mm, with results varying depending on the season.

Documenting the diurnal cycle of PWV during the growing season is of interest from a meteorological and climatological perspective. In particular, the identification of the physical processes responsible for the diurnal variation will expand our knowledge of the atmospheric component of the regional hydrologic cycle and aid in the understanding of summer severe weather associated with thunderstorms. Dai et al. (2002) examined multiyear (1996–2000) GPS and radiosonde-derived PWV in the United States to illustrate diurnal water vapor variability. Significant diurnal variations (24-h cycles) were found, with amplitudes of 1.0–1.8 mm over most of the central and eastern United States during summer, with a weaker signal in other seasons. The timing of PWV minima and maxima varied depending on the physiographic region and regionally dominant atmospheric processes that are associated with moisture variations. Other GPS-derived PWV diurnal cycle amplitude estimates in the midlatitudes on a global scale range from 0.5 to 2.0 mm, or about 5% of the annual mean, with the largest amplitudes occurring in summer and a gradual increase in PWV from late morning to evening local time (e.g., Wang and Zhang 2009; Wang et al. 2007; Jin and Luo 2009). Wang et al. (2002) documented the diurnal water vapor cycle using a profiling microwave radiometer in central Oklahoma, mainly during the spring and summer of 2000 and 2001. They found peaks in PWV occurred near 1700 local solar time (LST) with mean amplitudes of 0.08 mm in spring and 0.14 mm in summer, partly consistent with the results of Dai et al. (2002) for a similar location; Dai et al. had higher summer mean PWV amplitudes. Both studies highlighted increasing moisture within the summertime boundary layer as it built in thickness from ∼0900 LST (several hundred meters) to the peak in PWV at 1700 LST (>2 km thick), with the greatest moisture contributions within the lower 2–3 km of the troposphere. Strong (1997) was one of the first studies to investigate diurnal variations of water vapor and boundary layer evolution on the Canadian prairies, finding significant diurnal increases of about 0.4 mm h−1 of water vapor during the day and attributing much of the increase to local evapotranspiration depending on the background synoptic flow.

Dai et al. (1999, 2002) and Wang et al. (2002) all suggested likely reasons for the diurnal variations in PWV. In summary, these factors include 1) surface evapotranspiration, 2) mesoscale low-level horizontal convergence of moisture, 3) large-scale advection, 4) large-scale atmospheric vertical motion, and 5) localized moist convection. The latter two often lead to condensation or deposition and to precipitation. The relative contribution of each process is largely dictated by climatic and ecoclimatic factors associated with the location of interest such as degree of aridity, physiography, regional vegetation type(s), proximity to water bodies, and dominant air masses. Because the bulk of the PWV is concentrated in the lowest 2–3 km, factors 1, 2, and 5 largely affect the diurnal variation of moisture in the boundary layer.

The primary goals of this paper are 1) to document the observed diurnal cycle of PWV in Winnipeg, Manitoba, Canada (49°53′N, 97°10′W) during the 2003 and 2006 growing seasons using high-temporal-resolution measurements from a WVR-1100 (Radiometrics Corporation) Atmospheric Microwave Radiometer (AMR) and to provide insight into the various processes controlling this cycle and 2) to assess the accuracy of the regional GEM model analysis and forecasts of the diurnal cycle of PWV. The location of interest is a midcontinental setting within an agriculturally dominated prairie ecoclimatic region for which no previous study of this kind has taken place (Fig. 1).

Fig. 1.

Location of the study area (Winnipeg; 49°53′N, 97°10′W) and the major ecoclimatic regions. Most of the “grassland” regions in Manitoba are cultivated for annual field crops, and the southern half of the lower boreal zone is primarily agricultural land.

Fig. 1.

Location of the study area (Winnipeg; 49°53′N, 97°10′W) and the major ecoclimatic regions. Most of the “grassland” regions in Manitoba are cultivated for annual field crops, and the southern half of the lower boreal zone is primarily agricultural land.

The 2003 growing season was the tail end of a severe drought over much of the Canadian prairies. Winnipeg experienced 81% of its normal precipitation (55.7 mm below normal, with respect to the 1971–2000 climatic mean) for May–August. As of 27 August 2003, only 60% of normal precipitation was observed. The 2006 growing season experienced even less precipitation (46% of normal, or 159.7 mm below normal), with July being one of the driest during the past 110 years. The North American Regional Reanalysis (NARR) suggests that PWV in the summers (June–August) of both 2003 and 2006 over southern Manitoba was near normal to slightly above normal (anomalies of ≤1 mm relative to the 1971–2000 mean—not shown). With NARR suggesting near-normal PWV over the region and much-below-normal precipitation in both summers, this suggests that the precipitation efficiency was very low—in particular, in 2006. The significance of this situation is that sufficient moisture was present to produce precipitation but that the forcing mechanisms to produce precipitation were lacking in these two years.

2. Data and analysis methods

a. AMR data versus radiosonde observations

The AMR used in this study was the Radiometrics Corporation WVR-1100 (Fig. 2) that is tuned to measure the microwave emissions of the vapor and liquid water molecules in the atmosphere at specific frequencies (23.8 and 31.4 GHz). These two frequencies allow simultaneous determination of water vapor and liquid water burdens along a selected path, in this case nadir upward. The WVR-1100 (hereinafter referred to as AMR) was deployed at the University of Manitoba in Winnipeg. It provided PWV measurements every 30 s during the summer of 2003 (13 May–27 August) and over the identical time period in 2006. These types of AMRs have been in existence for many years and have been shown to provide PWV comparable in accuracy to that from radiosondes, with r2 up to 0.99 (Hogg et al. 1983; Liljegren 1994; Jarlemark and Elgered 2003, etc.).

Fig. 2.

The Radiometrics Corporation WVR-1100 total-column water vapor and liquid microwave radiometer deployed on a roof at the University of Manitoba. The radiometer has a built-in precipitation sensor for quality control (top of the unit). Other instrumentation (not shown) included standard meteorological measurements of air temperature, relative humidity, wind (speed/direction), and pressure and a tipping-bucket-type rain gauge.

Fig. 2.

The Radiometrics Corporation WVR-1100 total-column water vapor and liquid microwave radiometer deployed on a roof at the University of Manitoba. The radiometer has a built-in precipitation sensor for quality control (top of the unit). Other instrumentation (not shown) included standard meteorological measurements of air temperature, relative humidity, wind (speed/direction), and pressure and a tipping-bucket-type rain gauge.

During precipitating conditions, AMR data cannot be used because the microwave dielectric window becomes contaminated with liquid water. The AMR has a blower that evaporates water on the dielectric window once precipitation ceases. The data become useful once again when sufficient drying time has elapsed. This was diagnosed manually by observing the time series data and identifying when the PWV became stable over time (i.e., no obvious sharp increases or decreases). The AMR had a coincident precipitation sensor that also provided excellent quality control for diagnosis of rain events. Any data that showed signs of rain contamination were rejected from the analysis. For the AMR to become a real-time unattended system for use in data-assimilation applications, an automated quality-control system to mitigate erroneous data due to rain would need to be developed.

The hourly AMR PWV was calculated using 30-s data averaged over the 5 min prior to and after the top of each hour of each day from 13 May to 27 August 2003 and 2006. Averaging tends to increase the signal-to-noise ratio by smoothing out high-frequency oscillations. Out of a possible total of 2541 hourly observations in 2003, 2315 (or about 91%) were usable. Out of the 226 observations not included, 219 were rejected because of rain and the other 7 were data outages because of computer maintenance. In 2006, 363 hourly observations (∼14% of 2541 h) were not used because of either rain contamination or missing data.

Despite the AMR’s proven ability in other studies (e.g., Westwater et al. 2001; Jarlemark and Elgered 2003), analysis of the accuracy of the AMR used in this work was performed by comparing its PWV with values derived from once-per-day (1800 UTC) RS-80 radiosonde ascents (hereinafter referred to as sonde) during the summer of 2003. The sondes were launched at the Prairie and Arctic Storm Prediction Centre in Winnipeg (CXWI) located 8.7 km north of the AMR site. The AMR PWV measurements were averaged from sonde release time to 15 min after release time to approximate the time window that the sonde was in the lower half of the troposphere where the bulk of atmospheric moisture exists.

Of the 88 sonde ascents in 2003, 15 were not compared with the AMR because of rain on the AMR window (12 of the 15) or were incomplete soundings that only reached 2–3 km AGL (3 of 15). The sonde flights that were used reached 100 hPa or less (or higher altitudes) to estimate PWV. PWV was estimated from the sondes by integrating their water vapor densities, measured at 5-s intervals, up to their highest level. The 73 valid comparisons yielded an r2 of 0.98 (Fig. 3). The mean absolute difference (AMR − sonde) was 1.0 mm; the AMR’s PWV values were, on average, 0.2 mm larger than the sonde values. Hence, there is no indication of any systematic wet/dry bias of the AMR or the sondes for the range of PWVs observed in this study. Our radiometer–sonde comparison statistics are very similar to other studies conducted outside of Canada (e.g., Jarlemark and Elgered 2003; Niell et al. 2001). Thus, based upon our sonde-versus-AMR comparison, confidence in the AMR PWV measurements can be assumed.

Fig. 3.

Scatterplot of PWV (mm) observations from station CXWI compared with those from the AMR during the warm season of 2003. CXWI was located 8.7 km north of the AMR site.

Fig. 3.

Scatterplot of PWV (mm) observations from station CXWI compared with those from the AMR during the warm season of 2003. CXWI was located 8.7 km north of the AMR site.

b. GEM data

Hourly forecasts of PWV were interpolated from the 15-km GEM numerical model to the University of Manitoba location for both the 0000 and 1200 UTC model runs for the 2003 and 2006 growing seasons through the courtesy of the Canadian Meteorological Centre. There are inherent errors from interpolating the model data to the point in question, and the AMR’s sampling volume increases with height because of its sensor field of view (5.8°); however, it is difficult to quantify these potential errors. In 2003, each model run provided one analysis (time T + 0 h) and 36 forecast values (from T + 1 to T + 36 h; i.e., 1–36 h from T + 0 h) of PWV at 1-h intervals. In 2006, forecasts were available from T + 0 through T + 48 h; however, the study restricted itself to only using data out to T + 36 h to maintain consistency with 2003.

The hourly PWV forecasts from GEM were then statistically compared [mean bias, RMSE, mean absolute error (MAE), and maximum absolute error] with corresponding data obtained from the AMR for two scenarios: 1) for all GEM time periods (from T + 0 to T + 36 h) and 2) excluding hours for which GEM produced precipitation. The rationale for scenario 2 is that the statistical comparisons could be influenced by GEM producing erroneous precipitation (i.e., GEM may have produced precipitation when none actually occurred).

3. Results and discussion

a. Observed diurnal total-column PWV

The AMR PWV measurements were averaged over the growing season (∼3.5 months from mid-May through August) for each hour and revealed a diurnal cycle in 2003 and in 2006 (Fig. 4). The mean daily PWV was 22.6 (±8.3) mm in 2003 and 23.8 (±8.5) mm in 2006, with distinct diurnal amplitudes of about 1.5 and 1.8 mm, respectively. The mean diurnal amplitude was fairly consistent from month to month over each growing season, ranging from 1.4 to 2.0 mm.

Fig. 4.

PWV (mm) observations from 2003, 2006, and both years combined, along with the percentage of hourly Winnipeg airport observations that reported thunder (2003, 2006, and long term (1971–2000).

Fig. 4.

PWV (mm) observations from 2003, 2006, and both years combined, along with the percentage of hourly Winnipeg airport observations that reported thunder (2003, 2006, and long term (1971–2000).

In 2003, the average hourly PWV values increased above the mean daily value during the afternoon (2000 UTC), plateauing at around 23.4 mm in the evening from 2300 to 0200 UTC [note that UTC is 5 h later than local daylight time (LDT)]. PWV then fell to a minimum value of 21.9 mm by 1300 UTC. In 2006, the average hourly PWV values increased above the mean daily value during the early afternoon (1700 UTC), peaking near 24.6 mm in the late afternoon (from 2000 to 0000 UTC). PWV then fell to a minimum value of 22.8 mm at 0900 UTC. In 2006, the mean daily PWV was 1.2 mm higher than in 2003, the maximum values occurred 2–3 h earlier, and the amplitude of the diurnal cycle was greater. Combining both years (Fig. 4) showed that the highest average hourly PWV values occurred in the late afternoon and early evening and then decreased throughout the night to reach an early morning minimum. A maximum of 24.0 mm occurred from 2100 to 0000 UTC, and the minimum of 22.5 mm was observed from 1200 to 1300 UTC. The difference between the maximum and minimum was 1.7 mm, or about 7%, and is similar to many central and east-central U.S. locations based upon GPS-derived and radiometer-derived PWV (Dai et al. 2002; Wang et al. 2002; Wang and Zhang 2009). The mean daily PWV values cited here are less than in many U.S. locations but are 6–7 mm greater than were found in southern Alberta (Smith et al. 2008)—a region that has a drier climate and higher elevation than does the location in this study.

To explain the diurnal variations of PWV in Fig. 4, an examination of the key processes controlling its evolution was required. The PWV over a location for hour i is defined by the equation

 
formula

where PWVi−1 is the previous hour’s precipitable water vapor, ΔSi = [Fi + (ETiPi)L] is the change in the amount of water vapor stored over the location, Fi = [(Fi+ + Fi)/2] is the average horizontal flux over the location, L is the scale length of the area under consideration (Trenberth 1999; Raddatz 2005), Fi+ and Fi are respectively the average hourly horizontal moisture advection of water vapor into and out of the area, ETi is evapotranspiration, and Pi is precipitation. Although ETi has been used to represent the upward flux of water vapor, this term is negative or downward for the nighttime hours when dew occasionally forms. Here Ci is the condensate that is not rained out (cloud production).

On the eastern Canadian prairies, during the growing season, ET and precipitation due to moist deep convection (i.e., thunderstorms) are known to have diurnal cycles (Raddatz 1993; also the Environment Canada 1971–2000 Climate Data for Winnipeg dataset). Thunderstorms are included in about 80% of the significant (>10 mm) precipitation events (Raddatz and Hanesiak 2008). In contrast, the duration of the synoptic-scale weather patterns that are responsible for horizontal advection is several days (Andreas et al. 2002); hence, when averaging PWV over a season the diurnal effects of advection are essentially minimized (e.g., Strong 1997; Raddatz 2005). Temperature, which changes with the seasons and controls the saturation vapor pressure, sets the base seasonal level of atmospheric PWV. The ET is largely from the region’s annual field crops (spring wheat and other C3 crop types), and is generally responsible for 40%–65% of the average summer flux of water vapor over the prairies while advection is responsible for the remainder of the atmospheric water vapor flux (Raddatz 2005).

Evapotranspiration ETi has two components (Oke 1987)—one that is due to the air’s vapor density deficit and a second that is due to the difference between the saturation vapor density at the surface temperature and the saturation vapor density at the air’s temperature. The latter component is often referred to as the energy term, in which the saturation vapor density at the surface temperature is often larger than the air’s temperature during the day and colder at night. When the atmosphere is poorly coupled to the surface, the energy term dominates and the diurnal cycle of hourly ETi values tends to be in phase with the diurnal cycle of global insolation, which peaks, on average, around solar noon—about 1830 UTC (1330 LDT) for Winnipeg. Because the average hourly PWV values for 2003 and for 2006 peaked and reached a plateau later in the afternoon and evening, it can be safely concluded that the atmosphere was, on average, well coupled to the surface in both of these growing seasons and that the atmospheric vapor density deficit was the main force driving ET (Oke 1997). In addition, it has been well documented that ET from crops and prairie grasses has a distinct diurnal pattern (e.g., Baldocchi 1994; Wever et al. 2002; Hanan et al. 2005), with an afternoon hourly maximum and an early morning hourly minimum, very similar to the diurnal PWV cycle observed in Fig. 4. Thus, it is reasonable to assume that the diurnal pattern in PWV observed here is partially caused by the diurnal variations of ET from the region’s annual field crops.

There is also evidence that deep convection and local condensation processes can also influence diurnal PWV. In Winnipeg, the climatological frequency of thunder is the highest during the evening hours (Fig. 4). Thunder is reported (Environment Canada 1971–2000 Climate Data) most often from 0000 to 0600 UTC—peaking at about 0300 UTC. Thunder is reported the least often during the morning and early afternoon hours from 1300 to 1900 UTC, with a minimum at 1600 UTC (Fig. 4).

The hourly occurrences of thunder in 2003 and 2006 roughly followed the climatic norm, but with greater variability (Fig. 4). Occurrences in the 2003 growing season were relatively close to the long-term average, with the 2003 peak in thunder occurring 1 h later than the 1971–2000 average at 0400 UTC. In 2006, the peak in thunder occurrences was earlier than 2003, occurring in the late afternoon at 2200 UTC. The reason for the slight phase shift in the average hourly PWV values and maximum thunder-occurrence times between 2003 and 2006 is not known. However, a plateau in the hourly average PWV values immediately preceded the peak in the occurrence of thunder in both 2003 and in 2006. This suggests a link between the diurnal cycle of average hourly PWV values and the diurnal cycle of average hourly occurrences of thunder/deep convection. Similar observations of the PWV–precipitation/convection relationship have been noted in other studies as well (e.g., Foster et al. 2003; Champollion et al. 2004). It is important to note that the radiometer explicitly measures water vapor using its 23.8-GHz frequency; hence, it does not confuse or combine vapor and liquid phases when cloud is present within its measurement volume, unlike radiosondes in some cases. For this reason, atmospheric water vapor appears to have been reduced by the development of deep convective clouds accompanied by thunder with or without rainout; note that PWV is reduced not only by precipitation but by cloud production (condensation); hence, the production of cloud in the evening and overnight periods can cause PWV to decrease, with or without precipitation, as shown in Eq. (1). This suggests the importance of deep convection (local condensation processes), with or without rain, in the reduction of PWV late in the day and in the overnight periods.

From this analysis, it seems reasonable to associate the mean daily PWV value, or the daily level of atmospheric water vapor, with the seasonal temperature and the control that it has on the atmosphere’s capacity for water vapor. The slightly higher mean daily PWV value in 2006 versus 2003 was, therefore, attributed to an average atmospheric-column temperature difference between these two growing seasons. Because cycles of moisture advection have periodicities of several days (i.e., much longer than the diurnal cycle), it was reasonable to associate the average diurnal oscillations of the average hourly PWV values about the mean daily PWV value with the diurnal cycles of ET and of deep convection accompanied by thunder (i.e., thunderstorms).

To test the assumption that horizontal moisture advection had a minimal effect on the diurnal PWV variations in Fig. 4, the analysis used to generate Fig. 4 was redone for days with light winds (i.e., maximum winds from the surface to 850 hPa were below 20 km h−1). There were 21 such days that fit the criteria, 12 in 2003 and 9 in 2006. Note that moisture advection was not calculated explicitly, but rather, the light-wind cases were used as a proxy for low moisture advection. Redoing the analysis for light-wind days showed no major differences in the diurnal amplitude or evolution of PWV (not shown). Hence, the light-wind analysis further suggests that the daytime increase in PWV is due primarily to ET and the evening/overnight decline in PWV is due to deep convection, with or without precipitation, along with the nocturnal decrease in ET.

b. GEM versus AMR (model precipitation included)

The hourly AMR data were compared with the hourly GEM analysis and forecasts out to 36 h (as outlined in section 2b) to highlight any systematic biases in the model. Figure 5 shows the hourly model bias (GEM − AMR) for the 1200 and 0000 UTC model runs in 2003 and 2006 for analysis time (T + 0 h) out to the 36-h forecast time (T + 36 h). Each hour in Fig. 5 represents the mean of each hourly data point for each day and model run. Note that the AMR observations repeat every 24 h. The T + 0 and T + 24 h AMR values are identical because they represent the average of the exact same observations. This is also true for the group T + 1 and T + 25 h and so forth. It is important to realize that although the observations are being plotted in a repeating pattern they are in fact being compared with different model lead times. Note that the 0000 and 1200 UTC AMR plots for a particular year represent exactly the same values—just out of phase by 12 h.

Fig. 5.

GEM PWV bias (GEM − AMR; mm) for the 0000 and 1200 UTC model runs over the growing seasons of 2003 and 2006.

Fig. 5.

GEM PWV bias (GEM − AMR; mm) for the 0000 and 1200 UTC model runs over the growing seasons of 2003 and 2006.

In 2003, GEM had a slight moist bias (GEM − AMR) of 0.5–1.5 mm in most of its forecasts, except for the 0000 UTC model run which from T + 21 to T + 34 h was within a few tenths of a millimeter of either side of the zero bias line (Fig. 5). In 2006, the model generally exhibited a dry bias. The exception was from T + 0 to T + 12 h of the 1200 UTC run which showed a maximum moist bias of 1 mm at T + 4 h decreasing to near 0 by T + 13 h. The remainder of the 1200 UTC run showed a dry bias that averaged near 0.5 mm. The 0000 UTC run was generally too dry by about 0.5 mm from T + 0 to T + 22 h but then varied between ±0.4 mm afterward. When combining all of the 0000 and 1200 UTC forecasts in each year (not shown), GEM’s overall performance was very good, being slightly too moist (0.7 mm) in 2003 and marginally too dry in 2006 (0.2 mm).

Looking at GEM–AMR comparisons in more detail shows that the 0000 UTC GEM run in 2003 had a noticeable slight moist bias in the late evening and overnight periods of the first night, which then decreased during the day so that by late afternoon and through most of the following night this run showed no significant bias (Fig. 5). The small moist bias began to increase once again in the early morning hours of the following day. The 1200 UTC GEM in 2003 had a more consistent moist bias than the 0000 UTC run did; it had a moist bias of 0.5–1.0 mm from T + 0 through T + 12 h, but this decreased to a minimum in the midevening of the first day (from T + 13 through T + 15 h). In the late evening however, the moist bias increased again (to 1–1.5 mm) and persisted until T + 34 h (the afternoon of day 2).

Of interest is that in 2003, especially in the latter 18 h and to a lesser degree the first 18 h, the bias of the 0000 UTC GEM run is almost a mirror image of its 1200 UTC counterpart. This “mirror image” may be due to the same diurnal influences (errors in model parameterizations) acting on each model run, but 180° out of phase because of the 12-h time difference of GEM’s runs. For example, the larger moist bias in 2003 of the 0000 UTC run in the late evening and overnight hours may suggest that the model was ineffective in producing condensation and precipitation during the night. A similar bias also showed up in the 1200 UTC run from the late evening of day 1 through the early morning of day 2. In 2003, the 0000 UTC GEM run eliminated its moist bias in the late afternoon (T + 21 h) and showed a near-zero (±0.3 mm) bias until T + 35 h, almost reaching 1.0 mm again by T + 36 h. The 1200 UTC GEM run was never able to eliminate its moist bias, although it came close in the early evening of day 1 and in the late afternoon of day 2. These bias tendencies suggest that it is possible that GEM’s Fritsch–Chappell convective scheme in 2003 (e.g., Bélair et al. 2000) may have been too aggressive with surface-based convection during the afternoons and too conservative with elevated convection, which can dominate during the night.

For 2006, the 0000 UTC GEM was generally 0.2–0.5 mm too dry throughout most of the 36 h of the forecast (see Fig. 5). In contrast, the 2006 1200 UTC GEM started out very similar to 2003 in that its first 5 h of forecasts were too moist by 0.5–1.0 mm then began to dry out beyond this time, and by T + 15 h it had a dry bias that was similar to that of the 0000 UTC GEM. Note that the convective scheme in GEM was changed to Kain–Fritsch (e.g., Kain 2004) prior to 2006 and that it is not obvious (or known) what affect this may have had on the 2006 GEM PWV forecasts as compared with those in 2003.

In 2003, the overall GEM MAE was 2.7 mm and the RMSE was 3.6 mm; in 2006, the MAE was 2.8 mm and the RMSE was 3.8 mm (Table 1). The MAE (RMSE) generally increased from about 2.5 (3.3) mm at analysis time T to approximately 3.3 (4.5) mm by T + 33 h (Fig. 6). There were only slight differences between the 0000 and 1200 UTC model runs in each year (see Table 1).

Table 1.

GEM MAE and RMSE for all forecast hours combined (from T + 0 through T + 37) for 2003 and 2006. The “combined” row shows results when combining the 0000 and 1200 UTC model runs.

GEM MAE and RMSE for all forecast hours combined (from T + 0 through T + 37) for 2003 and 2006. The “combined” row shows results when combining the 0000 and 1200 UTC model runs.
GEM MAE and RMSE for all forecast hours combined (from T + 0 through T + 37) for 2003 and 2006. The “combined” row shows results when combining the 0000 and 1200 UTC model runs.
Fig. 6.

GEM MAE (mm) and RMSE (mm) for the growing seasons of 2003 and 2006 (0000 and 1200 UTC GEM runs combined).

Fig. 6.

GEM MAE (mm) and RMSE (mm) for the growing seasons of 2003 and 2006 (0000 and 1200 UTC GEM runs combined).

A comparison of 2006 with 2003 (Fig. 6) shows that the better GEM MAE scores were achieved earlier in 2006, at T + 6 h, as opposed to T + 14 h in 2003. However, the 2006 scores were generally lower during the time frame from T + 12 through T + 27 h. Similar results are found in RMSE scores; hence, only the MAE will be highlighted hereinafter. The shifting of the best scores to earlier times in 2006 may have been due to the implementation of the four-dimensional variational (4D-Var) data-assimilation system that is used by the GEM global model, from which the first-guess forecast for the GEM regional model is derived. As a result, it would be expected that the initial forecasts in 2006 should have improved relative to those of 2003, and this improvement should have been carried on throughout the forecast integration. Although we did observe this in the first 12 h of the forecast, it was not the case in the following 24 h. In fact, overall MAE scores were 4.7% lower in 2006. When GEM precipitation was factored out of the equation (see more in section 3c), the MAE in 2006 was only 0.7% lower than in 2003.

Figure 7 shows the single worst absolute error scores of 2003 and 2006 for each hour from T + 0 through T + 36 h. In 2003, The 0000 UTC run suffered from nine maximum absolute error scores over 15 mm during the summer, whereas the 1200 UTC run had only one. In 2006, both the 0000 and 1200 UTC GEM runs had eight maximum absolute error scores of more than 15 mm. In both years there is a tendency for the 1200 UTC GEM runs to experience poorer scores than the 0000 UTC runs during the first 21 h. After that time, however, the 0000 UTC run almost exclusively dominates the poorer score rankings. This may indicate that the 0000 UTC run has more trouble capturing nighttime (typically elevated) convection on day 2 than the 1200 UTC run has with the daytime (typically surface based) convection hours of day 2.

Fig. 7.

GEM maximum absolute errors (mm) for 2003 (gray line) and 2006 (black line). Solid circles or squares indicate that the worst error occurred during the 0000 UTC model run; open circles or squares indicate that the worst error was during the 1200 UTC model run.

Fig. 7.

GEM maximum absolute errors (mm) for 2003 (gray line) and 2006 (black line). Solid circles or squares indicate that the worst error occurred during the 0000 UTC model run; open circles or squares indicate that the worst error was during the 1200 UTC model run.

The 2003 GEM results presented here are similar to those of Smith et al. (2008) who showed a general increasing wet bias in GEM PWV from the T + 0 (+0.2 mm) to T + 9 (+0.9 mm) h time period during the summer of 2004 in southern Alberta using GPS-derived PWV. The same basic version of the GEM regional model would have been used for 2003 and 2004. The current study results are different than those of Deblonde et al. (2005) who showed a negative GEM bias but with an increasing wetness with prognosis time. It should be stressed that Deblonde et al. used the global version of GEM (coarser resolution of 100 km) and looked at different averaging periods over many different locations. Hence, it is not surprising that the results of Deblonde et al. are different from ours. For the current study, the nature of the GEM bias and other error statistics depends on the model run (0000 vs 1200 UTC) and year.

c. GEM versus AMR (model precipitation excluded)

For the assessments in section 3b, all PWV data (AMR and model) were excluded when the AMR was indicating precipitation. This was necessary because the AMR data are unreliable in precipitating conditions. GEM-versus-AMR PWV comparisons were still included, however, when GEM was forecasting precipitating conditions even when there was no observed precipitation. Thus, a perception may exist that GEM could be unfairly moist: a situation that perhaps could be largely responsible for its moist bias in 2003. To alleviate any such concerns, the data were reexamined excluding all hours for which GEM predicted precipitation. This resulted in the elimination of a further 15% of the observation–model comparisons in the 2003 data and 19% of the 2006 data.

As one may expect, there was a decrease in both observed and modeled PWV values—in this case, by <1 mm for both years. In 2003, the slight moist bias that GEM exhibited in the first 12 h was still there (Fig. 8). In fact, the bias varied little in the first 18 h. After that time however, there was an improvement, with the moist bias reduced by about 0.2 mm. The MAE and RMSE scores were also nearly identical in the first 18 h, but improvement was noted after that time (Figs. 9, 10). Eliminating model precipitation events improved the overall RMSE score by 3.1%, with most of the gains realized in the second 18 h of the forecast. In 2006, GEM that already had a very slight dry bias was marginally degraded (more of a dry bias) by about 0.1–0.2 mm in most of the forecast hours (Fig. 8). Error scores were nearly identical in the first 12 h; however, improvement was noted in the final 24 h for an overall RMSE improvement of 4.0% (Figs. 9, 10).

Fig. 8.

GEM bias (mm) for 2003 and 2006, with 0000 and 1200 UTC model runs combined. Results obtained when excluding GEM precipitation are compared with data that include GEM precipitation (used in section 3b).

Fig. 8.

GEM bias (mm) for 2003 and 2006, with 0000 and 1200 UTC model runs combined. Results obtained when excluding GEM precipitation are compared with data that include GEM precipitation (used in section 3b).

Fig. 9.

As in Fig. 8, but for MAE (mm).

Fig. 9.

As in Fig. 8, but for MAE (mm).

Fig. 10.

As in Fig. 8, but for RMSE (mm).

Fig. 10.

As in Fig. 8, but for RMSE (mm).

When model precipitation events are eliminated, it is perhaps not surprising to see a more substantial improvement in error scores with increasing forecast time since most of the model’s erroneous precipitation would be generated beyond the 12 h forecast and would have been removed from the dataset. For example, there were >70% more hours removed beyond the 12-h forecast period relative to the forecast period from T + 1 to T + 12 h.

4. Conclusions

Water vapor is a critical component in the atmosphere, and its spatial and temporal distributions have implications for cloud and precipitation production and for their forecasts. There have been very limited high-temporal-resolution measurements of water vapor because of the lack of techniques/instruments. Surface-based microwave or GPS instruments are useful tools to assess hourly variations in atmospheric water vapor to understand better the processes controlling its diurnal cycle and for validating numerical models. Their utility in model data assimilation schemes is also emerging, especially as the spatial network of these instruments grows and quality-control routines become automated.

This article focuses on high-temporal-resolution measurements of precipitable water vapor using a Radiometrics WVR-1100 microwave radiometer during the growing season (from mid-May to the end of August) in 2003 and 2006 over Winnipeg—a midcontinental setting within an agriculturally dominated prairie ecoclimatic region. The objectives of this article were 1) to document the diurnal cycle of PWV and to provide insight into the various processes controlling this cycle and 2) to assess the accuracy of the Canadian regional Global Environmental Multiscale model analysis and forecasts of PWV.

For objective 1, hourly AMR PWV averaged over the growing season revealed a diurnal cycle in 2003 and in 2006. The mean daily PWV was 22.6 mm in 2003 and 23.8 mm in 2006, with distinct diurnal amplitudes of 1.5 and 1.8 mm, respectively. In 2006, the mean daily PWV was 1.2 mm higher than in 2003, the maximum values occurred 2–3 h earlier, and the amplitude of the diurnal cycle was greater. It is not known why the two years were different other than the mean time of thunderstorm occurrence peak time of day was 2–3 h earlier in 2006 relative to normal and 2003; we also do not know why the mean thunderstorm occurrence peaked earlier in the day in 2006. When combining both years, the difference between the mean daily maximum and minimum PWV was 1.7 mm, or about 7%. The two years in the study were significantly dry in terms of precipitation, and it is unknown how different (if at all) these results would be for wetter and near-normal years—future work will aim to address this issue.

The PWV over a location for any hour is the result of 1) the previous hour’s PWV, 2) the horizontal moisture advection of water vapor into and out of the area, 3) the difference between the local vertical fluxes of evapotranspiration and precipitation, and 4) the conversion of water vapor into condensate (cloud and precipitation production). Because factor 2 (cycles of moisture advection) has periodicities of several days, factors 3 and 4 should dominate the diurnal signal in PWV when averaging the diurnal PWV over the long time periods used in this study (3 months). Thus, it was apparent that the diurnal cycle of the average hourly PWV values about the mean daily PWV value was controlled by ET and the occurrence/timing of local deep convection (i.e., thunderstorms). The PWV in both years was maximized late in the afternoon as opposed to at solar noon. This suggests that the surface and atmosphere were well coupled, with ET primarily being controlled by the vapor pressure deficit between the vegetation/surface and atmosphere—the underlying rootzone soil moisture, via the vegetation, being the primary moisture source. The decline in PWV in the evening-to-overnight periods is due to deep convection (precipitation and/or cloud production) and reduced ET contributions. PWV is reduced not only by precipitation but by cloud production (condensation); hence, the production of cloud in the evening and overnight periods can cause PWV to decrease, with or without precipitation.

For objective 2, comparison of AMR PWV with the GEM model indicted that the model performed well in some cases (with little or no bias) but can also be in error by 3 mm (13%) or more (as indicated by the RMSE), even in the first 12 h. These 3 mm can be crucial in accurately assessing atmospheric stability and rainfall potential; hence, the need for ongoing real-time observed 3D moisture measurements cannot be overstated. During the summer of 2003, GEM provided the best PWV forecasts from T + 13 to T + 17 h (MAE of about 2 mm and RMSE of 3 mm), after which time it slowly degraded (maximum MAE of 3–3.5 mm and RMSE of 4–4.5 mm). In the summer of 2006, GEM had its best performance much earlier, generally from T + 3 to T + 9 h (slightly lower MAE and RMSE than for 2003), but by T + 12 h and for most of the remaining hours out to T + 36 h errors scores were larger than for 2003. It is not known why the 2006 version of GEM performed better in the shorter time frame in comparison with the 2003 version; however, one major change between the two years was the addition of the 4D-Var system in the global GEM. It is not known if this change had any effect on the results shown here.

The authors are advocates for a small network of profiling radiometers strategically located to supplement the sparse Canadian upper-air observing network—in particular, in the southern prairies, where the greatest population resides. This network would greatly enhance Canada’s capacity for high-impact weather forecasting through improved monitoring of atmospheric thermodynamics as well as climatological analyses such as this study. A viability experiment of this network for improved operational model analyses and forecasts through data-assimilation experiments would also be useful for possible future operational long-term implementation.

Acknowledgments

This work was jointly funded through the Drought Research Initiative (DRI), which is a Canadian Foundation for Climate and Atmospheric Sciences (CFCAS) Network, and the Natural Sciences and Engineering Research Council (NSERC) Discovery Grant, both to JH. We thank David Baggaley of the Prairie and Arctic Storm Prediction Centre (Winnipeg) and Philip Harder (DRI data manager) of the University of Manitoba for data analysis assistance.

REFERENCES

REFERENCES
Andreas
,
E. L.
,
P. S.
Guest
,
P. O. G.
Persson
,
C. W.
Fairall
,
T. W.
Horst
,
R. E.
Moritz
, and
S. R.
Semmer
,
2002
:
Near-surface water vapor over polar sea ice is always near ice saturation.
J. Geophys. Res.
,
107
,
8033
.
doi:10.1029/2000JC000411
.
Baldocchi
,
D.
,
1994
:
A comparative study of mass and energy exchange over a closed C3 (wheat) and an open C4 (corn) canopy: I. The partitioning of available energy into latent and sensible heat exchange.
Agric. For. Meteor.
,
67
,
191
220
.
Bélair
,
S.
,
A.
Méthot
,
J.
Mailhot
,
B.
Bilodeau
,
A.
Patoine
,
G.
Pellerin
, and
J.
Côté
,
2000
:
Operational implementation of the Fritsch–Chappell convective scheme in the 24-km Canadian Regional Model.
Wea. Forecasting
,
15
,
257
274
.
Champollion
,
C.
,
F.
Masson
,
J.
Van Baelen
,
A.
Walpersdorf
,
J.
Chéry
, and
E.
Doerflinger
,
2004
:
GPS monitoring of the tropospheric water vapor distribution and variation during the 9 September 2002 torrential precipitation episode in the Cévennes (southern France).
J. Geophys. Res.
,
109
,
D24102
.
doi:10.1029/2004JD004897
.
Côté
,
J.
,
S.
Gravel
,
A.
Méthot
,
A.
Patoine
,
M.
Roch
, and
A.
Staniforth
,
1998
:
The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation.
Mon. Wea. Rev.
,
126
,
1373
1395
.
Dai
,
A.
,
F.
Giorgi
, and
K. E.
Trenberth
,
1999
:
Observed and model-simulated precipitation diurnal cycle over the contiguous United States.
J. Geophys. Res.
,
104
,
6377
6402
.
Dai
,
A.
,
J.
Wang
,
R. H.
Ware
, and
T.
Van Hove
,
2002
:
Diurnal variation in water vapor over North America and its implications for sampling errors in radiosonde humidity.
J. Geophys. Res.
,
107
,
4090
.
doi:10.1029/2001JD000642
.
Deblonde
,
G.
,
S.
Macpherson
,
Y.
Mireault
, and
P.
Heroux
,
2005
:
Evaluation of GPS precipitable water over Canada and the IGS network.
J. Appl. Meteor.
,
44
,
153
166
.
Fisico
,
T.
,
2008
:
Meteorological characteristics of the Canadian Arctic Shelf Exchange Study. M.S. thesis, Dept. of Environment and Geography, University of Manitoba, 129 pp. [Available from Dept. of Environment and Geography, University of Manitoba, Winnipeg, MB R3T 2N2, Canada]
.
Foster
,
J.
,
M.
Bevis
,
Y-L.
Chen
,
S.
Businger
, and
Y.
Zhang
,
2003
:
The Ka‘ū storm (November 2000): Imaging precipitable water using GPS.
J. Geophys. Res.
,
108
,
4585
.
doi:10.1029/2003JD003413
.
Guldner
,
J.
, and
D.
Spankuch
,
2001
:
Remote sensing of the thermodynamic state of the atmospheric boundary layer by ground-based microwave radiometry.
J. Atmos. Oceanic Technol.
,
18
,
925
933
.
Gutman
,
S. I.
, and
S. G.
Benjamin
,
2001
:
The role of ground-based GPS meteorological observations in numerical weather prediction.
GPS Solutions
,
4
,
16
24
.
Han
,
Y.
, and
E.
Westwater
,
1995
:
Remote sensing of tropospheric water vapor and cloud liquid water by integrated ground-based sensors.
J. Atmos. Oceanic Technol.
,
12
,
1050
1059
.
Hanan
,
N. P.
,
J. A.
Berry
,
S. B.
Verma
,
E. A.
Walter-Shea
,
A. E.
Suyker
,
G. G.
Burba
, and
A. S.
Denning
,
2005
:
Testing a model of CO2, water and energy exchange in Great Plains tallgrass prairie and wheat ecosystems.
Agric. For. Meteor.
,
131
,
162
179
.
Hogg
,
D. C.
,
F. O.
Guiraud
,
J. B.
Snider
,
M. T.
Decker
, and
E. R.
Westwater
,
1983
:
A steerable dual-channel microwave radiometer for measurement of water vapor and liquid in the troposphere.
J. Climate Appl. Meteor.
,
22
,
789
806
.
Jarlemark
,
P.
, and
G.
Elgered
,
2003
:
Retrieval of atmospheric water vapour using a ground-based single-channel microwave radiometer.
Int. J. Remote Sens.
,
24
,
3821
3837
.
Jin
,
S.
, and
O. F.
Luo
,
2009
:
Variability and climatology of PWV From global 13-year GPS observations.
IEEE Trans. Geosci. Remote Sens.
,
47
,
1918
.
doi:10.1109/TGRS.2008.2010401
.
Kain
,
J. S.
,
2004
:
The Kain–Fritsch convective parameterization: An update.
J. Appl. Meteor.
,
43
,
170
181
.
Knupp
,
K.
,
R.
Ware
,
D.
Cimini
,
F.
Vandenberghe
,
J.
Vivekanandan
,
E.
Westwater
, and
T.
Coleman
,
2008
:
Ground-based passive microwave profiling during dynamic weather conditions.
J. Atmos. Oceanic Technol.
,
26
,
1057
1073
.
Lesht
,
B. M.
, and
J. C.
Liljegren
,
1997
:
Comparison of precipitable water vapor measurements obtained by microwave radiometers and radiosondes at the Southern Great Plains Cloud and Radiation Testbed site.
Proc. Sixth Atmospheric Radiation Measurement (ARM) Science Team Meeting, San Antonio, TX, U.S. Dept. of Energy, CONF-9603149, 59–62
.
Liljegren
,
J. C.
,
1994
:
Two-channel microwave radiometer for observation of total column precipitable water vapor and cloud liquid water path.
Preprints, Fifth Symp. on Global Change Studies, Nashville, TN, Amer. Meteor. Soc., 262–269
.
Naito
,
I.
,
Y.
Hatanaka
,
R.
Ichikawa
,
S.
Shimada
,
T.
Yabuki
,
H.
Tsuji
, and
T.
Tanaka
,
1998
:
Global positioning system project to improve Japanese weather, earthquake predictions.
Eos, Trans. Amer. Geophys. Union
,
79
,
301
.
Niell
,
A. E.
,
A. J.
Coster
,
F. S.
Solheim
,
V. B.
Mendes
,
P. C.
Toor
,
R. B.
Langley
, and
C. A.
Upham
,
2001
:
Comparison of measurements of atmospheric wet delay by radiosonde, water vapor radiometer, GPS, and VLBI.
J. Atmos. Oceanic Technol.
,
18
,
830
850
.
Oke
,
T. R.
,
1987
:
Boundary Layer Climates.
2nd ed. Methuen, 435 pp
.
Oke
,
T. R.
,
1997
:
Surface climate processes.
Surface Climates of Canada, W. G. Bailey, T. Oke, and W. R. Rouse, Eds., McGill–Queens University Press, 202–204
.
Raddatz
,
R. L.
,
1993
:
Prairie Agroclimate Boundary-Layer Model: A simulation of the atmosphere/crop-soil interface.
Atmos.–Ocean
,
31
,
399
419
.
Raddatz
,
R. L.
,
2005
:
Moisture recycling on the Canadian prairies for summer droughts and pluvials from 1997 to 2003.
Agric. For. Meteor.
,
131
,
13
26
.
Raddatz
,
R. L.
, and
J. M.
Hanesiak
,
2008
:
Significant summer rainfall in the Canadian prairie provinces: Modes and mechanisms 2000–2004.
Int. J. Climatol.
,
28
,
1607
1613
.
Smith
,
C. D.
,
N.
Nicholson
,
S.
Skone
, and
G. S.
Strong
,
2008
:
Evaluating regional atmospheric water vapour estimates derived from GPS and short-range forecasts of the Canadian Global Environmental Multiscale model in southern Alberta.
Atmos.–Ocean
,
46
,
455
471
.
Strong
,
G. S.
,
1997
:
Atmospheric moisture budget estimates of regional evapotranspiration from RES-91.
Atmos.–Ocean
,
35
,
29
63
.
Trenberth
,
K. E.
,
1999
:
Atmospheric moisture recycling: Role of advection and local evaporation.
J. Climate
,
12
,
1368
1381
.
Vedel
,
H.
, and
X-Y.
Huang
,
2004
:
Impact of ground based GPS data on numerical weather prediction.
J. Meteor. Soc. Japan
,
82
,
459
472
.
Wang
,
J.
, and
L.
Zhang
,
2009
:
Climate applications of a global, 2-hourly atmospheric precipitable water dataset derived from IGS tropospheric products.
J. Geod.
,
83
,
209
217
.
Wang
,
J.
,
A.
Dai
,
D. J.
Carlson
,
R. H.
Ware
, and
J. C.
Liljegren
,
2002
:
Diurnal variation in water vapor and liquid water profiles from a new microwave radiometer profiler.
Preprints, Sixth Symp. on Integrated Observing Systems, Orlando, FL, Amer. Meteor. Soc., 5.5. [Available online at http://ams.confex.com/ams/pdfpapers/27917.pdf]
.
Wang
,
J.
,
L.
Zhang
,
A.
Dai
,
T.
Van Hove
, and
J.
Van Baelen
,
2007
:
A near-global, 2-hourly data set of atmospheric precipitable water from ground-based GPS measurements.
J. Geophys. Res.
,
112
,
D11107
.
doi:10.1029/2006JD007529
.
Westwater
,
E.
,
Y.
Han
,
M.
Shupe
, and
S.
Matrosov
,
2001
:
Analysis of integrated cloud liquid and precipitable water vapor retrievals from microwave radiometers during the Surface Heat Budget of the Arctic Ocean project.
J. Geophys. Res.
,
106
,
32019
32030
.
Wever
,
L. A.
,
L. B.
Flanagan
, and
P. J.
Carlson
,
2002
:
Seasonal and interannual variation in evapotranspiration, energy balance and surface conductance in a northern temperate grassland.
Agric. For. Meteor.
,
112
,
31
49
.

Footnotes

Corresponding author address: John Hanesiak, Dept. of Environment and Geography Centre for Earth Observation Science, Rm. 440, Wallace Bldg., 125 Dysart Rd., University of Manitoba, Winnipeg, MB R3T 2N2, Canada. Email: john_hanesiak@umanitoba.ca