• Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge measurements. J. Atmos. Oceanic Technol., 20 , 752759.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ciach, G. J., , and Krajewski W. F. , 1999: On the estimation of radar rainfall error variance. Adv. Water Resour., 22 , 585595.

  • Ciach, G. J., , Krajewski W. F. , , and Villarini G. , 2007: Product-error-driven uncertainty model for probabilistic quantitative precipitation estimation with NEXRAD data. J. Hydrometeor., 8 , 13251347.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Duchon, C. E., , and Essenberg G. R. , 2001: Comparative rainfall observations from pit and aboveground rain gauges with and without wind shields. Water Resour. Res., 43 , 32533263.

    • Search Google Scholar
    • Export Citation
  • Gourley, J. J., , Kaney B. , , and Maddox R. A. , 2003: Evaluating the calibrations of radars: A software approach. Preprints, 31st Int. Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., P3C.1. [Available online at http://ams.confex.com/ams/pdfpapers/64171.pdf].

    • Search Google Scholar
    • Export Citation
  • Humphey, M., , Istok D. J. D. , , Lee J. Y. , , Hevesi J. A. , , and Flint A. L. , 1997: A new method for automated dynamic calibration of tipping bucket rain gauges. J. Atmos. Oceanic Technol., 14 , 15131519.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, D., , Nelson B. , , and Cedrone L. , 2006: Reprocessing of historic Hydrometeorological Automated Data System (HADS) precipitation data. Preprints, 10th Symp. on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans, and Land Surface, Atlanta, GA, Amer. Meteor. Soc., 8.2. [Available online at http://ams.confex.com/ams/pdfpapers/100680.pdf].

    • Search Google Scholar
    • Export Citation
  • Kim, D., , Tollerud E. I. , , Vasiloff S. V. , , and Caldwell J. , 2009: Comparison of manual and automated quality control of operational hourly precipitation data of the National Weather Service. Preprints, 23rd Conf. on Hydrology, Phoenix, AZ, Amer. Meteor. Soc., J6.3. [Available online at http://ams.confex.com/ams/pdfpapers/146869.pdf].

    • Search Google Scholar
    • Export Citation
  • Maddox, R. A., 1983: Large-scale meteorological conditions associated with midlatitude, mesoscale convective complexes. Mon. Wea. Rev., 111 , 14751493.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Medlin, J. M., , Kimball S. K. , , and Blackwell K. G. , 2007: Radar and rain gauge analysis of the extreme rainfall during Hurricane Danny’s (1997) landfall. Mon. Wea. Rev., 135 , 18691888.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seo, D-J., 1998: Real-time estimation of rainfall fields using radar rainfall and rain gage data. J. Hydrol., 208 , 3752.

  • Seo, D-J., , and Breidenbach J. P. , 2002: Real-time estimation of spatially non-uniform bias in radar rainfall data using gauge measurements. J. Hydrometeor., 3 , 93111.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sieck, L. C., , Burges S. J. , , and Steiner M. , 2007: Challenges in obtaining reliable measurements of point rainfall. Water Resour. Res., 43 , W01420. doi:10.1029/2005WR004519.

    • Search Google Scholar
    • Export Citation
  • Steiner, M., , Smith J. A. , , Burges S. J. , , Alonso C. V. , , and Darden R. W. , 1999: Effect of bias adjustment and rain gauge data quality control on radar rainfall estimation. Water Resour. Res., 35 , 24872503.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tollerud, E., , Collander R. , , Lin Y. , , and Loughe A. , 2005: On the performance, impact, and liabilities of automated precipitation gauge screening algorithms. Preprints, 21st Conf. on Weather Analysis and Forecasting/17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., P1.42. [Available online at http://ams.confex.com/ams/pdfpapers/95173.pdf].

    • Search Google Scholar
    • Export Citation
  • Vasiloff, S. V., and Coauthors, 2007: Improving QPE and very short term QPF: An initiative for a community-wide integrated approach. Bull. Amer. Meteor. Soc., 88 , 18991911.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, J., , and Brandes E. A. , 1979: Measurement of rainfall—A summary. Bull. Amer. Meteor. Soc., 60 , 10481058.

  • Xu, X., , Howard K. , , and Zhang J. , 2008: An automated radar technique for the identification of tropical precipitation. J. Hydrometeor., 9 , 885902.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhang, J., , Howard K. , , and Gourley J. J. , 2005: Constructing three-dimensional multiple-radar reflectivity mosaics: Examples of convective storms and stratiform rain echoes. J. Atmos. Oceanic Technol., 22 , 3042.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Map of MT with the white rectangle indicating the area of focus. WSR-88D locations with 250-km range rings are shown. Also shown are HADS (H) and ASOS (A) gauge locations.

  • View in gallery

    Series of composite reflectivity images showing the storm episodes at (a) 1800 UTC 16 Jun, (b) 2200 UTC 16 Jun, (c) 0200 UTC 17 Jun, (d) 0600 UTC 17 Jun, and (e) 1000 UTC 17 Jun 2007.

  • View in gallery

    Preliminary storm reports from the SPC for 16–17 Jun 2007. Triangles denote hail >2 in. diameter and squares denote winds >65 kt. Rain gauge locations are indicated by the letter G. The location of the KGGW radar is also indicated.

  • View in gallery

    (a) The 24-h Q2RAD precipitation accumulations focused on the supercell path with gauge locations in Fig. 4b overlaid; (b) gauge circle map comparing 24-h Q2RAD and gauge data where circle size denotes gauge amount and circle color represents the radar–gauge ratio. Warm colors indicate radar values less than gauge values; cool colors indicate radar estimates greater than gauge values.

  • View in gallery

    Radar–gauge ratios for each storm period and 24-h totals ending 1200 UTC 17 Jun 2007.

  • View in gallery

    Time series of 5-min hybrid scan reflectivity, 1-h Q2RAD, and 1-h gauge amounts at (a) HWSM8, (b) MSWM8, (c) SACM8, (d) GGW, (e) GWSM8, (f) NSHM8, (g) OLF, (h) WPTM8, (i) CLBM8, and (j) JDN. The letter M indicates missed data transmission for the hour.

  • View in gallery

    Series of composite reflectivities from KGGW showing the passage of the supercell over the radar. (a) Composite reflectivity (maximum in column) at 2330 UTC 16 Jun 2007. (b) HSR at 2330 UTC 16 Jun 2007. (c) As in (a) but for 2355 UTC 16 Jun 2007. (d) As in (b) but for 2355 16 Jun 2007. (e) As in (a) but for 0020 UTC 17 Jun 2007. (f) As in (b) but for 0020 UTC 17 Jun 2007. Select gauge IDs are overlaid.

  • View in gallery

    (a) Regions between MT radars for which reflectivity differences are computed for the Radar Reflectivity Comparison Tool used by the WSR-88D Radar Operations Center. Each bin is 20 km × 20 km × 120 km. (b) Average reflectivity differences for the comparison region between KGGW and KTFX shown in Fig. 8a from 2200 UTC 16 Jun to 0200 UTC 17 Jun 2007. The red line indicates that KGGW is overall 2.5 dBZ higher than KTFX.

  • View in gallery

    Image from the QVS showing locations of the Maricopa County, AZ; Oklahoma Mesonet; and Lower Colorado River Authority (TX) gauge networks. (Information courtesy of the NMQ Web site: http://nmq.ou.edu/).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 163 163 10
PDF Downloads 140 140 6

Difficulties with Correcting Radar Rainfall Estimates Based on Rain Gauge Data: A Case Study of Severe Weather in Montana on 16–17 June 2007

View More View Less
  • 1 NOAA/National Severe Storms Laboratory, Norman, Oklahoma
© Get Permissions
Full access

Abstract

The principal source of information for operational flash flood monitoring and warning issuance is weather radar–based quantitative estimates of precipitation. Rain gauges are considered truth for the purposes of validating and calibrating real-time radar-derived precipitation data, both in a real-time sense and climatologically. This paper examines various uncertainties and challenges involved with using radar and rain gauge data in a severe local storm environment. A series of severe thunderstorm systems that occurred across northeastern Montana illustrates various problems with comparing radar precipitation estimates and real-time gauge data, including extreme wind effects, hail, missing gauge data, and radar quality control. Ten radar–gauge time series pairs were analyzed with most found to be not useful for real-time radar calibration. These issues must be carefully considered within the context of ongoing efforts to develop robust real-time tools for evaluating radar–gauge uncertainties. Recommendations are made for radar and gauge data quality control efforts that would benefit the operational use of gauge data.

Corresponding author address: Steven V. Vasiloff, NSSL, 120 David Boren Blvd., Norman, OK 73072. Email: steven.vasiloff@noaa.gov

Abstract

The principal source of information for operational flash flood monitoring and warning issuance is weather radar–based quantitative estimates of precipitation. Rain gauges are considered truth for the purposes of validating and calibrating real-time radar-derived precipitation data, both in a real-time sense and climatologically. This paper examines various uncertainties and challenges involved with using radar and rain gauge data in a severe local storm environment. A series of severe thunderstorm systems that occurred across northeastern Montana illustrates various problems with comparing radar precipitation estimates and real-time gauge data, including extreme wind effects, hail, missing gauge data, and radar quality control. Ten radar–gauge time series pairs were analyzed with most found to be not useful for real-time radar calibration. These issues must be carefully considered within the context of ongoing efforts to develop robust real-time tools for evaluating radar–gauge uncertainties. Recommendations are made for radar and gauge data quality control efforts that would benefit the operational use of gauge data.

Corresponding author address: Steven V. Vasiloff, NSSL, 120 David Boren Blvd., Norman, OK 73072. Email: steven.vasiloff@noaa.gov

1. Introduction

Warnings for short-term flash floods and longer-term river floods are derived principally from National Weather Service (NWS) Weather Surveillance Radar-1988 Doppler (WSR-88D) quantitative precipitation estimates (QPEs) and, to a lesser extent, rain gauge observations. The advantage of radar QPEs is their nearly continuous temporal and spatial coverage, especially in areas void of complex terrain. Rain gauge observations are considered ground truth for radar QPEs and are used routinely in the multisensor precipitation estimator (MPE; Seo and Breidenbach 2002) in both NWS Weather Forecast Offices (WFOs) and River Forecast Centers (RFCs) for bias adjustments. In the NWS Western Region, daily precipitation analyses consist of almost entirely gauge data. A key source of rain gauge data used in NWS operational applications is the Hydrometeorological Automated Data System [HADS; Kim et al. (2006); information online at http://www.nws.noaa.gov/oh/hads/], a real-time data acquisition and distribution system operated by the National Oceanic and Atmospheric Administration’s (NOAA) Office of Hydrologic Development (OHD). The HADS data originate from a variety of NWS and non-NWS gauge networks. Also used are gauge data from the Automated Surface Observing Systems (ASOS) program (information online at http://www.nws.noaa.gov/asos/), a complete weather observing system usually placed at airports. Both ASOS and HADS gauge data are collected by the National Centers for Environmental Protection (NCEP). The National Severe Storms Laboratory (NSSL) then obtains these gauges via ftp for inclusion in the National Mosaic and QPE system (NMQ [information online at http://www.nssl.noaa.gov/projects/q2/; Vasiloff et al. (2007)] with an average number of gauges per hour of ∼2280. WFOs and RFCs also utilize gauges from additional networks not in the HADS data stream [e.g., the Community Collaborative Rain, Hail and Snow Network (CoCoRaHS; infomation online at http://cocorahs.org/) and the Oklahoma Mesonet (information online at http://www.mesonet.org/)]. For the purposes of this study, however, our discussion is limited to HADS and ASOS gauges.

Radar and rain gauge data are susceptible to a variety of errors that have been under investigation for many years. Types of uncertainties range from basic gauge and radar mechanical functionality, to spatial and temporal differences, to microphysical effects. Kim et al. (2006) described various timing and transmission errors with HADS data. For instance, some gauges will miss a transmission but otherwise appear to function normally. Other times gauges can be clogged with water trickling in over time. (Melting ice/snow will also trickle in, although a treatment of winter weather is beyond the scope of this paper.) Ciach (2003) described ∼5% error due to random mechanical errors. Duchon and Essenberg (2001) reported a 4%–5% undercatch for nonshielded tipping-bucket gauges for southern plains storms. Meteorological effects can significantly degrade the confidence in gauge measurements. For instance, strong winds can reduce gauge collection efficiency by over 50% (Sieck et al. 2007). Medlin et al. (2007) found large gauge underestimates during Hurricane Danny and made an attempt at wind corrections. In addition to wind effects, the undercatch can be as much as 30% due to heavy rain (Humphey et al. 1997). There are also numerous challenges in relating volumetric radar data to point gauge data, including vertical storm structure and wind shear effects (e.g., Wilson and Brandes 1979). Seo (1998) used an optimal estimation method to account for temporal and spatial inconsistencies when combining the two data sources.

To effectively calibrate radar QPEs, potential errors must be carefully investigated. For instance, Steiner et al. (1999) showed how gauge false zero reports reduce the accuracy of bias adjustments. For 30 storms over the Goodland Watershed in Mississippi, they found that 80% of the storms had 70% of the gauges working properly. Only 4 out of 30 storms had all the gauges functioning.

This paper presents a case study of rain gauge and radar observations during a series of severe storm events that occurred in Montana on 16–17 June 2007. The area across northern Montana has few gauges, which presents a significant operational challenge for determining accurate radar rainfall rates. Furthermore, the effects of severe wind and hail reduce the confidence in both data sources. Heretofore, little has been documented in the literature on dealing with gauge performance in extreme local storm conditions. The NSSL NMQ system, which includes next-generation QPE algorithms (Q2), is used to demonstrate concepts and potential solutions. Finally, recommendations are made for various tools and methods that would improve the real-time assessment of gauge uncertainties. These include applications being developed collaboratively among NSSL, the Earth System Research Laboratory (ESRL), the National Climatic and Data Center (NCDC), and the Lower Mississippi River Forecast Center (LMRFC).

2. Storm overview

Three distinct convective systems traversed Montana on 16–17 June 2007. An examination of upper-air soundings (not shown) depicted an environment conducive to supercell storm development for which the NWS Storm Prediction Center issued a severe thunderstorm watch. While a detailed assessment of the mesoscale environment and subsequent storm evolution in the northern plains on this day is beyond the scope of the current paper, the storms of interest developed within a synoptic-scale zone of upper divergence and deep vertical motion attending the eastward-progressing short wave.

The study area is in northeast Montana where the few gauges that exist are along the Missouri River (Fig. 1). The city of Glasgow and the nearby KGGW WSR-88D are situated in the eastern portion of the Missouri River valley that extends into North Dakota. Figure 2 shows a series of composite reflectivity images at 4-h intervals from 1800 UTC 16 June 2007 to 1200 UTC 17 June. The initial convection occurred during the early afternoon as an isolated supercell that propagated east-southeastward along the Missouri River basin before dissipating along the Montana–North Dakota border. Note the large core of reflectivity greater than 70 dBZ, indicating the presence of large hail. The storm damaged nearly all the building and residential roofs in Glasgow, which is just a few miles from the radar site KGGW (B. Martin 2008, personal communication). Six flash flood warnings were issued after this storm passed to the east of Glasgow. Verification data are not available, so it is not known if there were any missed events. The second event began farther west as an area of convective storms in conjunction with elevated heating. These storms subsequently evolved into a mesoscale convective complex (MCC; Maddox 1983) (see Figs. 2c and 2d) that traversed the state before dissipating in northeastern North Dakota. Following in the wake of the MCC, a third wave of intense storms forming a mesoscale convective system (MCS) developed along a line southwest to northeastern across central Montana (Figs. 2d and 2e). These storms also moved east-northeast and persisted into North Dakota during the morning of 17 June. The locations of severe storm reports received at the NOAA/Storm Prediction Center are shown in Fig. 3. A swath of severe hail and wind reports across northern Montana coincide with the supercell and, to a lesser extent, the MCC that produced five of the severe wind reports. The proximity of the reports to gauges is used to infer potential gauge uncertainties.

Precipitation accumulations for the 24-h period ending 1200 UTC 17 June are shown in Fig. 4a for the NSSL Q2 radar-based analysis (i.e., Q2RAD; Vasiloff et al. 2007). The Q2RAD product is derived from hybrid scan reflectivity (HSR) data at 1-km resolution using dynamic reflectivity-to-rain rate relations (ZR; Zhang et al. 2005; Xu et al. 2008). A variety of precipitation-type flags (e.g., convective and stratiform) are automatically assigned and corresponding ZR relations are applied to create precipitation rates that can then be accumulated. Note the sliver of missing data ESE of KGGW due to beam blockage. This and other radar artifacts will be discussed in more detail in the following section. Corresponding gauge data are shown in Fig. 4b. The circle sizes in Fig. 4b correspond to gauge amounts and the circle colors represent the radar to gauge ratio (RGR) values, with warm colors indicating Q2RAD underestimation and cool colors indicating Q2RAD overestimation.

3. Individual radar–gauge comparisons

The heaviest precipitation was associated with the supercell followed by the MCC. Figure 5 shows RGRs for each gauge and storm period, as well as 24-h totals. The light-blue bar represents 24-h totals and corresponds to the circle RGR. There are large fluctuations in the RGRs for the different storm episodes and the 24-h RGR is a poor indicator of shorter-term event biases or uncertainties. For instance, the Malta, Montana, Great Plains Cooperative Agricultural Weather Network (AgriMet) weather station (MWSM8) gauge has a 24-h RGR of 0.83 while the individual episodes range from 0.7 to 2.0. The Wolf Point, Montana, ASOS (OLF) gauge has a 24-h RGR close to 1 but much different at shorter time scales. Missing bars are due to either the gauge or radar being zero or missing.

The Harlem, Montana, AgriMet weather station (HWSM8) gauge (Fig. 6a) was on the edge of the supercell’s path and at a relatively long range from the KGGW radar. Accumulations during this first episode were minor and the overall reflectivity values were moderate (mostly <40 dBZ), with an RGR of 1. During the MCC the gauge measured 10 mm while Q2RAD was nearly double (RGR = 2.25). The zero gauge value at 0100 UTC is perplexing since reflectivity values between 0100 and 0200 UTC were higher than those during the following hour yet the gauge accumulation was less. As usual, several factors could contribute to the discrepancies between the radar and gauge. Vertical reflectivity gradients could result in over/underestimates where the elevated beam samples different parts/intensities of the storm. It is also plausible that the gauge became clogged. In addition, large spatial variability could account for the differences since the gauge is only a point measurement while the radar is a volumetric estimate having variable dimensions [nominally 1° vertically and azimuthally and 1 km in range; see, e.g., Ciach and Krajewski (1999)].

For the 24-h total, the MWSM8 gauge report (Fig. 6b) was similar to the radar QPE amount with an RGR of 0.88. This gauge also matched the radar closely for each storm episode with some underestimating during the MCC (RGR = 0.7). Again, note that the hour-by-hour (and storm by storm) differences offset each other, resulting in a bias that is close to 1 for the 24-h period. The algorithmic detection of hail invoked a reflectivity cap of 49 dBZ to prevent overestimation. In this situation, the cap may have contributed to radar underestimation. Strong wind was also likely and may have resulted in gauge undercatch. Rainfall amounts were low during the MCS episode with a slight Q2RAD overestimate.

The time series for the AgriMet station near Saco, Montana (SACM8), is shown in Fig. 6c. Overall, the SACM8 gauge indicated Q2RAD overestimation for the 24-h total. However, like MWSM8, SACM8 was impacted by wind and hail during the supercell, likely causing gauge undercatch, with an RGR of 1.89. With the MCC passage the RGR was initially 3.12 due to missing gauge data for 0300 and 0400 UTC. The missing data were the result of a data transmission error of the type that Kim et al. (2006) described. Data results retrieved from an archived database for 0300 and 0400 UTC were 3.3 and 0.75 mm, respectively, and led to a corrected RGR of 1.52, shown in Fig. 5. Note that the circle plot uses the uncorrected RGR. A third brief storm episode had a closer match between the radar and gauge estimates with a bias of 0.68.

The Glasgow, Montana (GGW), gauge, located just a few kilometers from the KGGW radar, measured 31.5 mm during the passage of the supercell followed by 40.1 mm associated with the MCC (Fig. 6d). The radar estimates were missing during both episodes due to the “cone of silence” (COS) and missing data extremely close to the radar that is not always filled in by surrounding radars. The data hole over KGGW can be seen during the passage of the supercell over the radar as shown in Fig. 7.

Although very close to the GGW gauge, the AgriMet station near Glasgow, Montana (GWSM8), was not in the COS. The overall RGR was 0.47, indicating underestimation. For the supercell, the gauge recorded 34.3 mm while Q2RAD was only 12.2 mm (Fig. 6e). Initially, the gauge reported 12.6 mm at 0000 UTC while the Q2RAD estimate was zero. As seen in Fig. 7, at 2355 UTC the storm core was only a few kilometers to the west of the GWSM8 gauge. With a horizontal translation speed of nearly 60 km h−1 (1 km min−1), the storm effectively “jumped over” the Q2RAD observation time of 0000 UTC. Assuming a rain rate of 51 mm h−1, this would mean a loss of ∼5 mm of rain from the analysis. The discretization of radar analyses can often be seen as wave patterns in QPE fields associated with fast-moving storms. Also contributing to Q2 underestimation was the inadvertent removal of HSR data by the QC program for two analysis time intervals (not shown). This resulted in another 7.6 mm of missing Q2RAD precipitation. Both factors add up to 12.6 mm, accounting for the difference between Q2RAD and the gauge. The actual precipitation is unknown since the gauge likely underestimated due to strong winds and hail. The Q2RAD estimate was nearly one-half that of the gauge for the MCC, most likely due to a combination of effects noted for the supercell. No rain was observed for the third episode.

In addition to the COS, timing, and radar QC problems, it appears that, while not often associated with the WSR-88D system, attenuation appears to have reduced the Q2RAD values. As seen in Fig. 8b, a decrease in reflectivity values occurred during the passage of the supercell over the KGGW radar site. When the supercell was to the west of KGGW, (Fig. 7), the maximum reflectivity exceeded 70 dBZ. There were almost no reflectivity values greater than 60 dBZ when the core was directly over the radar, and reflectivity values increased as the storm moved away from the radar. A radar reflectivity comparison tool developed at NSSL for the WSR-88D operations center compares reflectivity values in collocated bins of similar size (Gourley et al. 2003). The bins in Montana are shown in Fig. 8a. Within each bin, individual beam volumes must be within certain spatial and temporal restrictions in order to be compared. The numbers of reflectivity pairs average in the thousands. A comparison between KGGW and the Great Falls, Montana radar (KTFX) during the supercell’s passage over KGGW is shown in Fig. 8b. Between 2330 and 2345 UTC there is a ∼9 dB reduction of KGGW reflectivity values as compared to KTFX, followed by a ∼13 dB increase as the storm core with large hail and heavy rain passed to the east of KGGW. Smaller fluctuations may be the result of nonstandard beam propagation paths and the position of the hail core between the radar and echoes in the comparison volume. A 9-dB decrease can have a dramatic effect on rain rates at the upper end of the reflectivity scale. For example, for an attenuation of 10 dB and a convective reflectivity rain-rate relation of Z = 300R1.4, the rainfall rate of 63 mm−1 at 50 dBZ is reduced to 12 mm h−1 at 40 dBZ. Assuming a 15-min duration, this results in an underestimate by the radar of nearly 13 mm. As indicated by the red line in Fig. 8b, KGGW is on average about 2.5 dB higher than KTFX. It is not known if there were calibration differences, an additional contribution to radar uncertainties.

The largest difference between a gauge observation and Q2RAD (given that both were nonzero) was for the AgriMet station gauge near Nashua, Montana (NSHM8). The time series (Fig. 6f) shows that, during the supercell, there was 60+ dBZ over the gauge, resulting in a 1-h Q2RAD amount of 34 mm while the gauge reported only 2.8 mm. Strong horizontal reflectivity gradients may have been responsible for instances of disparate radar and gauge pairs since the gauge is a point observation within the 1 km × 1 km radar pixel. Vertical wind shear along with strong winds and hail at the gauge may have also contributed to the severe mismatch. Further, the radar estimates were likely impacted due to the presence of hail although the extent is difficult to determine. Wind and/or hail effects also likely caused the undercatch during the MCC period at 0500 UTC. Note that the gauge reported no rain during the MCS passage, which leads one to suspect, considering earlier radar–gauge discrepancies, that the gauge may have been having problems.

OLF reported 29.5 mm compared to 26.7 mm for the Q2RAD estimate and an RGR of 0.91. However, the time series (Fig. 6g) shows an initial radar overestimate bias of 2.24 and a later underestimate bias of 0.63, nearly countering each other and resulting in a misleading 24-h RGR value. The gauge was very close to a sector of beam blockage but appears not to have been affected. The radar and gauge were nearly equal for the MCS. Overall, the effects of wind and hail are likely present but difficult to quantify.

The gauge accumulations for the Wolf Point, Montana, AgriMet station (WPTM8) are nearly identical to those for OLF, indicating that both gauges’ rainfall observations are likely reasonable. However, the lowest radar beam at the WPTM8 gauge is blocked, with greatly reduced reflectivity resulting in consistent Q2RAD underestimation for all three events (Fig. 6h). Given the small size of the blocked sector (Fig. 4a), a cross-radial interpolation should eliminate the gap and improve the radar QPE in the sector.

The time series for the Culbertson, Montana, AgriMet station (CLBM8; see Fig. 6i) indicates severe problems with the gauge, especially during the MCC where the gauge reported zero compared to 34.5 mm for Q2RAD. Again, this indicates a probable gauge transmission error.

The final station to be examined is the ASOS at Jordan, Montana (JDN), which observed 18.3 mm of rain primarily during the MCS. The radar estimated 32 mm during the MCS (Fig. 6j). It is possible that the gauge’s underestimation was due to hail (note the severe hail report near the gauge in Fig. 3).

4. Discussion

Knowledge of uncertainties in radar and rain gauge data is important in verifying radar estimates, providing bias corrections for flash floods and longer-term river flooding, and as input to hydrologic models. Time series of data from 10 rain gauges have been compared to radar-based precipitation estimates for a series of severe storms that occurred in Montana on 16–17 June 2007. The storms were accompanied by severe wind and large hail. Few of the gauges appeared to be free of adverse effects caused by radar or gauge issues. These comparisons illustrate a wide variety of potential problems with both the radar and gauge data. Uncertainties include a gauge’s position relative to a storm’s path, high winds, hail, and the ZR relation. While many of these problems have been described in the literature, their operational impacts are significant and have yet to be fully quantified. Furthermore, this study presents additional uncertainty examples regarding radar attenuation, quality control (QC), and analysis timing issues. Because of complex and unknown interrelationships among the various uncertainty factors, their quantification is very difficult and additional statistical tools are needed. For instance, Ciach et al. (2007) present a radar–rain-rate uncertainty model for probabilistic QPEs. Their study includes high quality gauge data from the Oklahoma Mesonet and the Agricultural Research Service Micronet. With the vast majority of gauges used operationally being of lesser-known quality and the need to assess gauge-specific uncertainties, the ability to rapidly assess quality issues is needed. RFCs have the ability to painstakingly edit and manipulate gauge and radar data to produce more accurate QPE fields (which in turn are used to verify numerical models). However, forecasters at WFOs have little time to evaluate data quality in short-fuse warning situations. Also, MPE, which also runs in the WFOs, does not provide quality controlled gauge data to flash flood warning products and tools such as the Flash Flood Monitoring and Prediction (FFMP) software. FFMP is driven by radar data that has only a mean field bias adjustment. Therefore, we suggest that future collaboration between WFOs and RFCs include use of quality controlled gauge data in gauge-adjusted radar products for very short-term applications.

Efforts are on going to use RFC-generated “bad” gauge lists in order to codify manual QC techniques. The NSSL, ESRL, NCDC, and LMRFC are collaborating on the development of automated gauge QC methods verified by manually edited datasets, as well as Web-based applications to assess gauge uncertainties within the context of radar adjustments (Kim et al. 2009). Currently, the automated methods use neighbor checks that fail easily in convective weather. An automated QC scheme developed at ESRL and subsequently adapted by NCEP (Tollerud et al. 2005) attempts to mitigate this difficulty by verifying 24-h accumulated hourly gauge reports against a larger set of higher quality daily data. However, the resulting >24 h lag may not be timely enough for many applications, or may miss gauge failures that last only a few hours. Thus, to improve the gauge quality assessment, it is important to compare radar reflectivity time series with gauge time series to diagnose gauge data dropouts and other performance issues as shown by the examples in this paper. Many of these diagnostic tools are in place at the NSSL, ESRL, and NCDC. For instance, HADS gauge monthly histories are available on the NCDC Reprocessed HADS Web site (http://www.ncdc.noaa.gov/hads/index.html). Histories for neighboring HADS gauges can be displayed on the same plot (“mass analysis”). The gauges used in this study were verified as functioning normally with this tool. The ESRL Web site (http://precip.fsl.noaa.gov/beta/precip7.html) displays hourly and 24-h gauge data that can be used for visual neighbor checks.

The NSSL real-time QPE Verification System (QVS; information online at http://nmq.ou.edu) provides access to radar, gauge, model, and satellite data for the primary purpose of validating multisensor QPE algorithms. A key component is access to dense, high quality gauge networks. In addition to HADS and ASOS gauges, the QVS ingests Oklahoma Mesonet, Lower Colorado River Authority, and Maricopa County, Arizona, gauge networks (Fig. 9). Another QVS tool is the time series (TS) tool for gauge data and many associated radar variables such as reflectivity, precipitation type, and rain rates. Since the TS plots must be manually requested, efforts are under way to automate the production of critical variable databases that can be more quickly accessed and displayed with enhanced interactive Web applications. Work is also under way to formulate key variables into a gauge uncertainty index. For instance, zero gauge amounts with nonzero radar amounts would yield a low index. Statistics of hourly RGR would also provide insight into a gauge’s validity. Robust and fast online visual presentations of these data and gauge QC algorithm output would greatly facilitate the rapid evaluation of gauge data and QC algorithms.

Acknowledgments

The authors are grateful to Edward Tollerud for his review and discussions on the use of gauge data. Larry Cedrone and Dongsoo Kim have also provided helpful information regarding the HADS data and processing methods. Carrie Langston and Brian Kaney were instrumental in the development of the QVS, which greatly facilitates case analyses such as that presented herein.

REFERENCES

  • Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge measurements. J. Atmos. Oceanic Technol., 20 , 752759.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ciach, G. J., , and Krajewski W. F. , 1999: On the estimation of radar rainfall error variance. Adv. Water Resour., 22 , 585595.

  • Ciach, G. J., , Krajewski W. F. , , and Villarini G. , 2007: Product-error-driven uncertainty model for probabilistic quantitative precipitation estimation with NEXRAD data. J. Hydrometeor., 8 , 13251347.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Duchon, C. E., , and Essenberg G. R. , 2001: Comparative rainfall observations from pit and aboveground rain gauges with and without wind shields. Water Resour. Res., 43 , 32533263.

    • Search Google Scholar
    • Export Citation
  • Gourley, J. J., , Kaney B. , , and Maddox R. A. , 2003: Evaluating the calibrations of radars: A software approach. Preprints, 31st Int. Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., P3C.1. [Available online at http://ams.confex.com/ams/pdfpapers/64171.pdf].

    • Search Google Scholar
    • Export Citation
  • Humphey, M., , Istok D. J. D. , , Lee J. Y. , , Hevesi J. A. , , and Flint A. L. , 1997: A new method for automated dynamic calibration of tipping bucket rain gauges. J. Atmos. Oceanic Technol., 14 , 15131519.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kim, D., , Nelson B. , , and Cedrone L. , 2006: Reprocessing of historic Hydrometeorological Automated Data System (HADS) precipitation data. Preprints, 10th Symp. on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans, and Land Surface, Atlanta, GA, Amer. Meteor. Soc., 8.2. [Available online at http://ams.confex.com/ams/pdfpapers/100680.pdf].

    • Search Google Scholar
    • Export Citation
  • Kim, D., , Tollerud E. I. , , Vasiloff S. V. , , and Caldwell J. , 2009: Comparison of manual and automated quality control of operational hourly precipitation data of the National Weather Service. Preprints, 23rd Conf. on Hydrology, Phoenix, AZ, Amer. Meteor. Soc., J6.3. [Available online at http://ams.confex.com/ams/pdfpapers/146869.pdf].

    • Search Google Scholar
    • Export Citation
  • Maddox, R. A., 1983: Large-scale meteorological conditions associated with midlatitude, mesoscale convective complexes. Mon. Wea. Rev., 111 , 14751493.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Medlin, J. M., , Kimball S. K. , , and Blackwell K. G. , 2007: Radar and rain gauge analysis of the extreme rainfall during Hurricane Danny’s (1997) landfall. Mon. Wea. Rev., 135 , 18691888.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seo, D-J., 1998: Real-time estimation of rainfall fields using radar rainfall and rain gage data. J. Hydrol., 208 , 3752.

  • Seo, D-J., , and Breidenbach J. P. , 2002: Real-time estimation of spatially non-uniform bias in radar rainfall data using gauge measurements. J. Hydrometeor., 3 , 93111.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sieck, L. C., , Burges S. J. , , and Steiner M. , 2007: Challenges in obtaining reliable measurements of point rainfall. Water Resour. Res., 43 , W01420. doi:10.1029/2005WR004519.

    • Search Google Scholar
    • Export Citation
  • Steiner, M., , Smith J. A. , , Burges S. J. , , Alonso C. V. , , and Darden R. W. , 1999: Effect of bias adjustment and rain gauge data quality control on radar rainfall estimation. Water Resour. Res., 35 , 24872503.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tollerud, E., , Collander R. , , Lin Y. , , and Loughe A. , 2005: On the performance, impact, and liabilities of automated precipitation gauge screening algorithms. Preprints, 21st Conf. on Weather Analysis and Forecasting/17th Conf. on Numerical Weather Prediction, Washington, DC, Amer. Meteor. Soc., P1.42. [Available online at http://ams.confex.com/ams/pdfpapers/95173.pdf].

    • Search Google Scholar
    • Export Citation
  • Vasiloff, S. V., and Coauthors, 2007: Improving QPE and very short term QPF: An initiative for a community-wide integrated approach. Bull. Amer. Meteor. Soc., 88 , 18991911.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wilson, J., , and Brandes E. A. , 1979: Measurement of rainfall—A summary. Bull. Amer. Meteor. Soc., 60 , 10481058.

  • Xu, X., , Howard K. , , and Zhang J. , 2008: An automated radar technique for the identification of tropical precipitation. J. Hydrometeor., 9 , 885902.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhang, J., , Howard K. , , and Gourley J. J. , 2005: Constructing three-dimensional multiple-radar reflectivity mosaics: Examples of convective storms and stratiform rain echoes. J. Atmos. Oceanic Technol., 22 , 3042.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Map of MT with the white rectangle indicating the area of focus. WSR-88D locations with 250-km range rings are shown. Also shown are HADS (H) and ASOS (A) gauge locations.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 2.
Fig. 2.

Series of composite reflectivity images showing the storm episodes at (a) 1800 UTC 16 Jun, (b) 2200 UTC 16 Jun, (c) 0200 UTC 17 Jun, (d) 0600 UTC 17 Jun, and (e) 1000 UTC 17 Jun 2007.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 3.
Fig. 3.

Preliminary storm reports from the SPC for 16–17 Jun 2007. Triangles denote hail >2 in. diameter and squares denote winds >65 kt. Rain gauge locations are indicated by the letter G. The location of the KGGW radar is also indicated.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 4.
Fig. 4.

(a) The 24-h Q2RAD precipitation accumulations focused on the supercell path with gauge locations in Fig. 4b overlaid; (b) gauge circle map comparing 24-h Q2RAD and gauge data where circle size denotes gauge amount and circle color represents the radar–gauge ratio. Warm colors indicate radar values less than gauge values; cool colors indicate radar estimates greater than gauge values.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 5.
Fig. 5.

Radar–gauge ratios for each storm period and 24-h totals ending 1200 UTC 17 Jun 2007.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 6.
Fig. 6.

Time series of 5-min hybrid scan reflectivity, 1-h Q2RAD, and 1-h gauge amounts at (a) HWSM8, (b) MSWM8, (c) SACM8, (d) GGW, (e) GWSM8, (f) NSHM8, (g) OLF, (h) WPTM8, (i) CLBM8, and (j) JDN. The letter M indicates missed data transmission for the hour.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 7.
Fig. 7.

Series of composite reflectivities from KGGW showing the passage of the supercell over the radar. (a) Composite reflectivity (maximum in column) at 2330 UTC 16 Jun 2007. (b) HSR at 2330 UTC 16 Jun 2007. (c) As in (a) but for 2355 UTC 16 Jun 2007. (d) As in (b) but for 2355 16 Jun 2007. (e) As in (a) but for 0020 UTC 17 Jun 2007. (f) As in (b) but for 0020 UTC 17 Jun 2007. Select gauge IDs are overlaid.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 8.
Fig. 8.

(a) Regions between MT radars for which reflectivity differences are computed for the Radar Reflectivity Comparison Tool used by the WSR-88D Radar Operations Center. Each bin is 20 km × 20 km × 120 km. (b) Average reflectivity differences for the comparison region between KGGW and KTFX shown in Fig. 8a from 2200 UTC 16 Jun to 0200 UTC 17 Jun 2007. The red line indicates that KGGW is overall 2.5 dBZ higher than KTFX.

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Fig. 9.
Fig. 9.

Image from the QVS showing locations of the Maricopa County, AZ; Oklahoma Mesonet; and Lower Colorado River Authority (TX) gauge networks. (Information courtesy of the NMQ Web site: http://nmq.ou.edu/).

Citation: Weather and Forecasting 24, 5; 10.1175/2009WAF2222154.1

Save