• Ahijevych, D. A., , R. E. Carbone, , J. D. Tuttle, , and S. B. Trier, 2001: Radar data and climatological statistics associated with warm season precipitation episodes over the continental U.S. NCAR Tech. Note TN-448+STR, 81 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307].

    • Search Google Scholar
    • Export Citation
  • Belair, S., , and J. Mailhot, 2001: Impact of horizontal resolution on the numerical simulation of a midlatitude squall line: Implicit versus explicit condensation. Mon. Wea. Rev., 129 , 23622376.

    • Search Google Scholar
    • Export Citation
  • Bluestein, H., 1993: Synoptic-Dynamic Meteorology in Mid-latitudes: Observations and Theory of Weather Systems. Vol. 2. Oxford University Press, 431 pp.

    • Search Google Scholar
    • Export Citation
  • Bryan, G. H., , J. C. Wyngaard, , and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Mon. Wea. Rev., 131 , 23942416.

    • Search Google Scholar
    • Export Citation
  • Carbone, R. E., , and J. D. Tuttle, 2008: Rainfall occurrence in the U.S. warm season: The diurnal cycle. J. Climate, 21 , 41324146.

  • Carbone, R. E., , J. D. Tuttle, , D. A. Ahijevych, , and S. B. Trier, 2002: Inferences of predictability associated with warm season precipitation episodes. J. Atmos. Sci., 59 , 20332056.

    • Search Google Scholar
    • Export Citation
  • Casati, B., and Coauthors, 2008: Forecast verification: Current status and future directions. Meteor. Appl., 15 , 318.

  • Clark, A. J., , W. A. Gallus, , and T. C. Chen, 2007: Comparison of the diurnal precipitation cycle in convection-resolving and nonconvection-resolving mesoscale models. Mon. Wea. Rev., 135 , 34563473.

    • Search Google Scholar
    • Export Citation
  • Dai, A. G., , F. Giorgi, , and K. E. Trenberth, 1999: Observed and model-simulated diurnal cycles of precipitation over the contiguous United States. J. Geophys. Res., 104 , 63776402.

    • Search Google Scholar
    • Export Citation
  • Davis, C. A., , K. W. Manning, , R. E. Carbone, , S. B. Trier, , and J. D. Tuttle, 2003: Coherence of warm-season continental rainfall in numerical weather prediction models. Mon. Wea. Rev., 131 , 26672679.

    • Search Google Scholar
    • Export Citation
  • Fritsch, J. M., , and G. S. Forbes, 2001: Mesoscale convective systems. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 323–357.

    • Search Google Scholar
    • Export Citation
  • Gallus Jr., W. A., 2002: Impact of the verification grid-box size on warm-season QPF skill measures. Wea. Forecasting, 17 , 12961302.

  • Germann, U., , and I. Zawadzki, 2002: Scale dependence of the predictability of precipitation from continental radar images. Part I: Description of the methodology. Mon. Wea. Rev., 130 , 28592873.

    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , V. E. Kousky, , and R. J. Joyce, 2005: Diurnal cycle of precipitation determined from the CMORPH high spatial and temporal resolution global precipitation analyses. J. Geophys. Res., 110 , D23105. doi:10.1029/2005JD006156.

    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , V. J. Dagostaro, , V. E. Kousky, , and R. J. Joyce, 2007: An examination of precipitation in observations and model forecasts during NAME with emphasis on the diurnal cycle. J. Climate, 20 , 16801692.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23 , 931952.

    • Search Google Scholar
    • Export Citation
  • Knievel, J. C., , D. A. Ahijevych, , and K. W. Manning, 2004: Using temporal modes of rainfall to evaluate the performance of a numerical weather prediction model. Mon. Wea. Rev., 132 , 29953009.

    • Search Google Scholar
    • Export Citation
  • Laroche, S., , P. Gauthier, , J. St.-James, , and J. Morneau, 1999: Implementation of a 3D variational data assimilation system at the Canadian Meteorological Center. Part II: The regional analysis. Atmos.–Ocean, 37 , 281307.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44 , 133149.

  • Mass, C. F., , D. Ovens, , K. Westrick, , and B. A. Colle, 2002: Does increasing horizontal resolution produce more skillful forecasts? The results of two years of real-time numerical weather prediction over the Pacific Northwest. Bull. Amer. Meteor. Soc., 83 , 407430.

    • Search Google Scholar
    • Export Citation
  • Parker, M. D., , and D. A. Ahijevych, 2007: Convective episodes in the east-central United States. Mon. Wea. Rev., 135 , 37073727.

  • Trier, S. B., , C. A. Davis, , D. A. Ahijevych, , M. L. Weisman, , and G. H. Bryan, 2006: Mechanisms supporting long-lived episodes of propagating nocturnal convection within a 7-day WRF model simulation. J. Atmos. Sci., 63 , 24372461.

    • Search Google Scholar
    • Export Citation
  • Tuttle, J. D., , and C. A. Davis, 2006: Corridors of warm season precipitation in the central United States. Mon. Wea. Rev., 134 , 22972317.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., 1975: Diurnal variations in precipitation and thunderstorm frequency over the conterminous United States. Mon. Wea. Rev., 103 , 406419.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., , W. C. Skamarock, , and J. B. Klemp, 1997: The resolution dependence of explicitly modeled convective systems. Mon. Wea. Rev., 125 , 527548.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in Atmospheric Sciences. Academic Press, 467 pp.

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

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Analysis domain. The dotted contours represent the coverage of the 2.5-km CAPPI maps, while the black rectangle corresponds to the domain on which all statistics are computed. The precipitation pattern observed in this figure is typical for the spring of 2008.

  • View in gallery

    (a) Average hourly rain rate and (b) fractional precipitation coverage over the entire domain (32°–45°N, 103°–78°W) and time period (24 days between 18 Apr 2009 and 6 Jun 2009) for radar observations (solid line) and GEM forecasts (dashed line) as a function of forecast time (30-h forecasts initialized at 0000 UTC). The gray shading represents the first 6 h of the forecasts.

  • View in gallery

    Regional diurnal cycle of average hourly rainfall rate over the subdomains presented in Fig. 1 for the entire time period (24 days between 18 Apr 2008 and 6 Jun 2008) for radar observations (solid line) and GEM forecasts (dashed line). The order of the graphs corresponds to the geographic position of the domains.

  • View in gallery

    As in Fig. 3, but for fractional precipitation coverage.

  • View in gallery

    Hovmöller diagrams of the average diurnal cycle of precipitation for the spring of 2008 (16 Apr–6 Jun 2009) for the (a),(b) average hourly rain rate and (c),(d) precipitation coverage for a threshold equivalent to 15 dBZ for (a),(c) radar observations and (b),(d) GEM forecasts in the range of 6–30 h. All values are normalized with respect to the daily mean, which is indicated in the bottom right corner of each diagram (mm h−1 in the case of the average hourly rain rate). The diagrams have been repeated twice along 0600 UTC for clarity. The graphs have been smoothed over a window of 96 km in longitude and 3 h in time.

  • View in gallery

    Hovmöller time series of average hourly rain rate (mm h−1) for 16 Apr–14 May 2008, averaged over the latitudinal range 32°–45°N. The gray gaps represent the days when data were not available.

  • View in gallery

    As in Fig. 6, but for 14 May–11 Jun 2008.

  • View in gallery

    Rainfall event corresponding to 7 May 2008 as depicted by (left) radar observations and (right) GEM forecasts. The black rectangle denotes the analysis domain and the radar contours are overlaid on the forecast image for comparison.

  • View in gallery

    Normalized power spectrum of the Hovmöller diurnal cycle (Fig. 5) of (a),(b) average hourly rain rate and (c),(d) precipitation coverage over a threshold equivalent to 15 dBZ for the spring of 2008 for (a),(c) radar observations and (b),(d) GEM forecasts. The scale represents fraction of the variance.

  • View in gallery

    Phase of the (a),(b) 24- and (c),(d) 12-h harmonics as function of longitude for radar (black) and model (gray) in terms of both (left) average rainfall intensity and (right) precipitation coverage for the spring of 2008. The dotted line means that the harmonic does not explain 10% of the observed variance.

  • View in gallery

    Average diurnal cycle of precipitation for summer 2008 for radar observations of (a) precipitation intensity and (b) precipitation coverage. The values are normalized with respect to the mean, which is indicated in the bottom right corner of (a),(b) (0.08 mm h−1 for average rainfall intensity, and 4% for precipitation coverage). The white ovals highlight the semidiurnal mode. The diagrams have been repeated twice along 0600 UTC for clarity.

  • View in gallery

    Normalized power spectrum of the Hovmöller diurnal cycle (Fig. 11) of the (a) average rainfall rate and (b) precipitation coverage for radar observations during the summer of 2008.

  • View in gallery

    As in Fig. 10, but for the summer of 2008.

  • View in gallery

    Average diurnal cycle of rainfall intensity for (a),(c) spring and (b),(d) summer 2008, computed over the (top) northern and (bottom) southern halves of the domain (delimited at 38.5°N). The values are normalized with respect to the mean, which is indicated in the bottom right corner of (a)–(d) in mm h−1. The white dashed lines highlight features discussed in the text.

  • View in gallery

    (a) The diurnal cycle of GEM-simulated average rainfall intensity (in normalized units) and (b) the corresponding normalized power spectrum for the summer of 2008. The white ovals highlight features discussed in text.

  • View in gallery

    As in Fig. 14, but for GEM forecasts. GEM-radar correlation coefficients computed in Hovmöller space are indicated in the bottom-left corner of each diagram.

  • View in gallery

    CSI computed for a threshold of 15 dBZ as a function of forecast time for the spring (solid line) and summer (dashed line) of 2008 over the six subdomains presented in Fig. 1.

  • View in gallery

    The diurnal cycle of average hourly rain rate for the (a) spring and (b) summer averaged for 12 yr between 1996 and 2007.

  • View in gallery

    The year-to-year variability of the diurnal cycle of average hourly rain rate shown in Fig. 18, expressed as percentage of the 1996–2007 mean.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 63 63 10
PDF Downloads 17 17 7

The Diurnal Cycle of Precipitation from Continental Radar Mosaics and Numerical Weather Prediction Models. Part I: Methodology and Seasonal Comparison

View More View Less
  • 1 Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada
© Get Permissions
Full access

Abstract

The diurnal cycle of precipitation over the continental United States is characterized through the analysis of radar rainfall maps and is used as a measure of performance of the Global Environmental Multiscale (GEM) model during the spring (April–May) and summer (July–August) of 2008. The main interest is to determine the effects of different types of forcing (synoptic versus thermal) on the average daily variability of precipitation and on the model’s representation of it. A secondary objective is to study the interannual variability of the diurnal cycle. The investigation is based on the analysis of time–longitude diagrams of precipitation fields, of average statistics, and of model skill scores.

The results show that the main differences between the spring and summer diurnal cycles are the duration of propagating systems, the frequency of convective events in the southeastern United States, and more interannual variability of the spring diurnal cycle. However, most interesting is that the timing of precipitation initiation over the Rockies is in phase with the cycle of solar warming for both seasons, despite the strong synoptic forcing during spring. Also, east of the Rockies, the diurnal cycle is mainly determined by transport mechanism and is consequently out of phase with the solar cycle.

While GEM represents fairly well the timing of precipitation initiation along the Rockies during both seasons, it fails to correctly depict the propagation characteristics of these systems. During spring, the simulated systems show more variability in propagation paths than observed, while during summer, the observed propagation is simply not captured by GEM. This is probably a consequence of different propagation mechanisms acting in the model and in the atmosphere, and between spring and summer.

Corresponding author address: Madalina Surcel, McGill University, Room 945, Burnside Hall, 805 Sherbrooke St. W, Montreal QC H3A 2K6, Canada. Email: madalina.surcel@mail.mcgill.ca

Abstract

The diurnal cycle of precipitation over the continental United States is characterized through the analysis of radar rainfall maps and is used as a measure of performance of the Global Environmental Multiscale (GEM) model during the spring (April–May) and summer (July–August) of 2008. The main interest is to determine the effects of different types of forcing (synoptic versus thermal) on the average daily variability of precipitation and on the model’s representation of it. A secondary objective is to study the interannual variability of the diurnal cycle. The investigation is based on the analysis of time–longitude diagrams of precipitation fields, of average statistics, and of model skill scores.

The results show that the main differences between the spring and summer diurnal cycles are the duration of propagating systems, the frequency of convective events in the southeastern United States, and more interannual variability of the spring diurnal cycle. However, most interesting is that the timing of precipitation initiation over the Rockies is in phase with the cycle of solar warming for both seasons, despite the strong synoptic forcing during spring. Also, east of the Rockies, the diurnal cycle is mainly determined by transport mechanism and is consequently out of phase with the solar cycle.

While GEM represents fairly well the timing of precipitation initiation along the Rockies during both seasons, it fails to correctly depict the propagation characteristics of these systems. During spring, the simulated systems show more variability in propagation paths than observed, while during summer, the observed propagation is simply not captured by GEM. This is probably a consequence of different propagation mechanisms acting in the model and in the atmosphere, and between spring and summer.

Corresponding author address: Madalina Surcel, McGill University, Room 945, Burnside Hall, 805 Sherbrooke St. W, Montreal QC H3A 2K6, Canada. Email: madalina.surcel@mail.mcgill.ca

1. Introduction

There are many different ways of approaching the evaluation of numerical weather prediction (NWP) models (an extensive review can be found in Casati et al. 2008), which either independently or combined can provide insight on certain aspects of model performance. For example, research has been done on assessing the advantages of increasing spatial resolution in NWP models (Mass et al. 2002; Bryan et al. 2003). Other studies were interested in finding out at what horizontal scale could convection be explicitly represented in NWP models (Weisman et al. 1997; Kain et al. 2008). We concentrate here and in a follow-up paper on the ability of NWP models to capture the diurnal cycle of precipitation. This aspect of NWP validation has become increasingly popular in the recent years (Davis et al. 2003; Knievel et al. 2004; Clark et al. 2007; Janowiak et al. 2007). As pointed out by Carbone et al. (2002) the cycle of solar heating is an important forcing mechanism in midlatitudes during summer, influencing the evolution of convective precipitation systems. Moreover, the skill of a model in the depiction of the diurnal cycle of precipitation is indicative of the quality of the model’s representation of certain physical processes. For instance, previous studies attributed the inaccurate description of the diurnal cycle of warm-season precipitation in NWP models to the failure of the models to properly characterize the initiation and propagation of organized convection (Davis et al. 2003; Clark et al. 2007), finding that improvements can be obtained through increasing horizontal resolution as to allow the use of explicit rather than implicit convective parameterization schemes. Knievel et al. (2004) also mentioned that an accurate depiction of the diurnal cycle of precipitation over the southeastern coast of the United States could indicate that the model is skillful in simulating the sea breeze, which influences the development of convective precipitation in that region. However, evaluating a model’s depiction of the diurnal cycle of precipitation involves first characterizing it from observations.

Efforts toward characterizing the diurnal cycle of precipitation over the continental United States have been made as early as the 1900s. Thanks to an extensive number of studies carried out over the years (see, among others, Wallace 1975; Dai et al. 1999; Carbone et al. 2002; Ahijevych et al. 2001; Janowiak et al. 2005; Trier et al. 2006; Tuttle and Davis 2006; Carbone and Tuttle 2008, and references therein), the diurnal variation of precipitation during the warm season in North America is well documented. Essentially, this diurnal cycle is characterized by a maximum in the afternoon across the continent, except over the central Plains, where there is an additional nocturnal maximum. Carbone et al. (2002) found that these patterns are generated by coherent episodes of warm-season rainfall, usually initiated at the foothills of the Rocky Mountains, which occur almost daily, span a good portion of the continent, and exhibit predictability beyond the time scales of mesoscale convective systems. They described these episodes as due to the successive dissipation and regeneration of convection, which in turn is not related to significant synoptic forcing, but appears to be associated with mesoscale dynamics. Moreover, Carbone and Tuttle (2008) showed that the semidiurnal signal over the central Plains is caused by the combined effects of convection initiation in that region and of the arrival of propagating events from the Rocky Mountains.

On the other hand, during spring, synoptically forced precipitation events are more frequent, and hence not necessarily phase locked with the diurnal cycle of solar heating (Bluestein 1993). In addition, research on the diurnal cycle of rainfall during spring is sparser. Analyzing 30 yr of gridded precipitation data derived from station records across the United States, Dai et al. (1999) found a smaller amplitude and more interannual variability of the diurnal cycle during spring compared to summer.

The goal of this paper is twofold: (i) to characterize the diurnal cycle of precipitation over the continental United States from radar observations, and (ii) to evaluate the ability of the Canadian Global Environmental Multiscale model (GEM; Mailhot et al. 2006) to capture the cycle. The methodology presented herein will then be employed in a follow-up paper, which will focus on intercomparing the skill of several forecasting systems in rendering the diurnal cycle of precipitation, in relation to their horizontal resolution and physical parameterizations. The organization of this paper is as follows. After introducing the data employed (section 2), the paper continues with the characterization of the diurnal cycle of precipitation during spring 2008, both from radar observations and from GEM forecasts (section 3). Section 4 will then address the seasonal variability of the observed and the model-simulated diurnal cycles. The results are finally summarized in section 5.

2. Data description

a. Precipitation forecasts

The precipitation forecasts evaluated in this paper were generated using the GEM model. GEM (see a complete description in Mailhot et al. 2006 and references therein) is a global, hydrostatic, variable resolution model, used operationally at the Canadian Meteorological Center (CMC) for regional forecasting since May 2004. In the central part of the domain, which covers North America, the horizontal resolution is uniform at 15 km and the vertical resolution is variable with 58 vertical levels and with the model lid being at 10 hPa. It is run twice a day, at 0000 and at 1200 UTC, and the initial conditions are provided by a three-dimensional variational data assimilation (3DVAR) Regional Data Assimilation System (RDAS; Laroche et al. 1999), which does not include radar observations. The GEM model employs the Kuo transient scheme for shallow convection and the Kain–Fritsch scheme for deep convection. In this study we have evaluated 30-h precipitation forecasts of hourly rainfall accumulations generated at 0000 UTC.

b. Radar observations

GEM precipitation forecasts have been validated against rainfall maps derived from U.S. composite radar reflectivity mosaics generated by the National Severe Storm Laboratory (NSSL; Zhang et al. 2005). NSSL produces two types of precipitation products at high temporal and spatial resolution (5 min and 1 km): 3D reflectivity mosaics covering the continental United States with a vertical resolution of 100 m at low heights, and several 2D quantitative precipitation estimation (QPE) products obtained by processing the 3D data. From the available products, we have used reflectivity maps at 2.5 km altitude rather than the 2D composite maps: the artifacts present in the 2D composite maps (strong echoes associated with nonmeteorological targets, especially birds and insects) were undermining the objective validation of the forecasts. Some of the disadvantages of using the 2.5-km data are the reduced area coverage due to beam blockage in areas of steep topography (e.g., around the Rocky mountains; Fig. 1), and the uncertainty associated with the variability of the vertical profile of reflectivity below 2.5 km. However, close inspection showed the superior quality of this product.

For the analysis presented in section 4b (the analysis of the interannual variability of the diurnal cycle of precipitation), we have also used the available time series of the WSI Corporation NOWrad (national composite) product for the period 1996–2007. These data have benefited from three levels of quality control. Original reflectivity mosaics have a 2-km spatial resolution and a 15-min temporal resolution, and they were generated by taking the maximum value (of dBZ) measured by any WSR-88D radar at any height in the vertical of each grid point (similar datasets have been used by Carbone et al. 2002; Germann and Zawadzki 2002). These reflectivity observations are not comparable to rainfall maps based on sophisticated algorithms that correct for the vertical profile of reflectivity, visibility, or ZR variability. However, the extensive coverage (both spatial and temporal) of these composite maps makes them unique for the study of continental scale phenomena.

Maps of reflectivity, Z, were converted to rain rate, R, using a standard ZR relationship, Z = 300R1.5 (used, among others, by Carbone et al. 2002) and instantaneous rain rate maps were accumulated to obtain hourly rainfall maps.

To avoid contamination of results due to radar data artifacts, a threshold of 15 dBZ was considered to discriminate between raining and nonraining areas.

c. Case studies and analysis domain

In this work, the validation of GEM rainfall forecasts has been done from 16 April 2008 to 6 June 2008 and from 1 July 2008 to 31 August 2008. From this point on, we shall refer to the time period from 16 April to 6 June as “spring” and from 1 July to 31 August as “summer.”

Given the difference in spatial resolution between the forecasts and the observations, precipitation fields have been averaged to 32-km resolution as to enable the comparison. The spatial domain was chosen to cover most of central and eastern United States, running from 103° to 78°W in longitude and between 32° and 45°N in latitude, in order to avoid the region around the Rocky Mountains where the quality of radar observations is poor. To characterize the diurnal cycle of precipitation in different regions, the domain has been evenly split into six subdomains as shown in Fig. 1.

3. Characterizing the diurnal cycle of precipitation during spring 2008 from radar observations and model forecasts

a. The diurnal cycle of average rainfall statistics

Figure 2 shows average values of mean rainfall intensity and coverage over the whole analysis domain (see Fig. 1) as a function of UTC time for spring 2008. Average hourly rainfall intensity over precipitating areas is 2.3 mm h−1, while GEM forecasts underestimate it by about 0.5 mm h−1. Precipitation intensity presents significant dependence on the time of the day: maximum intensity occurs around 0300–0400 UTC (in the figure the maximum appears twice—at forecast hours 0300–0400 and 2700–2800) and minimum intensity at 1900 UTC. There is a secondary relative maximum of precipitation at 1400 UTC, which is discussed later on in the text. Note that because of having used a dataset composed of nonsuccessive days, the average hourly rain rate and precipitation coverage during hours 0000–0600 and 2400–3000 are not identical.

Alternatively, GEM slightly overestimates precipitation coverage for lead times between 14 and 30 h. Besides, GEM coverage also shows a more significant diurnal variability than observations, which show rather constant precipitation coverage of around 7%.

However, the longitudinal extent of the domain may be filtering out the evidence of precipitation maxima due to the combined effect of precipitation propagation (with a dominant west–east component) and the impact of solar forcing in convection initiation/inhibition (happening earlier in the east than in the west). These two mechanisms acting together result in rainfall peaks occurring at different times in different locations that may interfere when rainfall is averaged over the entire domain. Figures 3 and 4 show the diurnal evolution of precipitation intensity and coverage for the six subdomains of Fig. 1. Radar observations show that the average rain rate and precipitation coverage clearly depend on the geographical subdomain (maximum daily averages in the northwest of around 1.5 mm h−1 and 10%, for rain rate and coverage, respectively, and minimum daily averages of 0.9 mm h−1 and 4% in the southeast), and the diurnal cycle shows significantly different amplitude and time of maximum precipitation in different regions: most clear diurnal signals are evident in the northwest and southeast subdomains, and the timing of the diurnal maximum differs across the domain (i.e., maximum coverage occurs at around 0500–0700 UTC in the western domains and at around 2300 UTC, in the eastern domains). The propagation of the diurnal maximum is not clear from Figs. 3 and 4, but it is further studied in section 3b. The analysis of the geographical dependence of the diurnal cycle of precipitation, explains Fig. 2: the rain-rate cycle in the west dominates the graph of Fig. 2a because of its higher intensities and amplitude, except for the secondary maximum that is mostly due to the 1100–1500 UTC maximum of the north-central domain (Fig. 3b). The latter can, thus, be associated with the nocturnal maximum over the central Plains reported by Dai et al. (1999). The combination of the off-phase diurnal cycles of precipitation coverage in different subdomains results in the almost flat diagram of Fig. 2b.

On the other hand, GEM forecasts need between 6 and 15 h (depending on the domain) to reach the values of observed rain rates (except for the southwestern domain, where rain rates are underestimated most of the time). This could in part be an effect of the spinup time (GEM forecasts are generated every day at 0000 UTC), and is similar to the results of Clark et al. (2007), who found effects of the model spinup present up to 24-h lead time. As previously mentioned, GEM forecasts of precipitation coverage show very good skill in the short term, which tends to worsen for longer lead times. It is also worth noting the general trend of GEM to anticipate the peak (and in a lesser extent, the minima) of precipitation coverage.

Synoptically forced precipitation events do not necessarily show significant dependence on the time of the day. However, the results presented above have shown the dependence of precipitation coverage and intensity on the time of the day, despite the fact that the analysis focused on spring, when synoptic conditions are expected to have a more dominant role than thermal forcing. More on this topic is provided in section 3b.

b. Time–longitude analysis of precipitation fields

As shown by previous studies, the diurnal cycle of warm-season rainfall is due to the combined effects of precipitation initiation and propagation. However, it is difficult to distinguish between the impacts of these two mechanisms solely from Figs. 2 –4. Therefore, we proceed with a more detailed analysis of precipitation systems in time–longitude coordinates (averaging over the latitudinal range 32°–45°N, similarly to Carbone et al. 2002). The reason for choosing longitude as the spatial coordinate is that precipitation systems in North America usually exhibit a stronger zonal than meridional component of propagation.

Figure 5 illustrates the diurnal cycle of precipitation averaged over 24 days between 16 April 2008 and 6 June 2008. To avoid the problems associated with the spinup time of the model (see section 3a), the first 6 h of the forecasts have been ignored. The main feature of the diurnal cycle is a rainfall maximum originating around 103°W at 1900 UTC (however, the sparse radar coverage west of 103°W might be masking precipitation systems being initiated or propagating through the westernmost boundary of our domain, see Fig. 1), which propagates along a corridor that extends to 85°W at about 0600 UTC. This means that, on average, for the time period of 25 days studied here, precipitation systems were generated in the west and propagated eastward, lasting close to 24 h. The zonal extent of individual events can be inspected in Figs. 6 and 7, which display the time series of longitudinally averaged rainfall for the entire study period. Indeed, precipitation systems often originated along the Rockies at about 0000 UTC as convective cells, which then merged together becoming more organized as they propagated eastward (e.g., 2, 4, 5, and 6 June; see Fig. 7). However, the main weather patterns during the studied time period were longer-lasting streaks of precipitation that occurred as the convective cells organized along fronts or other large-scale patterns (e.g., 2, 6–7 May, 5 June). Therefore, it appears that the signal observed in Fig. 5 should not be attributed to the same factors as the patterns found by Carbone et al. (2002) for the “warm season” (June–August), when the thermal forcing is stronger than the large-scale forcing. In our case, although thermally driven convection episodes did occur, their organization was synoptically influenced, resulting in larger-scale, longer-lived systems, consistent in terms of initiation timing. There are several characteristics of the propagating band in Fig. 5 that are worth discussing. While the shape of the band is similar in fractional coverage and in average intensity, the evolution of the two variables within the streak is different: precipitation intensity peaks at 0000–0100 UTC along longitude 100°W, significantly earlier than precipitation coverage. A possible explanation for this pattern is the occurrence of convective systems along longitude 100°W, which have a limited spatial extent, but intense rainfall rates. The organization of such systems as they propagate causes a subsequent increase in precipitation coverage.

It appears from Figs. 6 and 7 that, during the studied time period, stationary convective activity did not occur systematically in the southeast, a region that during summer exhibits a very strong diurnal cycle. Therefore, the weak maximum detected within the longitudinal range 84°–78°W in the diurnally averaged Hovmöller diagram (Fig. 5) is likely to be caused by propagating precipitation systems rather than stationary convection. Parker and Ahijevych (2007) have reported that such events occur in the afternoon in the Appalachian Mountains, and then propagate eastward toward the coast.

Through the entire period of analysis there appears to be more variability in the propagation paths forecasted by GEM, as the simulated average diurnal cycle (Figs. 5b,d) exhibits several diagonal streaks rather than a single well-defined streak, as in the observations (Figs. 5a,c). This is perhaps due to the forecasts being affected by location and timing errors. Indeed, Figs. 6 and 7 reveal several aspects concerning GEM’s performance. First, GEM fails to predict the timing of convection initiation in most days, forecasted systems developing later and to the east of the observed (e.g., 29 May, 2 and 3 June; Fig. 7). Second, when convection is forecasted, the position of the precipitation areas is inaccurate. Finally, in synoptically driven cases, GEM overestimates the amplitude of the events (e.g., 18 April and 7–8 May; Fig. 6). These errors generated model precipitation similar to radar patterns, but not showing the time–location consistency described for observations.

An illustration of this behavior is presented in Fig. 8, which shows the evolution of a precipitation event on 7 May 2008 from radar observations and GEM forecasts. At 0000 UTC, the radar observations show convective cells developing in Texas, Kansas, Nebraska, and Iowa, that are not captured by GEM, neither at initial time, nor at forecast hour 24 (forecast initialized at 0000 UTC on 6 May—not shown). At 0600 UTC, GEM-simulated precipitation systems are somewhat larger and displaced with respect to the observations. At 1200 UTC, some organization becomes apparent over northern Texas, Oklahoma, and Kansas along a large-scale feature, while the precipitation systems over Iowa are in decay. While GEM captures the new organization, it has difficulties reproducing the decay over Iowa. Finally at 1800 and 2200 UTC, the radar observations illustrate a typical midlatitude cyclone, with the most intense precipitation bands occurring along the edge of the cold front. GEM seems to predict fairly well the position of the center of the low, while it most evidently overestimates precipitation intensities.

The delay in precipitation initiation and decay in GEM could be a consequence of the spinup time. GEM does employ a 3DVAR regional data assimilation system with 6-h cycles and digital filter initialization, and, indeed, it produces precipitation fairly fast. However, the forecasted precipitation systems suffer of significant displacement (and less significant timing) errors, which seem to affect the model’s environment (e.g., the spatial distribution of moisture), and therefore the evolution of precipitation systems at later times. Radar data assimilation could be useful in this context, by correcting positional errors at initial time. The overestimation of precipitation in synoptic events could perhaps be related to a too moist planetary boundary layer, or to a lack of convective inhibition in the model’s environment, but these issues are not investigated here.

c. Frequency analysis of daily averaged Hovmöller diagrams

Figure 9 shows the normalized power spectra corresponding to the diurnal time series of average rainfall intensity and coverage of Fig. 5 as a function of longitude. In the case of radar observations, the 24-h harmonic dominates through nearly the whole longitudinal domain, being most important in the range 103°–92°W, where it explains over 90% of the observed variance. As the dataset was not controlled by particular events (section 3b; Figs. 6 and 7), we consider the 24-h harmonic as characteristic to the periodicity of rainfall during the study period. Along 103°W, rainfall occurrence is associated with precipitation initiation, while east of 103°W propagation was the main mechanism of precipitation occurrence. Therefore, the dominance of the 24-h harmonic signifies two things: (i) the timing of precipitation initiation over the foothills of the Western Cordillera is relatively constant from day to day, and (ii) the zonal propagation characteristics (direction and speed) are consistent between the precipitation systems developed in this area.

Carbone et al. (2002) noted the importance of the 12-h harmonic in the longitudinal range 98°–90°W, associated with the combined effects of precipitation initiation in this region and of the arrival of propagating episodes from the Rocky Mountains. This type of signal is hardly visible in our case, and we believe it to be due to the lack of thermally forced convective events during the period of study in these longitudes. Both the average rainfall intensity and precipitation coverage plots indicate a minor significance of higher-frequency harmonics (Fig. 9), especially east of 92°W. However, the discrepancies between the average rain-rate and precipitation coverage diagrams, and the lesser occurrence of rainfall in this region, make us hesitate to attribute any physical meaning to these modes: they may be merely an artifact of the Fourier analysis or a measure of the variability of rainfall occurrence as a consequence of synoptic forcing. In addition, while the diurnal mode is dominant around 80°W, its amplitude is not very strong, as indicated by Fig. 5. This diurnal mode is mostly associated with the arrival of propagating precipitation systems (as it was observed in the time series of Figs. 6 and 7), and not due to the well-documented stationary maximum in the southeastern United States.

GEM demonstrates a distribution of power across the frequency domain different than the one of the radar observations (Fig. 9): while radar observations show the prevalence of the 24-h harmonic in the western half of the domain, GEM forecasts attributed, in addition, significant importance to the 12- and 8-h harmonics in that region, in agreement with the previous remark that the model describes more variability in propagation paths than observed (as previously discussed for Fig. 5). On the other hand, in the longitudinal band 92°–85°W, GEM attributes more than 90% of the variance to the 24-h harmonic, contrarily to the observations. The difference between the radar and GEM power spectra underlines the inability of the model to capture the time–space consistency of precipitation systems, as a result of phase errors affecting the precipitation forecasts.

Figure 10 shows the phase of the 24-h harmonic as a function of longitude for both radar observations and GEM forecasts. We first note that the propagating signal is clear in the shift of the phase with longitude from 0300 UTC at 100°W to 0000 UTC at 89°W, indicating that precipitation systems take around 24 h to propagate across this longitudinal range. East of 89°W, the timing of the maximum of the 24-h harmonic oscillates within a 3-h range around 0000 UTC but the scarcity of rainfall in this region questions the significance of this pattern. In terms of average rainfall intensity, GEM depicts well the timing of the maximum of the 24-h harmonic in most regions, but the shift in phase with longitude does not quite match that of the observations.

4. Seasonal and interannual analysis

a. Comparison of the diurnal cycle during spring and summer 2008

1) The radar-observed diurnal cycle

The analysis of radar-observed and model-simulated diurnal cycles of precipitation during spring 2008 showed that, while thermally driven convection did occur regularly in the western foothills during this time period, its organization was influenced by large-scale phenomena, resulting in long-lived, large-scale systems. Here, we focus on the characteristics of the diurnal cycle of precipitation in summer 2008 to highlight the differences with that of spring 2008. This will also allow us to study the different performance of GEM in summer and spring [section 4a(2)]. Figure 11 illustrates the Hovmöller diurnal cycle of average rainfall intensity and precipitation coverage for summer 2008. There are three main features that stand out in these diagrams:

  • a propagating rainfall band originating over the Rockies at about 2000 UTC;
  • a second maximum between 93° and 85°W at 1000 UTC;
  • a stationary maximum centered at 2200 UTC located between longitudes 85° and 78°W.
We will discuss each of these features with respect to the diurnal cycle observed for spring 2008, referring to the frequency analysis of the Hovmöller diagrams (Figs. 12 and 13) for a more quantitative description.

The propagating band originating over the Rockies at about 2000 UTC has also been observed during the spring of 2008 (Fig. 5). The frequency analysis of the Hovmöller diagrams of Fig. 11 (presented in Fig. 12) shows the dominance of the diurnal mode in the range 105°–95°W. Comparing the phase of the 24-h harmonic between the two seasons (Figs. 10 and 13), we conclude that the timing of precipitation occurrence at 103°W is invariable with season, but that the zonal propagation speed between 103° and 95°W is slightly higher during summer. Moreover, as also seen in Figs. 5 and 11, the extent of the band is shorter during summer, propagation ceasing at about 95°W after close to 12 h, contrary to 88°W for the spring of 2008, when it lasted almost 24 h. This illustrates the difference in the types of precipitation systems during spring and summer: our analysis revealed that precipitation systems during spring 2008 were strongly influenced by synoptic forcing, to which they owed their longevity, while on the other hand, such strong influence was not detected for the summer period, radar fields showing mostly mesoscale self-organization of weakly or nonforced convection (e.g., mesoscale convective complexes). These last results are in agreement with those of Carbone et al. (2002), who described the observed propagation signal as not related to large-scale disturbances, but as characteristic to mesoscale dynamics. While the exact propagation processes are not well known, they suggested density current propagation in the boundary layer and inertia–gravity wave propagation in the free troposphere as plausible mechanisms. In addition, Fritsch and Forbes (2001) have mentioned large-scale ascent over a surface front, a substantial low-level jet, and possibly weak midtropospheric short waves as important factors affecting the propagation of mesoscale convective systems.

The longitudinal range 95°–85°W exhibits a particular regime: the diagrams of precipitation coverage (and, to a lesser extent, of average rain rate) show a dominating semidiurnal component (white oval in Fig. 11, quantified in Fig. 12), with one maximum occurring around 1000 UTC, and the other around 2200 UTC (in this range of longitudes, Figs. 13c,d show the phase of the 12-h harmonic roughly constant at 1000 or 2200 UTC). To investigate the possible cause of this semidiurnal mode, we have analyzed the diurnal cycle of precipitation in the southern and northern halves of the domain separately. Figure 14d indicates that the 2200 UTC maximum is characteristic to the southern portion of the domain (marked by a white dashed line), being related to convection occurring along the Gulf Coast in phase with the diurnal cycle of solar heating and associated with the land/sea-breeze circulation. The 1000 UTC maximum (white dashed line in Fig. 14b), while in phase with the arrival of propagating episodes from the Rockies, seems independent of it, and is stronger in the northern half of the domain. Carbone and Tuttle (2008) have also observed the 1000 UTC maximum, explaining it as being a result of the reversal of the mountain–planes solenoidal circulation, which becomes stronger in relation to moisture transport by the Great Plains low-level jet. Therefore, the daily evolution of precipitation in this longitudinal range during summer is very different than during spring, when precipitation events from the western Cordillera propagated over this region and sea-breeze convection was not observed.

The maximum previously observed at 2200 UTC and associated with sea-breeze and airmass convection in the southeastern United States extends into the range between 86° and 78°W longitude, as shown by both the Hovmöller diagrams of Fig. 11 and the frequency analysis of these diagrams shown in Figs. 12 and 13.

The latitudinal dependence of the diurnal cycle of precipitation is also different between spring and summer (Fig. 14). During spring, the northern half of the domain has more rainfall than the southern half, with a mean of 0.17 mm h−1 in the north compared to 0.11 mm h−1 in the south. Also, both regions show very little rainfall east of 85°W (Figs. 14a,c), with the main mechanism of rainfall occurrence being west–east propagation both in the north and in the south. On the other hand, during summer, the two regions show a different spatial distribution of daily averaged precipitation, while the mean values of rainfall are identical (0.08 mm h−1). In the north (Fig. 14b), there is more rainfall occurring in the western part, while the opposite happens in the south (Fig. 14d). Also, in the north, precipitation mostly occurs along the Rockies, and it then propagates eastward, while in the south, rainfall is mostly associated with local convective activity in the southeastern United States.

2) Model performance

As seen in the previous section, GEM was able to reproduce the propagation of rainfall systems during the spring of 2008, although with the aforementioned differences in timing and location. In this section, we investigate how the difference in the type of forcing between spring and summer translates to the model’s representation of the diurnal cycle between the two seasons.

Figure 15a illustrates the GEM-simulated Hovmöller diurnal cycle of average rainfall intensity for the summer of 2008. By comparison to the radar observed diurnal cycle (Fig. 10), we can see that GEM is completely unable to reproduce the propagating signal in the western part of the domain. Instead, in this region, GEM shows a stationary diurnal maximum at 2200 UTC. This type of performance of convection-parameterized models has been previously reported by Davis et al. (2003) and Clark et al. (2007). Using idealized simulations or by intercomparison with convection-resolving models, these studies have shown that the inability of models to reproduce propagating convection is associated with the convective parameterization scheme. Figure 15b shows the power spectrum associated with Fig. 15a. In agreement to what was observed for the radar data, the 24-h harmonic dominates throughout the longitudinal domain, with the exception of the range 95°–85°W. Within this range, the 12-h harmonic is the most significant, and the 24-h harmonic is not considered a representative feature of the diurnal cycle.

Figure 13 shows the comparison of the phases of these two harmonics between radar and GEM. At 103°W, the timing of the diurnal maximum agrees between the two (the black and the gray lines stay close together), meaning that GEM forecasts well the precipitation initiation. However, between 103° and 95°W, the radar data indicates a propagating signal (shift of the phase with longitude), while GEM shows a phase constant with longitude. This is due to the fact that, while GEM forecasts convection initiation at 103°W at 2200 UTC, the model-simulated systems are shorter lived that in reality, and therefore of shorter longitudinal extent, hence causing the lack of rainfall in this longitudinal range between 0400 and 1900 UTC. This lack of rainfall also causes an artificial maximum in the GEM diurnal cycle of Fig. 15a at 2200 UTC (white oval in Fig. 15a) along longitude 99°W, which further acts to maintain the phase of the 24-h harmonic close to 0°. Between longitudes 85° and 78°W, where the diurnal mode also clearly dominates, GEM represents relatively well (within 1–2 h) the timing of the daily maximum, while between 95° and 85°W GEM is able to reproduce the dominance of the semidiurnal mode and the timing of the maxima (within a 2-h range). Here the diurnal mode is weak and, therefore, the model-radar disagreement is not significant.

We have also evaluated GEM’s ability to differentiate between diurnal cycle in the northern and southern regions (Fig. 16). During spring, GEM agrees fairly well with the observations in both regions, the correlation coefficients computed in Hovmöller space being 0.75 in the northern half and 0.70 in the southern half.

During summer, GEM performance is much poorer and shows a clear difference between the two domains, which could be attributed to the difference in the type of precipitation systems between the north and south. In the south, GEM captures well both the timing of the diurnal maximum east of 87°W (cf. Figs. 16d and 14d) and the timings of the semidiurnal maxima between longitude 94° and 87°W, although it only has a correlation coefficient of 0.54 with the radar observed cycle. This is due to GEM’s inability to capture the initiation and propagation in the westernmost side of the domain. This problem is more pronounced in the northern half of the domain, which is dominated by propagating systems: GEM’s depiction of the diurnal cycle is radically different from the one observed, with the correlation coefficient a mere 0.24 (cf. Figs. 16b and 14b). GEM does forecast a propagating signal, but one that is shifted east of the observed, maybe an indication of systematic positional errors.

Finally, we have also investigated the diurnal variation of forecast skill at the regional scale in terms of the Critical Success Index (CSI; Wilks 1995). The CSI is a contingency table based score defined as
i1520-0493-138-8-3084-e1
where
  • H is the number of hits (data points where both forecasts and observations indicate rain over a given threshold),
  • F is the number of false alarms (data points where the forecasts indicate rain over a given threshold while the observations do not), and
  • M is the number of misses (data points where the observations indicate rain over a given threshold, while the forecasts do not).

The CSI values presented here should not be regarded in absolute terms. Biases in the radar rainfall estimates (e.g., due to the variability of the ZR relationship) or in the model forecasts could negatively affect the scores, but misrepresent the actual skill of the model. In addition, the verification gridbox size also affects the values of the CSI (as shown by Gallus 2002), and while more elaborate methods could be used to account for the differences in horizontal resolution between forecasts and observations, verification methodology in itself is not a primary focus of this paper. In turn, we regard these CSI values for comparison between GEM’s performance at various lead times, in different regions and during different seasons. Figure 17 shows the CSI for a threshold of 0.2 mm h−1 for the six regions of Fig. 1. During spring, model performance varies according to the region but there is no clear diurnal signal of CSI (see solid line in Fig. 17). The summer scores (dashed line) are evidently lower than those for spring 2008 (thin line), especially in the westernmost portion of the domain. The southeastern region does not show any difference in scores between spring and summer mostly because of the lack of rainfall in that region during spring. However, the summer scores in that region (Fig. 17f) are low (maximum 0.2), despite the qualitative agreement indicated by the Hovmöller diagrams (cf. Figs. 14d and 16d). This also happens in the northeastern side of the domain (cf. Figs. 17c, 14b, and 16b) and could be due to the displacement errors associated with forecasts of convective rainfall (especially in the areas of convection initiation), whose effects are diminished in the Hovmöller domain due to averaging. In the central regions, the CSI varies with lead time, such that during the first 16 h the scores are much higher than during the last 8 h. We recall that two different precipitation regimes occur in the central United States: the arrival of propagating events from the west (with an associated maximum at 1200 UTC), and the initiation of local convection (with an associated maximum at 2200 UTC). Therefore, the variation of skill could mean that GEM has skill in forecasting the already developed propagating events, but not the initiation of convection in this area.

b. The interannual variability of the diurnal cycle

The question of how representative is the diurnal cycle during 2008 for the spring and summer seasons in general motivates the investigation of its interannual variability. Figure 18 shows the Hovmöller diurnal cycle of precipitation intensity for spring (Fig. 18a) and summer (Fig. 18b) averaged over the years 1996–2007. These graphs have been generated using the available data corresponding to the NOWrad product of WSI (as described in section 2). First, the diurnal cycles of precipitation during spring and summer are similar through the initiation timing and slope of the propagating band in the west, and through the maximum centered at 86°W at 2100 UTC. This pattern of convective rainfall in western longitudes is consistent with the position of the low-level jet, which is believed to favor the intensification and organization of convection in this region (Tuttle and Davis 2006). In agreement with Dai et al. (1999), the amplitude of the diurnal cycle is lower during spring than during the summer for any given longitude (notice the magnitudes of the maxima in Fig. 18), but this could also be an indication of more interannual variability during the spring. In addition, the propagating band during spring is wider than during summer, which shows less consistency of the propagating characteristics than during the summer. The stationary convection maximum occurring during summer afternoons in the eastern portion of the domain is apparent in the spring diurnal cycle as well, but it is significantly weaker, as thermally forced convective events are sparser during spring over that region.

By analyzing the Hovmöller diagrams for each of the 12 yr for spring and summer (not shown), we have concluded that there is significantly more year-to-year variability in spring average diurnal cycles than for summer (as shown in Fig. 19): during certain years, thermal forcing is prevalent resulting in a diurnal cycle of precipitation similar to summer, while in years of prevalent synoptic forcing, precipitation systems show more variability in terms of propagation paths [this also explains the smooth pattern of the 12-yr average Hovmöller diagram for spring cf. summer (Figs. 19a,b, respectively)]. However, it is remarkable that the timing of precipitation initiation over the Western Cordillera is fairly consistent from year to year and it is likely due to the very stable effect of solar heating.

The diurnal cycles of precipitation for 2008 differ from the long-term average through several features. During the spring of 2008, while the initiation timing and propagation duration of precipitation systems originated over the Rockies were in fair agreement with the long-term mean, there was no evidence of a stationary diurnal signal in the eastern part of the domain. To better understand the reasons for this, the strength of the forcing (primarily the land/sea-breeze circulations) should be investigated, but this is outside of the scope of the current paper. The diurnal cycle of precipitation during the summer of 2008 differs from the long-term mean mostly through the strength of the semidiurnal mode in the central part of the domain. Carbone and Tuttle (2008) also reported the interannual variability of the semidiurnal mode in the 95°–90°W longitudinal range, but again, a closer analysis should be attempted to provide more insight into the matter.

5. Conclusions

We have quantified the diurnal cycle of precipitation during spring and summer over the continental United States both from radar observations and from model forecasts. The main objectives of this paper were to assess the skill of GEM in reproducing this variability and to relate its performance to the type of dominant forcing, which has motivated the characterization of the diurnal cycle during the two seasons from radar observations.

While our results for the summer of 2008 are mostly in agreement with previous studies, we have found that over the central United States, the semidiurnal mode was more significant than expected, the north–south separation of the analysis domain indicating that this signal was occurring north of 35°N.

For the spring of 2008, we have found that, despite the major influence of synoptic disturbances on the initiation and development of precipitation systems, these systems have, on average, a variability consistent with the diurnal cycle of solar heating and subsequent propagation. In other words, the diurnal variability of precipitation averaged over the 25 days of spring 2008 in our analysis appears to have been caused by a combination of weak thermal forcing and strong synoptic-scale forcing, resulting in large-scale precipitation systems consistent in terms of initiation timing and propagation characteristics. On the average time–longitude diagrams, these systems have appeared as a clearly propagating rainfall band.

GEM has performed very differently during the two seasons. For the spring of 2008, GEM has depicted the propagation of the systems, even though temporal and displacement errors in precipitation forecasts resulted in a wider band than in the observations in terms of the Hovmöller diagrams. The daily time–longitude diagrams have demonstrated that the differences between the model-simulated and the observed diurnal cycles were representative of most of the study period, and not a consequence of a few precipitation events that skewed the results. Additional analysis of hourly rainfall accumulation maps showed that there were two causes for this signal: first, the forecasts suffered from positioning and timing errors in convective precipitation systems; second, GEM precipitated too easily along regions of synoptic forcing, resulting in the overestimation of rainfall occurrence. As mentioned by Belair and Mailhot (2001), the overestimation could be related to the Kain–Fritsch parameterization, which might not remove enough energy, but which allows for grid-scale mesoscale circulations. Other reasons for overestimation could be the misrepresentation of convective processes along the edge of precipitation systems and imperfect initial conditions.

On the other hand, during the summer of 2008 GEM proved no skill in reproducing the propagating signal in the western part of the domain. This result agreed with Davis et al. (2003) and Clark et al. (2007), who attributed this problem to the inability of the convective parameterization scheme to reproduce propagating convection that is phased locked with the diurnal cycle of solar heating. As the precipitation forecasts used in the study were generated in operational mode, we could not test the impact of using different cumulus parameterizations [or no parameterization, even though 15-km resolution is not considered high enough for explicit representation of convection; Weisman et al. (1997)]. However, Part II (M. Surcel, M. Berenguer, and I. Zawadzki 2010, unpublished manuscript) will discuss different model configurations with different horizontal resolutions and convection parameterizations. The poor performance of GEM was also clear through the low CSI that the precipitation forecasts registered in the western part of the analysis domain. The situation was different on the eastern side of the domain, where GEM depicted fairly well the timing of the stationary diurnal maximum. This variability of model performance was related to the differences in the type of forcing between the two regions (west–east) and the two seasons (spring–summer), which indicates that some form of adaptive forecasting (e.g., convection-allowing resolution during the summer over the Rockies, and lower resolution over the eastern side of the domain) could be an alternative approach for rainfall forecasting. This issue conveys the importance of horizontal resolution in NWP (Mass et al. 2002), which will be further investigated in the follow-up paper.

We have also briefly looked at the interannual variability of the diurnal cycle, and at the differences between spring and summer. As expected, the diurnal cycle of precipitation during spring showed more interannual variability than during summer, as synoptic forcing does not directly depend on the diurnal cycle of solar heating. However, a deeper look into this issue should focus on the variability of the forcing itself, which was not attempted here.

Some shortcomings of the analysis presented herein are related to using radar-derived precipitation fields for model validation and to the limited size of our dataset. Radar estimated rain rates are affected by the variability of the ZR relationship; however, in most of our analysis we made no distinction between values exceeding 15 dBZ (except in the average rainfall rate diagrams, which are consistent with the rest), and hence we do not expect the results to be significantly affected by the variability of the ZR relationship.

Acknowledgments

The authors are grateful to an anonymous reviewer whose thorough remarks helped to considerably improve the text. This work was made possible by the support of Environment Canada to the J. S. Marshall Radar Observatory.

REFERENCES

  • Ahijevych, D. A., , R. E. Carbone, , J. D. Tuttle, , and S. B. Trier, 2001: Radar data and climatological statistics associated with warm season precipitation episodes over the continental U.S. NCAR Tech. Note TN-448+STR, 81 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307].

    • Search Google Scholar
    • Export Citation
  • Belair, S., , and J. Mailhot, 2001: Impact of horizontal resolution on the numerical simulation of a midlatitude squall line: Implicit versus explicit condensation. Mon. Wea. Rev., 129 , 23622376.

    • Search Google Scholar
    • Export Citation
  • Bluestein, H., 1993: Synoptic-Dynamic Meteorology in Mid-latitudes: Observations and Theory of Weather Systems. Vol. 2. Oxford University Press, 431 pp.

    • Search Google Scholar
    • Export Citation
  • Bryan, G. H., , J. C. Wyngaard, , and J. M. Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Mon. Wea. Rev., 131 , 23942416.

    • Search Google Scholar
    • Export Citation
  • Carbone, R. E., , and J. D. Tuttle, 2008: Rainfall occurrence in the U.S. warm season: The diurnal cycle. J. Climate, 21 , 41324146.

  • Carbone, R. E., , J. D. Tuttle, , D. A. Ahijevych, , and S. B. Trier, 2002: Inferences of predictability associated with warm season precipitation episodes. J. Atmos. Sci., 59 , 20332056.

    • Search Google Scholar
    • Export Citation
  • Casati, B., and Coauthors, 2008: Forecast verification: Current status and future directions. Meteor. Appl., 15 , 318.

  • Clark, A. J., , W. A. Gallus, , and T. C. Chen, 2007: Comparison of the diurnal precipitation cycle in convection-resolving and nonconvection-resolving mesoscale models. Mon. Wea. Rev., 135 , 34563473.

    • Search Google Scholar
    • Export Citation
  • Dai, A. G., , F. Giorgi, , and K. E. Trenberth, 1999: Observed and model-simulated diurnal cycles of precipitation over the contiguous United States. J. Geophys. Res., 104 , 63776402.

    • Search Google Scholar
    • Export Citation
  • Davis, C. A., , K. W. Manning, , R. E. Carbone, , S. B. Trier, , and J. D. Tuttle, 2003: Coherence of warm-season continental rainfall in numerical weather prediction models. Mon. Wea. Rev., 131 , 26672679.

    • Search Google Scholar
    • Export Citation
  • Fritsch, J. M., , and G. S. Forbes, 2001: Mesoscale convective systems. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 323–357.

    • Search Google Scholar
    • Export Citation
  • Gallus Jr., W. A., 2002: Impact of the verification grid-box size on warm-season QPF skill measures. Wea. Forecasting, 17 , 12961302.

  • Germann, U., , and I. Zawadzki, 2002: Scale dependence of the predictability of precipitation from continental radar images. Part I: Description of the methodology. Mon. Wea. Rev., 130 , 28592873.

    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , V. E. Kousky, , and R. J. Joyce, 2005: Diurnal cycle of precipitation determined from the CMORPH high spatial and temporal resolution global precipitation analyses. J. Geophys. Res., 110 , D23105. doi:10.1029/2005JD006156.

    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , V. J. Dagostaro, , V. E. Kousky, , and R. J. Joyce, 2007: An examination of precipitation in observations and model forecasts during NAME with emphasis on the diurnal cycle. J. Climate, 20 , 16801692.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23 , 931952.

    • Search Google Scholar
    • Export Citation
  • Knievel, J. C., , D. A. Ahijevych, , and K. W. Manning, 2004: Using temporal modes of rainfall to evaluate the performance of a numerical weather prediction model. Mon. Wea. Rev., 132 , 29953009.

    • Search Google Scholar
    • Export Citation
  • Laroche, S., , P. Gauthier, , J. St.-James, , and J. Morneau, 1999: Implementation of a 3D variational data assimilation system at the Canadian Meteorological Center. Part II: The regional analysis. Atmos.–Ocean, 37 , 281307.

    • Search Google Scholar
    • Export Citation
  • Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44 , 133149.

  • Mass, C. F., , D. Ovens, , K. Westrick, , and B. A. Colle, 2002: Does increasing horizontal resolution produce more skillful forecasts? The results of two years of real-time numerical weather prediction over the Pacific Northwest. Bull. Amer. Meteor. Soc., 83 , 407430.

    • Search Google Scholar
    • Export Citation
  • Parker, M. D., , and D. A. Ahijevych, 2007: Convective episodes in the east-central United States. Mon. Wea. Rev., 135 , 37073727.

  • Trier, S. B., , C. A. Davis, , D. A. Ahijevych, , M. L. Weisman, , and G. H. Bryan, 2006: Mechanisms supporting long-lived episodes of propagating nocturnal convection within a 7-day WRF model simulation. J. Atmos. Sci., 63 , 24372461.

    • Search Google Scholar
    • Export Citation
  • Tuttle, J. D., , and C. A. Davis, 2006: Corridors of warm season precipitation in the central United States. Mon. Wea. Rev., 134 , 22972317.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., 1975: Diurnal variations in precipitation and thunderstorm frequency over the conterminous United States. Mon. Wea. Rev., 103 , 406419.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., , W. C. Skamarock, , and J. B. Klemp, 1997: The resolution dependence of explicitly modeled convective systems. Mon. Wea. Rev., 125 , 527548.

    • Search Google Scholar
    • Export Citation
  • Wilks, D. S., 1995: Statistical Methods in Atmospheric Sciences. Academic Press, 467 pp.

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

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

Analysis domain. The dotted contours represent the coverage of the 2.5-km CAPPI maps, while the black rectangle corresponds to the domain on which all statistics are computed. The precipitation pattern observed in this figure is typical for the spring of 2008.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 2.
Fig. 2.

(a) Average hourly rain rate and (b) fractional precipitation coverage over the entire domain (32°–45°N, 103°–78°W) and time period (24 days between 18 Apr 2009 and 6 Jun 2009) for radar observations (solid line) and GEM forecasts (dashed line) as a function of forecast time (30-h forecasts initialized at 0000 UTC). The gray shading represents the first 6 h of the forecasts.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 3.
Fig. 3.

Regional diurnal cycle of average hourly rainfall rate over the subdomains presented in Fig. 1 for the entire time period (24 days between 18 Apr 2008 and 6 Jun 2008) for radar observations (solid line) and GEM forecasts (dashed line). The order of the graphs corresponds to the geographic position of the domains.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 4.
Fig. 4.

As in Fig. 3, but for fractional precipitation coverage.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 5.
Fig. 5.

Hovmöller diagrams of the average diurnal cycle of precipitation for the spring of 2008 (16 Apr–6 Jun 2009) for the (a),(b) average hourly rain rate and (c),(d) precipitation coverage for a threshold equivalent to 15 dBZ for (a),(c) radar observations and (b),(d) GEM forecasts in the range of 6–30 h. All values are normalized with respect to the daily mean, which is indicated in the bottom right corner of each diagram (mm h−1 in the case of the average hourly rain rate). The diagrams have been repeated twice along 0600 UTC for clarity. The graphs have been smoothed over a window of 96 km in longitude and 3 h in time.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 6.
Fig. 6.

Hovmöller time series of average hourly rain rate (mm h−1) for 16 Apr–14 May 2008, averaged over the latitudinal range 32°–45°N. The gray gaps represent the days when data were not available.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 7.
Fig. 7.

As in Fig. 6, but for 14 May–11 Jun 2008.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 8.
Fig. 8.

Rainfall event corresponding to 7 May 2008 as depicted by (left) radar observations and (right) GEM forecasts. The black rectangle denotes the analysis domain and the radar contours are overlaid on the forecast image for comparison.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 9.
Fig. 9.

Normalized power spectrum of the Hovmöller diurnal cycle (Fig. 5) of (a),(b) average hourly rain rate and (c),(d) precipitation coverage over a threshold equivalent to 15 dBZ for the spring of 2008 for (a),(c) radar observations and (b),(d) GEM forecasts. The scale represents fraction of the variance.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 10.
Fig. 10.

Phase of the (a),(b) 24- and (c),(d) 12-h harmonics as function of longitude for radar (black) and model (gray) in terms of both (left) average rainfall intensity and (right) precipitation coverage for the spring of 2008. The dotted line means that the harmonic does not explain 10% of the observed variance.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 11.
Fig. 11.

Average diurnal cycle of precipitation for summer 2008 for radar observations of (a) precipitation intensity and (b) precipitation coverage. The values are normalized with respect to the mean, which is indicated in the bottom right corner of (a),(b) (0.08 mm h−1 for average rainfall intensity, and 4% for precipitation coverage). The white ovals highlight the semidiurnal mode. The diagrams have been repeated twice along 0600 UTC for clarity.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 12.
Fig. 12.

Normalized power spectrum of the Hovmöller diurnal cycle (Fig. 11) of the (a) average rainfall rate and (b) precipitation coverage for radar observations during the summer of 2008.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 13.
Fig. 13.

As in Fig. 10, but for the summer of 2008.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 14.
Fig. 14.

Average diurnal cycle of rainfall intensity for (a),(c) spring and (b),(d) summer 2008, computed over the (top) northern and (bottom) southern halves of the domain (delimited at 38.5°N). The values are normalized with respect to the mean, which is indicated in the bottom right corner of (a)–(d) in mm h−1. The white dashed lines highlight features discussed in the text.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 15.
Fig. 15.

(a) The diurnal cycle of GEM-simulated average rainfall intensity (in normalized units) and (b) the corresponding normalized power spectrum for the summer of 2008. The white ovals highlight features discussed in text.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 16.
Fig. 16.

As in Fig. 14, but for GEM forecasts. GEM-radar correlation coefficients computed in Hovmöller space are indicated in the bottom-left corner of each diagram.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 17.
Fig. 17.

CSI computed for a threshold of 15 dBZ as a function of forecast time for the spring (solid line) and summer (dashed line) of 2008 over the six subdomains presented in Fig. 1.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 18.
Fig. 18.

The diurnal cycle of average hourly rain rate for the (a) spring and (b) summer averaged for 12 yr between 1996 and 2007.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Fig. 19.
Fig. 19.

The year-to-year variability of the diurnal cycle of average hourly rain rate shown in Fig. 18, expressed as percentage of the 1996–2007 mean.

Citation: Monthly Weather Review 138, 8; 10.1175/2010MWR3125.1

Save