• Alexander, G. D., , J. A. Weinman, , V. M. Karyampudi, , W. S. Olson, , and A. C. Lee, 1999: The effect of assimilating rain rates derived from satellites and lightning on the forecasts of the 1993 superstorm. Mon. Wea. Rev., 127 , 14331457.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. E., , and K. E. Mitchell, 1998: Progress on the NCEP hourly multi-sensor U.S. precipitation analysis for operations and GCIP research. Preprints, Second Symp. on Integrated Observing Systems, Phoenix, AZ, Amer. Meteor. Soc., 10–11.

  • Benjamin, S. G., and Coauthors, 2004: An hourly assimilation–forecast cycle: The RUC. Mon. Wea. Rev., 132 , 495518.

  • Benjamin, S. G., , S. S. Weygandt, , S. E. Koch, , and J. M. Brown, 2006: Assimilation of lightning data into RUC model convection forecasting. Preprints, Second Conf. on Meteorological Applications of Lightning Data, Atlanta, GA, Amer. Meteor. Soc., CD-ROM, 4.3.

  • Betts, A. K., 1986: A new convective adjustment scheme. Part I: Observational and theoretical basis. Quart. J. Roy. Meteor. Soc., 112 , 677692.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , and M. J. Miller, 1986: A new convective adjustment scheme. Part II: Single column tests using GATE wave, BOMEX, and arctic air-mass data sets. Quart. J. Roy. Meteor. Soc., 112 , 693709.

    • Search Google Scholar
    • Export Citation
  • Boccippio, D. J., , K. L. Cummins, , H. J. Christian, , and S. J. Goodman, 2001: Combined satellite- and surface-based estimation of the intracloud–cloud-to-ground lightning ratio over the continental United States. Mon. Wea. Rev., 129 , 108122.

    • Search Google Scholar
    • Export Citation
  • Bukovsky, M. S., , J. S. Kain, , and M. E. Baldwin, 2006: Bowing convective systems in a popular operational model: Are they for real? Wea. Forecasting, 21 , 307324.

    • Search Google Scholar
    • Export Citation
  • Chang, D-E., , J. A. Weinman, , C. A. Morales, , and W. S. Olson, 2001: The effect of spaceborne microwave and ground-based continuous lightning measurements on forecasts of the 1998 Groundhog Day storm. Mon. Wea. Rev., 129 , 18091833.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1996: Sensitivity of moist convection forced by boundary layer processes to low-level thermodynamic fields. Mon. Wea. Rev., 124 , 17671785.

    • Search Google Scholar
    • Export Citation
  • Cummins, K. L., , M. J. Murphy, , E. A. Bardo, , W. L. Hiscox, , R. B. Pyle, , and A. E. Pifer, 1998: A combined TOA/MDF technology upgrade of the U.S. national lightning detection network. J. Geophys. Res., 103 , 90359044.

    • Search Google Scholar
    • Export Citation
  • Errico, R. M., , K. D. Raeder, , and L. Fillion, 2003: Examination of the sensitivity of forecast precipitation rates to possible perturbations of initial conditions. Tellus, 55A , 88105.

    • Search Google Scholar
    • Export Citation
  • Fillion, L., , and S. Bélair, 2004: Tangent linear aspects of the Kain–Fritsch moist convective parameterization scheme. Mon. Wea. Rev., 132 , 24772494.

    • Search Google Scholar
    • Export Citation
  • Fillion, L., , M. Tanguay, , N. Ek, , C. Pagé, , and S. Pellerin, 2005: Balanced coupling between vertical motion and diabatic heating for variational data assimilation. Preprints, Symp. on Nowcasting and Very Short Range Forecasting, Toulouse, France, World Weather Research Programme, CD-ROM, 3.10. [Available online at http://www.meteo.fr/cic/wsn05/DVD/index.html.].

  • Fritsch, J. M., , and C. F. Chappell, 1981: Preliminary numerical tests of the modification of mesoscale convective systems. J. Appl. Meteor., 20 , 910921.

    • Search Google Scholar
    • Export Citation
  • Fulton, R. A., , J. P. Breidenbach, , D. J. Seo, , D. A. Miller, , and T. O’Bannon, 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13 , 377395.

    • Search Google Scholar
    • Export Citation
  • Gallus Jr., W. A., , and M. Segal, 2001: Impact of improved initialization of mesoscale features on convective system rainfall in 10-km Eta simulations. Wea. Forecasting, 16 , 680696.

    • Search Google Scholar
    • Export Citation
  • Hodur, R. M., 1997: The Naval Research Laboratory’s Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS). Mon. Wea. Rev., 125 , 14141430.

    • Search Google Scholar
    • Export Citation
  • Holle, R. L., , and M. W. Maier, 1982: Radar echo height related to cloud-ground lightning in south Florida. Preprints, 12th Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 330–333.

  • Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122 , 927945.

    • Search Google Scholar
    • Export Citation
  • Jones, C. D., , and B. Macpherson, 1997: A latent heat nudging scheme for the assimilation of precipitation into an operational mesoscale model. Meteor. Appl., 4 , 269277.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1992: The role of the convective “trigger function” in numerical forecasts of mesoscale convective systems. Meteor. Atmos. Phys., 49 , 93106.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Covection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

  • Lhermitte, R., , and P. R. Krehbiel, 1979: Doppler radar and radio observations of thunderstorms. IEEE Trans. Geosci. Electron., 17 , 162171.

    • Search Google Scholar
    • Export Citation
  • MacGorman, D. R., , and W. D. Rust, 1998: The Electrical Nature of Storms. Oxford University Press, 422 pp.

  • Marshall, C. H., , K. C. Crawford, , K. E. Mitchell, , and D. J. Stensrud, 2003: The impact of the land surface physics in the operational NCEP Eta model on simulating the diurnal cycle: Evaluation and testing using Oklahoma Mesonet data. Wea. Forecasting, 18 , 748768.

    • Search Google Scholar
    • Export Citation
  • Papadopoulos, A., , T. G. Chronis, , and E. N. Anagnostou, 2005: Improving convective precipitation forecasting through assimilation of regional lightning measurements in a mesoscale model. Mon. Wea. Rev., 133 , 19611977.

    • Search Google Scholar
    • Export Citation
  • Pereira Fo, A. J., , K. C. Crawford, , and D. J. Stensrud, 1999: Mesoscale precipitation fields. Part II: Hydrometeorologic modeling. J. Appl. Meteor., 38 , 102125.

    • Search Google Scholar
    • Export Citation
  • Rison, W., , R. J. Thomas, , P. R. Krehbiel, , T. Hamlin, , and J. Harlin, 1999: A GPS-based three-dimensional lightning mapping system: Initial observations in central New Mexico. Geophys. Res. Lett., 26 , 35733576.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, cited. 1998: Changes to the NCEP operational “early” Eta analysis/forecast system. NWS Tech. Procedures Bull. 447, National Oceanic and Atmospheric Administration/National Weather Service, 36 pp. [Available online at http://www.nws.noaa.gov/om/tpb/447.htm.].

  • Rogers, R. F., , J. M. Fritsch, , and W. C. Lambert, 2000: A simple technique for using radar data in the dynamic initialization of a mesoscale model. Mon. Wea. Rev., 128 , 25602574.

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

  • Stensrud, D. J., 1996: Effects of persistent, midlatitude mesoscale regions of convection on the large-scale environment during the warm season. J. Atmos. Sci., 53 , 35033527.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J-W. Bao, 1992: Behaviors of variational and nudging assimilation techniques with a chaotic low-order model. Mon. Wea. Rev., 120 , 30163028.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J. M. Fritsch, 1994a: Mesoscale convective systems in weakly forced large-scale environments. Part II: Generation of a mesoscale initial condition. Mon. Wea. Rev., 122 , 20682083.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J. M. Fritsch, 1994b: Mesoscale convective systems in weakly forced large-scale environments. Part III: Numerical simulations and implications for operational forecasting. Mon. Wea. Rev., 122 , 20842104.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , G. S. Manikin, , E. Rogers, , and K. E. Mitchell, 1999: Importance of cold pools to NCEP Mesoscale Eta model forecasts. Wea. Forecasting, 14 , 650670.

    • Search Google Scholar
    • Export Citation
  • Thomas, R. P., , P. R. Krehbiel, , W. Rison, , S. Hunyady, , W. Winn, , T. Hamlin, , and J. Harlin, 2004: Accuracy of the lightning mapping array. J. Geophys. Res., 109 .D14207, doi:10.1029/2004JD004549.

    • Search Google Scholar
    • Export Citation
  • Warner, T. T., , and H-M. Hsu, 2000: Nested-model simulation of moist convection: The impact of coarse-grid parameterized convection on fine-grid resolved convection. Mon. Wea. Rev., 128 , 22112231.

    • Search Google Scholar
    • Export Citation
  • Ziegler, C. L., 1999: Issues in forecasting mesoscale convective systems: An observational and modeling perspective. Storms, R. Pielke Jr. and R. Pielke Sr., Eds., Vol. 2, Routledge Press, 26–42.

    • Search Google Scholar
    • Export Citation
  • Ziegler, C. L., , W. J. Martin, , R. A. Pielke Sr., , and R. L. Walko, 1995: A modeling study of the dryline. J. Atmos. Sci., 52 , 263285.

  • Ziegler, C. L., , T. J. Lee, , and R. A. Pielke Sr., 1997: Convective initiation at the dryline: A modeling study. Mon. Wea. Rev., 125 , 10011026.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The decision process for each grid column during assimilation.

  • View in gallery

    Lightning data for the 15-in period starting at 2145 UTC 7 Jul 2000. (a) The CG detections by the NLDN. (b) Source points from the STEPS LMA. Contour levels are 1, 5, 10, 20, 50, 100, 250, 500, and 1000 counts per grid box per 15 min. [The lowest level in (a) is 0.8 to make single flashes visible]

  • View in gallery

    The coarse (90 km), intermediate (30 km), and fine (10 km) mesh areas of the model domain. The circle indicates the approximate range of the LMA.

  • View in gallery

    Surface analysis at 0000 UTC 21 Jul 2000, the starting time for the forecast period. Conventional station model format includes temperature (°C) over dewpoint (°C), mean sea level pressure (mb) at upper right, and wind barb (full barb = 5 m s−1, half barb = 2.5 m s−1). Instantaneous base radar reflectivity is shown by gray fill. The site of the DDC sounding discussed in the text is indicated by DDC.

  • View in gallery

    Observed and modeled total precipitation (mm) for a 6-h period starting 0600 UTC 20 Jul 2000 during the spinup period. (a) Rain gauge data with sampled NLDN strikes. (The 1st and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast (no lightning assimilation). (c) With assimilation of CG lightning from NLDN, moisture forcing, and full suppression. (d) With assimilation of both CG and total (LMA) lightning, moisture forcing, full suppression, and 25% feedback of KF precipitation.

  • View in gallery

    Modeled total precipitation (mm) for 0600–1200 UTC 20 Jul 2000 (during the spinup period). These simulations have no precipitation feedback from the CPS to the resolved scale. (a) Same as in Fig. 5b (control run). (b) Normal KF trigger with forcing from lightning assimilation. (c) Same as in (b) adding nudging of boundary layer moisture. (d) Same as in (c), but suppresses the KF trigger where no lightning occurred (no precipitation feedback). The region of heaviest rainfall in (d) has a substantial resolved-scale contribution.

  • View in gallery

    Air temperature (2 m) at 0000 UTC 21 Jul 2000, the beginning of the forecast period. (a) Warm start with 24-h spinup, 12-hourly data update cycle. (b) The 24-h assimilation of NLDN with only full suppression and 25% feedback of precipitation from KF to the resolved scale. (c) The 24-h assimilation of all lightning (NLDN and LMA) data to force convection (no suppression of KF nor feedback of precipitation). (d) Same as in (b), but with assimilation of both NLDN and LMA data.

  • View in gallery

    National Weather Service sounding for DDC at 0000 UTC 21 Jul 2000.

  • View in gallery

    Model soundings for DDC at 0000 UTC 21 Jul 2000. Solid gray and black curves show dewpoint and potential temperature, respectively, of the 12-h forecast. Long dashed curves are the MVOI-adjusted values for the initialization of the next forecast cycle. Black wind barbs are from the forecast, gray barbs are from the subsequent analysis adjustment. (a) Control run (no lightning assimilation). (b) Simulation with assimilation of lightning (NLDN and LMA), full suppression, and 25% feedback of precipitation from the KF scheme to the resolved scale.

  • View in gallery

    Observed and forecast total precipitation (mm) for 0000–0600 UTC 21 Jul 2000. (a) Rain gauge data with sampled NLDN strikes. (The first and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast from standard warm start (no lightning assimilation). (c) Forecast from initialization with assimilation of NLDN and LMA data, moisture forcing, and no suppression nor feedback. (d) Forecast from initialization with assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback.

  • View in gallery

    Observed and forecast total precipitation (mm) for 0600–1200 UTC 21 Jul 2000. (a) Rain gauge data with sampled NLDN strikes. (The 1st and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast from standard warm start (no lightning assimilation). (c) Forecast from initialization with assimilation of NLDN and LMA data, moisture forcing, and no suppression nor feedback. (d) Forecast from initialization with assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback.

  • View in gallery

    Radar composites and convective precipitation forecasts on 21 Jul 2000. (a) Radar composite at 0100 UTC. (b) The 0000–0100 UTC precipitation forecast from initial condition generated by assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback. (c) The 0000–0100 UTC control forecast from standard warm start (no lightning assimilation). (d), (e), (f) Same as in (a), (b), (c) but for 0200 (0100–0200) UTC. Observed and forecast cold pool boundaries are overlaid, the latter determined from the forecast surface temperature field associated with convective rainfall cores. For reference, reflectivity at 0000 UTC was depicted in Fig. 4.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 48 48 9
PDF Downloads 45 45 10

A Lightning Data Assimilation Technique for Mesoscale Forecast Models

View More View Less
  • 1 Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, and NOAA/OAR/National Severe Storms Laboratory, Norman, Oklahoma
  • | 2 NOAA/OAR/National Severe Storms Laboratory, Norman, Oklahoma
© Get Permissions
Full access

Abstract

Lightning observations have been assimilated into a mesoscale model for improvement of forecast initial conditions. Data are used from the National Lightning Detection Network (cloud-to-ground lightning detection) and a Lightning Mapping Array (total lightning detection) that was installed in western Kansas–eastern Colorado. The assimilation method uses lightning as a proxy for the presence or absence of deep convection. During assimilation, lightning data are used to control the Kain–Fritsch (KF) convection parameterization scheme. The KF scheme can be forced to try to produce convection where lightning indicated storms, and, conversely, can optionally be prevented from producing spurious convection where no lightning was observed. Up to 1 g kg−1 of water vapor may be added to the boundary layer when the KF convection is too weak. The method does not employ any lightning–rainfall relationships, but rather allows the KF scheme to generate heating and cooling rates from its modeled convection. The method could therefore easily be used for real-time assimilation of any source of lightning observations. For the case study, the lightning assimilation was successful in generating cold pools that were present in the surface observations at initialization of the forecast. The resulting forecast showed considerably more skill than the control forecast, especially in the first few hours as convection was triggered by the propagation of the cold pool boundary.

Corresponding author address: Edward Mansell, NOAA/NSSL/National Weather Center, 120 David L. Boren Blvd., Norman, OK 73072. Email: mansell@ou.edu

Abstract

Lightning observations have been assimilated into a mesoscale model for improvement of forecast initial conditions. Data are used from the National Lightning Detection Network (cloud-to-ground lightning detection) and a Lightning Mapping Array (total lightning detection) that was installed in western Kansas–eastern Colorado. The assimilation method uses lightning as a proxy for the presence or absence of deep convection. During assimilation, lightning data are used to control the Kain–Fritsch (KF) convection parameterization scheme. The KF scheme can be forced to try to produce convection where lightning indicated storms, and, conversely, can optionally be prevented from producing spurious convection where no lightning was observed. Up to 1 g kg−1 of water vapor may be added to the boundary layer when the KF convection is too weak. The method does not employ any lightning–rainfall relationships, but rather allows the KF scheme to generate heating and cooling rates from its modeled convection. The method could therefore easily be used for real-time assimilation of any source of lightning observations. For the case study, the lightning assimilation was successful in generating cold pools that were present in the surface observations at initialization of the forecast. The resulting forecast showed considerably more skill than the control forecast, especially in the first few hours as convection was triggered by the propagation of the cold pool boundary.

Corresponding author address: Edward Mansell, NOAA/NSSL/National Weather Center, 120 David L. Boren Blvd., Norman, OK 73072. Email: mansell@ou.edu

1. Introduction

Recent studies have shown that forecasts can be improved by incorporating the effects of deep convection during the initialization period of mesoscale forecast models. For example, based on model experiments that used subjective analyses to improve initial conditions, Stensrud and Fritsch (1994a) suggested that forecast skill could be enhanced by using data assimilation procedures that include “the effects of parameterized convection, as indicated by radar or satellite during the assimilation period . . .” as well as explicit representation of boundary layer cold pools from ongoing storms as diagnosed from surface observations. Stensrud and Fritsch (1994b) demonstrated that explicitly introducing storm-induced cold pools into the mesoscale initial condition improved the mesoscale quantitative precipitation forecast by focusing the triggering of ongoing convection forced by those cold pools. It is recognized, however, that data assimilation is not a panacea for all problems of forecast models. The greatest improvements in forecasts from assimilating data that depict convection should occur in environments where storms have a significant impact on near-future convection and the mesoscale environment of the convection, such as by generation of outflow boundaries and mesoscale upper-tropospheric outflow (anvil) plumes (as in the case studied by Stensrud and Fritsch 1994b).

A later study by Stensrud et al. (1999) found that initializing cold pools into the forecast resulted in minor changes except in one case with weaker large-scale forcing. That study used the Betts–Miller–Janjić (BMJ) convection scheme (Betts 1986; Betts and Miller 1986; Janjić 1994), which responds weakly to cold pool effects. Gallus and Segal (2001) also tested cold pool initialization with both the BMJ and the Kain–Fritsch (KF; Kain and Fritsch 1993) schemes and found that it tended to worsen results with BMJ but improved the skill with KF for the two lowest precipitation amounts. Gallus and Segal (2001) found the most improvement from adding water vapor to dry layers in the low to middle troposphere where convection was indicated by radar.

Although static initialization of cold pools seems to have low impact generally, a dynamic initialization through data assimilation may improve results by allowing the model to respond to the cold outflows in a preforecast period. Although Stensrud and Fritsch (1994a) suggested assimilating radar or satellite data, it is possible to use any type of data that reveals the location of convection. Lightning data satisfy this requirement and have the following additional advantages: compactness (i.e., low bandwidth) and reliability (real-time availability with few gaps); ability to unambiguously locate deep convection; detection in mountainous areas and beneath high clouds; and long-range detection of storms over oceans beyond radar network coverage. Furthermore, technologies capable of delineating lightning activity over the entire earth, including over all oceans, have already been demonstrated. Thus, techniques for assimilating lightning data could be applied in extensive regions where radar coverage does not exist, such as over oceans.

Relatively little has been done, however, to develop techniques for assimilating lightning data. Two studies (Alexander et al. 1999; Chang et al. 2001) demonstrated an improvement in the 12–24-h forecast of rainfall and location of convection in an intense extratropical cyclone when lightning data were assimilated along with other satellite data during model initialization. Their assimilation scheme used occasional microwave data from a low-earth-orbiting satellite to estimate the amount of rainfall per cloud-to-ground flash, used this relationship to estimate convective rainfall during all assimilation times, converted rainfall to latent heating rates, and then used latent heating to nudge the model (Jones and Macpherson 1997). This assimilation significantly improved the forecast for the case study. Lightning–rainfall relationships, however, can vary by more than an order of magnitude in warm-season continental storms and by several orders of magnitude for storms in different climatological regimes (e.g., MacGorman and Rust 1998, 225–229), requiring that this method of assimilating lightning data be calibrated for each day and region in which it is applied.

More recently, Papadopoulos et al. (2005) assimilated ground-strike lightning data to nudge relative humidity through the use of sample profiles of humidity associated with deep convection. The study used a version of the Eta model (see, e.g., Janjić 1994; Rogers et al. 1998) with the BMJ scheme. The humidity was adjusted in proportion to the ground flash rate reported by the ZEUS long-range detection system in Europe. During the assimilation periods for three cases, the nudging technique generally resulted in improved precipitation rates, especially for larger thresholds. Bias increased at most thresholds, which worsened the high (>1) control bias at lower rates (6-h accumulations of 12 mm or less), but improved the low bias (<1) at higher thresholds, so that skill was improved especially at higher rates. Forecasts of 6 and 12 h also showed similar trends compared to a pure forecast with no assimilation. The method of Papadopoulos et al. (2005) is roughly analogous to a simple initialization technique tested by Gallus and Segal (2001), who found similar operational Eta Model forecast impacts with the BMJ scheme when the model humidity was increased where organized radar reflectivity was observed.

The Rapid Update Cycle (RUC) model cloud analysis (Benjamin et al. 2004) now includes a preliminary use of lightning data (Benjamin et al. 2006). The scheme estimates rainfall per lightning flash as a supplement to other estimated hourly rainfall rates. The estimated rain rates are then passed to the model, which tries to activate convection for sufficiently high rainfall rates. The actual impact of the lightning data, however, has not been thoroughly investigated (S. Weygandt 2006, personal communication).

The present study uses an approach similar to those recently developed for assimilating radar data (e.g., Rogers et al. 2000) by allowing lightning data to control the “trigger function” of a convective parameterization scheme during the assimilation period leading up to the forecast period. The focus of this assimilation research is to use lightning data to activate or deactivate subgrid-scale, deep, moist convection during the data assimilation cycle of the mesoscale model. This is particularly important in situations in which past convection has modified the troposphere on scales anywhere from storm scale through synoptic scale in ways that influence the subsequent evolution of convection [e.g., by moistening the boundary layer, forming surface cold pools, or modifying synoptic troughs; Stensrud (1996)]. The timing and location of convection may have a significant adverse affect on the forecast (Fritsch and Chappell 1981). For example, Stensrud and Bao (1992) compared a convective parameterization trigger to a decision point in a chaotic system, and Kain and Fritsch (1992) demonstrated the dramatic differences in rainfall estimates that can result from different trigger schemes. Therefore, corrected triggering of convection alone should have a positive impact on a forecast.

The purpose of this study is to demonstrate a new lightning assimilation technique that can introduce significant changes in the initial condition produced by the mesoscale model spinup period. Furthermore, the assimilation has a substantial positive impact on the model forecast of subsequent convective activity. The forecast time in the case study was chosen as a time when significant convection was already present, which is a challenge to initiate statically.

2. Assimilation method

The method of lightning assimilation is similar to the technique used by Rogers et al. (2000), who used radar data to determine the occurrence of convection. Lightning observations in the present technique are similarly used to control the activation of the convection parameterization scheme (CPS), which in the present work is the KF scheme. This method uses the forecast model’s physics to estimate the effects (including latent heating) of the deep convection inferred from lightning. This differs from the method of Alexander et al. (1999) and Chang et al. (2001), who used satellite data to estimate the rainfall per cloud-to-ground flash during the assimilation period, and then used the cloud-to-ground flash rates to determine a rate of latent heat release. Their use of latent heating replaces the convection parameterization scheme during the assimilation period.

The general outline of the decision process for assimilation is shown in Fig. 1. At 10-min intervals of model time, each grid column is checked for activation of the KF scheme. If the KF scheme is not active, the model decides whether or not it needs to be activated. An input threshold Tflash (with units of number of flashes per time interval per grid cell) is used to determine whether the observed lightning rate is locally high enough to infer the presence of deep convection. The lightning data could also be filtered for noise in the gridding process. (In future applications, Tflash could be made dependent on the grid spacing, as more noise points could be accumulated in a larger box.) If Tflash is met or exceeded during the assimilation period, but KF is not active, then an attempt is made to force KF to activate. Conversely, if the lightning counts are below Tflash, then KF may be hindered or completely prevented from activating, according to the selected level of suppression.

The KF trigger function tests successive mixed layers of air for instability, starting at the surface, using enough model levels to have a minimum pressure depth of 60 mb. A mixed parcel is given some upward momentum to see if it can reach its level of free convection (LFC). If it can, then the KF cloud model determines the cloud depth as the difference between the lifting condensation level (LCL) and the level of zero upward momentum for that mixed parcel. The standard scheme requires a minimum cloud depth of 4 km to produce precipitation (i.e., 4 km is the threshold for deep convection). [The version of KF used here did not have the shallow (nonprecipitating) convection component that is available in more recent versions.] The KF scheme uses a one-dimensional cloud model to determine the vertical distribution of condensational latent heating and evaporative cooling rates. It uses a stability-based closure assumption to determine total heating, drying, and precipitation rates. The scheme includes entrainment of environmental air and detrainment to the environment.

If convection is indicated in a grid column by lightning during assimilation, the most unstable mixed parcel in that column is found and forced to its LFC by ignoring any negative buoyancy (convective inhibition) and entrainment below the LFC. Updrafts in storms that produce lightning, however, must be strong enough to extend well above the freezing level to produce the graupel that is necessary for strong electrification (e.g., Lhermitte and Krehbiel 1979; Holle and Maier 1982). Therefore, to attempt to maintain consistency with the presence of lightning, an option was added to increase moisture in the parcel source layers by small increments (up to 1 g kg−1 as in Rogers et al. 2000) as necessary to produce a minimum cloud depth of 7 km and peak updraft of 10 m s−1. The depth and updraft thresholds were chosen as reasonable values that would be attainable on average with moisture adjustments of less than 1 g kg−1. The depth criterion was more easily attained (about 98% of the time and requiring less than 0.1 g kg−1 on average) than the minimum updraft speed threshold. On average, a moisture increase of 1 g kg−1 produced an updraft speed increase of about 3.5 m s−1, and the 10 m s−1 threshold was reached for about only about half of the columns that were forced, with an average moisture addition of 0.85 g kg−1.

The decision to increase moisture rather than adjust temperature to enhance convection follows from Crook (1996), who found that boundary layer moisture had a larger impact on mature convection than temperature. For conditions marginal for storm initiation, however, Crook (1996) found a greater sensitivity to boundary layer temperature. Variational data assimilation studies have also found forecast sensitivities to temperature and moisture. For example, Fillion and Bélair (2004) noted that the KF scheme was more responsive to low-level moisture than temperature, but Errico et al. (2003) found a primary influence of temperature perturbations on cyclone development using a non-KF convection scheme. Thus, one could also attempt to modify convection through the temperature or a combination of moisture and temperature in a future study.

For the case in which lightning is not observed in a grid column, three options were created for suppressing KF during lightning data assimilation: no suppression (s0), partial suppression (s1), and complete suppression (s2). With no suppression, the KF scheme is allowed to run without interference. Choosing the second option (s1) diminishes the likelihood of KF activation by limiting the “boost” given to parcels by the trigger function (thereby making it harder to reach the LFC) and by reducing the width of the KF updraft, thereby increasing entrainment and dilution of updraft air. By choosing the third option (s2), any grid column in which deep convection is not indicated by lightning is simply skipped by KF; the KF scheme is not allowed to run at all in that column.

A final option allowed for feedback of some convective precipitation to the resolved scale. This idea was suggested by J. Kain (2004, personal communication) as a possible means to generate stronger cold pools though evaporation in the resolved-scale microphysics. When activated, a fraction (typically set at 25%) of the KF-generated rain and snow at each level in the column is transferred to the resolved-scale grid. (Since only an array for rain was available, snow is actually melted first into rain, with appropriate latent heating effect. The rain can then be refrozen by the microphysics scheme.) Feedback is enabled during assimilation only where lightning was observed.

3. Data sources

Lightning observations were taken from two platforms: 1) the National Lightning Detection Network (NLDN; Cummins et al. 1998), which detects cloud-to-ground lightning over the 48 contiguous states to an accuracy of 500 m; and 2) the Lighting Mapping Array (LMA; Rison et al. 1999; Thomas et al. 2004), which operated in northwestern Kansas and northeastern Colorado during the Severe Thunderstorm Electrification and Precipitation Study (STEPS) field program in the summer of 2000. The LMA locates sources of very high frequency (VHF) radio emissions from both intracloud (IC) and cloud-to-ground (CG) lightning flashes, but does not automatically distinguish between the two, nor does it automatically group source points into flash events. Location accuracy varies from about 20 m above the network to on the order of 500 m at 100 km from the network center (Thomas et al. 2004). Each lightning flash may generate tens to thousands of source points in the LMA data.

The NLDN and the LMA provide point data that must be gridded for ingest by the model for the present assimilation scheme. The altitude information in the LMA data are ignored at present (i.e., “2D mode”), since future satellite-based optical lightning detectors will not have altitude information and therefore an algorithm incorporating altitude would not translate to such a data source. The two lightning data sources are each gridded into separate arrays that match the domains of the nested grid configuration (discussed in section 4). Data are accumulated for 15-min periods over a full 12-h update cycle, and each detected lightning point (from either the NLDN or LMA) simply increments the count in the grid box in which it falls. A 15-min data integration period was chosen because it gave good temporal resolution while providing enough samples to alleviate the patchiness that can result from gridding point data.

The KF scheme was modulated on the basis of observed lightning activity within 15 min of the model time (i.e., aggregate two 15-min periods to control the KF routine). (A typical time scale for KF convection is 15–30 min.) For NLDN data, the threshold Tflash to force KF was set at 1 strike per grid box during the look-ahead period. In the future, it may be desirable to use a threshold of 2 to avoid occasional activation of KF by spurious noise. For LMA data, Tflash was set at 10 points per grid box per look-ahead period, which was sufficient for removing noise points. Point sources with greater uncertainty were also removed at the gridding stage by requiring a minimum of seven stations and a reduced chi square (goodness of fit) of less than 5.0.

The NLDN has the important advantage of large area coverage, but detects primarily CG lightning with very limited IC lightning detection. Cloud-to-ground flashes make up only a fraction of total lightning (averaging roughly 25% nationally, but 10%–15% over the inner grid used in this study; e.g., Boccippio et al. 2001). The coverage of the NLDN makes it an excellent platform for determining the occurrence of deep electrified convection, especially of long-lived large systems that produce many CG flashes. The LMA, on the other hand, detects total lightning activity (tens to hundreds of points per individual flash), but provides 2D locations only out to roughly 200 km from the network center. (In the STEPS field program, the network center was in far northwestern Kansas.)

An example of NLDN and LMA data for a 15-min period during the case study period illustrates typical differences in the detail of the ongoing convection available from each source, as well as the spatial coverage of the two networks (Fig. 2). The LMA data have far greater detail, giving a better picture of the electrical intensity, cellular structure, and coverage of individual storms within the LMA detection range (eastern Colorado, western Kansas, and southwestern Nebraska). The NLDN indicated storms in central Kansas and northern New Mexico that were out of LMA coverage. Storms in the high plains region of the United States tend to have a lower percentage of CG flashes than the U.S. average (e.g., Boccippio et al. 2001), so the difference in this case may be greater than what is typical for other regions. Since the first flashes in storms are usually IC discharges, the LMA can determine the timing of initial strong electrification more accurately than the NLDN can.

4. Model setup and initialization

The lightning assimilation technique was developed for and applied to version 2 of the Coupled Ocean– Atmosphere Mesoscale Prediction System (COAMPS1) mesoscale model software (Hodur 1997) in research mode. All model runs in the present study employed a (conterminous United States) CONUS-scale outer grid and two finer-scale nested domains (Fig. 3) having grid spacings of 90, 30, and 10 km. Thus, “resolvable scale” on the innermost grid implies the full representation of meso-β-scale (about 20–200 km) circulations associated with a forecasted mesoscale convective system (MCS; Ziegler 1999). The chosen innermost grid contained the STEPS program region and most of the area affected by the observed convection. All simulations had 30 sigma-z levels, with the uppermost mass point at 31.05 km and the uppermost w point at 34.8 km.

The KF CPS was enabled on all grids, and COAMPS was initiated at 0000 UTC 20 July 2000 (cold start) from analyses, with boundary conditions from the Navy Operational Global Atmospheric Prediction System (NOGAPS). The 24-h spinup period was performed for all forecasts, including a 12-hourly ingest of atmospheric observations via the built-in multivariate optimal interpolation (MVOI). Data sources for MVOI include automated aircraft reports, radiosondes, pilot balloons, surface data (land and sea), and satellite data. For all experiments other than the control run, lightning data assimilation options were enabled during the spinup period. Assimilation of NLDN data was always enabled on the outermost grid, and the KF trigger was never suppressed on the outermost grid, as it extended beyond the range of the NLDN. The middle grid, however, always had the same KF suppression option as the innermost grid. Due to the limited spatial coverage of the LMA, its data were assimilated only on the innermost grid and always in combination with NLDN data. A 12-h pure forecast was then initiated from warm-start conditions at 0000 UTC 21 July 2000.

The default COAMPS model version deactivates advection and turbulent mixing of rain and snow for horizontal grid spacings greater than 9 km (i.e., at scales at which the CPS operates). Because some of the experiments inject rain and snow directly from the KF scheme to the resolved scale, advection and mixing were enabled for snow and rain at all scales and for all simulations. This maintained consistent model physics between the control and experimental forecasts.

5. Case study

The lightning assimilation method was tested with a case from July 2000 in the U.S. central plains. The STEPS field program operated in the region of western Kansas, eastern Colorado, and southwest Nebraska, and coverage by an LMA extended approximately 200 km from a point near the Kansas–Colorado border (Fig. 3). Widespread convection occurred on each of three successive days (20–22 July 2000). On each day, convection initiated in Colorado and/or Nebraska and developed into convective systems that traversed Kansas into Oklahoma, Missouri, and Arkansas. Convection also developed in a similar manner on 19 July, but was not as extensive or long lived.

Since the major objective of the study was to improve the forecast initial condition through the generation of cold outflow boundaries and dynamic circulations from previous convection, a 24-h assimilation spinup period was run from 0000 UTC 20 July to 0000 UTC 21 July 2000 (verified in Fig. 4). Deep convection had initiated in eastern Colorado by 0000 UTC 20 July, and squall lines had developed in Nebraska and Kansas by 0600 UTC. By 1200 UTC, a large system covered southeastern Kansas and parts of Oklahoma and Missouri. The system moved into Missouri and Arkansas by 1600 UTC, and new storms began forming in Colorado and Kansas by 2000 UTC. A vigorous system was in place in northeastern Colorado by 0000 UTC 21 July, with convection also evident in southern Colorado and north-central Kansas–south-central Nebraska (Fig. 4). The spinup period thus included both earlier convection and new, ongoing convection and a combination of old and new outflow boundaries (Fig. 4).

6. Results during assimilation

a. Precipitation

Lightning data assimilation substantially improved the location and amount of precipitation during the spinup period. Figures 5 and 6 display the precipitation accumulation during the period 0600–1200 UTC (20 July 2000) as reported by rain gauges and from different forecast experiments.

The gauge data are from the stage-IV National Centers for Environmental Prediction (NCEP) rainfall analysis (Baldwin and Mitchell 1998; Fulton et al. 1998; Seo 1998). (Unfortunately the preferable multisensor analyses were unavailable for the dates of interest.) The control run (Fig. 5b) had the lowest rainfall amounts, although it exhibited skill in producing its greatest values accurately within the larger observed rainfall values in Kansas.

The cases with lightning assimilation produced more rain in Kansas as well as capturing some convection in southeastern Nebraska and northeastern Kansas. Water vapor nudging substantially enhanced the amounts of precipitation (cf. Figs. 6b,c), but the rainfall was still less than was observed. Transferring 25% of convective precipitation to the resolved scale resulted in somewhat lower total precipitation, suggesting that the diverted precipitation evaporated more on the resolved scale than it would within the KF scheme. The quantitative precipitation estimate during the lightning assimilation period was up to approximately 40% of observed precipitation amounts, considerably more than in the control run. This supports the conclusion that forcing subgrid convection when lightning is present maintains much more realistic timing, intensity, and coverage of convection.

Although assimilation of NLDN ground-strike data alone provided considerable improvement (Fig. 5c), some further improvement occurred when LMA total lightning data were assimilated with NLDN ground-strike data (Fig. 5d). The addition of LMA data enhanced rainfall in northwestern Kansas. The enhancement may appear to be a little too much in the extreme northwestern part of the state, but it is probable that the rain gauge network poorly sampled those isolated storms.

Suppressing convection from the KF scheme where no lightning was observed helped to remove the spurious precipitation seen in the control forecast in Nebraska and in the Oklahoma panhandle region. The experiments that did not actively suppress KF were also able to reduce the frequency of spurious convection (Figs. 6b,c) that was present in the control forecast. As shown by Fritsch and Chappell (1981), triggering can be naturally suppressed by the compensating subsidence from previous convection, since the trigger function normally requires some positive vertical motion to activate. Warner and Hsu (2000) also demonstrated that parameterized convection on outer coarse grids can strongly influence resolved-scale precipitation on a fine inner grid. For future study, the effects of diabatic heating by the convection scheme could be studied from a variational data assimilation perspective using quasigeostrophic balanced equations (Fillion et al. 2005). Additionally, the assimilation may have improved the boundary conditions provided from the outer grids.

b. Effects on forecast initial conditions

A particular interest of this research is the generation of mesoscale boundaries by convective outflows. Surface and Weather Surveillance Radar-1988 Doppler (WSR-88D) radar mosaic observations at 0000 UTC 21 July indicate a strong, cold outflow forced by convection in northeastern Colorado, as well as boundaries in southeastern Colorado, north-central Kansas, and across Oklahoma (Fig. 4). The surface temperature fields from four model experiments are shown for comparison in Fig. 7. (A cold-start analysis had an obvious cold bias and is not shown.) The control case (Fig. 7a) did not generate the observed convection in northeastern Colorado during the spinup period and, therefore, failed to build the observed surface cold outflow.

A clear difference from the control run is seen in the experiments with lightning assimilation (Figs. 7b–d): a convectively generated cold pool is evident in northeastern Colorado as seen in the surface analysis. The case with assimilation of NLDN data only (Fig. 7b) developed a cold pool where convection was observed in northeastern Colorado, but it is slightly warmer and has a weaker gradient than when the same options were used with total lightning assimilation (NLDN plus LMA; Fig. 7d). This is a result of the sparseness of the NLDN ground strikes compared to the LMA total lightning data (seen in Fig. 2). A simulation with total lightning assimilation and full suppression but no rain feedback had results (not shown) very similar to Fig. 7d but with a slightly higher (by 0.5°C) minimum cold pool temperature. Of the two examples with total lightning assimilation, a stronger thermal gradient around the cold pool can be seen in the experiment in which spurious convection was actively suppressed (cf. Figs. 7c,d), the difference being due to suppression rather than the precipitation feedback.

Soundings at Dodge City, Kansas (DDC), also illustrate differences in the initial conditions generated by the control and assimilation experiments. The National Weather Service sounding from DDC at 0000 UTC 21 July 2000 is plotted in Fig. 8 (the sounding location is shown in Fig. 4). Model-generated soundings at the DDC location are shown in Fig. 9 from before and after the MVOI analysis. The soundings before the MVOI update are the 12-h model forecasts, whereas the postMVOI soundings show the new initial condition. The control run forecast sounding was saturated from 300 mb up to about 175 mb due to anvil outflow of spurious convection to the southwest of DDC. On the other hand, the forecast sounding from the lightning assimilation case is drier and more unstable above the moist boundary layer in agreement with the observed sounding (i.e., it does not exhibit contamination by convection) and, except for the near-surface winds, compares more favorably with the observed sounding.

Examination of ground layer conditions in the model output data (not shown) indicate that increased convective and total precipitation caused a significant increase in soil moisture in areas of antecedent convection. Given the demonstrated ability of assimilating lightning data to improve quantitative precipitation estimates during the assimilation period, one would expect more reliable soil moisture availability in areas that received heavy precipitation. The spatial soil moisture availability field is highly relevant to the determination of mesoscale surface layer fluxes (Marshall et al. 2003). Local soil moisture variations due to factors such as precipitation from previous convection may assist in forcing boundary layer evolution and convection initiation during subsequent diurnal cycles (e.g., Ziegler et al. 1995, 1997).

7. Results: Forecast

A main hypothesis for this study was that correctly locating soil moisture and outflow boundaries for the initial condition should improve model forecasts by improving the placement of physical mechanisms for triggering convection. The analysis above indicated that lightning data assimilation was successful in reproducing observed cold outflows at the end of the assimilation period, so the remaining test is whether forecast skill was similarly enhanced.

Observed and forecast rainfall accumulations for 6-h forecast periods are shown in Fig. 10 (0000–0600 UTC 21 July 2000) and Fig. 11 (0600–1200 UTC 21 July 2000). The larger observed rainfall accumulations from 0000 to 0600 UTC stretched from east-central Colorado and northwestern Kansas to south-central Kansas (Fig. 10a). The forecasts based on lightning assimilation (Figs. 10c,d) produced a more accurate pattern of the larger rainfall accumulations than the control forecast in this region, particularly in northwestern Kansas, where the control forecast had no precipitation. The assimilation-based forecast that did not suppress the KF scheme (Fig. 10c) had less spurious convection in Nebraska than the other experimental forecast but still had an overestimate of rainfall in northeastern Kansas. It also showed convection in north-central Oklahoma, consistent with lightning reports.

In the second 6-h period of the forecast, from 0600 to 1200 UTC, the heaviest observed accumulations had moved into Oklahoma, and relatively large values extended into south-central and southeastern Kansas (Fig. 11a). By this period, the pattern and amount of rainfall accumulations in Kansas from the experimental forecasts were converging on the pattern from the control forecast. In Oklahoma, however, the pattern and amounts of rainfall from the experimental forecasts appear to be improved in comparison with the control run.

The decrease in the forecasted rainfall accumulation relative to observations can be understood better by focusing on the early hours of the 0000–0600 UTC forecast period. A comparison of the hourly evolution of the observed radar reflectivity and radar-derived outflow boundaries with that of the forecasted convective rainfall and the associated cold pool boundaries suggests that assimilation of lightning data into the initial conditions did, in fact, improve the first several hours of the forecast mesoscale evolution (Fig. 12).

The observations show that a convective line from northeastern Colorado (cold pool 1 in Figs. 12a,d) propagated roughly toward the southeast, with other storm elements going eastward just north and south of the Kansas–Nebraska border. The observed storms in southeast Colorado (cold pool 2) weakened slightly and moved to the east over the 1-h period. At 0200 UTC, the radar showed a hint of an outflow boundary heading southward though east-central Colorado. Cold pool 3, associated with other forecast convection in northeast New Mexico, could not be evaluated due to sparse surface observations and radar blockage by terrain.

The control forecast (Figs. 12c,f) failed to generate any significant convection in northwestern Kansas or along the Kansas–Nebraska border, but produced convection along a temperature gradient that arched through southeastern Colorado and extended farther into southwestern Kansas than observed (Fig. 7b). In the experimental forecast (Figs. 12b,e), on the other hand, propagation of the two outflow boundaries (1 and 2) was similar to the observed behavior. This improvement is analogous to the results of Pereira Fo et al. (1999), who assimilated rainfall rate data to initialize a mesoscale forecast model. During the first hour, convection was triggered by the assimilation-produced cold pools in eastern Colorado, southwestern Nebraska, and northwestern Kansas, similar to observations.

The main convective line induced in northeast Colorado by the assimilation propagated southward instead of southeastward, but nevertheless the forecast was more skillful than the control run. The convection in southeastern Colorado was also more realistic in the experimental forecast than in the control run, in terms of placement, rainfall amounts, and the extent of propagation eastward into Kansas. The experimental forecast, however, also had some spurious convection in Nebraska and southwestern Iowa.

Convection in the second hour of the experimental forecast weakened relative to the observed convection. This weakening is particularly noticeable in the decreasing area of larger rainfall accumulations (cf. Figs. 12b,e), whereas the observed area of larger reflectivity was relatively unchanged (cf. Figs. 12a,b). It is possible that this weakening could be related to the already-discussed tendency for all activated subgrid-scale convection in the model to produce too little rainfall. As discussed in the last section, rainfall accumulations were only 40% of observed values, even when the convection was being nudged continually by observations, and the underestimating of rainfall increased with time in the forecast period (Figs. 10 and 11).

8. Conclusions

Assimilating lightning data to control parameterized convection in the spinup cycle of a forecast model has promise in improving the effects of prior convection on the initial condition of the forecast period. In particular, lightning data assimilation facilitates more accurate representation of cold pools in the boundary layer, the reduction of convective contamination of the environment where convection did not occur, and a more accurate distribution of soil moisture availability. The results suggest an optimal approach is to force convection where lightning was observed and to suppress it in the absence of lightning. The mere forcing of prior convection without suppression, however, can help to prevent some spurious triggering of convection in the forecast period. The feedback of 25% of precipitation from the convective scheme to the resolved scale had a minor effect on the resulting cold pool and forecast.

A continental warm-season forecast from an initial condition that included such effects of prior convection showed more subjective agreement with observations than a control forecast, at least over a short term. The assimilation scheme implemented in this study ingests lightning data directly, without additional analysis to estimate rainfall per flash as in alternate approaches. Direct ingest makes the scheme more appropriate for use in a RUC forecast or ensemble data analysis. It has the advantage over current implementations of variational data assimilation (VDA) schemes in being able to place convection correctly by forcing the model physics when needed, whereas VDA schemes presently can only adjust precipitation (e.g., based on instantaneous or surface accumulation data) where it was predicted by the forecast model. It is not clear whether VDA schemes could use a similar technique to force convection in the forward model without modifying the model physics, though it may be worth investigating the possibilities.

The results support the conclusion that forcing subgrid convection by lightning can improve representation of the intensity and coverage of convection and implies that the underprediction of convective precipitation during the forecast period is due at least in part to difficulty in maintaining cold convective outflows and triggering new convection along cold pool boundaries. The latter difficulty in maintaining forecast storm intensity may be related to the trigger function formulation and the microphysics in the trailing stratiform (resolved scale) region of the MCSs during the assimilation and forecast periods, among other factors (see Bukovsky et al. 2006). Since spatial coverage is rather accurately specified during assimilation, the underdiagnosed precipitation circumstantially implies that the subgrid convection scheme underpredicts convective rainfall, as it seems unlikely that the model environment has such a large significant dry (precipitable water) bias. Since precipitation is intimately connected to development and maintenance of both subgrid- and resolved-scale boundary layer cold pools in the assimilation and forecast cycles, a low bias in rainfall amounts relative to the observed MCS is consistent with accelerated weakening of the forecast MCS in our case study.

The results suggest that the assimilation may be more effective with total lightning data, such as from the ground-based lightning mapping array or data that could be acquired by a satellite-based optical system. Total lightning provides more detail of the convective cores, particularly in the geographic region of the test case where storms tend to have larger fractions of intracloud lightning than most storms elsewhere.

Acknowledgments

The authors thank Dr. J. Kain for providing code as well as suggestions and assistance in understanding the convective parameterization. The precipitation data and guidance in their application were provided by Dr. M. Baldwin. The authors are also grateful to two anonymous reviewers for improvements to the paper. We also thank Drs. D. Stensrud, B. Fiedler, and W. Beasley for useful discussions and comments. Funding for this study was provided by the Office of Naval Research Grant N00014-00-1-0525. Support was also provided to the lead author through a National Research Council associateship at the National Severe Storms Laboratory in Norman, Oklahoma. Additional funding for this research was provided by NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA17RJ1227. The data from the Lightning Mapping Array were provided by New Mexico Tech as part of the STEPS field program. Radar composites were obtained from the archive maintained by the National Center for Atmospheric Research.

REFERENCES

  • Alexander, G. D., , J. A. Weinman, , V. M. Karyampudi, , W. S. Olson, , and A. C. Lee, 1999: The effect of assimilating rain rates derived from satellites and lightning on the forecasts of the 1993 superstorm. Mon. Wea. Rev., 127 , 14331457.

    • Search Google Scholar
    • Export Citation
  • Baldwin, M. E., , and K. E. Mitchell, 1998: Progress on the NCEP hourly multi-sensor U.S. precipitation analysis for operations and GCIP research. Preprints, Second Symp. on Integrated Observing Systems, Phoenix, AZ, Amer. Meteor. Soc., 10–11.

  • Benjamin, S. G., and Coauthors, 2004: An hourly assimilation–forecast cycle: The RUC. Mon. Wea. Rev., 132 , 495518.

  • Benjamin, S. G., , S. S. Weygandt, , S. E. Koch, , and J. M. Brown, 2006: Assimilation of lightning data into RUC model convection forecasting. Preprints, Second Conf. on Meteorological Applications of Lightning Data, Atlanta, GA, Amer. Meteor. Soc., CD-ROM, 4.3.

  • Betts, A. K., 1986: A new convective adjustment scheme. Part I: Observational and theoretical basis. Quart. J. Roy. Meteor. Soc., 112 , 677692.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , and M. J. Miller, 1986: A new convective adjustment scheme. Part II: Single column tests using GATE wave, BOMEX, and arctic air-mass data sets. Quart. J. Roy. Meteor. Soc., 112 , 693709.

    • Search Google Scholar
    • Export Citation
  • Boccippio, D. J., , K. L. Cummins, , H. J. Christian, , and S. J. Goodman, 2001: Combined satellite- and surface-based estimation of the intracloud–cloud-to-ground lightning ratio over the continental United States. Mon. Wea. Rev., 129 , 108122.

    • Search Google Scholar
    • Export Citation
  • Bukovsky, M. S., , J. S. Kain, , and M. E. Baldwin, 2006: Bowing convective systems in a popular operational model: Are they for real? Wea. Forecasting, 21 , 307324.

    • Search Google Scholar
    • Export Citation
  • Chang, D-E., , J. A. Weinman, , C. A. Morales, , and W. S. Olson, 2001: The effect of spaceborne microwave and ground-based continuous lightning measurements on forecasts of the 1998 Groundhog Day storm. Mon. Wea. Rev., 129 , 18091833.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1996: Sensitivity of moist convection forced by boundary layer processes to low-level thermodynamic fields. Mon. Wea. Rev., 124 , 17671785.

    • Search Google Scholar
    • Export Citation
  • Cummins, K. L., , M. J. Murphy, , E. A. Bardo, , W. L. Hiscox, , R. B. Pyle, , and A. E. Pifer, 1998: A combined TOA/MDF technology upgrade of the U.S. national lightning detection network. J. Geophys. Res., 103 , 90359044.

    • Search Google Scholar
    • Export Citation
  • Errico, R. M., , K. D. Raeder, , and L. Fillion, 2003: Examination of the sensitivity of forecast precipitation rates to possible perturbations of initial conditions. Tellus, 55A , 88105.

    • Search Google Scholar
    • Export Citation
  • Fillion, L., , and S. Bélair, 2004: Tangent linear aspects of the Kain–Fritsch moist convective parameterization scheme. Mon. Wea. Rev., 132 , 24772494.

    • Search Google Scholar
    • Export Citation
  • Fillion, L., , M. Tanguay, , N. Ek, , C. Pagé, , and S. Pellerin, 2005: Balanced coupling between vertical motion and diabatic heating for variational data assimilation. Preprints, Symp. on Nowcasting and Very Short Range Forecasting, Toulouse, France, World Weather Research Programme, CD-ROM, 3.10. [Available online at http://www.meteo.fr/cic/wsn05/DVD/index.html.].

  • Fritsch, J. M., , and C. F. Chappell, 1981: Preliminary numerical tests of the modification of mesoscale convective systems. J. Appl. Meteor., 20 , 910921.

    • Search Google Scholar
    • Export Citation
  • Fulton, R. A., , J. P. Breidenbach, , D. J. Seo, , D. A. Miller, , and T. O’Bannon, 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13 , 377395.

    • Search Google Scholar
    • Export Citation
  • Gallus Jr., W. A., , and M. Segal, 2001: Impact of improved initialization of mesoscale features on convective system rainfall in 10-km Eta simulations. Wea. Forecasting, 16 , 680696.

    • Search Google Scholar
    • Export Citation
  • Hodur, R. M., 1997: The Naval Research Laboratory’s Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS). Mon. Wea. Rev., 125 , 14141430.

    • Search Google Scholar
    • Export Citation
  • Holle, R. L., , and M. W. Maier, 1982: Radar echo height related to cloud-ground lightning in south Florida. Preprints, 12th Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., 330–333.

  • Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122 , 927945.

    • Search Google Scholar
    • Export Citation
  • Jones, C. D., , and B. Macpherson, 1997: A latent heat nudging scheme for the assimilation of precipitation into an operational mesoscale model. Meteor. Appl., 4 , 269277.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1992: The role of the convective “trigger function” in numerical forecasts of mesoscale convective systems. Meteor. Atmos. Phys., 49 , 93106.

    • Search Google Scholar
    • Export Citation
  • Kain, J. S., , and J. M. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Covection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.

  • Lhermitte, R., , and P. R. Krehbiel, 1979: Doppler radar and radio observations of thunderstorms. IEEE Trans. Geosci. Electron., 17 , 162171.

    • Search Google Scholar
    • Export Citation
  • MacGorman, D. R., , and W. D. Rust, 1998: The Electrical Nature of Storms. Oxford University Press, 422 pp.

  • Marshall, C. H., , K. C. Crawford, , K. E. Mitchell, , and D. J. Stensrud, 2003: The impact of the land surface physics in the operational NCEP Eta model on simulating the diurnal cycle: Evaluation and testing using Oklahoma Mesonet data. Wea. Forecasting, 18 , 748768.

    • Search Google Scholar
    • Export Citation
  • Papadopoulos, A., , T. G. Chronis, , and E. N. Anagnostou, 2005: Improving convective precipitation forecasting through assimilation of regional lightning measurements in a mesoscale model. Mon. Wea. Rev., 133 , 19611977.

    • Search Google Scholar
    • Export Citation
  • Pereira Fo, A. J., , K. C. Crawford, , and D. J. Stensrud, 1999: Mesoscale precipitation fields. Part II: Hydrometeorologic modeling. J. Appl. Meteor., 38 , 102125.

    • Search Google Scholar
    • Export Citation
  • Rison, W., , R. J. Thomas, , P. R. Krehbiel, , T. Hamlin, , and J. Harlin, 1999: A GPS-based three-dimensional lightning mapping system: Initial observations in central New Mexico. Geophys. Res. Lett., 26 , 35733576.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, cited. 1998: Changes to the NCEP operational “early” Eta analysis/forecast system. NWS Tech. Procedures Bull. 447, National Oceanic and Atmospheric Administration/National Weather Service, 36 pp. [Available online at http://www.nws.noaa.gov/om/tpb/447.htm.].

  • Rogers, R. F., , J. M. Fritsch, , and W. C. Lambert, 2000: A simple technique for using radar data in the dynamic initialization of a mesoscale model. Mon. Wea. Rev., 128 , 25602574.

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

  • Stensrud, D. J., 1996: Effects of persistent, midlatitude mesoscale regions of convection on the large-scale environment during the warm season. J. Atmos. Sci., 53 , 35033527.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J-W. Bao, 1992: Behaviors of variational and nudging assimilation techniques with a chaotic low-order model. Mon. Wea. Rev., 120 , 30163028.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J. M. Fritsch, 1994a: Mesoscale convective systems in weakly forced large-scale environments. Part II: Generation of a mesoscale initial condition. Mon. Wea. Rev., 122 , 20682083.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , and J. M. Fritsch, 1994b: Mesoscale convective systems in weakly forced large-scale environments. Part III: Numerical simulations and implications for operational forecasting. Mon. Wea. Rev., 122 , 20842104.

    • Search Google Scholar
    • Export Citation
  • Stensrud, D. J., , G. S. Manikin, , E. Rogers, , and K. E. Mitchell, 1999: Importance of cold pools to NCEP Mesoscale Eta model forecasts. Wea. Forecasting, 14 , 650670.

    • Search Google Scholar
    • Export Citation
  • Thomas, R. P., , P. R. Krehbiel, , W. Rison, , S. Hunyady, , W. Winn, , T. Hamlin, , and J. Harlin, 2004: Accuracy of the lightning mapping array. J. Geophys. Res., 109 .D14207, doi:10.1029/2004JD004549.

    • Search Google Scholar
    • Export Citation
  • Warner, T. T., , and H-M. Hsu, 2000: Nested-model simulation of moist convection: The impact of coarse-grid parameterized convection on fine-grid resolved convection. Mon. Wea. Rev., 128 , 22112231.

    • Search Google Scholar
    • Export Citation
  • Ziegler, C. L., 1999: Issues in forecasting mesoscale convective systems: An observational and modeling perspective. Storms, R. Pielke Jr. and R. Pielke Sr., Eds., Vol. 2, Routledge Press, 26–42.

    • Search Google Scholar
    • Export Citation
  • Ziegler, C. L., , W. J. Martin, , R. A. Pielke Sr., , and R. L. Walko, 1995: A modeling study of the dryline. J. Atmos. Sci., 52 , 263285.

  • Ziegler, C. L., , T. J. Lee, , and R. A. Pielke Sr., 1997: Convective initiation at the dryline: A modeling study. Mon. Wea. Rev., 125 , 10011026.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The decision process for each grid column during assimilation.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 2.
Fig. 2.

Lightning data for the 15-in period starting at 2145 UTC 7 Jul 2000. (a) The CG detections by the NLDN. (b) Source points from the STEPS LMA. Contour levels are 1, 5, 10, 20, 50, 100, 250, 500, and 1000 counts per grid box per 15 min. [The lowest level in (a) is 0.8 to make single flashes visible]

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 3.
Fig. 3.

The coarse (90 km), intermediate (30 km), and fine (10 km) mesh areas of the model domain. The circle indicates the approximate range of the LMA.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 4.
Fig. 4.

Surface analysis at 0000 UTC 21 Jul 2000, the starting time for the forecast period. Conventional station model format includes temperature (°C) over dewpoint (°C), mean sea level pressure (mb) at upper right, and wind barb (full barb = 5 m s−1, half barb = 2.5 m s−1). Instantaneous base radar reflectivity is shown by gray fill. The site of the DDC sounding discussed in the text is indicated by DDC.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 5.
Fig. 5.

Observed and modeled total precipitation (mm) for a 6-h period starting 0600 UTC 20 Jul 2000 during the spinup period. (a) Rain gauge data with sampled NLDN strikes. (The 1st and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast (no lightning assimilation). (c) With assimilation of CG lightning from NLDN, moisture forcing, and full suppression. (d) With assimilation of both CG and total (LMA) lightning, moisture forcing, full suppression, and 25% feedback of KF precipitation.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 6.
Fig. 6.

Modeled total precipitation (mm) for 0600–1200 UTC 20 Jul 2000 (during the spinup period). These simulations have no precipitation feedback from the CPS to the resolved scale. (a) Same as in Fig. 5b (control run). (b) Normal KF trigger with forcing from lightning assimilation. (c) Same as in (b) adding nudging of boundary layer moisture. (d) Same as in (c), but suppresses the KF trigger where no lightning occurred (no precipitation feedback). The region of heaviest rainfall in (d) has a substantial resolved-scale contribution.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 7.
Fig. 7.

Air temperature (2 m) at 0000 UTC 21 Jul 2000, the beginning of the forecast period. (a) Warm start with 24-h spinup, 12-hourly data update cycle. (b) The 24-h assimilation of NLDN with only full suppression and 25% feedback of precipitation from KF to the resolved scale. (c) The 24-h assimilation of all lightning (NLDN and LMA) data to force convection (no suppression of KF nor feedback of precipitation). (d) Same as in (b), but with assimilation of both NLDN and LMA data.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 8.
Fig. 8.

National Weather Service sounding for DDC at 0000 UTC 21 Jul 2000.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 9.
Fig. 9.

Model soundings for DDC at 0000 UTC 21 Jul 2000. Solid gray and black curves show dewpoint and potential temperature, respectively, of the 12-h forecast. Long dashed curves are the MVOI-adjusted values for the initialization of the next forecast cycle. Black wind barbs are from the forecast, gray barbs are from the subsequent analysis adjustment. (a) Control run (no lightning assimilation). (b) Simulation with assimilation of lightning (NLDN and LMA), full suppression, and 25% feedback of precipitation from the KF scheme to the resolved scale.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 10.
Fig. 10.

Observed and forecast total precipitation (mm) for 0000–0600 UTC 21 Jul 2000. (a) Rain gauge data with sampled NLDN strikes. (The first and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast from standard warm start (no lightning assimilation). (c) Forecast from initialization with assimilation of NLDN and LMA data, moisture forcing, and no suppression nor feedback. (d) Forecast from initialization with assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 11.
Fig. 11.

Observed and forecast total precipitation (mm) for 0600–1200 UTC 21 Jul 2000. (a) Rain gauge data with sampled NLDN strikes. (The 1st and 30th strikes are plotted in each 10-km grid box.) Gray-filled areas indicate data voids. (b) Pure forecast from standard warm start (no lightning assimilation). (c) Forecast from initialization with assimilation of NLDN and LMA data, moisture forcing, and no suppression nor feedback. (d) Forecast from initialization with assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

Fig. 12.
Fig. 12.

Radar composites and convective precipitation forecasts on 21 Jul 2000. (a) Radar composite at 0100 UTC. (b) The 0000–0100 UTC precipitation forecast from initial condition generated by assimilation of both CG and total (LMA) lightning, moisture forcing, and full suppression and 25% feedback. (c) The 0000–0100 UTC control forecast from standard warm start (no lightning assimilation). (d), (e), (f) Same as in (a), (b), (c) but for 0200 (0100–0200) UTC. Observed and forecast cold pool boundaries are overlaid, the latter determined from the forecast surface temperature field associated with convective rainfall cores. For reference, reflectivity at 0000 UTC was depicted in Fig. 4.

Citation: Monthly Weather Review 135, 5; 10.1175/MWR3387.1

1

COAMPS is a registered trademark of the Naval Research Laboratory.

Save