A Methodological Study on Using Weather Research and Forecasting (WRF) Model Outputs to Drive a One-Dimensional Cloud Model

Accurate prediction of the microphysical structures and evolutions of targeted clouds is essential for cloud-seeding operations and research, and remains a very challenging task. Aided by increasingly powerful computing capability, high-resolution mesoscale numerical weather prediction (NWP), especially ensemble-based weather prediction, has shown promise in regional and storm-scale quantitative precipitation forecasting (QPF) (e.g., Walser et al., 2004; Kong et al., 2006, 2007a, 2007b, 2008, 2009; Xue et al., 2007, 2008, 2009; Clark et al., 2009; Zhu et al., 2012, 2013). Ensemble-based weather forecasting can not only provide probabilistic guidance for future weather system development, but also forecast uncertainties. Employing ensemble forecasting techniques to produce high quality probabilistic QPF could become a feasible and innovative approach to improve efficiency in cloud-seeding operations (Zhu et al., 2012, 2013).

Nevertheless, enormous computing resources are required because of the need to resolve highly sophisticated microphysical processes in the high-resolution mesoscale NWP models involved, making it impractical to carry out real-time cloud-seeding numerical simulations to determine the optimal seeding time, location, and amount of seeding material, as well as provide decision-making guidance for cloud-seeding operations.

On the contrary, one-dimensional (1D) cloud models, owing to their simplified treatment of cloud dynamic interactions and sophisticated representation of microphysics processes, are ideal tools for studying cloud and precipitation generation in detail, as well as evolution processes. (Hu et al., 1983) developed a 1D bulk microphysics cloud model to simulate cold stratiform precipitating clouds. (Guo et al., 1999) developed a stratiform cloud model with sophisticated bin microphysics for raindrop distribution. Liu and Niu (2009, 2010) added ice crystal multiplication process to (Guo et al., 1999) 1D model. (Yang et al., 2007) further refined the treatment of the auto-conversion of cloud droplets to embryonic raindrops. (Hong and Zhou, 2005), based on a 3D hailstorm cloud model (Kong, 1991; Hong, 1997), developed a 1D stratiform cold cloud model (1DSC) including comprehensive mixed-phase microphysics processes with additional parameterization of homogeneous nucleation processes of cloud water to form ice crystals at -40°C. (Hu et al., 2007) used the 1DSC model to study a stratiform precipitation event observed on 5 July 2004, while (Hou et al., 2011) later refined the treatment of ice crystals in the model.

1D cloud models require certain atmospheric profiles for initialization. However, operationally available sounding data are far from adequate, both in spatial and temporal domains, to facilitate the use of 1D cloud models to forecast systematic precipitation events. Furthermore, traditional 1D cloud modeling studies in the literature assume constant atmospheric soundings and updrafts throughout the simulation duration, with the latter prescribed based on certain assumptions without downdraft flow (see Hong and Zhou, 2005; Hu et al., 2007). Such assumptions and approaches result in unrealistic outcomes in 1D cloud modeling, with all simulated clouds finally reaching steady states without decaying. Another drawback of traditional 1D cloud modeling is the absence of initial hydrometeor content, causing a prolonged cloud initiation period. In brief, traditional 1D cloud modeling studies emphasize cloud and precipitation characteristics in their steady states, and are unable to simulate realistically the generation and decay processes of clouds and precipitation. These drawbacks prevent the meaningful application of traditional 1D cloud models in operational cloud-seeding forecasting.

To develop a probabilistic-based operational cloud-seeding system, a two-tier ensemble forecasting system that employs the ensemble forecasting concept, and combines a dynamically sophisticated and computationally intensive 3D mesoscale NWP system [e.g., the Weather Research and Forecasting (WRF) model] with detailed microphysics treatment but a computationally efficient 1D cloud model, could be an innovative approach to produce reliable model prediction of rain enhancement potentials from cloud seeding. Such a two-tier system should consist of a comprehensive fine-scale regional short-range ensemble weather forecasting component and a multi-column ensemble 1D cloud-seeding modeling component. In this context, work is underway to use a WRF-based mesoscale ensemble forecasting system called IAP_REFS, developed at the Institute of Atmospheric Physics, Chinese Academy of Sciences (Zhu et al., 2012, 2013), to first provide regional probabilistic precipitation forecasts, and then to approximately determine the preliminary cloud-seeding target area from the mesoscale ensemble forecasts. Multiple model soundings, updraft velocities and hydrometeor profiles across the preliminary target area can be extracted from the mesoscale ensemble members and used to drive an ensemble of 1D cloud-seeding model runs. Various seeding strategies (seeding time, seeding height, seeding rate etc.) could then be simulated to produce corresponding rain enhancement probability maps for each seeding strategy. Such real-time products could help cloud-seeding operators to make optimal seeding plans for whether and how to perform seeding operations.

However, a key issue to address prior to the design of such a system is how to use WRF model outputs effectively to drive a 1D cloud-seeding model. In the present reported work, we aimed to determine an optimal methodology for using WRF outputs to drive a 1DSC, and to demonstrate that 1D cloud models do indeed have the potential to predict realistic precipitation patterns when properly driven by accurate atmospheric profiles derived from a regional short-range forecasting system. The main questions we sought to answer were: What WRF model output fields should be selected to initialize the 1D cloud model? What technical approaches are most efficient in driving the 1D cloud model? How do forecast uncertainties in WRF mesoscale forecasts affect the 1D cloud-seeding results? A WRF-driven 1DSC modeling system (WRF-1DSC) is presented in this paper and the results from validating the system through a real stratiform precipitation case are presented.

The remainder of the paper is organized as follows. Section 2 provides a brief description of the real stratiform precipitation case, which took place on 5 July 2004 and is the same case used in (Hu et al., 2007), and the numerical models used in the study. Section 3 gives details of WRF and WRF-1DSC configurations, experimental designs and simulation results. The simulated precipitation lag issue is discussed in section 4, followed by conclusions and a discussion of directions for future work in section 5.

The Advanced Research version of the WRF model (WRF-ARW) is a widely used, 3D, fully compressible and non-hydrostatic numerical weather modeling system designed for both idealized case studies and operational NWP purposes (Michalakes et al., 2001; Skamarock et al., 2001). WRF-ARW adopts the Arakawa C-grid in the horizontal direction, and the terrain-following height coordinate in the vertical direction. The model dynamic core consists of fully compressible Euler non-hydrostatic equations with a hydrostatic option and the Runge-Kutta third-order time integration scheme. WRF-ARW version 3.1 was used for this study. Compared to previous versions, version 3.1 has a new double-moment bulk microphysics scheme, and the Morrison two-moment scheme, with explicit prediction of the mixing ratios and number concentrations of five hydrometeor species (cloud droplets, cloud ice, snow, rain, and graupel). The Morrison scheme offers a more detailed microphysics treatment in clouds and the raindrop size distribution is assumed to follow a gamma distribution function (Morrison and Pinto, 2005; Morrison et al., 2009).

The 1DSC model used in this study was a simplified version of the 2D slab-symmetric mixed cloud system (MCS) model originally developed by Hong (1997, 1998). It is a time-independent cloud model with detailed parameterized mixed-phase microphysics processes. The hydrometeors in the model are divided into five categories: cloud water, cloud ice, rain, snow and graupel. The model assumes that clouds are evenly distributed horizontally, with a given distribution of vertical flow. The vertical domain height and resolution are 11 km and 0.2 km, respectively, with a time-step of 5 seconds. It is driven by an observed sounding profile, with model cloud initiated by a prescribed upward motion according to the formula (Hu et al., 2007)

where w₀ is the maximum vertical velocity, k is the grid ordinal number, and (h+m) is the grid level of w₀. (Hu et al., 2007) assumed k21, h=30 and m=10. Both the sounding profile and updraft are kept constant throughout the entire simulation duration. Owing to a lack of vertical velocity observations, the value w₀ in Eq. (1) is mainly determined by trial and error by comparisons between simulation results and observed radar echoes. As discussed in the Introduction, the main shortcomings of traditional 1DSC models are: (1) the cloud environment remains constant during the entire simulation; (2) the ignorance of downward flow; (3) the process to determine the value of w₀ is time consuming; and (4) the accumulated precipitation amount has no real meaning due to the fact that the model can only seek the final steady states. One major effect of these shortcomings is a lack of decaying process for any simulated cloud. Past studies (e.g., Chen, 2002; Liu and Niu, 2009, 2010; Chen and Siao, 2010) also pointed out that 1D cloud models cannot reproduce realistic cloud and precipitation intensity changes, because of the constant environment and lack of energy exchange between clouds and their environment. Such models are, at best, used for cloud and precipitation mechanistic studies, and are not capable of cloud and precipitation prediction.

The case used for validation occurred on 4-5 July 2004. It was associated with a Northeast China low, and was a typical large-scale stratiform precipitation event affecting Jilin Province. Observations based on rain-gauge stations surrounding the Changchun municipal area indicate the precipitation started at 2000 UTC 5 July 2004 and lasted for 8 hours with a total of 5.92 mm of precipitation having fallen. This typical stratiform precipitation system was characterized by uneven radar bright band echoes. The primary observation data used in the study were obtained by Doppler radar and twenty rain gauges. Both Plan Position Indicator (PPI) and Range Height Indicator (RHI) images were provided by a C-band Doppler weather radar located at Changchun Station (43.9°N, 125.3°E; height above sea level: 294.6 m). Twenty tilting bucket rain gauges with an average spacing of 10 km were used to collect the precipitation data. Unrealistic precipitation values were removed through quality-control procedures.

Figure 1 shows the PPI and RHI radar images obtained at Changchun at various times throughout the precipitation event. The PPI image obtained at 2114 UTC 4 July 2004 (Fig. 1a) indicates large areas of moderate reflectivity values between 15-25 dB Z embedded with a stronger northeast-southwest radar reflectivity band of 30-35 dB Z. The radar echoes moved toward the northwest, while the whole precipitation system moved anticlockwise. By 0250 UTC 5 July 2004, the reflectivity at Changchun Station became weaker, and decreased to 10 dB Z from 35 dB Z at 0209 UTC, indicating a weakening of ground precipitation. The RHI images at five valid times with the azimuth of 170° (Fig. 1b) indicate that the cloud top height was between 8-9 km, and a clear 0°C isotherm bright band with a reflectivity value of 35 dB Z was at the height of 3.2 km. The unevenness of the bright band echoes suggested an uneven cloud microphysical structure above the 0°C layer (Huang et al., 1987).

Figure 1.  Plan Position Indicator (PPI) (a) and Range Height Indicator (RHI) (b) images of radar reflectivity (units: dB Z) obtained at different valid times. The range markers are both 30 km from each other for PPI, and the maximum radius is 150km. The central location marked as a white square in each image of (a) indicates the Changchun site.

The case was observed relatively well, and is a good candidate for case studies needed for numerical model simulations of stratiform precipitation process. It was used by (Hu et al., 2007) in their numerical simulation study in which the original (traditional) 1DSC model was used along with airborne measurements. The same case was chosen in this study as a basis for comparison. In addition to the traditional 1DSC approach, different techniques to drive the 1DSC model from the WRF forecast outputs were examined and the best technique identified and validated. The resulting new WRF-driven 1D cloud model (WRF-1DSC) will be used in combination with an ensemble-based regional short-range forecasting system to study the feasibility and capability of stratiform cloud seedability prediction under a probabilistic framework in subsequent work.

WRF-ARW version 3.1 was used to simulate the stratiform precipitation case over the northeast of China. The model domain, centered at (46.10°N, 123.10°E), had 155(lon) ×135(lat) grid points at a 12 km grid spacing in the horizontal direction and 40 vertical eta-levels (Fig. 2). A nested grid was used at a 3:1 nesting ratio with a resolution of 4 km. The model was integrated from 1200 UTC 4 July 2004 to 1200 UTC 5 July 2004 using a time step of 90 s, with the initial and boundary data interpolated from the National Centers for Environmental Prediction’s (NCEP) 1°×1°resolution global analysis (GDAS) dataset. Physical parameterization options used include the Kain-Fritsch (new eta) cumulus scheme, the RRTM (Rapid and accurate Radiative Transfer Model) longwave (Mlawer et al., 1997) and Dudhia shortwave radiation schemes, the Noah land surface model, the MYJ (Miller-Yamada-Janji c’) planetary boundary layer scheme (Janji c’, 1994), and the two-moment Morrison microphysics scheme. Being able to explicitly predict both the mixing ratio and number concentration of cloud water, rain water, cloud ice, snow, and graupel, the advantage of a two-moment bulk scheme to represent most of the cloud microphysics processes was documented by Morrison and Grabowski (2007).

Upward flows in stratiform clouds are generally very weak with vertical velocities of approximately 10^-3-10^-1 m s^-1, and up to 0.05-0.3 m s^-1 in cyclonic convergent zones (Huang et al., 1999). Owing to a lack of vertical velocity observations, the value of w₀ in the traditional 1DSC model is often determined by trial and error from comparisons between simulated results and observed radar echo structures (e.g., Hu et al., 1983; Hu and Yan, 1986, 1987; Hong, 1997, 1998; Hu et al., 2007; Hou et al., 2011). For the stratiform precipitation event in this study, w₀ was determined in this way to have a value of 10^-2 m s^-1 at the 3 km height level. Figure 3 shows the profile of the vertical velocity (w) in the traditional 1DSC model according to Eq. (1). Such a hypothetical vertical velocity profile is highly unrealistic for ignoring downward air motion; and, together with a constant atmosphere (sounding), simulated clouds using the traditional 1DSC model always follow the same stages of cloud initiation, adjust and grow, and reach a steady state after a 4-5 hour spin-up period. No cloud and precipitation dissipation process can be simulated with this type of model, and clouds take very long—often several hours—to generate (see Hu et al., 1983; Hong and Zhou, 2005; Hu et al., 2007; Hou et al., 2011

).

Figure 2.  The domain of WRF (Changchun Station is marked with a star).

Figure 3.  The vertical velocity profile (units: cm s^-1) used in the traditional 1DSC model.

Figure 4.  Hourly evolutions of the vertical velocities (units: cm s^-1) derived from the WRF model for the period of 1300 UTC 4 July-1200 UTC 5 July 2004. The y-axis in each figure represents altitude (units: km).

Applied to the same case, (Hu et al., 2007) conducted a 15-hour simulation using the original 1DSC model starting at 1200 UTC 4 July 2004, with initial precipitation occurring after 4 hours into the simulation. In their simulation, the initial mixing ratios of various hydrometeor species and the grid volume average number concentration were all set to zero. Cloud water appeared after 180 min into the simulation, and became intensified during 180-240 min. Rain started to fall slowly at 240 min. The simulated cloud system became steady around 2100 UTC 4 July 2004 to the end of the simulation. The spin-up process—a measurement of how long the model took to reach the steady state—took more than half of the total simulated time in their experiments. With a constant atmosphere and no downward motion, the modeled cloud became stationary. Such traditional 1DSC models are only able to analyze general structure and microphysical characteristics of stationary stratiform clouds, but are not suitable for predicting realistic cloud and precipitation evolution, or for obtaining ground precipitation accumulation, as desired in the present study.

Figure 4 shows the time evolution of the vertical velocity at Changchun Station extracted from the WRF-ARW model output. It can be seen that the updraft velocity first increases and then decreases from 1300 UTC 4 July to 1200 UTC 5 July 2004. Between 2200 UTC 4 July 2004 and 0400 UTC 5 July 2004, the maximum values of upward velocities are all exceeding 20 cm s^-1, peaking at 0100 UTC 5 July 2004 with a maximum of 34.4 cm s^-1 at a height of 8.3 km. Downdrafts become dominant in the mid and lower troposphere from 1100 UTC 5 July 2004. These vertical velocity profiles are far different from the one shown in Fig. 3. It was one of the primary goals of this study to discover whether such WRF-forecasted vertical motion profiles can be used efficiently for directly driving 1D cloud models in order to produce realistic stratiform cloud and precipitation structures and evolutions. Two approaches of using WRF forecasts to drive the 1DSC model were studied through various sensitivity experiments and the results will now be discussed in the remainder of this section. The first approach was to apply WRF outputs as the initial conditions for the 1DSC model, and the second approach was to frequently update the 1DSC profiles with WRF outputs.

Five numerical experiments were conducted to use WRF outputs to initialize the 1DSC model. Table 1 outlines the configurations for each sensitivity experiment. Experiment a1 was the control run, with exactly the same configuration as the original 1DSC model in (Hu et al., 2007), which was conducted under environment conditions given by observed soundings and w depicted in Fig. 3. Experiment a2 was the same as a1 except that sounding data derived from the WRF model were used. Experiment a3 differed from a1 in that WRF output data were used as the w profile. Experiments a4 and a5 were driven with both soundings and the w profile from the WRF forecast, with the latter (Experiment a5) also taking into consideration the influences of WRF-forecasted hydrometeor and number concentration profiles. The vertical profiles of hydrometeors extracted from the WRF model were introduced into the 1DSC model using a fitting method that firstly applied a cubic spline function to the WRF output, and then estimated the input for the 1DSC model according to the fitting results. In Table 1, Q_x and N_x are the grid volume average mixing ratios and number concentrations, respectively, of hydrometeor species in cloud, where the subscript x stands for cloud water (c), rain water (r) , ice crystal (i), snow (s) and graupel (g), and in the case of mixing ratios also for vapor (v) (Hong, 1997). All experiments were started at 2000 UTC 4 July 2004, and ended at 1100 UTC 5 July 2004. For experiments a1 and a3, we reconstructed a new sounding by interpolating the 1200 UTC 4 July 2004 and 0000 UTC 5 July 2004 observed Changchun soundings.

The observed surface precipitation data were used as validation in this study. Figure 5 illustrates the time evolutions of simulated precipitation rates at Changchun Station from all the experiments listed in Table 1. After two hours into the simulations, there was no rain generated in Experiments a1, a2 and a4. The simulated precipitation rates in Experiments a2, a4 and a5 were smaller than those in a1 and a3. This was largely due to errors in the input dew-point temperature, T_d, extracted from the WRF model at 2000 UTC 4 July 2004. The input T_d was much lower than observed (not shown), resulting in relatively small rainfall rates in Experiments a2, a4 and a5. It can also be seen that, in Experiment a5, the precipitation occurred right after the model initiation, much sooner than the other experiments. This indicates that adding initial values of N_x and Q_x extracted from the WRF model to the 1DSC model can indeed reduce the spin-up process.

Experiments a2-a5 employed the same traditional initialization approach as a1 (traditional 1DSC) in that environment profiles were only provided at the starting time and remained constant throughout the entire simulation. As a result, they still failed to simulate cloud and precipitation decaying processes. Past studies, such as those by (Hu et al., 1983), Hu and Yan (1986, 1987), (Hu et al., 2007), (Hou et al., 2011) and (Yang and Lei, 2012), all encountered similar problems using 1D cloud models with a constant cloud environment. As such, the second set of experiments was performed by continuously updating the 1DSC model with WRF forecast outputs.

Figure 5.  Time evolutions of modeled precipitation intensity (colored lines) at Changchun Station from sensitivity experiments of different schemes shown in Table 1. The observational data are presented as the red line.

Experiments b1, b2 and b3 were the profile-updating counterparts of Experiment a5 (see Table 2), each with a different profile update frequency ranging from 90 seconds to 1 hour. Instead of only adopting sounding data and w profiles at the initialization time, this set of experiments updated the temperature, dewpoint and w profiles at a given time interval (frequency) continuously throughout the entire WRF-1DSC simulation period using WRF forecast outputs. Results showed (see Fig. 6) that unlike the traditional 1DSC model(e.g., Hong and Zhou, 2005; Hu et al., 2007) from which the simulated precipitation rate always reaches a steady state until the end of the simulation, by initializing Q_x and N_x from WRF, and continuously updating soundings and w profiles with a given time interval thereafter, WRF-1DSC can not only reduce the spin-up period, but also successfully reproduce the generation and dissipation of precipitation processes. Compared to the observed precipitation rate (red line in Fig. 6), the results from Experiments b1-b3 seem alike, all showing a time-lag behind the observation by about two hours. Comparison between the simulated reflectivity field and observed radar echo indicates that Experiment b3 (with a 90-second update frequency) was more realistic than the other two (not shown).

Figure 6.  Time evolutions of modeled precipitation rates (colored lines) at Changchun Station from sensitivity experiments of different schemes shown in Table 2. The observational data are presented as the red line.

Vertical air motions play a very important role in determining the formation and evolution of precipitation process. Updrafts supply and lift moisture into clouds to form precipitation particles, and help precipitating drops and ice particles to maintain for longer in the cloud, and grow sufficiently to fall. On the contrary, downdrafts can block moisture supply and repress precipitation. As discussed in previous sections, with a pre-specified static updraft, and lack of downdraft, there is no dissipation of cloud and precipitation in traditional 1DSC simulations. It has been demonstrated in this section that, with frequent updating of simulated vertical velocity profiles that include both updraft and downward motions, WRF-1DSC can reproduce a much more realistic precipitation pattern with dissipation, albeit with some time-lag. Furthermore, the spin-up period in WRF-1DSC is remarkably reduced compared to the traditional 1DSC model.

To better understand the time-lag in this set of experiments, Fig. 7 plots the precipitation rates from b3 and from WRF model outputs (interpolated to Changchun Station), along with observations. As can be seen, the curve from WRF-1DSC (black line) is very close to that from the WRF mesoscale model simulation (green line). Examination of different grid points reveals similar agreements in precipitation rate patterns between WRF and WRF-1DSC simulations (not shown). This leads us to believe that the possible cause of the time-lag problem shown in b3 is the WRF model simulation errors that place the precipitation system a couple of hours upstream (farther southeast in this case) compared to the actual situation. Such model-forecasted misplacement of precipitation systems is not uncommon owing to various forecast uncertainties associated with regional mesoscale NWP models. Such precipitation system location errors resulting from the regional WRF forecasts may directly contribute to time-lag (in this case) or time-lead situations for the simulated precipitation rate patterns from the WRF-1DSC simulations that are driven using WRF profiles at fixed grid locations. Furthermore, the comparison of model results and measurements could suffer from representativeness issues since measurements are point values, while the model results represent a volume corresponding to the grid box size (Dierer et al., 2009).

As discussed above, the time-lag problem in Experiment b3 can be attributed to the location errors in the WRF simulated precipitation system. Precipitation simulations from any NWP model such as WRF suffer from various uncertainties. These uncertainties result from a combination of NWP model errors, observation data errors, and physics parameterization errors. The WRF model uncertainties may result in a phase shift in the simulated mesoscale precipitation system, causing unrealistic model soundings and vertical profiles at a given location (in this study, Changchun Station). Taking into consideration the precipitation system movement errors (moving too fast or too slow) in the WRF simulation, it would be safe to assume some nearby grid points may be more representative than the Changchun location in the WRF output profiles.

Figure 7.  Time evolutions of precipitation intensities (units: mm h^-1) at Changchun Station from the WRF model (dashed line) and WRF-1DSC (dotted line). The observed rainfall rate is presented as the solid line.

Figure 8.  Hourly maps of modeled precipitation rates (mm h^-1) for 2000 UTC 4 Jul-0300 UTC 5 Jul 2004. The pentagrams represent Changchun, and the red circles show the radar range (the radius is 150 km).

Replacing single (deterministic) forecasts with ensemble-based probabilistic forecasts is an effective technique to cope with the mesoscale NWP uncertainty problem. This study is part of a broader research initiative with ultimate objectives to develop a novel ensemble-based cloud-seeding operation system in a probabilistic framework (Zhu2012, 2013). As a methodological study at this stage of the research, all experiments reported in the present paper have not yet involved ensemble simulations. However, it is still meaningful to see if, by selecting a nearby grid point within the precipitation area that is more meteorologically representative of the observed Changchun sounding based on the movement path of the simulated precipitation system, WRF-1DSC could reproduce a realistic precipitation pattern measured at Changchun Station. It should be noted that choosing representative points was neither a goal nor means of the present study; instead, it was simply a way to demonstrate the ability of WRF-1DSC for potential use in an ensemble-based regional short-range forecasting system to realistically forecast precipitation rate patterns.

WRF simulations provide full information about the precipitation areal distribution. The WRF-simulated hourly precipitation rates are shown in Fig. 8. Comparing Fig. 8 with the radar reflectivity PPIs obtained from a C-band radar at Changchun Station at different times (Fig. 1a) shows that there is indeed a time-lag in WRF-simulated precipitation over Changchun Station. The simulated precipitation rates southeast of Changchun seem more consistent with observations. There are several ways to select nearby grids (see Table 3). Grid points P1, P2 and N5 were chosen by means of Root Mean Square Deviation (RMSD).

Spatial locations of nearby grid points chosen for WRF-1DSC are marked in Fig. 9. The comparisons between observations and WRF simulations at these grid points are shown in Fig. 10. As can be seen from Fig. 10, the simulated precipitation rate at point N5 is the closest to the Changchun observation. Experiment b3 was then re-run with soundings, w, Q_x and N_x profiles derived from the WRF output at grid point N5. The simulation was started at 2000 UTC 4 July 2004, when precipitation at Changchun began. The model soundings, mixing ratios of cloud water, rain water, cloud ice, snow, and graupel, number concentrations of rain, ice, snow, and graupel derived at grid point N5 were used as inputs for WRF-1DSC, and WRF-simulated temperature, dew point temperature, and vertical velocity profiles at N5 were continuously updated every 90 seconds.

Figure 9.  Spatial distribution of grid points from the WRF simulation area chosen for 1DSC simulations.

A comparison of precipitation rates between WRF-1DSC and observations for the re-run experiment (see Fig. 11) shows that the WRF-1DSC-modeled precipitation started immediately upon model initiation, and the WRF-1DSC model is capable of reproducing the observed precipitation rate pattern with two peaks. The first peak matches the observed in terms of time very well, although is somewhat weaker in magnitude. The second peak has a small time-lag but a well-matched magnitude. More impressively, the simulated and observed precipitation patterns agree well in their start and end times. We also compared the WRF-1DSC results of N5 with those of other grid points (not shown), and found that the simulated precipitation rate at N5 was the closest to observed. This might be mainly due to the N5 point being more meteorologically representative of Changchun Station. The weather system in this case moved from southeast to northwest across Changchun Station, as seen in the radar reflection images. The WRF-modeled precipitation over the area south of Changchun was comparable to Changchun observations in terms of intensity and duration (such as N1 and N5), and the distance between N5 and Changchun was long enough to reduce the time-lag error in the WRF model.

An important objective of the study was to enable 1D cloud models to be capable of reproducing realistic precipitation patterns and intensities for various types of weather systems. It has been demonstrated that, by properly driving 1D cloud models with WRF output profiles, WRF-1DSC does indeed possess the skill to predict precipitation process.

An efficient WRF-model-driven 1D cloud modeling system, called WRF-1DSC, was developed and validated by simulations of a real stratiform precipitation event that took place in Northeast China on 4-5 July 2004. Simulation results showed that WRF-1DSC overcomes the drawbacks of traditional 1DSC models, which are incapable of simulating the precipitation decaying process and are characterized by the start-up of precipitation being too slow. When supplied with sounding data and hydrometeor profiles extracted from a WRF simulation as the initial inputs, and continuously updated with soundings and vertical velocity profiles (including downdrafts) derived from WRF simulations at 90-second intervals, WRF-1DSC was not only capable of reproducing the generation and decay process of the precipitation event, but also capturing the major evolution pattern of the precipitation rate.

Figure 10.  Time evolutions of observed (solid line) precipitation intensities at Changchun Station and rainfall rates modeled by the WRF model (dashed line) at newly chosen grid points (units: mm h^-1).

Figure 11.  Time evolutions of WRF-1DSC-simulated (dotted line) rainfall rates (units: mm h^-1) at point N5. The observed precipitation intensity at Changchun Station is presented as the solid line.

This study also suggests that if the WRF model can predict precipitation systems with reduced uncertainties and provide accurate atmospheric conditions, including the time-varying profiles of soundings and vertical velocities, WRF-1DSC possesses the skill to predict the precipitation process. The uncertainties in WRF forecasts can be quantitatively sampled by a well-designed regional ensemble prediction system. The implication of this study is that it is feasible, under the ensemble forecasting framework, to combine mesoscale NWP models and microphysically comprehensive 1D cloud models to predict stratiform precipitation events, and to fully utilize WRF-1DSC’s capability in cloud-seeding research and operation areas.

Although a single case study does not provide enough significance in a statistical sense, the promising results obtained in this study suggest that WRF-1DSC has good potential for use with a WRF-based weather forecasting system. Nevertheless, further research work is required for real application of WRF-1DSC in cloud-seeding forecasting (Jin et al., 2013). In addition, two-moment microphysics schemes, such as that used in the 1DSC in this study are insufficient to describe the variability of the size distributions in clouds. In further research, we will explore the use of a detailed spectral (bin) microphysical 1D cloud model (Yang and Lei, 2012) instead of the two-moment 1DSC to fully include the size distributions for hydrometeors.