Below is a concise review of my examination into the potential improvement radar-retrieved wind fields can make on mesoscale model simulations by modifying the model's initial conditions. The event I'm studying is the same presented in earlier work regarding a mesoscale convective system (MCS) that formed over southwestern Kansas just prior to 0000 UTC July 22, 1996. The convection immediately initiated an outflow region that propagated eastward through all of Kansas and southeastward through a large portion of Oklahoma. This outflow boundary reached the Twin Lakes WSR-88D radar (KTLX) immediately east of Oklahoma City by 0900 UTC on the 22nd. Remember, my previous work showed 2-dimensional wind retrieval results concerning this outflow boundary as it passed over the radar site. All simulations of this MCS are made with MM5 version 2 using a Kain-Fritsch convective parameterization scheme and Burk-Thompson PBL scheme. All experiments begin at 00 UTC on the 22nd and are run for 12 hours. The model grid is 100 x 100 points with 20 km resolution. All figures show only the region of interest (Kansas and Oklahoma) which is a 40 x 40 point portion of the domain.

Figure 1 shows wind field and temperature initial conditions over Kansas and Oklahoma produced using NCEP analysis. Surface data (0000 UTC) across Kansas and from the Oklahoma Climatological Survey mesonet is inserted into the initial condition through the NBOGUS method provided through RAWINS. This produces a significant change across western Kansas as winds are shifted to an easterly and northeasterly direction (Fig. 2a). In addition, the surface temperature gradient is shifted southwestward and compacted in this same area. The experiment using data inserted through this bogussing method is the control run (CNTL) for the following set of experiments. I am focussing primarily on two parts of the simulation: 1) propagation of the outflow boundary (will mainly look at the outflow boundary location 9 hours into the simulation) and 2) 12-hour cumulative rainfall amounts across Kansas. Figure 2b shows the temperature and wind fields 9 hours into the simulation. The Twin Lakes Nexrad radar (KTLX) is located roughly at grid point X=49, Y=41. (X values change in the east-west direction whily Y values change in the north-south direction). In reality the outflow boundary crosses KTLX at 0900 UTC, however, in the CNTL simulation, the outflow boundary is approximately 70 km northwest of this position. (As an interesting sidenote, notice the heat burst over south-central Kansas in the wake of the convection). Figure 2c shows the 12-hr precipitation totals (from 0000 - 1200 UTC on the 22nd) associated with the MCS. A precipitation max of 55.4 mm is located near X=41, Y=56. This is associated with the initial convection that first produces the large outflow region. All precipitation across eastern Kansas is associated with convection that formed along the leading edge of the outflow boundary which propagated eastward at about 40 km/hr. A region of 20+ mm is located in east-central Kansas.

Fig. 3 shows a 12-hour composite of radar-estimated rainfall based on data provided by the WSR-88D network. According to this estimate, a maximum of over 100 mm of rainfall feel very near Dodge City, Kansas. The center of this maximum is located at X=37.5,Y=54.1 on the model grid which is approximately 80 km southwest of the precip. maximum in the CNTL run. MM5 does a very good job predicting the areal extent of the rainfall. Moreover, according to Fig. 3, there are two maximums: one located near Dodge City while the other is in southeastern Kansas. The CNTL run also reproduces two regions of maximum rainfall total in roughly the same locations although the amount of rainfall is underpredicted. According to radar estimates, over 4 inches (well over 100 mm) fell in southwestern and southeastern Kansas. In the CNTL run, MM5 predicts 55.4 and 31.7 mm respectively.

The impact of radar data on the simulation of the MCS and associated features is examined through several experiments that adjust model initial conditions using static initialization procedures. The Clear-Air Adjoint-Method (CAAM) is utilized to retrieve wind information surrounding multiple Nextrad sites in Kansas and Oklahoma. Specifically, data from Dodge City, KS (KDDC); Witchita, KS (KICT); Tulsa, OK (KINX); Twin Lakes, OK (KTLX); and Frederick, OK (KFDR) are used. (Note: additional data from Goodland (KGLD) and Topeka KS (KTWX) will be added in the next couple of days) The CAAM is used to retrieve information over a circular region 90 km in diameter (centered on the radar site) at 14 horizontal levels spaced 250 m apart and up to 3500 m AGL. (Note: Even at vertical spacing as small as 100m, there is enough vertical resolution within the raw WSR-88D data to obtain important information on the atmosphere. For instance, take a look at the following set of images showing radial wind data for KICT (Wichita, KS). These images are actually horizontal slices of the atmosphere from 500m to 1000m (AGL) in height and are produced using the pre-processing routines of CAAM. Notice the strong veering in the wind field from easterly to southwesterly over this 500m (in depth) portion of the atmosphere.)

The retrieved wind field at each of the 14 vertical levels is smoothed by averaging data in a 5 km radius for each grid point. Then through tri-linear interpolation, wind information is obtained at model grid points. Output from the RAWINS MM5 pre-processing routine is adjusted by substituting CAAM-retrieved wind information at model grid points surrounding each of the Nexrad radar sites. Then INTERP is run to obtain the model initial conditions. Figure 4a shows the resulting wind and temperature initial conditions. This procedure modifies grid points over circular regions encompassing approximately 17 grid points over the lowest 10 half-sigma model levels.

In this experiment (RADWND), no substantial improvement is made in the position of the outflow boundary 9 hours into the simulation (Fig. 4b). In addition, no improvement in the location of the precipitation maximum is made in southwestern Kansas(Fig. 4c). In fact, the maximum is now located 20 km further eastward resulting in a spatial error of 85 km from the true location of the precip. maximum (Fig. 3). The eastern precip. maximum has shifted further northwestward with the area of 20+ mm rainfall located over the same general region.

Because Exp. RADWND only modifies grid points corresponding in location with retrieved wind data, only a few grid points scattered across the model domain are modified. Therefore, a different approach is taken to adjust model initial conditions that uses radar-retrieved information. Radar observations can precisely locate boundaries or differing air-mass regimes as far as 100 km from the radar in clear-air situations. By combining observations from KDDC, KICT, and KINX, a continuous picture of the location of the surface frontal feature can be obtained. This continuous feature separates two different air-masses containing opposing surface wind fields. To the south of the front southerly to southwesterly wind directions exist while the wind field north of the front is out of the northeast. Based upon wind retrievals from KDDC and KICT, the wind field behind the front is roughly from 70 degrees. Therefore, all grid points that have wind directions greater than 90 degrees behind the front and across much of central Kansas were set to 70 degrees to match the radar-retrieved wind field. Hence, a much sharper mesoscale front can be defined through a wind discontinuity instead of a broad diffuse region containing a gradual shift in wind direction as in Figure 1.

Therefore, radar data is now being used to create modifications to model initial conditions over various 3-dimensional air-mass regions (encompassing many model grid points) which surround the surface frontal feature. This type of modification is still based solely upon radar data. Radar observations can precisely identify important boundaries and radar-retrieved winds give a very good indication of the general wind directions and wind speeds ahead of and behind these boundaries. Furthermore, when CAAM is run over several vertical levels, a composite, 3-dimensional picture can then be derived and built into model initial conditions. Although the radar may not provide direct observations at all grid points that are modified in the model, the information that a radar can provide allows one to make logical deductions about the chartacteristics of the varying air-masses that have combined to produce and/or affect the meteorological phenomenon being investigated.

By using CAAM, a 3-dimensional picture of the surface front is produced. Winds within the air-mass north of the front are also changed at vertical levels above the surface based on CAAM's wind-retrieved results. These results show an easterly wind-field exists at the surface up to about 500m where the wind shifts abruptly to the southeast. From about 700 m to 900 m the wind gradually veers until it is out of the southwest above the frontal zone. Therefore, wind up to the 900 m level are modified to match with results provided by the CAAM. (I will load figures from the KICT site to show this)

One additional air-mass is also seen by radar data. Although the Nexrad radar at Vance Air Force Base in Oklahoma (KVNX) was not operating at 0000 UTC on the 22nd, it did operate up until 2130 UTC on the 21st. By examining radar reflectivity observations from 2027 to 2119 UTC, another boundary is detected. This boundary separates the south-southwesterly wind field from a east-southeasterly wind field. This boundary moves northward so that by 0000 UTC the boundary is entirely located in Kansas. This feature is observed by the KDDC radar although it is not as easily detected. By 2200 UTC, convection had initiated along this boundary near Alva, OK. This particular airmass, which is 1300 m deep according to the CAAM-retrieved wind field using KVNX radar observations, continues to get pinched-off as it moves northward while the meandering surface frontal feature over Kansas slowly moves southward. Based upon the KVNX radar data, the wind directions on the model grid within this airmass are set to 120 degrees if they are greater than 120 degrees. The final initial condition field for this experiment (AMWIND) based on modifying the wind field within varying air-masses using radar observations is shown in Figure 5a.

The propagation of the outflow boundary is still only changed very little (Fig. 5b). However, the 12-hour total precipitation max. in southwestern Kansas has shifted 25 km southwestward (Fig. 5c). This shift is likely due to the increased amount of convergence along the frontal boundary not only at the surface but in the lowest 900 m of the atmosphere. An additional improvement is made in the precipitation field along the secondary boundary extending southeast of DDC towards the Kansas-Oklahoma border. The increased convergence along this boundary has led to an increase in the amount of precipitation along this feature which agrees with the radar-derived rainfall in Fig. 3. Radar retrieved wind fields from the KVNX Nexrad radar prove valuable in illuminating secondary boundaries that play a role in convection.

Because radar observations can be used to identify surface boundaries, the opportunity to modify mass-field information also exists. Although explicit mass-field observations are not observed by radar, the use of surface temperatures from standard hourly observing stations as well as Oklahoma mesonet sites in conjunction with knowledge of existing boundaries provided by radar allow a more accurate representation of the air-masses surrounding the surface front. By referring back to Figure 2a, it is seen that the temperature gradient is smeared-out across a good portion of Kansas. In reality, temperatures are as much as 4-6 degrees C cooler immediately behind the front. Therefore, temperatures are reduced immediately behind the front by up to 3 degrees C to produce a more compact temperature gradient across the surface feature (Fig. 6a).

Further improvements are made in the simulation using the modifications to the temperature field. The initial propagation speed of the outflow boundary is increased so that by the ninth hour of the simulation, the outflow boundary has progressed 20 km further than in earlier simulations (Fig. 6b). Therefore, the error in the outflow boundary location is reduced to 40 km at 0900 UTC. In addition, the location of the 12-hour total precipitation maximum is shifted further westward (Fig. 6c) and now matches the location given by radar estimates (Fig. 3).