Assimilation of DAWN Doppler Wind Lidar Data During the 2017 Convective Processes Experiment (CPEX): Impact on the Precipitation and Flow Structure

An improved representation of the 3-D air motion and precipitation structure through forecast models and assimilation of observations is vital for improvements in weather forecasting capabilities. However, there is little independent data to properly validate a model forecast of precipitation structure when the underlying dynamics are evolving on short convective times scales. Using data from the JPL Ku/Ka-band Airborne Precipitation Radar (APR-2) and the 2-um Doppler 15 Aerosol Wind (DAWN) lidar collected during the 2017 Convective Processes Experiment (CPEX), the NASA Unified Weather Research and Forecasting (WRF) Ensemble Data Assimilation System (EDAS) modeling system was used to quantify the impact of the high resolution, sparsely-sampled DAWN measurements on the analyzed variables and on the forecast when the DAWN winds were assimilated. Overall, the assimilation of the DAWN wind profiles had a discernible impact to the wind field and the evolution and timing of the 3-D precipitation structure. Analysis of individual variables revealed that the 20 assimilation of the DAWN winds resulted in important and coherent modifications of the environment. It led to increase of the near surface convergence, temperature and water vapor, creating more favorable conditions for the development of convection exactly where it was observed (but not present in the control run). Comparison to APR-2 and observations by the Global Precipitation Measurement (GPM) satellite shows a much-improved forecast after the assimilation of the DAWN winds – development of precipitation where there was none, more organized precipitation where there was some, and a much more 25 intense and organized cold pool, similar to the analysis of the dropsonde data. Onset of the vertical evolution of the precipitation showed similar radar-derived cloud top heights, but delayed in time. While this investigation was limited to a single CPEX flight date, the investigation design is appropriate for further investigation of the impact of airborne Doppler wind lidar observations upon short-term convective precipitation forecasts. https://doi.org/10.5194/amt-2020-503 Preprint. Discussion started: 23 December 2020 c © Author(s) 2020. CC BY 4.0 License.

structure and the evolution of the cold pools, and the precipitating systems in general, is the microphysical characteristics of the precipitation (Hristova-Veleva et al., 2021), which strongly affects the evaporation rates. Morrison et al. (2012), among others, found that numerical simulations with higher evaporation had stronger cold pools, faster propagation, larger storm size, 65 greater updraft mass flux (but weaker convective updrafts at mid-and upper levels), and greater total condensation that compensates for the increased evaporation to give more surface precipitation. In turn, the structure and the intensity of the divergent near-surface cold pools modify the morphology of the convective systems. These joint processes affect the vertical growth and glaciation of water-abundant clouds, and further aggregation and organization of individual cumulus clouds into much larger mesoscale convective systems (Rowe et al., 2012;Houze, 2018). 70 While the overall processes responsible for these interactions have been identified for some time, their precise nature and interactions remains under-constrained by observations; in particular the uncertainty regarding convection and cloud processes directly results in much of the uncertainty in both weather and climate prediction. Further constraining the uncertainty in convective cloud processes linking 3-D air motion and cloud structure through models and observations is vital for 75 improvements in weather forecasting and understanding limits on atmospheric predictability. To date, there is little independent validation data to properly validate a model forecast of precipitation structure when the underlying dynamics are evolving on convective time scales.
Many years of NASA-sponsored airborne field campaigns have focused on the microphysical processes linking clouds, 80 convection and precipitation, as well as ground validation, following the deployment of the Tropical Rainfall Measuring Mission (TRMM) satellite in 1997 and the Global Precipitation Measurement (GPM) mission (2014-current). These airborne campaigns featured narrow swath precipitation profiling radars, such as the JPL Ku/Ka-band Airborne Precipitation Radar (APR-2) (Durden et al., 2012). However, the Doppler capability of these radars is intended for estimating the vertical Doppler velocity within precipitating clouds, and are not capable of capturing vertically resolved observations of 3-dimensional wind 85 structure in close proximity (10-km or less) to cloudy regions. A space-based Doppler wind lidar (DWL) capability has been envisioned as one means to overcome this observational shortcoming (Okamoto et al., 2018;Baker et al., 2014). The current Atmospheric Dynamics Mission (ADM)-Aeolus wind lidar (Stoffelen et al., 2005) has been successfully collecting satellitebased line-of-sight profiles (Lux et al., 2020), at a synoptic scale suitable for global numerical weather prediction (NWP) data assimilation, rather than the spatial scale of cloud-resolving mesoscale models (Šavli et al., 2018;Horányi et al., 2015). 90 Previous DWL-based airborne campaigns lacked scanning Doppler precipitation radar capabilities on the same aircraft, whose data collection was synchronized with the DWL operations. During the May-June 2017 Convective Processes Experiment (CPEX), joint observations were collected from the APR-2 and the 2-um Doppler Aerosol Wind (DAWN) lidar Kavaya et al., 2014) during approximately 100 flight hours of the NASA DC-8 aircraft (Turk et al., 2020). The  2 radar operates at the same frequencies as the GPM Dual-Frequency Precipitation Radar (DPR), proving reflectivity products approximately every 360-m along track. The multi-beam measurements from the DAWN lidar were processed into highresolution vertical wind profiles spaced as finely as 3-7-km along-track , including the environment close to where the clouds develop. To date, there has been relatively little analysis of the assimilation impact of airborne Doppler wind lidar data upon the joint evolution of the mesoscale model-forecasted 3-D precipitation structure together with the 100 associated 3-D wind field (Cui et al., 2019).
In this manuscript, the impact of assimilating the high resolution, sparsely-sampled airborne DAWN measurements upon the forecasted precipitation structure are examined with the NASA Unified Weather Research and Forecast (NU-WRF) Ensemble Data Assimilation System (EDAS) modeling system (Zhang et al., 2017;Zhang et al. 2013). A previous study of the impact 105 of assimilating DAWN data from CPEX was carried out by Cui et al. (2019), who examined how different assimilation methods affected the forecasted wind and 2-D precipitation structure inferred from the gridded GPM IMERG (Tan et al., 2019) precipitation dataset. A unique aspect of this study is that both the horizontal and vertical evolution of the forecasted precipitation field is compared with near-simultaneous data from APR-2 radar data, and from DPR data from overpasses of GPM. This manuscript is a direct follow-on to the recently published manuscript by the authors (Turk et al., 2020), which 110 describes in detail the APR-2 and DAWN data for the 10 June 2017 flight date investigated here. In particular, the graphics and discussion in the Turk et al. (2020) manuscript specifically tailored the DC-8 flight segments on June 10 into four onehour defined segments. Each of those one-hour segments corresponds to the same assimilation time window used in the NU-WRF data assimilation cycles. The forecast impact is examined with and without (i.e., a control run) the assimilation of the DAWN wind profiles into the model. The role of the data assimilation process is to adjust the model forecast based on any 115 observed data, accounting for errors in the forecast and the observations. The assimilation impact is assessed in two steps.
First, the forecasted precipitation field is compared between the NU-WRF control run and the DAWN assimilation run for each of the four one-hour segments. For both runs, the forecasted precipitation field is compared to the observed APR-2 precipitation structure. Times and areas where the assimilation demonstrated an improved 3-D representation of the precipitation structure are identified. In the second step, the model environmental state fields are compared between the control 120 run and the analysis, to determine how the model state (wind, temperature, moisture) changed in the model as a result of the assimilation of DAWN wind profiles. While this investigation and its conclusions are limited to a single CPEX flight date, the investigation design is appropriate for further investigation of the impact of airborne Doppler wind lidar observations upon short-term convective precipitation forecasts.

125
For the sake of not replicating a large number of figures in this manuscript, the discussion in this manuscript will make frequent reference to specific figure numbers from Turk et al. (2020) (full open access, so all can refer to it). To simplify the nomenclature, the term T2020 is used to cite that manuscript.
During CPEX, NASA DC-8-based airborne observations were collected from the JPL Ku/Ka-band Airborne Precipitation 130 Radar (APR-2) and the 2-µm Doppler Aerosol Wind (DAWN) lidar during approximately 100 flight hours. The performance of DAWN during CPEX is presented by Greco et al. (2020), and the complementary observations of APR-2 and DAWN during CPEX, tailored to this 10 June 2017 case, are presented in T2020. Therefore, only a brief description is provided here.
For CPEX, the APR-2 provided vertical air motion and structure of the cloud systems in nearby precipitating regions where 135 DAWN is unable to sense. Conversely, DAWN sampled vertical wind profiles in aerosol-rich, clear or broken cloud regions surrounding the convection, but is unable to sense the wind field structure within cloud. Figure 1 of T2020 shows the scanning operations of both instruments onboard the DC-8 for CPEX.
APR-2 acquires simultaneous measurements of multiple parameters at both Ku-and Ka-band (14 and 35 GHz), including co-and cross-polarized backscatter, and line of sight (LOS) Doppler velocities of hydrometeors. APR-2 scans cross-track to resolve the 3-D nature of precipitating clouds. (For more 140 recent field campaigns, the APR-2 was modified into APR-3 with the inclusion of a W-band (94 GHz) radar, but this capability was not available for CPEX). APR-2 range (vertical) resolution of 37-m and cross-beam (horizontal) resolution of ≈ 800m at 9-km distance are more than adequate to capture cloud features down to the resolution typical of high-resolution models, and appropriate for comparison in the vicinity of DAWN wind profiles.

145
DAWN is NASA's highly capable airborne wind-profiling lidar with a 2-micron laser that pulses at 10 Hz. DAWN can provide wind profiles view the local wind field from multiple azimuth angles, these LOS profiles are further processed to estimate the 150 profile of the horizontal wind components (u, v) at different pressure levels . In this presentation, these profile data are used for the data assimilation impact studies.

3
NU-WRF configuration and simulations for the June 10 case.
The NASA-Unified Weather Research and Forecasting (NU-WRF) modeling system was used for all cloud-resolving modeling and data assimilation tasks (Zhang et al., 2017). NU-WRF is an observation-driven regional earth system modeling 155 and assimilation system, including physics modules, a satellite data simulation unit (G-SDSU) capable of simulating modernera NASA satellite observations, including the GPM DPR Ku/Ka-band (14/35 GHz) equivalent radar reflectivity profiles, and the GPM microwave imager (GMI) (Matsui et al., 2014), and an ensemble data assimilation system that can assimilate conventional state variables such as wind, temperature and moisture as well as cloud/precipitation affected microwave radiances. 160 For simulations of this flight date, the NU-WRF EDAS model and analysis was configured as specified in Table 1.
Considering the data assimilation approach, the lateral boundary and initial conditions from NCEP already have the conventional data and all other operational data streams assimilated (DAWN data is not part of this). This is a standard and necessary procedure for a regional model to run. It should be pointed that in this investigation, an additional data assimilation of conventional data is conducted also in the regional system NU-WRF EDAS. This ensemble data assimilation is carried out 165 in domain 1. The reasoning for the additional data assimilation is outlined as follows: The WRF model forward integration is configured as 1-way nesting; when the regional model integrates forward, the domain interior states evolve differently and could drift away comparing to the global analysis. Thinking this way, the data impact (such as from the conventional data) at the boundary is lost in the domain interior, thus justifying the existence of regional data assimilation in the domain (i.e., no assimilation at or near the boundary). Technically, one could say that the conventional data are thereby assimilated twice. A 170 more meaningful way would be to view this as "a re-enforcement of the data constraint in the regional model interior".
For this investigation, NU-WRF EDAS was specifically adapted for assimilation also of DAWN profile winds in domain 1 of the regional model. We conducted the DAWN data assimilation to test the impact of these observationsan impact that comes in addition to that of the standard assimilation of the conventional data. 175 The GPM data and the APR-2 radar data were not assimilated in this study. Both of these datasets were used only for model validation. To support this validation, we used the NU-WRF forward simulations (integrations/forecasts) of the geophysical fields as inputs to instrument simulators to produce the synthetic satellite-like (GPM-specific) observablesthe passive MW brightness temperature (TB) at the 13 GMI channels (10.7 through 183.31 GHz), and the DPR equivalent radar reflectivity 180 factor profiles at same frequencies as DPR and APR-2 (14/35 GHz). The only assimilated data in this study, in addition to the NCEP conventional observations, were the DAWNv3 wind profiles. Hence, the improved representation of the precipitation structure in the simulations with DAWNv3 data assimilation is solely the results of assimilating the DAWN winds.
In   The NU-WRF EDAS consists of ensemble forecasts and a central forecast. The ensemble forecasts are used to estimate the flow-dependent background error covariance. The analysis updates the initial conditions for the subsequent central forecast 195 that is not the ensemble mean, though very close to it. The analysis error covariance generates the ensemble perturbations for the ensemble forecasts in the next cycle. The results presented in the paper are from the central analyses and forecasts.
A 3-km inner grid was used for the comparisons with the APR-2 data. In particular, the model assimilation cycle was hourly, incorporating all observations (conventional observations such as radiosondes) and DAWN winds within a 30-min window, 200 centered on the top of the hour. The one-hour forecast was then carried forward for the next hourly assimilation cycle.
Precipitation is accumulated over one hour of model integration, and output at hourly intervals. For example, a 0600 UTC assimilation cycle would incorporate all observations from 0530-0630 UTC. The resultant one-hour forecast at 0700 UTC is used as the background in the next (0700 UTC) assimilation cycle. The precipitation at 0700 UTC represents a one-hour integration from 0600-0700 UTC. 205

210
The four panels of Figure 2 show the cross-section of the DAWN wind zonal (u-component) vector wind field, processed for the one-hour assimilation cycles centered at each of the four one-hour assimilation cycles. In general, the Doppler lidarderived wind vectors are more abundant near upper levels (higher signal-to-noise ratio) and closer to the surface (more aerosols, larger backscatter), with a general loss of signal and less data in the mid-levels. Areas of cloud contamination are shaded in blue color. The NU-WRF EDAS was run in two modes: 1) a control run where only conventional observations (e.g., 215 radiosondes, clear-sky radiances) are assimilated in the National Center for Environmental Prediction (NCEP) model that provides the initial boundary conditions; 2) a second data assimilation run where the DAWN wind profiles were assimilated in addition to the conventional observations.

255
Fortuitously, there was a GPM overpass directly over this region that occurred near 1852 UTC. Figure 5a shows the GMI 89H GHz image, showing the well-developed convection (TB < 200 K) to the north of the target area, but some indication (only a few pixels in GMI) of developing convection inside of the target area (boxed area). In order to provide resolution, Figure 5b shows the DPR Combined Radar-Radiometer Algorithm (CORRA) (Grecu et al., 2016) precipitation rate inside of the boxed area (DPR has 4-km pixel size; individual pixels are plotted for detail). The CORRA product has two variants, a 260 single frequency (Ku-band only) product covering the full 245-km swath, and a dual-frequency (Ku-and Ka-band) product which encompasses the central 120-km swath area where both radars jointly scan. The dual-frequency product capitalizes upon relationships between the different path integrated attenuation to mitigate ambiguities in the assumed hydrometeor size distribution. For purposes of maximizing coverage, the Ku-band product is depicted in Figure 5b  In Figure 5a, there is a thin magenta colored line that runs near the sub-track of the GPM satellite, that lies within the swath coverage of both the DPR Ku-and Ka-band radars. Figure 6 shows the DPR cross section along this line. The resolution of 280 the DPR data has been averaged over a 3x3 area to match better with the resolution of the GMI 89 GHz channel (100 scan lines of GMI, corresponding to about 1000-km along-track distance). The top panel and middle panels of Figure 6 show the Ku-and Ka-band reflectivity profiles. The lower panel shows the trace of each of the 13 GMI channels under this same cross section. Near GMI scan 20, the radar tops are near 10-km, with significant attenuation of the Ka-band profile relative to Kuband below the 4.5-km freezing level (blue dashed line). The developing cell in the boxed area of Figure 5a near (25N, 73W) 285 is near scan line 50, and the widespread convection above 28N is near scan line 20. The area near scan line 50 has less developed cloud above the freezing level, but also significant Ka-band attenuation relative to Ku-band. The passive MW TB for channels < 89 GHz are not fairly similar for these two areas, but the ice scattering signatures at the GMI highest frequency channels (166 and 183.31 GHz) are more evident (significant TB depression) for the developed convection near scans 15-25.

295
These GPM radar and radiometer observations show good agreement with the location of the precipitation after the data assimilation of DAWN profile winds. However, this alone is insufficient to explain what changed, in the model state variables, as a result of the assimilation. Interpretation of the GPM data imply fairly high liquid water contents below the freezing level, 300 indicative of isolated, growing small-scale convection. In the next section, the vertical structure of the NU-WRF simulations is analyzed and contrasted to profile characteristics from this DPR and actual APR-2 radar data.

Comparison of simulated radar profiles and model 2-D and 3-D fields.
As mentioned, NU-WRF provides post-processing options including an instrument simulator option, to forward-simulate satellite observations. This option was used to simulate radar observations at DPR frequencies and passive microwave TB at 305 the 13 GMI channels. For purposes of this study, only the radar simulations derived from the NU-WRF run with DAWN assimilation are shown below.
The simulation of the DPR Ku-band radar observations using the microphysics, water vapor and temperature structure from the NU-WRF analysis at 2000 UTC is shown in Figure 7. The overall extent of the image is the same as the panels in Figures Figure 7 near (25.5N, 72.9W). While the actual developing convection shows > 35 dBZ max reflectivity in the N-S Ku-band crosssection), NU-WRF modeled these as shallow clouds, limited to < 3-km vertical extent in simulated DPR cloud tops. For the 340 N-S cross-section, even within these very shallow all-liquid clouds, 10 dB Ka-band path attenuation is present relative to Kuband, in accord with the associated DPR overpass indicating the presence of very high liquid water content.
To compare these simulated profiles with APR-2 observed profiles, Figure 9 shows an APR-2 cross section (Ku-and Kaband) between 2000-2010 UTC, about midway through the DC-8 flight segment (red dashed line) in Figure 7. Essentially, 345 Figure 9 is a close-up of the APR-2 profile shown in Figure 12 of T2020, but showing both radar frequencies. The clouds in this area are more mature and developed than what NU-WRF had forecasted, with Ku-band cloud top exceeding 8-km level (owing to the APR-2 radar configuration, the radar was unable to sense in the 1.8-km zone below the DC-8 flight altitude).
Strong differential attenuation (i.e., Ku-minus Ka-band difference increasing closer to the surface) was noted near scan 170, indicative of high liquid water content below 4-km. 350  Figure 11, the associated vertical cross-sections show the 360 rapid growth of the cloud near (25.5N, 73W), with 45 dB radar tops near 10-km, more in accord with the APR-2 structure during this time (see Figure 15 in T2020). In the right column of Figure 11, the cloud near (25N, 73.3W) has intensified to near 45 dB but has developed to only 5-km radar tops, in both the N-S and E-W directions. The Ka-band attenuation is severe in both cells, with the radar signal being lost (below simulated DPR detection limits) before reaching the surface.

375
To compare these simulated profiles with nearby observed profiles, Figure 12 shows an APR-2 cross section (Ku-and Kaband) between 2115-2125 UTC, near the end of the DC-8 flight segment (blue dashed line) in Figure 10 (essentially, Figure   12 is a close-up of the APR-2 profile shown in Figure 15 of T2020). The clouds in this area are more mature and developed 380 than what NU-WRF had forecasted, with Ku-band cloud top exceeding 8-km level and especially strong differential attenuation (below 4-km level) near scan 200. In summary, the control NU-WRF simulations failed to produce precipitation inside the box that was sampled during the June-10 flight mission. In contrast, according to both APR-2 and the GPM satellite observations, convection was observed in that 390 box. After the assimilation of the DAWN winds, NU-WRF developed convection in the places where it was observed.
However, the development of the convection was delayed by about 1-hour. While the NU-WRF simulations well-represented the location of the developing precipitation (even though delayed in time), the associated growth in the heights of vertical precipitation structure also evolved slower. The NU-WRF simulated Ku/Ka-band radar tops did not reach the vertical development noted by APR-2, but they had better agreement to APR-2 cloud structure in the 2030-2130 period (an hour behind 395 the observed precipitation). Hence, the assimilation of the DAWN winds resulted in the development of clouds and precipitation, even though delayed, where it was observed. Interestingly, NU-WRF succeeded in producing the observed characteristics of the clouds and precipitationpredominantly shallow, non-glaciated clouds with high liquid water content, noticed in the strongly attenuated Ka-band radar profile.

4.2
Impact of the DAWN data assimilation on the model wind, temperature and moisture structure: Analyzing 400 the analysis increments The analysis above has focused upon the change to the convective structure that occurred following the assimilation of the DAWN wind profiles in the NU-WRF EDAS. As noted in Figure 4, the assimilation of the DAWN winds, beginning with the 1900 UTC assimilation cycle, produced subsequent precipitation in the area where it was observed, whereas the control run produced no precipitation in the same region. A relevant question is how the assimilation of the DAWN winds contributed 405 to the subsequent development of precipitation in the area where it was observed in reality. While this study is not of a scope to fully answer these questions, one can compare the environmental state (structure of wind, potential temperature and water vapor) that was produced by the control model forecast, to the analysis produced after the assimilation cycle, to address the impact of the DAWN data assimilation on modifying the initial environment in which the subsequent convection develops.
Indeed, the assimilation of the DAWN winds, even at a single time step, produced a very significant impact to the associated 410 wind, temperature and moisture structure, as further illustrated below.
To examine, the direct impact of the DAWN data assimilation on the wind structure of the model is presented. The environmental wind shear conditions that were present at the time of the first assimilation period (a ±30 min assimilation window centered at 1900 UTC) are illustrated in Figure 7 in T2020, which depicts the vertical wind shear conditions inferred 415 solely from DAWN profiles during this period. There was sustained directional wind shear between 2-km and 8-km levels in the area west of 73W, oriented from west to east, whereas between 2-km and 6-km the shear was weaker and oriented more south to north. For a particular example, Figure 2  These wind conditions were provided to the NU-WRF EDAS in the DAWN assimilation run, and absent in the control run. 420 Figure 13 presents the analysis increments (assimilation-minus-control) introduced in the vertical profiles of the zonal (u) and meridional (v) wind components after the 1900 UTC assimilation cycle. A close look suggests the assimilation of the DAWN winds resulted in a decrease of the zonal wind shear immediately next to the convective development, at the end of the DC-8 track (80-90 km distance). This is manifested by more positive valued increments near the surface and more negative increments at upper levels (above 6.25 km). The meridional component increments suggested the reverse; i.e., an increase in 425 meridional shear in the 80-90 km range (near the subsequent precipitation). This is manifested by the more negative increments at that range near the surface versus the more positive increments at the same range but at higher altitudes.  Figure 14 shows the NU-WRF 500-m height model wind field after the first assimilation cycle (centered at 1900 UTC). The wind increments are broken down into their zonal and meridional components and plotted as a difference field (assimilation-435 minus-control). The red curves indicate the approximate boundary (high gradient) between the negative-and positive-valued contours, with a focus on the area where the subsequent precipitation would develop in the data assimilation run. The zonal winds (left panel) reveal positive differences to the west of this boundary (stronger westerly winds) and negative differences to the east (stronger easterly winds), both indicative of stronger low-level convergence of the zonal wind in the assimilation run versus the control, i.e.a stronger zonal forcing after the assimilation. The meridional winds (right panel) reveal positive 440 differences (increments) to the south of the boundary line (stronger southerly winds) and negative differences to the north (stronger northerly winds), also indicative of stronger low-level meridional convergence in the vicinity of the subsequent development of precipitation. Therefore, the assimilation of the DAWN winds modified the low-level wind field in such a way as to strengthen convergence (both zonal and meridional, almost at the same location) in a narrowly-focused zone. This line of convergence provided favorable dynamical conditions to promote further vertical cloud development. 445

455
In a similar fashion to Figure 14, Figure 15 shows the NU-WRF water vapor mixing ratio increments (left panel) and the potential temperature increments (right panel) resulting from this same DAWN assimilation cycle (centered at 1900 UTC), also plotted as a difference field (assimilation-minus-control, i.e., an analysis increment). The region of strongest positive increments (contours) is shown in the red shaded area. It can be seen from Figure 15 that the assimilation of the DAWN winds during this time produced positive moisture and temperature increments in highly-overlapping areas. While only new 460 wind data were assimilated (no new moisture data), the resulting increments in moisture and temperature were produced through the background error covariances, generated by the model ensemble. Both the higher moisture and the warmer temperatures resulted in enhancing the convective potential in these regions.

470
Further observational evidence of the enhanced moisture in the area of the observed convective development comes from combining a number of retrievals of the total precipitable water (TPW) and a passive MW rain index (RI), provided by a variety of NASA, NOAA and EUMETSAT satellite systems (Figure 16). The RI is a multi-channel index combining brightness temperatures (TB) in the 10-90 GHz range (Hristova-Veleva et al., 2020). The RI from the same GMI overpass 475 (1852 UTC) shown in Figure 5, and the 6-hour composite (14-20 UTC) of the total precipitable water vapor (TPW) produced from the Microwave Integrated Retrieval System (MiRS) , are shown in the left and right panels of Figure

485
These overall conditions (enhanced near-surface wind convergence, accompanied by enhanced low level moisture and temperature) that resulted from the DAWN assimilation provided favorable conditions for eventual vertical development.
After the assimilation, the subsequent forecast produced precipitation where there was none in the control run. Where the precipitation developed, it appeared more organized. Figure 17 provides a conceptual interpretation of the underlying drivers and consequences. This figure summarizes differences in precipitation in relation to the analysis increments introduced by the 490 assimilation of DAWN data during the 1900 UTC assimilation cycle. The features in Figure 14 (zonal and meridional wind convergence) and in Figure 15 (increased moisture, temperature) are co-registered in Figure 17a. The analysis increment produced surface convergence, co-located with increased moisture and temperature. Figure 17b overlays these features on top of the resultant precipitation in the model integration interval (1900( -2000, following this period of DAWN data assimilation, shown earlier in Figure 4b. These dynamic and thermodynamic components increased in a very coherent 495 way, strongly suggesting that the resultant precipitation in the model integration interval was the consequence of the DAWNassimilation inducing increasing of the convective potential exactly where the precipitation was observed. This resulted in convective initiation and the subsequent development of precipitation where there was none in the control run. From this, can one identify the important self-aggregating processes that allowed this initial convection grow upscale to produce the radar extensive area of precipitation, within the box of the flight area (DC-8 flight line in Figure 17b). 500

510
As precipitation develops it produces precipitation-loading-driven downdrafts that lead also to the entrainment of drier midlevel air -ready to evaporate the precipitation when the two come in contact. When precipitation evaporates, it further enhances the downdrafts, making them more negatively buoyant because of the loss the latent heat needed for the evaporation. When these downdrafts reach the surface, they spread out, being colder and denser than the surrounding air. This leads to the creation 515 of the so-called cold poolsareas that are colder than the surrounding air, causing them to spread radially outward (Schlemmer and Hohenegger, 2014) as density currents. The cold pools created by the individual downdrafts interact with each other and the mesoscale flow organizes them into bigger entities. In turn, these precipitation-induced cold pools lead to the initiation of new convection along their leading edged by creating favorable conditions of forced lifting of the environmental air, affecting the location, strengths and organization of the convection that develops later on. As this environmental air is warmer and has 520 more moisture, the induced lifting comes as an additional benefitting component, further improving the chances for the development of new convection and precipitation.
These mechanisms behind the formation and dissipation of the cold pool process (Zuidema et al., 2017;Grant and van den Heever, 2016), and its identification in cloud resolving model simulations (Drager and van den Heever, 2017) are beyond the 525 scope of this investigation. Here, the role of the cold pools in terms of their structure and relationship to the precipitation development is addressed. The control run and the assimilation run are compared and contrasted in terms of the near-surface temperature anomalies that develop. The two model forecasts are compared using the 2-m air temperature anomaly (difference from the initial state) as a footprint of the cold pool structure. Figure 18 shows the 2000 UTC analysis 2-m temperature anomaly (shaded), precipitation (contoured in thin red lines) and wind at 500-m level (vectors). The thick solid red line denotes 530 the approximate boundary of the cold pool that was detected in dropsonde observations taken during the June 10 flight (Zipser and Rajagopal, 2018).

540
Both control and assimilation run show cold anomalies (cold pools) that are closely related to the precipitating areas, as it should be expected. However, the assimilation run shows much more intense and bigger cold pools. Two maxima are of interest. First, the smaller one to the north that is closely related to the precipitating area in the assimilation run (Figure 18b) 545 that was not present in the control (Figure 18a). This cold pool, while not very big in areal extent, is closely related to the one observed in the dropsonde data as marked by the thick red line. Second, the much bigger, better organized and stronger cold pool found to the south-east. This extensive area of cold anomalies is related to the much bigger and organized precipitating system there. Interestingly, this area of organized precipitation shows further signs of upscale growth as revealed by the precipitation structure at the later 2000-2100 UTC and the 2100-2200 UTC periods revealed in Figure 4. 550
This manuscript has presented the results of the impact to the forecasted precipitation structure that resulted when DAWN wind vector profiles were assimilated by the NASA NU-WRF EDAS. This study is a direct follow-on to the recently published manuscript by the authors (Turk et al., 2020), which describes in detail the DAWN observations during each of the one-hour 555 periods used in the assimilation, and the APR-2 data for the 10 June 2017 flight date used for this impact study. The study focused on (a) understanding whether (and if so, how) the assimilation of the DAWN winds impacted the subsequent development of convection and precipitation, and (b) determining what environmental factors were modified by the assimilation and understanding how would they have possibly impacted the development of precipitation.

560
The impact was examined from two directions. First, the structure and timing of the model precipitation field was examined relative to that observed by the APR-2 radar data collected coincidently with the DAWN data. The Goddard SDSU instrument simulator was used to simulate DPR (Ku/Ka-band) radar profiles for direct comparison to APR-2 and a GPM overpass that occurred during the first data assimilation cycle. Second, the structure of the NU-WRF model winds, temperature and moisture was contrasted between the model control run and the model data assimilation run. With these prognostic variables, the pattern 565 of convergence of moister air was examined, to explain the role of the thermodynamics in the evolution of the resultant model precipitation horizontal and vertical structure, and how the vertical structure evolved in time relative to the APR-2 observations. During 1830-1930 UTC time interval, dense DAWN observations were sampled in the surrounding environment, notably in the cloud-free region just W-SW of the area of interest (Figure 4 in T2020), showing fairly strong wind shear between the 570 upper and lower levels. This is the time interval just preceding the onset of the precipitation within the DC-8 coverage area noted in the model data assimilation run. While the NU-WRF simulations well-represented the location of the developing precipitation in the subsequent 1900-2000 UTC period, the associated growth in the heights of vertical precipitation structure evolved slower, with better agreement to APR-2 cloud structure in the 2030-2130 period. In accord with actual DPR data collected earlier (1852 UTC), NU-WRF produced shallow, non-glaciated clouds with high liquid water content, noticed in the 575 strongly attenuated Ka-band radar profile.
Assimilation of the DAWN winds in NU-WRF EDAS, even at a single time step, produced a very significant impact. Analysis of individual variables revealed that the assimilation of the DAWN winds resulted in important and coherent modifications of the environment. It led to increase of the near surface convergence, 2-m air temperature and water vapor, creating more 580 favorable conditions for the development of convection exactly where it was observed. The realism of the forecasted precipitation structure was shown by comparisons with nearby satellite and aircraft observations. Comparison to observations from APR-2 (and a fortuitous GPM satellite overpass) shows a much-improved precipitation forecast after the assimilation of the DAWN windsdevelopment of precipitation where it was observed but not present in the control, and more organized structure where the precipitation eventually developed. Most importantly, the assimilation produced a much more intense and 585 organized cold pool, similar to the one detected in a separate analysis of the dropsonde data collected during the DC-8 mission flight on that day. It should be pointed that a similar result was noted by Cui et al. (2019) in their DAWN assimilation study (modification of the near surface wind convergence field), taken from a different modeling system and two CPEX flight dates different than the date studied here.

590
These findings add to the growing body of evidence that suggests that assimilation of high-resolution, high-precision Doppler wind lidar profiles into convection-allowing models improves the analysis of the environment, creating conditions that favor convective storm development and upscale growth. While encouraging, these findings represent a limited number of cases.
A longer assimilation period, and more flight dates, are needed to establish any repeatable impact from which to draw conclusions. This is challenging, given limited duration flight dates and fairly short (typically 3-4 hours) aircraft on-station 595 times that encounter convection in its early formation stages. Future research can also address many other important questions.
Are cold pools more effective at initiating new convection in some environmental conditions versus others, including variable aerosol loading? (aerosol effects were not addressed here). In each case, one can relate the environmental parameters to the strength and the structure of the cold pools, and then their ability to generate and continuously support new convection at their leading lines, eventually resulting in an upscale growth of the system. The proposed NASA-ESA CPEX-AW field campaign 600 will provide the opportunity to fly the APR-3, DAWN and the High Altitude Lidar Observatory (HALO) (Bedka et al., 2020) alongside available ADM-Aeolus observations, in the eastern Atlantic where African easterly waves interact with the Saharan air layer (Zipser et al., 2009). The HALO instrument provides aerosol and water vapor profiles (a missing component of these airborne data), to complement the DAWN wind-sensing capability. The investigation design presented here, based on the availability of concurrent precipitation radar observations, is appropriate for further investigation of the impact of airborne 605 Doppler wind lidar observations upon short-term convective precipitation forecasts.