Research‎ > ‎

Pacific Northwest

Working closely with the Cloud and Precipitation Processes and Patterns Group headed by Dr. Sandra Yuter of NCState, I contributed to the modeling component of a dual observational-modeling study of embedded convection within stratiform precipitation centered on the Pacific Northwest. I worked on this research during my first year and a half of graduate school while taking classes and thinking more in-depth about what research I'd like to conduct for my dissertation. That soul-searching and the fact that the grant ran out on this project led me down alternate paths so that I did not continue the investigation after the start of 2013. Here is a summary of my findings between 2011-2012.

Discussion of Modeling Intermittent Embedded Convection in the Pacific Northwest using the WRF Model

 1. Introduction

     The observational study of Cunningham and Yuter (2013) centered on the Portland, OR radar (KPDX) examined the frequency and environment of intermittent embedded convection.  Embedded convective cells within stratiform precipitation are cells that locally increase the rainfall rate and are indicated by higher radar reflectivities than the surrounding stratiform region of precipitation. Embedded convection was found to occur for approximately 20% of observed categorized time periods from 2002-2008. Through a statistical analysis of the environment to determine the sources of instability that led to the likely generation of these cells, the instability as calculated from most unstable parcel height had a mean value of 2.5 km, which is 2.0 km higher than that of purely convective periods. Therefore, mid-level instability is an important ingredient of the environment of the embedded convective cells.

    There is a limited observational network to study the environment for times during which embedded convection was observed. The available gridded data has coarse horizontal and vertical resolution and may not provide enough detail to understand the reasons why embedded convection occurs. Therefore, the Weather Research and Forecasting (WRF) Model was employed to simulate a selection of cases and further examine the environment.

2. Methodology

a. Model Specifics

    The WRF version 3.4 was used to simulate various cases for 24-36 h. Four domains with one-way nesting are shown in Fig. 1 and are 12, 4, 1.33, 0.44 km horizontal resolution from the outer-most to the inner-most. All domains contained 39 vertical levels. The initial and boundary conditions, snow cover and soil information came from the 6-hourly 1.0° GFS gridded analyses. The planetary boundary layer was parameterized using the YSU scheme (Hong et al. 2006). The microphysical parameterization was the WSM 6-class scheme (Hong and Lim 2006). The Kain-Fritsch cumulus scheme (Kain 2004) was used only for the 12-km domain.

Fig 1. Domains used in the WRF simulations.                             Fig 2. Location and Orientation of Cross section
                                                                                                                         used for evaluation.

b. Selection of Cases

    Cases were selected from the database created by Jeffery Cunningham of three hour periods of observed embedded convection within stratiform precipitation. 10 cases were selected from 2004-2008 based on the availability of data (prior to 2003 was mostly unavailable) and chosen based on the number of 3-hour periods associated with an event, with an average of three 3-hour periods per event. Events counted separately if they occurred more than 24 hours apart. Table 1 shows the cases chosen and includes information on both the observation and simulation. Simulations were initialized at least 9 h prior to the time of the embedded observation to allow for the model to spin-up from its initial state and hopefully agree with the timing of the observation.

Table 1. Selected cases of observed embedded convection within stratiform precipitation and the WRF simulation times.

Case Number


Date and Time of

Observed Embedded Convection


Simulation Date and Time

Case 1

6-7 November 2006




12 UTC 6 Nov - 12 UTC 7 Nov

Case 2

14-15 December 2006




00 UTC 13 Dec - 00 UTC 15 Dec

Case 3

18-19 December 2007


00 UTC 18 Dec - 00 UTC 19 Dec

Case 4

20-21 December 2005


00 UTC 20 Dec - 00 UTC 21 Dec

Case 5

02-03 November 2006


06 UTC 2 Nov - 06 UTC 3 Nov

Case 6

16-17 November 2007


06 UTC 16 Nov - 06 UTC 17 Nov

Case 7

10-11 January 2006


12 UTC 10 Jan - 12 UTC 11 Jan

Case 8

26-27 December 2005



18 UTC 25 Dec - 18 UTC 26 Dec

Case 9

23-24 March 2008


06 UTC 23 Mar - 06 UTC 24 Mar

Case 10

18-19 November 2004


00 UTC 18 Nov - 00 UTC 19 Nov


c. Evaluation of Cases

    The cases were first subjectively evaluated by plotting reflectivity at various vertical levels less than 3 km high using RIP version 4 and later NCL. Vertical cross-sections of reflectivity were also plotted (Fig.2). Notes were taken on the structure and duration of the identified embedded convection, if it was able to be simulated. For a subset of cases for which embedded cells were observed, reflectivity was plotted at several levels less than 3 km in time-longitude diagrams or Hovmoller diagrams. The latitude band over which reflectivity was averaged was narrowed from over 3° to approximately 0.5° to compare results and try to resolve the movement of the simulated and observed cells (Fig. 5).

    The environment of the cases was analyzed comparing the simulated variables to observations at two sounding locations, Salem, OR (SLE) and Quillayute, WA (UIL) seen in Fig. 1. The vertical profile of relative humidity and stability as analyzed by the vertical difference in ϴ, ϴe and ϴes and the moist squired Brunt-Vaisala frequency were compared. Infrared satellite imagery was used to identify the location of SLE relative to the synoptic-scale feature the embedded convective cells were attributed to.

    A case study was performed in-depth on a relatively “good” case in which embedded convective cells were representatively simulated. The surface time-series of accumulated precipitation, temperature, relative humidity, wind speed and direction were compared. Some of the results will be presented in Section 3b.

3. Results and Model Evaluation

a. Overall

    Out of the 10 simulated cases, 3 were subjectively identified as somewhat accurately simulating the embedded convection within stratiform precipitation in the vicinity of the KRTX radar. The structure of the embedded cells in the horizontal reflectivity plots gained more detail as the horizontal resolution increased from 12 km to 1.33 km but there was little change in structure between 1.33 km and 0.444 km (Fig. 3). Notes on each case that were made from the plan-view and vertical cross-section plotting of simulated reflectivity are found in Table 2 with the three “best” simulations marked with an asterisk (Cases 3, 6 and 7).

Figure 3. Finescale structure became more defined with an increase in horizontal resolution from 4 km to 1.33 km, but not much was gained from 1.33 km to 0.444 km for a representative snapshot from Case 6: 16-17 November 2007.

Table 2. Notes on the simulated reflectivity as compared to observations with the KRTX radar.


Precipitation Character

(Horizontal dBZ)

Vertical Cross Section of dBZ

Case 1: 6-7 Nov 2006

KRTX: Cells observed as slow and tied to terrain moving WSW-ENE.

KRTX: Embedded cells seen with possible seeder/feeder structure.


WRF: Simulated dBZ showed linear convection coming onshore and then additional linear convection forming. With higher resolution (1.33-km and 444m), can pick out individual cells within linear feature.

WRF: 6 Nov 12-18 UTC can see embedded structure, and convection is deep (extending to > 5.0 km). A few cells can be picked out propagating across terrain from 18-00 UTC.

Case 2: 13-15 Dec 2006

KRTX: 13/0900: NW-SE linear embedded feature observed. 14-15/1800-0000: KRTX had cells from SW-NE

KRTX: Embedded cells observed.


WRF: 13/0900: NW-SE linear embedded feature observed. 14-15/1800-0000:  Larger enhanced areas of dBZ overdoing convective features.

WRF: No embedded evidence only isolated cells.

*Case 3: 18-19 Dec 2007

KRTX:  SSW-NNE movement of about 40% embedded cells and then scattered cells later in the period.

KRTX: Embedded cells observed.


WRF:  More isolated than embedded cells simulated.

WRF:  Weak embedded signatures became weaker lee of the Costal Range and became isolated cells upon approaching the Cascades.

Case 4: 20-21 Dec 2005

KRTX:  Linear feature moving WSW-ENE over a period of about 3 hours with embedded cells.

KRTX: Embedded cells observed.


WRF:  Lacks linear embedded feature and only simulates scattered, isolated cells.

WRF:  Strongest embedded cells occurred at Coastal Range crest and weakened in the Willamette Valley but became enhanced again at Cascades.

Case 5: 2-3 Nov 2006

KRTX: SW-NE movement of embedded cells.

KRTX: Scattered cells progress into more organized precipitation structure with embedded cells.


WRF:  Slower with onset of heavier precipitation and has a large clump of convective features but then becomes smaller scattered cells rather than embedded.

WRF: Strong cells approach Coastal Range, weaken in lee, but cut off due to timing of loops. May want to extend longer to 0300 UTC on the 3rd.

*Case 6: 16-17 Nov 2007

KRTX: SE of KRTX shows embedded features moving SW-NE.

KRTX: Embedded cells observed and possible mid-level feeder cell from weak echo between 4-6 km.


WRF: Embedded cells SW of location of KRTX moving SW-NE.

WRF: Possible seeder-feeder mechanism (4-6 km), however predominately shallow convective cells seen with dBZ enhanced at Coastal Range then decreased in Willamette Valley then enhanced at Cascades.

*Case 7: 10-11 Jan 2006

KRTX:  Large fraction of observable reflectivity embedded with cells that move from W-E.

KRTX: WRF weaker embedded cells than observed at KRTX.


WRF: Slow on the onset of precipitation and simulates embedded cells coming onshore. Frontal passage is a little slower and all embedded convection is pre-frontal.

WRF: Weak embedded cells, possible seeder/feeder lee of Coastal Range and windward of Cascades.

Case 8: 25-26 Dec 2005

KRTX: 26/1200: (better) Clear frontal structure with cells moving SSW-NNE.

KRTX: Embedded cells observed.


WRF: 26/1200: Clear frontal structure with embedded cells out ahead of it.

WRF: Weak enhanced reflectivity tied to the terrain and cells are isolated and all tied to the windward Coastal Range.

Case 9: 23-24 Mar 2008

KRTX:  Cells moving from SW-NE embedded in the flow.

KRTX: Embedded cells observed.


WRF: Weak embedded cells form but mostly isolated convection is simulated.

WRF: Weak cells with flow, mostly tied to terrain. Slight evidence of seeder/feeder.

Case 10: 18-19 Nov 2004

KRTX:  Embedded cells moving NW-SE with frontal boundary.



WRF: Good timing of FROPA but failed to resolve any embedded enhancements of simulated reflectivity as in it was much weaker than observed KRTX.

WRF: Enhanced reflectivity at coastal range then decreased in valley. Slight evidence of feeder cell.

    There were subtle differences in the environment between cases, but it is unclear why the WRF was able to resolve the cells for one case but not for another similar case. The cells were resolved best for the warm sector or pre-cold frontal environment (5 cases) and worse for cold frontal passages (2 cases). There were no significant differences, via rough calculations at SLE and UIL, between observed and simulated surface temperatures, 600-hPa temperatures and surface relative humidity values. A selection of the calculations at SLE for all 10 cases follow. The average of the mean absolute error between the simulated and observed (WRF – Obs) surface temperature at SLE was 2.70833°C. The average of mean error for 600-hPa temperature was 0.8333°C. The difference in the average of the raw surface relative humidity values was -5.64286% which meant that the WRF was tending to be too dry at the surface. To isolate the calculations to just the three “best” cases, the surface temperature mean absolute error was 1.25°C, the 600-hPa temperature mean absolute error was 1°C and the surface relative humidity was -5.69231%. These calculated values were similar for “good” cases and “bad” cases. In noting the importance of a mid-level unstable layer, vertical profiles of dtheta-e/dz were calculated using GFS 0.5° gridded data at first and then raw radiosonde observations second and compared to the simulated profile. An unstable layer was defined as a layer ≥ 15-hPa where dtheta-e/dz < 0 K km-1. Qualitatively, between the “good” and “bad” cases the WRF resolved the profile of dtheta-e/dz similarly and as such inconsistently. The mid-level unstable layers were not always captured for the “good” cases. What may be the difference is the surface stability. The WRF was more consistently unstable near the surface while the observations were more stable (not shown).


b. Case Study of Case 6: 16-17 November 2007

    The case of 16-17 November 2007 (Case 6) was subjectively determined to be the “best” case out of the 10 selected cases mainly based on simulated reflectivity cell structure and duration and therefore received more attention for further analysis. As Figure 3 showed above, there were clearly defined simulated embedded cells. Weak and shallow embedded cells were seen crossing the coastal range and into the Willamette Valley in the cross-section (not shown). An analysis of the time-height of reflectivity at the site in Portland, OR during this event shows that between 1800-2300 UTC the WRF captures some embedded cells with an indication of the vertical extent of higher reflectivity values approaching 3-4 km which may show evidence of seeder cells (Fig. 4).

Figure 4. Time-height reflectivity at Portland, OR for the WRF simulation and the observed values.

    Another way of showing evidence of the cells is by plotting reflectivity within Hovmoller diagrams. Although plots were made with a variation of the latitude band used, only the plots from the region indicated by the thin embedded box in Figure 5 are shown. The reflectivity data were plotted < 1 km for both the 4 and 1.33-km domains of the WRF simulation and KRTX data in Figure 6. Comparing the WRF results with that of KRTX, the cells are not as clearly defined as progressing to the east-northeast as they are in the observation. The increase in resolution from 4 km to 1.33 km also does not increase the definition of the transient cells in these diagrams. This may be due to the simulated cells being smaller (untested) or having a shorter duration than observed (also untested).

5. Area over which reflectivity values were calculated and plotted for the time-longitude diagrams.

Figure 6. Hovmoller diagrams for the latitude region indicated in Fig. 5 for two WRF domains and KRTX data at about 0.5 km for the WRF simulations and the base level for KRTX.

    The time-series of accumulated precipitation at SLE is shown in Figure 7. The 4 and 1.33-km domains both over-simulated the precipitation received at SLE but the 0.444-km domain precipitation was the closest to the observed value. A reason for this may be the simulated cells were too large in the relatively larger domains or perhaps there were more of them passing over SLE (untested).

Figure 7. Accumulated precipitation for Case 6 for d02, d03 and d04 (4, 1.33, 0.44 km respectively) and the observed amounts (ob).

    The time-series of surface temperature and relative humidity at both SLE and UIL are useful for analysis (not shown). The WRF simulated values from multiple domains show consistently warmer values at both locations by little more than 1°C. The WRF simulation also trended to be drier with time than observed, ending each simulation about 10% drier at both locations.

    Comparing the simulated environment in the vertical, the sounding-calculated vertical profiles of ϴ and ϴe were compared at SLE and UIL and the results are provided in Figure 8 for 00 UTC 17 November 2007. The top panels show that the various horizontal resolution interpolations to those points agreed with the generally stable trend, but missed a 1.0-km deep layer of weak stability and shallower layer of instability at SLE from roughly 3-4 km. The simulation also underestimated the stability between 2.5-3.5 km at UIL. These discrepancies are more evident in the vertical profiles of equivalent potential temperature (bottom panels of Fig. 8). What these results show is that while the WRF can produce embedded cells, they may not be directly related to the mid-level instability that is hypothesized to be necessary for their formation but possibly other processes. Other cases that performed worse than this case were able to resolve mid-level unstable or weakly stable layers but weak embedded cells were observed from the horizontal simulated reflectivity (not shown).

Figure 8. 0000 UTC 17 November 2007 vertical profiles at SLE (left panels) and UIL (right panels) of potential temperature (top panels) and equivalent potential temperature (bottom panels).

4. Summary and Future Work

    The evaluation of the 10 cases in which embedded convection within stratiform precipitation was observed but not necessarily simulated proved less straightforward than initially thought. The increase in horizontal resolution of the model improved the detailed structure of the resolved cells from 12 to 4 km and from 4 to 1.33 km, but little change was observed in the increase from 1.33 to 0.444 km. The environment of the simulations was not consistently biased towards being too warm, too dry or too stable across the board when comparing “good” cases with “bad” cases. Available moisture and stability are likely the reasons why the cells were too isolated in nature (e.g. near-surface environment was too unstable) rather than embedded. However, quantifying those issues were not very conclusive.

    This study started out with comparing two PBL schemes (YSU and MYJ) and three different microphysical schemes (WSM6, SBU-Lin, Thompson) for one single case (Case 1). The 10 simulations were then run just once with the YSU PBL scheme and WSM6 microphysical scheme only. Perhaps if an ensemble of the different combinations was created there may be “better” simulated cases or more than 3 out of 10 may pass the subjective test of resolving the cells. Another issue is that although the KRTX radar cannot scan the cells as they approach from the west over the Pacific Ocean until they are closer to the radar, the WRF seemed to always have them originating from far over the ocean (i.e. wherever the larger synoptic system was at the initialization time). Unfortunately, there were no observations outside of using the Quillayute, WA station to represent the ambient oceanic environment. Perhaps the NARR or GFS 1.0° analyses could be used to stand-in for observations to analyze the oceanic environment in more detail and maybe there are more differences between simulated and “observed” values of stability and moisture. In summary, the WRF can simulate embedded convection within stratiform precipitation however it is unclear at this time why it can versus why it can’t for similar synoptic environments.