An Evaluation of Simulated Microphysics over Terrain during the OLYMPEX Field Campaign. Robert John Cuson Conrick

Transcription

1 An Evaluation of Simulated Microphysics over Terrain during the OLYMPEX Field Campaign Robert John Cuson Conrick A thesis Submitted in partial fulfilment of the requirements for the degree of Master of Science University of Washington 2018 Committee: Clifford F. Mass Lynn McMurdie Robert A. Houze Program Authorized to Offer Degree: Atmospheric Sciences 1

2 Copyright 2018 Robert John Cuson Conrick 2

3 University of Washington Abstract An Evaluation of WRF Microphysics over Terrain during the OLYMPEX Field Campaign Robert John Cuson Conrick Chair of the Supervisory Committee: Professor Clifford F. Mass Department of Atmospheric Sciences The OLYMPEX field campaign of winter offers the opportunity to assess the performance of model physics over a coastal mountain range. This thesis explores precipitation and microphysics simulations, and investigates the ability of full-physics simulations from the Weather Research and Forecasting Model (WRF) to simulate Kelvin-Helmholtz waves. First, moist physics is assessed from the perspective of precipitation for a variety of microphysics and planetary boundary parameterizations. Over the period from November 2015 to February 2016, WRF and the 13-km NOAA/NWS GFS model precipitation accumulation was underpredicted over the windward (western) side and crests of the Olympic Mountains. Underprediction was greatest where observed precipitation was largest. Two major atmospheric river type events (November 12-15, November 16-19) were examined in detail using a variety of physics choices. Improved simulation skill was generally limited to increasing resolution from 36 to 12-km, with substantial variability among microphysics and PBL choices. For the two cases noted above, warm-period precipitation maxima occurred farther up the valley than observed, with a tendency for underprediction. Differences in simulated microphysics are also examined, with 3

4 results indicating the tendency of WRF to produce rain particles that are larger and less numerous than observations. Next, two Kelvin-Helmholtz wave (KH; December 12 th and December 17 th ) events were simulated to further evaluate model physics at short spatial and temporal scales. Waves were realistically simulated at 444-m grid spacing, including reproducing the location and structure of waves. Waves were shown to be resolution-dependent and only adequately represented at 444-m grid spacing due to their 3-5 km wavelengths. In both cases, waves developed as the result of an intense shear layer, caused by low-level easterly flow. The Olympic Mountains enhanced wave amplitudes, and removing the Olympic Mountains eliminated wave activity in the December 12 th case. When waves were within the melting level (December 12 th ), simulated microphysical fields experienced considerable oscillatory behavior; when waves were below the melting level (December 17 th ), the microphysical response was attenuated. Turning off moist physics and latent heating resulted in weaker KH waves, while varying physics choices resulted in variability in the amount of hydrometeors produced and the strength of the waves vertical velocities. 4

5 Table of Contents i. List of Figures... 7 ii. List of Tables Chapter 1. Introduction Previous Evaluations of Simulated Microphysics in Complex Terrain OLYMPEX in the Context of Past Field Campaigns Kelvin-Helmholtz Waves over Complex Terrain Purpose and Scope Chapter 2. Model Configuration and Data Model Configuration Microphysical Data Chapter 3. Results from Precipitation Evaluations UW real-time WRF and GFS Evaluation Fidelity of incoming moisture flux during OLYMPEX UW WRF and GFS forecast evaluation: November 01, February 01, Results from Two Atmospheric River Events Description of Events Precipitation Evaluation Area-Averaged Precipitation Evaluation by Elevation Windward vs. Leeward Slopes Precipitation in the Quinault Valley Microphysical Evaluation Chapter 4. Results from Two Kelvin-Helmholtz Wave Events Case I: December 12, Observed and simulated mesoscale conditions and wave generation Impact of KH waves on microphysics

6 Impact of choice of physics schemes Case II: December 17, Observed and simulated mesoscale conditions Impact of KH waves on microphysics Impact of choice of physics schemes Comparison to a conceptual model of KH wave microphysics Chapter 5. Concluding Remarks and Future Work Acknowledgements Bibliography

7 i. List of Figures Figure 2.1: Map of the WRF-ARW domains used in this study. The outer domain has 36-km grid spacing, with d02, d03, d04, and d05 indicating the 12, 4, 1.33-km, and 444-m domains, respectively. Figure 2.2: Map of OLYMPEX observing stations used in this thesis, including the NPOL radar. Elevation is shaded. The blue dashed line is a cross section used for analyses in future sections. Figure 3.1: hpa moisture flux (IVT) observed at the NPOL location compared to UW real-time WRF for (a) November February 2016, and from an ensemble of microphysics schemes for (b) November and (c)november The black lines are from observing soundings, and the blue lines are from the WRF model. Figure 3.2: Accumulated precipitation error averaged all stations in the 1.33 km domain (Fig. 4) for November February 2016 as a function of resolution or model. Figure 3.3: November February 2016 mean precipitation error of the UW WRF for 36, 12, 4, and 1.33-km (WRFGFSD1, WRFGFSD2, WRFGFSD3, WRFGFSD4, respectively) and the 12-km MM5 model (MM5ETAD2). Values are averaged over all stations within a 1-by-1 degree box. Figure 3.4: November February 2016 mean coefficient of efficiency for precipitation of the UW WRF for 36, 12, 4, and 1.33-km (WRFGFSD1, WRFGFSD2, WRFGFSD3, WRFGFSD4, respectively) and the 12-km MM5 model (MM5ETAD2). Values are averaged over all stations within a 1-by-1 degree box. 7

8 Figure 3.5: Plot of observed vs. simulated precipitation in the UW WRF 1.33-km domain for the period November 2015 February The dashed line represents a perfect forecast, while the blue and black lines represent the regression line for the 4-km and 1.33-km domains, respectively. Figure 3.6: Precipitation forecast accuracy by station elevation as a percent of observed precipitation for the UW WRF 1.33 km and the GFS 13 km forecasts. Figure 3.7: Synoptic conditions for the November case study: (a,b) 500 hpa and surface analyses from the US National Weather Service at 0000 UTC November 13, and (c) sounding from the NPOL radar at 0305 UTC November 13. Figure 3.8: Synoptic conditions for the November case study: (a,b) 500 hpa and surface analyses from the US National Weather Service at 1200 UTC November 17, and (c) NPOL sounding at 1115 UTC November 17. Figure 3.9: Total precipitation accumulations of selected microphysics scheme simulations in the 1.33 km domain for (a) November and (b) November 16-19, and boundary-layer scheme simulations for (c) November and (d) November Figure 3.10: Map of observing station locations and sites of NPOL and UIL soundings, with the windward (blue) and leeward (green) domains marked. Symbols indicate which network the observations are from (circle = ASOS, square = MesoWest, diamond = OLYMPEX) Figure 3.11: Total precipitation error from all stations shown in Fig. 3.8 for (a) microphysics and (b) boundary-layer scheme simulations for November and November Figure 3.12: Same as Figure 3.6, except for varied microphysics scheme simulations during (a) November and (b) November

9 Figure 3.13: Same as Figure 3.6, except for varied boundary-layer scheme simulations during (a) November and (b) November Figure 3.14: Percentage of observed precipitation for varied microphysics simulations over the windward and leeward regions as defined in Fig. 3.8 for (a) November and (b) November Figure 3.15: Same as Figure 3.14, except for the boundary-layer simulations. Figure 3.16: Simulated (1.33 km domain) and observed total precipitation at the Quinault River Valley OLYMPEX observing sites in Fig. 2.2) for different storm sectors of November and November Distance from the Pacific Ocean increases to the right in each panel. Note the differing vertical axes. Figure 3.17: Histograms of simulated and observed warm-period precipitation rates at select Quinault River Valley stations (Beach, Bishop Field, and Graves Creek) during (a) UTC November 13 and (b) UTC November 17. Figure 3.18: Frequency distributions of November-February 6-hourly errors (Forecast Observed) of (a) rain rate, (b) liquid water content (LWC), (c) D0, and (d) Nw for the regimes of overprediction, underprediction, and accurate prediction. Figure 3.19: Frequency distributions for the microphysics schemes tested of 10-minute errors during the November 13 and November 17 warm periods of (a) rain rate, (b) liquid water content (LWC), (c) D0, and (d) Nw for the regimes of overprediction, underprediction, and accurate prediction. 9

10 Figure 3.20: Warm period average cross sections in the Quinault Valley of (a) radar-derived hydrometeor ID with green squares indicating where graupel frequency exceeded the 90 th percentile of total observations noted by the green squares; (b-h) various microphysics schemes hydrometeor mass mixing ratios (red = rain; yellow = graupel; blue = snow). Figure 3.21: November 13 warm period averaged cross sections of (a) supercooled liquid water and (b) cloud ice mass mixing ratios along the line in Fig. 2. The dashed line represents the average melting level height. Figure 4.1: Geopotential height (black contours, 20-m interval), wind barbs, and wind speed (shading) at 925 hpa for the 36-km domain for the (a) NARR and (b) WRF valid 1200 UTC December 12, Soundings at NPOL, valid at ~2135 UTC December 12, 2015 of (c) observed and (d) 444-m simulated. Figure 4.2: (a) Reflectivity (shading) in dbz; (b) radial velocity in m s -1 ; (c) spectrum width; (d) differential reflectivity in db; (e) correlation coefficient from a RHI scan from the Ku-band of the DOW radar at an azimuthal angle of 54 at 2110 UTC on December 12, 2015: In each panel, the black dots show the location of the maximum spectrum width in each vertical column. (based on Figure 9 in Barnes et al. (2018), used with permission) Figure 4.3: Vertical cross sections of (a) Richardson number, (b) vertical wind shear, and (c) stability (vertical potential temperature gradient) along the cross section in Fig. 2 at various stages of the December 12, 2015 event. The dashed line in (a) is the melting level height. Figure 4.4: Vertical cross sections of (a) observed, (b) 4-km, (c) 1.33-km, and (d) 444-m simulated radial wind to the northeast along the cross section in Fig. 2, valid 2100 UTC December 12,

11 Figure 4.5: Vertical cross sections of (a) 4 -km and (b) 444-m simulated radial wind along the cross section in Fig. 2, valid 2100 UTC December 12, (c) and (d) are the same cross sections as (a) and (b), but the topography of the Olympic Mountains has been removed in those simulations. Figure 4.6: Vertical cross-section from NPOL to Bishop along the cross section in Fig. 2 of (a) vertical velocity, (b) reflectivity, (c) mass-weighted mean rain drop diameter, and (d) rain number concentration at 2010 UTC December 12. The dashed line is the simulated melting level. Figure 4.7: Simulated vertical profiles at Bishop Field during the period UTC of: (a) cloud water mixing ratio (grey fill; grey dashed contours), rain water mixing ratio (red), graupel mixing ratio (black), and snow mixing ratio (blue), all in 0.1 g kg -1 ; (b) 2.0 km vertical velocity (solid), km maximum vertical velocity (dashed) ; (c) massweighted mean rain drop diameter at the lowest model height; and (d) precipitation rate (solid) and observed precipitation rate (dashed) from a tipping bucket gauge. Figure 4.8: Time series of 2 km vertical velocity for the control (Thompson) run and the dry run during the period UTC December 12, 2015 at (a) Fishery and (b) Bishop Field Figure 4.9: Percent change from the control simulation of mixing ratios and 2-km vertical velocity averaged over the upward branch of the waves at (a) Fishery and (b) Bishop Field for the period UTC December (R=Rain, G=Graupel, C=Cloud). Figure 4.10: (a) NARR and (b) WRF 925 hpa geopotential height (black contours, 20-m interval), wind barbs, and wind (shading) for the 36-km domain, valid 0600 UTC December 17, (c) Observed and (d) 444-m simulated soundings at NPOL, valid at ~1430 UTC December 17,

12 Figure 4.11: (a) Reflectivity (shading) in dbz; (b) radial velocity in m s -1 ; (c) spectrum width; (d) differential reflectivity in db; (e) correlation coefficient from a RHI scan from the Ku-band of the D3R radar at an azimuthal angle of 254 at 1320 UTC on December 17, 2015: In each panel, the black dots show the location of the maximum spectrum width in each vertical column. (Based on Figure 10 in Barnes et al. (2018), used with permission) Figure 4.12: Vertical cross sections of (a) observed and (b) 444-m simulated radial wind extending northeast along the cross section in Fig. 2, valid 1200 UTC December 17, Figure 4.13: Vertical cross sections of (a) Richardson number, (b) vertical wind shear, and (c) stability along the cross section in Fig. 2 at various stages of the December 17, 2015 event. Figure 4.14: Vertical cross sections of (a) 4 -km and (b) 444-m simulated radial wind along the cross section in Fig. 2, valid 1200 UTC December 17, (c) and (d) are the same cross sections as (a) and (b), but the topography of the Olympic Mountains has been removed in those simulations. Figure 4.15: Vertical cross sections from NPOL to Bishop along the line in Fig. 2 of (a) vertical velocity, (b) reflectivity, (c) mass-weighted mean rain drop diameter, and (d) rain number concentration at 1320 UTC December 17. The dashed line is the melting level. Figure 4.16: Simulated vertical profiles at the Fishery site during the period UTC December 17 of: (a) cloud water mixing ratio (grey fill; grey dashed contours), rain water mixing ratio (red), graupel mixing ratio (black), and snow mixing ratio (blue) ; (b) 0.5 km vertical velocity (solid), km maximum vertical velocity (dashed), with contours every 0.1 g kg -1 ; (c) mass-weighted mean rain drop diameter at the lowest model height; and (d) precipitation rate (solid) and observed precipitation rate (dashed). 12

13 Figure 4.17: Time series of 0.5 km vertical velocity during the period UTC December 17, 2015 at (c) Fishery and (d) Bishop; all plots show the dry simulation and control (Thompson) simulation. Figure 4.18: Percent change from the control simulation of mixing ratios and 0.5-km vertical velocity averaged over the upward branch of the waves at (a) Fishery and (b) Bishop Field for the period UTC December (R=Rain, G=Graupel, C=Cloud). 13

14 ii. List of Tables Table 1: Physics experiments that were conducted for both Kelvin-Helmholtz wave cases. Experiment Name Microphysics PBL Latent Heat Control Run Thompson YSU On WSM6 WSM6 YSU On MORR2 Morrison 2-moment YSU On SHIN-HONG Thompson Shin-Hong On MYNN Thompson MYNN3 On Dry Run None YSU Off 14

15 Table 2: Statistics for the microphysics and PBL experiments run over November and November The greatest values are italicized. Statistics include only model grid points where precipitation was present in the 1.33 km domain. November Microphysics: 1.33 km THOMPSON WSM6 WDM6 MORR2 MY2 SBUYLIN P3 Total (mm) 1.03E E E E E E E+07 Mean (mm) Max (mm) % pts > 200 mm PBL: 1.33 km ACM2 MYJ MYNN3 MRF Total 1.01E E E E+06 Mean Max % pts > 200 mm November Microphysics: 1.33 km THOMPSON WSM6 WDM6 MORR2 MY2 SBUYLIN P3 Total (mm) 8.40E E E E E E E+06 Mean (mm) Max (mm) % pts > 200 mm PBL: 1.33 km ACM2 MYJ MYNN3 MRF Total 8.54E E E E+06 Mean Max % pts > 200 mm

16 Chapter 1. Introduction The evaluation of moist physics in numerical weather prediction models is a priority for the research and operational weather prediction communities. The fidelity of model moist physics in areas of orography is of particular interest for several reasons. First, the prediction of orographic precipitation is an important societal issue, with terrain-forced precipitation providing much of the world s water resources. This is particularly true in the western U.S., where orographic precipitation stored by an extensive dam/reservoir system is the main source of water for agriculture and human consumption. Second, heavy orographic precipitation can produce flooding and slope failures, thus threatening life and property. Finally, areas of terrain are natural laboratories for studying moist physics, since precipitation and clouds are reliably locked to specific locations, allowing the pre-positioning of observational assets prior to weather events. Several factors contribute to deficiencies in the simulation of precipitation and associated moist physics in complex terrain, including inadequacies of model microphysics and other parameterizations, insufficient horizontal and vertical resolution, and the complex interactions between baroclinic and orographic forcing (Stoelinga et al., 2003; Stoelinga et al., 2010). The challenges associated with the evaluation of moist physics were recognized by both the Eighth and Ninth Prospectus Development Teams of the U.S. Weather Research Program, whose reports (Fritsch et al., 1998 and Droegemeier et al., 2000, respectively) placed priority on the observational evaluation of cloud and precipitation microphysics parameterizations in numerical weather prediction (NWP) models. 16

17 1.1 Previous Evaluations of Simulated Microphysics in Complex Terrain Surface precipitation is often the first and most accessible moist physics variable to be evaluated in numerical models. From the perspective of the operational meteorological community, precipitation is the primary output from a microphysics scheme, and due to the impacts that poses for society, a number of studies have focused on precipitation. Colle and Zeng (2004), using the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5) model and several bulk microphysical schemes (BMPs), found overprediction over the windward slopes of the Sierra Nevada. Similar results were noted for a case study in the Oregon Cascades, where MM5 overpredicted precipitation on windward slopes by as much as 100% (Garvert et al., 2005a,b). Substantial precipitation biases have persisted in the Weather Research and Forecast (WRF) model, the successor to MM5. Lin and Colle (2009) showed that WRF precipitation was excessive on both windward and leeward slopes. In one study, Hahn and Mass (2009) showed that overprediction could be reduced with use of a positive-definite (PD) moisture advection scheme (Hahn and Mass, 2009). Without such a scheme, advection of moisture across small scales can produce small, but not inconsequential negative values; using such a PD scheme eliminates these negative moisture values. Thus, precipitation evaluations can offer valuable insight into how microphysics parameterizations are functioning and which deficiencies exist. The accurate simulation of snow has been a particular challenge for BMP schemes, and deficiencies in snow representations have been shown to contribute to precipitation biases over terrain. Snow shape influences the distribution and concentration of snow and, consequently, precipitation; thus, poor snow parameterization can lead to unrealistic precipitation distributions (e.g., Colle and Zeng, 2004; Colle et al., 2005; Woods et al., 2007; Thompson et al., 2008; Lin and Colle, 2009). In one sudy, excessive snow production aloft was shown to be in part the result 17

18 of parameterized snow shape, and contributed to the overprediction of precipitation via excessive generation of small snow particles on leeward slopes of the Cascades (Lin and Colle, 2009). Even more sophisticated multi-moment schemes share these biases, with the number of predicted moments having little impact on the accuracy of simulated snow in a global model (Milbrandt et al., 2008; Milbrandt et al., 2010). Another obstacle to accurately simulating orographic precipitation is the representation of graupel and riming (Lin and Colle, 2009), which can have consequences on the simulated availability of supercooled liquid water. Observations by Hobbs et al. (1973) confirmed that particle size and degree of riming modulates orographic precipitation distributions, and thus deficiencies in simulated riming can lead to poor precipitation forecasts (e.g., Rutledge and Hobbs, 1983; Reisner et al. 1998; Gilmore et al., 2004; Thompson et al., 2004; Colle et al., 2005; Morrison and Pinto, 2005). Other major issues in BMP schemes include the assumption of fixed graupel density (e.g., Ferrier, 1994; Myers et al., 1997; Geresdi, 1998; Thompson et al. 2004; Milbrandt and Yau, 2005a; Morrison et al., 2009) and the representation of partial riming (Lin et al., 2013). Most BMPs divide ice into discrete categories (e.g., graupel, cloud ice, etc.), but such a division is artificial and does not reflect the real-world spectrum of riming and other particle characteristics. Recently, there has been a focus on developing ice parameterizations with a continuum of properties to better simulate the diversity of ice that is observed. The Stony Brook University (SBUYLIN; Lin and Colle, 2011) and Predicted Particle Properties (P3; Morrison et al., 2015) schemes have pioneered this approach. SBUYLIN maintains one precipitating ice category and one riming intensity metric, while P3 simulates rime ice mass, volume mixing ratios, and ice density. An important question is whether such improved microphysics schemes impact the simulated precipitation biases noted in the past. 18

19 Boundary-layer parameterizations also influence the distribution of microphysical variables in regions of orography (Garvert et al., 2007). Such impacts can be produced in a number of ways, including changing the strength of orographic uplift, altering the amplitude of mountain waves, changing fluxes at the surface, and influencing the strength of vertical mixing. Thus, it is important to also evaluate the impacts of boundary layer schemes when exploring the deficiencies in the moist physics of the current generation of mesoscale models. 1.2 OLYMPEX in the Context of Past Field Campaigns A number of field campaigns have examined microphysics in orography. The first generation of such field programs [e.g., the Cascade Project (Hobbs et al., 1971), CYCLES (Hobbs, 1978), the Sierra Cooperative Pilot Project (Reynolds and Dennis, 1986), COAST (Bond et al., 1997), CASPII (Cober et al., 1995), WISP (Rasmussen et al., 1992), and MAP (Binder et al., 1996)], although of great value, lacked sufficient observing assets for the thorough evaluation of three-dimensional microphysical distributions in mesoscale models. A more recent campaign focusing on orographic microphysics, the Improvement of Microphysical PaRameterization through Observational Verification Experiment (IMPROVE), took place in the Pacific Northwest during 2001 (Stoelinga et al., 2003). Data gathered during IMPROVE had a substantial impact on numerical weather prediction, leading to the improvement and development of several current microphysical parameterizations (e.g., Thompson et al., 2004; Thompson et al., 2008; Lin and Colle, 2011; Morrison and Milbrandt, 2014; Milbrandt and Yau, 2005; Morrison et al., 2015; Sarkadi et al., 2016). While not necessarily a primary goal, past campaigns limitations of observing capabilities reduced their capacity to provide the data necessary for a comprehensive evaluation of 19

20 parameterizations in mesoscale models. For instance, IMPROVE had only one radar viewing windward slopes, lacked radar coverage over the ocean or leeside, and had no disdrometer data on upwind slopes to document orographic influences on particle size and fall speed. While many of these deficiencies were addressed during OLYMPEX, the primary objectives of the campaign were: (1) document precipitation processes in complex terrain, and (2) provide physical validation for the satellites of the NASA Global Precipitation Mission (GPM; Hou et al. 2014, Skyfronick-Jackson, 2017), the OLYMPEX field program was conducted from November 2015 to February 2016 (Houze et al. 2017). OLYMPEX was a success in terms of diversity of observations. A substantial number of frontal systems crossed the OLYMPEX domain during the campaign s intensive observing period (IOP) of November-December 2015, including three atmospheric rivers, four warm fronts, eight cold fronts, and six occluded fronts. Observations of these events were comprehensive, including aircraft, radar, and surface-observing platforms. Three aircraft were fielded, with the NASA DC- 8 and ER-2 providing multi-wavelength remote sensing capabilities aloft, and the University of North Dakota Citation securing in-situ microphysical data over and around the Olympic Mountains. Multiple ground-based radars observed precipitation/clouds at a variety of wavelengths and locations. On the coast, the dual-polarized National Weather Service (NWS) Langley Hill and NASA NPOL radars surveyed offshore and over windward slopes of the barrier. To the lee of the barrier, there were the NWS Camano Island and Canadian X-band radars, and the Doppler on Wheels (DOW) was utilized within a major windward valley (the Quinault). Four vertically pointing Micro Rain Radars (MRRs) documented the vertical structure and fall velocities in precipitating clouds over the windward side and crest of the Olympic Mountains. Furthermore, OLYMPEX included more frequent radiosonde coverage on the Washington coast (at Quillayute), 20

21 as well as soundings at the NPOL site and on southern Vancouver Island. Fourteen disdrometers, over twenty rain gauges, and several snow cameras for measuring accumulating snowfall were positioned on the windward side of the Olympics. Finally, there was access to data from the new GPM satellite, with its suite of precipitation radars and microwave sensors. Additional details on the OLYMPEX field campaign can be found in Houze et al. (2017). 1.3 Kelvin-Helmholtz Waves over Complex Terrain The performance of model physics during particular events is of considerable importance to the forecasting and NWP communities. Specifically, the ability of a model to reproduce microphysical details of an event is a critical question. Despite previous efforts to simulate Kelvin- Helmholtz (KH) waves, few studies have focused on their microphysical impact over complex terrain nor their sensitivity to choice of physics parameterizations. The KH wave events during OLYMPEX provide excellent examples to evaluate the ability of NWP models to simulate smallscale phenomena, including their simulated microphysical impact. The observing assets in place during OLYMPEX provided the data necessary to evaluate this phenomena. Kelvin-Helmholtz (KH) waves have been observed in a variety of atmospheric settings, including cumulonimbus anvils (Petre and Verlinde, 2004), sea breeze circulations (Sha et al., 1991), and within mid-latitude baroclinic systems and fronts (e.g. Houze and Medina, 2005; Friedrich et al., 2008; Houser and Bluestein, 2011; Medina and Houze, 2015; Medina and Houze, 2016; Barnes et al., 2018). KH waves have also been reported in regions of complex terrain, where substantial terrain-induced vertical wind shear can increase or initiate KH instability (e.g. Medina and Houze, 2016). Terrain-induced shear can originate from a number of mechanisms, including the lifting of a low-level shear zone (Houze and Medina, 2005), deceleration of low-level flow by 21

22 terrain (Medina et al., 2005; Medina and Houze, 2016) and downslope air flows (Geerts and Miao, 2010). An important question regards the influence of KH waves on moist physics. Based on observations from two field campaigns in regions of complex terrain, Houze and Medina (2005) suggested that the upward motion of KH waves leads to enhanced riming and aggregation that enhances precipitation over orography. Houser and Bluestein (2011), examining dual-polarimetric radar over Oklahoma, found that the upward branch of KH waves has a significant impact on microphysical processes, with riming enhanced due to either an increase in turbulent motions or from the introduction of supercooled water into a region of frozen particles. Neither of these studies, however, quantified the impacts of KH waves on precipitation at the surface. Barnes et al. (2018) described three Kelvin-Helmholtz wave events during OLYMPEX. They detailed the influence of KH waves on precipitation and radar-observed fields, showing KHrelated modulation of rain rate (amplitude of 0.3 mm hr -1 ) when the waves were below the melting level, and impacts on mass-weighted mean drop diameter, reflectivity, and fall speed. Additionally, Barnes et al. (2018) described KH waves occurring within and above the melting layer, but did not explore the microphysical implications in these cases. A number of studies have applied numerical modeling to the understanding of KH instability, with most using idealized or dry models (e.g., Fritts, 1979; Sykes and Lewellen, 1982; Droegemeier and Wilhelmson, 1986; Fritts et al., 1996; Smyth, 2004; Zhou and Chow, 2013; Ji, 2014; Nakanishi, 2014). Large Eddy Simulation (LES) experiments have investigated KH instability in a variety of cases, including a mesoscale convective system over southern England (Browning, 2012), within a hurricane boundary layer (Nakanishi, 2012), during frontogenesis (Samelson, 2016), and for stratified flow over terrain (Sauer 2016). Recent studies have used full- 22

23 physics NWP models to simulate KH waves. Kudo (2013) used an LES resolution (50-m grid spacing) model to illustrate that sublimating snow can decrease stability below cloud base to generate KH waves. Other full-physics studies include Efimov (2017), which used the WRF model to simulate KH waves over Crimea, and Trier (2012), which simulated turbulence arising from KH instability in a winter cyclone. Finally, Thompson (2008) simulated the formation of KH instability in a sea breeze front using the U.S. Navy's Coupled Ocean/Atmosphere Mesoscale Prediction System model. 1.4 Purpose and Scope The purpose of this thesis is to meet the charge of the Eighth and Ninth Prospectus Development Teams of the U.S. Weather Research Program (Fritsch et al., 1998 and Droegemeier et al., 2000, respectively) by contributing a modern evaluation of moist physics in numerical weather prediction models. Emphasis is given to two atmospheric river cases in November 2015 and two Kelvin-Helmholtz wave cases in December The influence and sensitivity of model microphysics on these events will be examined. This document is organized as follows: Chapter 2 details the model used, its configuration, and the data utilized for evaluations; Chapter 3 offers an evaluation of precipitation forecasts for the entirety of the OLYMPEX field campaign and describes results from evaluations of two atmospheric river events; Chapter 4 presents simulations of two KH wave events; Chapter 5 offers discussion and concluding remarks, and proposes future work. 23

24 Chapter 2. Model Configuration and Data 2.1 Model Configuration The Weather Research and Forecasting Model (WRF; Skamarock et al., 2005) versions and are used in this study. All model runs use a km domain configuration with 51 vertical levels, similar to the University of Washington (UW) real-time WRF forecast system 1, with the 1.33-km domain centered over the Olympic Peninsula (Fig. 2.1). A km domain, nested within the 1.33-km domain and centered over the Quinault Valley on the Olympic Peninsula, is used for the evaluations of Kelvin-Helmholtz waves in Chapter 5. All simulations use the NOAH-MP land surface model (Ek et al. 2003) and the RRTMG radiation scheme (Iacono et al., 2008). A cumulus parameterization scheme (Grell-Freitas; Grell et al. 2013) is utilized for all but the and km domains. Beyond the standard output fields, WRF was modified to output rain size distribution parameters, including drop diameter and number concentration. Furthermore, all simulations used NOAA/National Weather Service Global Forecast System (GFS) gridded analyses (0.25-degree) for initial and 3-hourly boundary conditions, with the 36-km domain nudged toward the GFS every 3 hours to ensure that synoptic conditions are well represented. Since new GFS runs are available every 6 h, only hours 00 and 03 of runs within a particular period are used. Chapters 3 and 4 use WRF version for the evaluation of precipitation and moist physics. For these experiments, the Yonsei University (YSU; Hong et al., 2006) boundary/surfacelayer PBL scheme is used in conjunction with several microphysical parameterization schemes, 1 The UW WRF has a proven record of success in the Pacific Northwest. 24

25 including: the Milbrandt-Yau double-moment (MY2; Milbrandt and Yau, 2005), Morrison doublemoment (Morrison et al., 2009), P3 (Morrison et al., 2015), Stony Brook University (SBUYLIN; Lin and Colle, 2011), Thompson (Thompson et al., 2008), WSM6 (Hong and Lim, 2006), and WDM6 (Lim and Hong, 2010) schemes. Forecasts using the P3 scheme applied WRF version due to compatibility issues 2. Sensitivity to choice of PBL scheme was also evaluated, with Thompson microphysics used alongside the ACM2 (Pleim, 2007), MRF (Hong and Pan, 1996), MYNN3 (Nakanishi and Niino, 2006), MYJ (Janjic, 1994), and YSU (Hong et al., 2006) schemes. Chapter 5 uses WRF version to simulate Kelvin-Helmholtz waves and examine their microphysical implications. Control runs used Thompson microphysics and the YSU PBL scheme. Additional physics choices were tested for the KH wave events, including the Morrison and WSM6 microphysics schemes. A suite of carefully-chosen PBL schemes were also tested for their impact on KH waves: the MYNN3 3 (Nakanishi and Niino, 2006), Shin-Hong 4 (Shin and Hong, 2015), and the YSU 5 (Hong et al., 2006) schemes. Additionally, a dry run using YSU PBL, but no microphysics or latent heating, was completed. Table 1 outlines the model sensitivity experiments for the KH wave analyses. 2.2 Microphysical Data Precipitation data were obtained from the National Weather Service (NWS) Automated Surface Observing System (ASOS), the MesoWest cooperative mesonet (Horel et al., 2002), and 2 The P3 scheme was written to be included with WRF version MYNN3 is known for its parameterization of stable boundary layers, which is critical for KH wave development. 4 The Shin-Hong PBL scheme was designed for sub-kilometer horizontal resolution. 5 YSU has been used in the University of Washington real-time forecasting system for many years over the Pacific Northwest with proven success. YSU was developed to accurately describe vertical mixing in weak and strong wind environments. 25

26 OLYMPEX rain gauges. For verification purposes, WRF precipitation was interpolated to the observation locations through a Cressman approach. Stations with more than 5% of hourly data missing during a given period were excluded from analysis. Barnes et al. (2018) and Houze et al. (2017) describe the instrumentation deployed during OLYMPEX, of which a subset is used in this work. The NPOL radar near the Washington coast (149 m ASL) is used in conjunction with tipping bucket rain gauge and Parsivel disdrometer observations at the stations in Figure 2.2. Figure 2.1: Map of the WRF-ARW domains used in this study. The outer domain has 36-km grid spacing, with d02, d03, d04, and d05 indicating the 12, 4, 1.33-km, and 444-m domains, respectively. 26

27 Figure 2.2: Map of OLYMPEX observing stations used in this thesis, including the NPOL radar. Elevation is shaded. The blue dashed line is a cross section used for analyses in future sections. 27

28 Chapter 3. Results from Precipitation Evaluations 3.1 UW real-time WRF and GFS Evaluation Fidelity of incoming moisture flux during OLYMPEX Vertically integrated moisture flux (integrated water vapor transport; IVT) plays a central role in modulating West Coast precipitation and is a key parameter in defining and forecasting atmospheric rivers (e.g., Newell et al., 1992; Zhou and Newell, 1998). Since IVT is closely correlated with West Coast orographic precipitation (Neiman et al., 2008), it is important to ensure simulated IVT accuracy before evaluating model microphysics. To determine whether the flow impinging on the Olympic Peninsula during OLYMPEX is accurately simulated, the hpa IVT from the UW real-time WRF, driven by the NWS GFS model, was compared against observed values at the Quillayute (UIL) and NPOL rawinsondes from November 2015 to February 2016 (Fig. 3.1a). There is excellent agreement (r 2 = 0.95; p < 0.001) in both magnitude and timing over the entire period, with a mean error of 0.9 kg m -1 s -1 and maximum error of kg m -1 s -1. The IVT simulated at NPOL by an ensemble of microphysics schemes perform similarly well for the two atmospheric river case studies (for both, r 2 > 0.9 with p < (Fig. 3.1b,c). Thus, the observed environment in the context of incoming moisture is well simulated during the OLYMPEX period. Because WRF moist physics are initialized with dynamical fields and water vapor, the accurate simulation of IVT ensures that microphysics initializations are similar across schemes. 28

29 3.1.2 UW WRF and GFS forecast evaluation: November 01, February 01, 2016 The UW real-time WRF was used with other operational models for decision making during OLYMPEX. While qualitative performance was considered good by project scientists, quantitative evaluation of this operational configuration of WRF for various resolutions is provided below. Figure 3.2 shows the 6-24 h average cumulative precipitation error for all stations in the 1.33-km domain for November 1, 2015 through February 1, 2016 of the full resolution (13- km equivalent) NOAA/NWS GFS model and the various WRF domains. Precipitation was generally underpredicted, with the error substantially reduced between the 36-km (-89.3 mm) and 12-km (-28.7 mm) domains and no improvement at higher resolutions. In contrast, the NCEP GFS 13-km forecasts overpredicted precipitation (33.5 mm), due to excessive precipitation on windward slopes of the Cascade Mountains and along the Washington coast (not shown). One notes that the GFS uses a simple cloud model without the advection of frozen hydrometeors (Zhao and Carr, 1993), which may contribute to these biases. The spatial distributions of November February 2016 precipitation forecast errors over the innermost domain at various resolutions are shown in Fig. 3.3a. The errors are defined as total (November through February) forecast precipitation subtracted from total observed precipitation, and thus serve to remove errors in model timing. This metric, therefore, represents the model departure from observations for this period. At all resolutions, WRF underpredicts precipitation over the windward slopes and crests of the Olympics and coastal mountains. Elsewhere, precipitation tends to be overpredicted. The coefficient of efficiency (CE ; Sutcliffe and Nash, 1970) is a commonly used measure in the hydrologic community which describes how well a prediction corresponds to an observation. The measure is similar to the coefficient of determination (r 2 ), with a value of CE=1 corresponding 29

30 to a perfect forecast, though CE < 0 indicates poor performance. Fig. 3.4 shows CE for the OLYMPEX winter. Like the precipitation error maps, CE is largest near the Pacific coast and tends to decrease inland. This indicates that while precipitation is underpredicted along coastal regions, accumulation tendencies in those areas tend to be more accurate. In other words, the predictability of observed precipitation is greater in Western Washington than in Eastern Washington, though, as shown previously, the magnitudes are not. Of interest is the prevalence of negative values of CE at 1.33-km grid spacing. Because CE measures accuracy of timing and magnitude, and because convection is no longer parameterized within this domain, it is logical that explicitly resolved convective or small-scale precipitation elements may be experiencing timing issues. Thus, to assess the accuracy of precipitation in this domain, we inspect the relationship between observed and simulated precipitation totals in the 4- km and 1.33-km domains. Figure 3.5 shows these relationships. An improvement is noted between the 4- and 1.33-km domains, though by using a two-sided T-test for the means of these domains, we conclude that there is no statistical difference between the accumulation within the domains (p = 0.89). Thus, errors in scale and timing are the likely contributors to the low CE. Finally, the relative error of precipitation from the UW WRF at 1.33-km grid spacing was examined as a function of elevation over the same domain (Fig. 3.6). The relative precipitation error is defined as the sum of all simulated precipitation at particular elevation divided by the sum of all observations at that same elevation (i.e. the percent of observed). The WRF bias was minimal for the m and m elevation bins, with modest (20%) underprediction for intermediate altitudes ( m) and substantial (40%) overprediction at the highest elevations (> 1000 m). The GFS was similarly biased, but with greater tendency for excessive precipitation at all elevations. 30

31 Figure 3.1: hpa moisture flux (IVT) observed at the NPOL location compared to UW real-time WRF for (a) November February 2016, and from an ensemble of microphysics schemes for (b) November and (c)november The black lines are from observing soundings, and the blue lines are from the WRF model. 31

32 Figure 3.2: Accumulated precipitation error averaged all stations in the 1.33 km domain (Fig. 4) for November February 2016 as a function of resolution or model. 32

33 Figure 3.3: November February 2016 mean precipitation error of the UW WRF for 36, 12, 4, and 1.33-km (WRFGFSD1, WRFGFSD2, WRFGFSD3, WRFGFSD4, respectively) and the 12-km MM5 model (MM5ETAD2). Values are averaged over all stations within a 1-by-1 degree box. 33

34 Figure 3.4: November February 2016 mean coefficient of efficiency for precipitation of the UW WRF for 36, 12, 4, and 1.33-km (WRFGFSD1, WRFGFSD2, WRFGFSD3, WRFGFSD4, respectively) and the 12-km MM5 model (MM5ETAD2). Values are averaged over all stations within a 1-by-1 degree box. 34

35 Figure 3.5: Plot of observed vs. simulated precipitation in the UW WRF 1.33-km domain for the period November 2015 February The dashed line represents a perfect forecast, while the blue and black lines represent the regression line for the 4-km and 1.33-km domains, respectively. 35

36 Figure 3.6: Precipitation forecast accuracy by station elevation as a percent of observed precipitation for the UW WRF 1.33 km and the GFS 13 km forecasts. 36

37 3.2 Results from Two Atmospheric River Events Two OLYMPEX events associated with heavy precipitation were selected for additional study, with simulations varying microphysical and boundary-layer schemes. Both events were considered to be atmospheric rivers (e.g., Newell et al., 1992; Zhu and Newell, 1994, 1998; Ralph et al., 2004), which are known for producing long-duration precipitation events characterized by high melting levels and warm-rain processes. Thus such events provide an ideal testbed for evaluating simulated warm-rain microphysics and associated precipitation accumulations Description of Events The first event included the passage of a midlatitude cyclone and associated fronts on November 12-15, 2015 (Fig. 3.7). At 0000 UTC November 13, a 500-hPa cutoff low was located over the Gulf of Alaska with a jet of 85 kts (45 ms -1 ) impacting the British Columbia coast. At the same time, a cold front was positioned to the northwest of the Olympic Peninsula, with a warm front just west of the peninsula. The warm sector of this system was characterized by a moist-neutral environment in the lower troposphere, with substantial vertical shear. Following cold frontal passage, precipitation intensity declined in the post-frontal environment. Accumulations in the windward Quinault River valley exceeded 300 mm during the 24-h period ending 0000 UTC November14, which includes the warm period of the event. Following Zagrodnik et al. (2018), we define the warm period as UTC November 13, the cold frontal period as 1800 UTC November UTC November 14, and the cold period as UTC November 14. The second event (November 16-19) was associated with a shortwave trough embedded in strong 500-hPa westerly flow (Fig. 3.8). At 0000 UTC November 17, a surface low pressure center 37

38 was northwest of the OLYMPEX domain. After the passage of the associated warm front, the warm period was characterized by a moist neutral environment with strong (45 kts) 850 hpa westsouthwesterly flow impinging upon the Olympic Mountains. Around 2200 UTC November 17, the system s cold front passed the OLYMPEX domain and began post-frontal conditions. For the 24-h beginning 0000 UTC November 17, over 200 mm of precipitation was observed on windward slopes of the Olympic Mountains. For this event, we define the warm period as UTC November 17, the cold frontal period as 2000 UTC November UTC November 18, and the cold period as UTC November 18, based on soundings taken at NPOL and radar imagery of cold frontal passage. 38

39 Figure 3.7: Synoptic conditions for the November case study: (a,b) 500 hpa and surface analyses from the US National Weather Service at 0000 UTC November 13, and (c) sounding from the NPOL radar at 0305 UTC November

40 Figure 3.8: Synoptic conditions for the November case study: (a,b) 500 hpa and surface analyses from the US National Weather Service at 1200 UTC November 17, and (c) NPOL sounding at 1115 UTC November

41 3.2.2 Precipitation Evaluation Area-Averaged Precipitation For both events, the spatial distributions of precipitation were similar among the microphysics schemes, with some (e.g., WSM6 and WDM6 [not shown]) producing more precipitation over the high terrain and windward slopes of the Cascades and Olympics (Fig. 3.9a,b). WSM6 and SBUYLIN (not shown) had the largest area-integrated precipitation (Table 2). The spatial distributions of precipitation varied less among the boundary-layer schemes (Fig. 3.9c,d), with ACM2 and MYNN3 having the greatest area-integrated precipitation (Table 2). The simulated average precipitation error over the 1.33-km domain, using the stations shown in Figure 3.10, is presented in Figure 3.11 for experiments in which microphysics and boundary-layer (PBL) schemes were varied. Considering average errors across all microphysics and boundary layer schemes, WRF underpredicted precipitation in the domain by between 13 mm and 28 mm. Thompson, WSM6, MORR2 and P3 were the best performing microphysics schemes for both events in terms of average error. For the boundary-layer schemes (all using Thompson microphysics), YSU and MRF 6 were best for the November event, with all PBL schemes comparable for the November case. Overall, all schemes underpredict precipitation, and similar levels of variability in the average error were produced by varying microphysics or PBL schemes Evaluation by Elevation Model precipitation was further analyzed by station elevation over the 1.33-km domain in order to investigate whether elevation-dependent biases exist (Fig. 3.12). For November 12-15, 6 YSU was developed from MRF, so are closely related. 41

42 simulated precipitation for the various microphysics schemes was % of observed between m, with underprediction increasing above 250 m. In contrast, for the November simulations, underprediction was the rule at all elevations, with simulated precipitation being 40-80% of observed. Boundary layer schemes showed a general increase in underprediction with elevation for both events (Fig. 3.13) Windward vs. Leeward Slopes Previous studies have found substantial differences in model precipitation/moist physics performance between the windward and leeward slopes of barriers (Garvert et al., 2005a; Lin and Colle, 2009); thus, windward and leeward sides of the Olympex mountains were evaluated separately for the OLYMPEX cases (see Fig for specified regions). For both OLYMPEX case studies, there is a considerable difference in accuracy between windward and leeward regions of the Olympic Mountains (Figs and 3.15). During the November event, precipitation on the leeward slopes was better simulated, with substantial improvement between 36 and 12-km grid spacing, and little bias except for the SBYLIN scheme (~ 15% underprediction). In contrast, the windward slopes experienced substantial underprediction, with little improvement as gridspacing was decreased below 12-km. On the windward slopes, WSM6 was generally the best, while MORR2 had the best performance on the leeward side. The results for the November event were considerably different than the earlier event, with underprediction dominating both sides of the barrier at 12-km grid spacing. On the windward side, underprediction by approximately 20% was the general rule, with the coarsest grid (36-km) having the worst performance. SBUYLIN was generally the worst, with WDM6 and WSM6 having the best performance. On leeward slopes, the coarse 36-km domain had the lowest error, possibly because the lack of resolution resulted in attenuated lee rain shadow drying, while 42

43 a substantial underprediction (about 30%) was evident at the highest resolutions. Varying the PBL schemes produced a similar trend in precipitation with model resolution, with ACM2 and YSU generally producing the lowest bias (Fig. 3.15). The consistent underprediction on windward slopes is of particular interest when viewed in the context of past literature. During the IMROVE campaigns, precipitation was shown to be overpredicted on windward slopes (Garvert et al., 2005a; Lin and Colle, 2009) Precipitation in the Quinault Valley During OLYMPEX, a dense network of rain gauges and other instrumentation was positioned along the Quinault River Valley, providing a uniquely comprehensive description of orographic enhancement over a windward valley. Focus is on stations in the Quinault Valley, oriented southwest-to-northeast up the valley. Fig shows the precipitation accumulation up the Quinault valley of varying microphysics simulations for the two case studies. The results are divided between the warm period, the period surrounding cold frontal passage, and the post-frontal period, when cold advection and post-frontal convection were taking place. The magnitude of simulated precipitation totals were sensitive to storm sector. Because the warm period of these events saw substantial precipitation in the Quinault Valley, and this period was of considerable duration, emphasis is given to this periods. During the warm periods, precipitation generally increased with elevation for observed and simulated precipitation from the Beach (4.6 m) to Bishop Field (86.9 m) sites. At the lowest elevations, simulated precipitation was less than observed. For the November event, observed precipitation declined at the two highest sites, yet simulations overpredicted precipitation there. For the November period, observed precipitation also declined at the upper two sites, with WRF generally equal to or less 43

44 than observed values. The shift in precipitation maximum upstream may be indicative of longer hydrometeor residence times in WRF compared to reality. During the short frontal passage period, observations indicated increasing precipitation with elevation up the valley. Accumulations were less than the warm period, with overprediction at most locations for November and underprediction for November Finally, the cooler, less stable post-frontal period exhibited opposite characteristics: there was nearly no increase up the valley and considerably less precipitation than the other periods. This is consistent with the localized and transient nature of post-frontal showers. Underprediction was apparent for November and general overprediction noted for November during the post-frontal periods. In terms of overall performance, no single scheme outperformed during both events. For November 12-15, SBUYLIN performed best across all periods. Results for November depended on storm sector, with Thompson performing well in the warm period, WDM6 most accurate during the frontal passage period, and SBUYLIN in the cold period. It is of interest to look at precipitation rates, particularly their frequency and intensity, during the warm period to determine whether schemes produced consistently heavy precipitation rates or if large accumulations were dominated by a few instances of heavy precipitation. Figure 3.17 shows histograms of precipitation rate during these periods for select sites along the transect. In both cases, observed precipitation rates reach a maximum in intensity and frequency at Bishop Field, located mid-way up the valley near the observed winter-long precipitation maximum (cf. Fig. 7 in Houze et al., 2017). In contrast, simulated precipitation rates reached their maximum farther up the valley at Bunch Field (not shown) and Graves Creek. The shift of distributions from low precipitation rates at the Beach site to more positively skewed distributions farther inland was 44

45 evidence of orographic enhancement taking place in the model, though for the sites considered in Fig. 3.17, rain rates are more frequently less intense when compared to observations. This finding is explored from a microphysics perspective in the next section, Figure 3.9: Total precipitation accumulations of selected microphysics scheme simulations in the 1.33 km domain for (a) November and (b) November 16-19, and boundary-layer scheme simulations for (c) November and (d) November

46 Figure 3.10: Map of observing station locations and sites of NPOL and UIL soundings, with the windward (blue) and leeward (green) domains marked. Symbols indicate which network the observations are from (circle = ASOS, square = MesoWest, diamond = OLYMPEX) 46

47 Figure 3.11: Total precipitation error from all stations shown in Fig. 3.8 for (a) microphysics and (b) boundary-layer scheme simulations for November and November

48 Figure 3.12: Same as Figure 3.6, except for varied microphysics scheme simulations during (a) November and (b) November

49 Figure 3.13: Same as Figure 3.4, except for varied boundary-layer scheme simulations during (a) November and (b) November

50 Figure 3.14: Percentage of observed precipitation for varied microphysics simulations over the windward and leeward regions as defined in Fig. 3.8 for (a) November and (b) November

51 Figure 3.15: Same as Figure 3.12, except for the boundary-layer simulations. 51

52 Figure 3.16: Simulated (1.33 km domain) and observed total precipitation at the Quinault River Valley OLYMPEX observing sites in Fig. 2.2) for different storm sectors of November and November Distance from the Pacific Ocean increases to the right in each panel. Note the differing vertical axes. 52

53 a. b. Figure 3.17: Histograms of simulated and observed warm-period precipitation rates at select Quinault River Valley stations (Beach, Bishop Field, and Graves Creek) during (a) UTC November 13 and (b) UTC November

54 3.2.3 Microphysical Evaluation Because the warm periods were noted as having large precipitation errors and were of long duration, the simulated microphysics of these periods will next be evaluated. A common characteristic of this storm sector is a high melting level, and thus a preference for liquid, rather than frozen, precipitation. As a result, warm-rain processes are prevalent and in-situ observations from disdrometers can offer valuable insight into model performance at the surface and aloft. All bulk microphysical parameterization schemes use an assumed size distribution and predict one or more moments of that distribution, with most schemes predicting either one (hydrometeor mass mixing ratio) or two moments (mixing ratio and number concentration). Nearly all processes represented in these schemes rely on a combination of mixing ratio, number concentration, or the mean diameter of particles in that distribution (Khain et al., 2015). It is important to note that the mean particle diameter and number concentration, rather than details of the entire distribution, are used in these schemes to represent quantities critical to precipitation accumulations. Of considerable importance to local hydrology is model performance during the wet season (winter) over windward slopes of the Olympic Mountains. During the November 2015 to February 2016 period, surface disdrometers captured particle size distribution characteristics of rain. Figure 3.18 shows frequency distributions of 6-hourly microphysical errors (forecast observed) during that period. First, we use the UW real-time WRF with Thompson microphysics to describe the microphysical characteristics of the OLYMPEX winter with a robust sample size (n > 800). Results indicate an overall tendency for the underprediction of rain rates and liquid water content (LWC), though there were occurrences of accurate LWC forecasts corresponding to underprediction of rain rate. 54

55 Next, the median volume diameter 7 (D0) and normalized intercept parameter 8 (Nw) are computed using parameters derived from the Thompson et al. (2008) microphysics scheme and adjusted to match the size range of the Parsivel disdrometer (D > 0.3 mm) in order to provide a more direct comparison. D0 and Nw are important for microphysical evaluations, as they provide a convenient way to characterize particle size distributions and allow for direct comparison between any model distribution and observations. For the OLYMPEX period, D0 was overpredicted and Nw underpredicted during all rain rate regimes, which indicates a tendency toward fewer but larger rain in the Thompson scheme during the warm period. This bias, however, is not limited to extended periods of time nor to the Thompson scheme. Figure 3.19 shows frequency distributions of 10-minute microphysical errors during the November 13 and November 17 warm periods for the various microphysical schemes tested. The aforementioned finding of larger and fewer rain drops is consistent among schemes: distributions of D0 and Nw are positively and negatively skewed, respectively. A notable exception is WDM6, which produces rain particles which are too few and too small. These differences will be explored in future work, though a potential explanation for these findings is that the number of cloud condensation nuclei (CCN) in WRF microphysics schemes is too small for the given model environment. Such a condition would explain the excessive drop sizes (drops can grow larger and are less numerous when fewer CCN are present). 7 The median volume diameter (D0) is defined as the diameter at which half of the particles in a given volume of air are smaller than D0 and half are larger. 8 The normalized intercept parameter (Nw) is defined as the intercept of an exponential-shaped particle size distribution that corresponds to a gamma distribution with the same number of particles. Thus, Nw is closely related to the particle number concentration of a distribution. (Ulbrich 1983; Willis 1984; Testud et al. 2001; Bringi et al. 2002; Bringi et al. 2009; Thompson et al. 2015). 55

56 Next, simulated mass mixing ratios for the November 13 warm period are evaluated. Figure 3.20 shows average cross sections of various mixing ratios over the Quinault valley compared qualitatively with hydrometeor identification output from the NPOL radar. All schemes produced a defined melting level at approximately the same altitude, which reiterates that the synoptic and mesoscale environments were consistent and well-represented among schemes. Among the simulations, snow and graupel are most variable. Qualitatively similar distributions of snow aloft are noted, with the exception of the Morrison scheme, which produced the greatest quantity of snow, and WSM6/WDM6, which produced the least. Graupel in schemes was confined primarily to the melting level, where supercooled liquid water is most likely to rime onto snow. Morrison scheme produced the least graupel and WSM6/WDM6 produced the most indicating either that the parameterization of depositional snow growth was too eager to grow snow in Morrison (therefore less small water droplets, less graupel, and more snow) and/or that WSM6/WDM6 produced and retained excessive quantities of supercooled water (therefore more small droplets, more graupel, and less snow). These mixing ratio distributions are similar for the November 17 case as well. Riming of supercooled water is necessary for graupel production and the aggregation of cloud ice produces snow in microphysics schemes. Thus, these quantities are considered critical for the existence of other hydrometeor species, and ultimately rainfall during these warm atmospheric river event. Three modes of simulated supercooled water and cloud ice were present during the November 13 th warm period, displayed in Fig. 3.21: (1) WDM6 and WSM6 produced excessive cloud ice above the melting level, but the least supercooled water; (2) MORR2, MY2, and SBUYLIN all produced only a modest layer of cloud ice around 8 km, but a considerable amount of supercooled liquid water above the melting level; (3) The Thompson scheme 56

57 produced almost no cloud ice, but had supercooled water values between first and second regimes. Thus, schemes with excessive graupel appear to have less supercooled water; whereas schemes with excessive snow have less cloud ice. 57

58 a. b. c. d. Figure 3.18: Frequency distributions of November-February 6-hourly errors (Forecast Observed) of (a) rain rate, (b) liquid water content (LWC), (c) D0, and (d) Nw for the regimes of overprediction, underprediction, and accurate prediction. 58

59 Figure 3.19: Frequency distributions for the microphysics schemes tested of 10-minute errors during the November 13 and November 17 warm periods of (a) rain rate, (b) liquid water content (LWC), (c) D0, and (d) Nw for the regimes of overprediction, underprediction, and accurate prediction. 59

60 Figure 3.20: Warm period average cross sections in the Quinault Valley of (a) radar-derived hydrometeor ID with green squares indicating where graupel frequency exceeded the 90 th percentile of total observations noted by the green squares; (b-h) various microphysics schemes hydrometeor mass mixing ratios (red = rain; yellow = graupel; blue = snow). 60

61 Figure 3.21: November 13 warm period averaged cross sections of (a) supercooled liquid water and (b) cloud ice mass mixing ratios along the line in Fig. 2. The dashed line represents the average melting level height. 61

62 Chapter 4. Results from Two Kelvin-Helmholtz Wave Events Of particular importance to the NWP community is the performance of models during specific and localized events. Barnes et al. (2018) described the occurrence of Kelvin-Helmholtz (KH) waves over the Quinault Valley during the OLYMPEX field campaign. Using a suite of ground-based observations, they concluded that such events do influence microphysics, including precipitation and rain drop sizes. The aim of this chapter is to examine the findings of Barnes et al. (2018) using WRF to determine the model s ability to simulate these events in a full-physics framework and what sensitivities exist for KH wave events in complex terrain. Using two KH events from OLYMPEX and detailed in Barnes et al. (2018), December 12 and 17, 2015, the following questions will be addressed: (1) How do synoptic conditions initiate and control the evolution of KH waves? (2) How are KH waves modified by different mesoscale environments (e.g., over the ocean, windward slopes, crests)? Are there key mesoscale controls on KH wave formation? (3) What role does topography (the Olympic Mountains) play in KH wave formation? (4) What resolution is required to realistically simulate KH waves in WRF? Are KH waves sensitive to the selection (or absence) of microphysics or PBL schemes? (5) What are the microphysical impacts of KH waves during orographic precipitation events? How do simulated modulations in precipitation and microphysical quantities compare to those observed during the OLYMPEX project? 62

63 4.1 Case I: December 12, Observed and simulated mesoscale conditions and wave generation The first event on December 12, 2015 was characterized by a midlatitude cyclone that approached from the west. At 1200 UTC December , prior to the period of observed KH waves, the NARR analysis shows a 925-hPa low with strong winds (50 kt, 25ms -1 ) off the British Columbia coast (Fig. 4.1). Although WRF realistically captured the position and timing of the system, the simulated low center was deeper than the NARR reconstruction by 80 m, with 925- hpa wind speeds that were modestly greater (5 kt; 2.5 ms -1 ) than observed. The 444-m simulated vertical profile during the period of KH waves (2135 UTC) is consistent with observations at the NPOL radar, including a stable saturated layer below 700 hpa and substantial wind shear below 900 hpa. As noted in Chapter 3, the incoming moisture flux during the OLYMPEX project was well-simulated by WRF. Figure 4.2 displays the DOW radar observations from Barnes et al. (2018), which documented KH waves within the melting level of stratiform precipitation in the vicinity of Bishop Field on December 12 th. Waves were observed in the 1-3 km AGL layer, with perturbations most pronounced in the reflectivity, velocity, and spectrum width fields. The waves had a wavelength of ~5 km and a period of ~7 minutes. The Richardson number (Ri) is a common diagnostic for KH waves, with waves developing when Ri < 0.25 (Miles and Howard, 1964) and Ri < 1 being sufficient for wave maintenance once initiated (Weckwerth and Wakimoto, 1992). Model vertical cross sections showed a decrease in Ri below 2 km prior to the onset of simulated waves around 1800 UTC (Fig. 63

64 4.3a). At 0900 UTC, a thin layer of Ri < 1 was noted around 500 m ASL; by 1200 UTC, this layer had grown considerably in magnitude, but not in depth, with critical (< 0.25) values noted over Bishop Field. Beginning at 1800 UTC, wave structures appeared in the km layer, and progressively grew in the vertical, with some approaching 2.5 km ASL by 2100 UTC over Bishop Field. Such vertical expansion was noted in the observations of Barnes et al. (2018) and appeared to perturb the melting level between Fishery and Bishop at the final time. Because Ri reflects the ratio of atmospheric stability to vertical wind shear, the atmosphere must become less stable and/or vertical shear must increase to obtain critical values of Ri. Both of these factors contributed to wave development/maintenance in this case, though increasing shear was dominant, a situation that has been observed over the Sierra Nevada Mountains (Medina and Houze, 2016). Figure 4.3b shows the evolution of simulated vertical wind shear during this event. Initially, vertical shear was greatest over land because of surface drag, but then increased and expanded vertically in the lower troposphere between 0900 and 1800 UTC as low-level easterly flow developed beneath westerly flow aloft. Environmental stability (defined as the vertical gradient of potential temperature; Fig. 4.3c) decreased considerably near the surface prior to wave generation as the front approached. The appearance of waves around 1800 UTC was coincident with the decrease in low-level stability between Bishop and Fishery. Overturning waves were evident in the stability field, evinced by unstable lapse rates at 2100 UTC near and upstream of Bishop. The appearance of KH waves in the simulation was highly sensitive to horizontal grid spacing. Figure 4.4 shows radial winds along the cross section at 2100 UTC from the NPOL radar and WRF at various horizontal grid spacings. In all domains, WRF was able to reproduce observed 64

65 mesoscale conditions, including the transition at approximately 1 km AGL from low-level easterly flow to westerly flow aloft. No wave activity was evident using 4-km grid spacing (despite there being easterly low-level flow) and only weak KH waves were noted in the 1.33-km domain. Only at 444-m grid spacing did the KH waves have realistic amplitudes and wavelengths (e.g., a wavelength of ~3.1 km at 2 km AGL). Such grid-spacing dependence follows from basic principles, since the effective resolution of WRF is roughly 6 times the grid spacing (Skamarock et al., 2004; Skamarock et al., 2014), which would be 2.6 km for the 444-m grid spacing and 7.8 km for the 1.3-km domain. Thus, only the 444-m domain had sufficient resolution to adequately resolve the observed KH waves, which had wavelengths of 4-5 km, according to Barnes et al. (2018). The simulated KH structures are most prominent immediately upstream of and over the Olympic foothills, with some wave activity up to ~40 km offshore, where low-level wind shear begins to increase as incoming flow is blocked by the Olympic Mountains (Fig. 4.5a,b). Furthermore, waves appear to weaken after traversing higher terrain over the eastern portion of the domain. To evaluate the impact of terrain blockage on the development of KH waves during this event, the Olympics were removed in an additional WRF simulation (Fig. 4.5c,d), while retaining the changes in surface characteristics between water and land. For the December 12 th event, removing the mountains greatly lessens the flow deceleration, removes the low-level flow at the surface, and results in no KH wave activity over or near the Olympics. Some small-amplitude oscillations are present over western portions of the cross section, resulting from localized ascent along the occluded front. Thus, it appears that orography is necessary for the development of KH waves in this case. The importance of orography for the development of KH waves was suggested by Medina et al. (2005). 65

66 It is important to note that these waves should be classified as Kelvin-Helmholtz waves and not waves resulting from symmetric instability. Despite symmetric instability occurring in similar synoptic conditions as these cases (i.e., an approaching frontal boundary, approaching upper-level trough, and moist neutral conditions.; Moore and Lambert, 1993), reasons for this distinction are: (1) symmetric instability is characterized by slantwise vertical motions (Wiesmueller and Zubrick, 1998); (2) symmetric instability is preferred in regions of speed shear, rather than directional shear (Wiesmueller and Zubrick, 1998); (3) the wavelengths of waves generated via symmetric instability are much longer than KH waves, with one study reporting wavelengths of km (Seltzer et al., 1985); and (4) the moist potential vorticity was nonnegative in both of these cases (Schultz and Schumacher, 1999). 66

67 Figure 4.1: Geopotential height (black contours, 20-m interval), wind barbs, and wind speed (shading) at 925 hpa for the 36-km domain for the (a) NARR and (b) WRF valid 1200 UTC December 12, Soundings at NPOL, valid at ~2135 UTC December 12, 2015 of (c) observed and (d) 444-m simulated. 67

68 Figure 4.2: (a) Reflectivity (shading) in dbz; (b) radial velocity in m s -1 ; (c) spectrum width; (d) differential reflectivity in db; (e) correlation coefficient from a RHI scan from the Ku-band of the DOW radar at an azimuthal angle of 54 at 2110 UTC on December 12, 2015: In each panel, the black dots show the location of the maximum spectrum width in each vertical column. (based on Figure 9 in Barnes et al. (2018), used with permission) 68

69 Figure 4.3: Vertical cross sections of (a) Richardson number, (b) vertical wind shear, and (c) stability (vertical potential temperature gradient) along the cross section in Fig. 2 at various stages of the December 12, 2015 event. The dashed line in (a) is the melting level height. 69

70 Figure 4.4: Vertical cross sections of (a) observed, (b) 4-km, (c) 1.33-km, and (d) 444-m simulated radial wind to the northeast along the cross section in Fig. 2, valid 2100 UTC December 12,

71 Figure 4.5: Vertical cross sections of (a) 4 -km and (b) 444-m simulated radial wind along the cross section in Fig. 2, valid 2100 UTC December 12, (c) and (d) are the same cross sections as (a) and (b), but the topography of the Olympic Mountains has been removed in those simulations. 71

72 Impact of KH waves on microphysics Analyzing polarimetric radar data, Barnes et al. (2018) found that the observed KH waves during the December 12 th event influenced microphysics, particularly when the waves were near the melting level. Specifically, enhanced reflectivity and turbulence (spectral width) within the upward branch of the waves were associated with greater riming. The impact of simulated KH waves on model microphysics in the 444-m domain is explored below. At 2010 UTC, waves were present and influenced microphysical fields between the NPOL site and Bishop Field (Fig. 4.6). Simulated reflectivity, mass-weighted mean rain diameter (D mw ), and rain number concentration were modulated by the waves, with enhancements immediately downstream of local vertical velocity maxima, indicating that model microphysics were influenced by the waves. Enhancement and wave-like modulation of reflectivity (dbz) was found in the lowest 2 km of the simulation, with particularly large values just below the melting level. At Bishop Field, where waves were observed in Barnes et al. (2018), simulated vertical velocity and microphysical quantities showed wave-like variability (Fig. 4.7). For UTC, simulated vertical velocity oscillations exceeded 5 ms -1 within the km layer, which impacted surface D mw and mixing ratios of cloud and rain water. Precipitation in the 444 m domain possessed substantial modulation, albeit of lesser amplitude than observed. Barnes et al. (2018) does not investigate the impact of these waves on surface-observed microphysics. 72

73 Figure 4.6: Vertical cross-section from NPOL to Bishop along the cross section in Fig. 2 of (a) vertical velocity, (b) reflectivity, (c) mass-weighted mean rain drop diameter, and (d) rain number concentration at 2010 UTC December 12. The dashed line is the simulated melting level. 73

74 Figure 4.7: Simulated vertical profiles at Bishop Field during the period UTC of: (a) cloud water mixing ratio (grey fill; grey dashed contours), rain water mixing ratio (red), graupel mixing ratio (black), and snow mixing ratio (blue), all in 0.1 g kg -1 ; (b) 2.0 km vertical velocity (solid), km maximum vertical velocity (dashed) ; (c) massweighted mean rain drop diameter at the lowest model height; and (d) precipitation rate (solid) and observed precipitation rate (dashed) from a tipping bucket gauge. 74

75 Impact of choice of physics schemes The removal of moist physics and latent heating from the model had a substantial influence on wave activity between the control and dry runs for the December 12, 2015 case, when waves were within the melting level. At the Fishery site, the control simulation had vertical velocity amplitudes associated with KH waves in excess of 2 ms -1 (Fig. 4.8). In contrast, there was no wave activity in the dry run, with only minor oscillations less than 0.5 ms -1. At Bishop Field, wave amplitudes were large in the control simulation between 2000 and 2200 UTC, while in the dry run the waves were attenuated after 2100 UTC. The results of the dry runs for both cases are consistent with the findings of Kudo (2013), who showed that latent heat can create reduce stability, providing more favorable conditions for the waves. The period of the waves in the dry run was greater (22 minutes) than that of the control run and with a longer wavelength (5 km compared to 3 km). Varying physics schemes had a significant impact at both sites for this event. Figure 4.9 shows changes relative to the control run in vertical velocity and mixing ratios of rain, graupel, and cloud water averaged over the waves areas of positive vertical velocity from UTC. Examining the impact of various microphysics schemes at Fishery, vertical motion was reduced by approximately 40% in the Morrison scheme and 10% by the WSM6 scheme compared to the control. Graupel mixing ratios were enhanced by approximately 140% in both the Morrison and WSM6 schemes, while rain was reduced by approximately 70% and cloud water by 40%. At Bishop, changing microphysics scheme had little impact on graupel or cloud water mixing ratios, but decreased rain by approximately 70% and reduced vertical motions by about 20% compared to the control. For varying PBL schemes at Fishery, the MYNN reduced vertical motions by about 25%, and rain mixing ratios by 50%, while the ShinHong PBL scheme increased rain and graupel 75

76 mixing ratios by approximately 125%. At Bishop, the MYNN PBL scheme had a major impact on vertical motion, with reductions of approximately 75%. Figure 4.8: Time series of 2 km vertical velocity for the control (Thompson) run and the dry run during the period UTC December 12, 2015 at (a) Fishery and (b) Bishop Field Figure 4.9: Percent change from the control simulation of mixing ratios and 2-km vertical velocity averaged over the upward branch of the waves at (a) Fishery and (b) Bishop Field for the period UTC December (R=Rain, G=Graupel, C=Cloud). 76

77 4.2. Case II: December 17, Observed and simulated mesoscale conditions During the second event on December 17, 2015, KH waves were observed for 6 hours ( UTC) over the Quinault Valley (Barnes et al. 2017). At 0600 UTC, the NARR analyzed a 925-hPa low over the Gulf of Alaska, with an occluded front extending southward offshore of the coast of Washington State and strong (> 40 kt) winds along and north of the front (Fig. 4.10). The synoptic conditions of this event were well simulated by WRF, although the simulated low was slightly deeper, with low-level winds stronger than the NARR. The simulated frontal position agreed with NARR, and simulated and observed vertical profiles at the NPOL site were in good agreement at the beginning of the event (1430 UTC). Specifically, WRF accurately simulated moist neutral conditions above 700 hpa, a stable layer below the melting level, and vertical wind shear exceeding 40 ms -1 below 1.5 km. Radar observations in Barnes et al for this event (reproduced in Fig. 4.11) showed KH waves located offshore and throughout the valley below 1.5 km, which was beneath the observed melting level (2.5 km). Observed waves had a wavelength of 4-5 km and period of ~10 minutes. These waves were found in the km layer and had a smaller vertical extent than the December 12 th event (1-3 km AGL). Radar cross sections indicated that waves amplified upon nearing the Olympic Mountains (Fig. 14 in Barnes et al., 2018). In the 444 m simulation, KH waves occurred between 0.5 and 2 km, with the greatest amplitudes within a layer of vertical shear between low-level easterly flow and westerly flow aloft (Fig. 4.12). This shear layer was evident around 1200 UTC in both the NPOL radial winds and in the 444-m simulation. Simulated waves had a 3-km wavelength and a period of 17 minutes. 77

78 Wave onset coincided with Ri declining to criticality (below 0.25) around 1200 UTC within the low-level shear layer. Figure 4.13a shows cross sections of Ri before and during the wave event. At 0500 UTC, Ri was < 0.5 in the lowest 300 m ASL, with some wave-like perturbations approximately 100 m above coastal ocean. Between UTC, Ri and vertical shear (Fig. 4.13b) were relatively unchanged at low level, with little evidence of the growth of KH waves and the low-level flow becoming more stable. Only aloft (2 km ASL) was shear increasing. By 1200 UTC, the front had passed Fishery with a dramatic increase in wind shear below 2 km and KH waves appeared between 0.5 and 1.5 km. By 1500 UTC, the layer of strong vertical wind shear became shallow and the wave amplitude began to decline. Thus, it appears that KH waves in this case were initiated by the increasing wind shear associated with the approaching front. Like the December 12 th case, waves expanded vertically as they approached terrain. At 1200 UTC, waves near the coastal NPOL site were evident in a layer approximately 500 m deep and increased to 1 km in vertical extent over Fishery, with lessened wave activity over higher terrain. Unlike the first case, however, removal of the Olympic Mountains did not eliminate the easterly surface flow (Fig. 4.14) and wave activity remained without terrain, confirming that the primary cause of the KH waves during this event was the synoptic environment and not terrain blocking. Wave amplitude appears, however, to be unchanged when the topography is removed. 78

79 Figure 4.10: (a) NARR and (b) WRF 925 hpa geopotential height (black contours, 20-m interval), wind barbs, and wind (shading) for the 36-km domain, valid 0600 UTC December 17, (c) Observed and (d) 444-m simulated soundings at NPOL, valid at ~1430 UTC December 17,

80 Figure 4.11: (a) Reflectivity (shading) in dbz; (b) radial velocity in m s -1 ; (c) spectrum width; (d) differential reflectivity in db; (e) correlation coefficient from a RHI scan from the Ku-band of the D3R radar at an azimuthal angle of 254 at 1320 UTC on December 17, 2015: In each panel, the black dots show the location of the maximum spectrum width in each vertical column. (based on Figure 10 in Barnes et al. (2018), used with permission) 80

81 Figure 4.12: Vertical cross sections of (a) observed and (b) 444-m simulated radial wind extending northeast along the cross section in Fig. 2, valid 1200 UTC December 17,

82 Figure 4.13: Vertical cross sections of (a) Richardson number, (b) vertical wind shear, and (c) stability along the cross section in Fig. 2 at various stages of the December 17, 2015 event. 82

83 Figure 4.14: Vertical cross sections of (a) 4 -km and (b) 444-m simulated radial wind along the cross section in Fig. 2, valid 1200 UTC December 17, (c) and (d) are the same cross sections as (a) and (b), but the topography of the Olympic Mountains has been removed in those simulations. 83

84 Impact of KH waves on microphysics Vertical velocity modulations associated with the KH waves had less impact on model microphysics than in the December 12 th case, likely because waves were weaker and occurred below the melting level. Figure 4.15 shows cross sections between the NPOL and Bishop sites at 1320 UTC, when Barnes et al. (2018) observed KH waves in the vicinity. While there were modest modulations in vertical velocity, D mw, simulated reflectivity, and rain number concentration, the amplitudes were considerably weaker than during the previous event. Turning to the observations and model output at the Fishery site, modest modulations were noted between UTC in several observed microphysical quantities and precipitation rate (0.3 mm hr -1 ), with periods similar to the observed KH waves (Barnes et al., 2018). Simulated mixing ratios at the Fishery site showed minor variations in rain, cloud water, and snow during the same period, with little graupel present (Fig. 4.16a). Simulated vertical velocities oscillated with an amplitude of ~1 ms -1 (Fig. 4.16b) several ms -1 less than observations from vertically-pointing radar. In contrast to the December 12 th case, time series of simulated precipitation rates and D mw showed less influence from KH waves during this event (Fig. 4.16c,d). 84

85 Figure 4.15: Vertical cross sections from NPOL to Bishop along the line in Fig. 2 of (a) vertical velocity, (b) reflectivity, (c) mass-weighted mean rain drop diameter, and (d) rain number concentration at 1320 UTC December 17. The dashed line is the melting level. 85

86 Figure 4.16: Simulated vertical profiles at the Fishery site during the period UTC December 17 of: (a) cloud water mixing ratio (grey fill; grey dashed contours), rain water mixing ratio (red), graupel mixing ratio (black), and snow mixing ratio (blue) ; (b) 0.5 km vertical velocity (solid), km maximum vertical velocity (dashed), with contours every 0.1 g kg -1 ; (c) mass-weighted mean rain drop diameter at the lowest model height; and (d) precipitation rate (solid) and observed precipitation rate (dashed). 86