Simulation of seasonal snowfall over Colorado

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

http://www.elsevier.com/copyright

Author's personal copy

Simulation of seasonal snowfall over Colorado

Kyoko Ikeda a, Roy Rasmussen a,⁎, Changhai Liu a, David Gochis a, David Yates a, Fei Chen a,Mukul Tewari a, Mike Barlage a, Jimy Dudhia a, Kathy Miller a, Kristi Arsenault c, Vanda Grubišić b,Greg Thompson a, Ethan Guttman a

a National Center for Atmospheric Research, United Statesb University of Vienna, Austriac George Mason University, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 9 January 2010Received in revised form 3 April 2010Accepted 19 April 2010

This paper presents results of four high resolution simulations of annual snowfall overColorado, U.S.A. The results are verified using SNOTEL data. Sensitivity to model resolution isalso explored. The results show that proper spatial and temporal depictions of snowfalladequate for water resource and climate change purposes can be achieved with the appropriatechoice of model resolution and physical parameterizations.

© 2010 Elsevier B.V. All rights reserved.Keywords:SnowfallSeasonalSNOTELColoradoHeadwatersRocky mountains

1. Introduction

Water is a key global resource that is essential to thedevelopment and sustainability of human civilization. Egyp-tian and Mesopotamian societies, for instance, were devel-oped along the Nile and Tigris and Euphrates river systems tobe close to a reliable water source. These river systems, inturn, are driven by precipitation in upstreammountains, suchas the Ethiopian Highlands. Modern societies, such as theWestern U.S., remain largely dependent on river flows drivenby orographic precipitation. The Colorado River is one suchsystem that is critical for a significant fraction of Western U.S.water needs.

Despite the critical importance of water in theWestern U.S.,current modeling systems do not accurately simulate seasonalsnowfall or snowpack. For instance, Leung et al. (2003) showthat current regional climate models typically underestimate

precipitation by ∼25% in the Western U.S. The headwatersregion of the Colorado River (Upper Colorado River Basin)seems to be a particularly difficult area for climate models toproperly handle, with inconsistent snowfall trends in thisregion from both the 3rd and 4th IPCC reports (2001, 2007,respectively), despite consistent predictions of temperatureincreases in this region from all climate models. With theincreasing recognition of global and regional climate warming,water managers are rightly concerned about the potentialimpact of climate change on water in the Western U.S.,especially given that recent studies suggest that globalwarming may lead to unprecedented drought conditions inthe Southwest U.S. (4th IPCC Assessment). The ColoradoHeadwaters region is particularly important, since ∼85% ofthe streamflow for the Colorado River comes from snowmelt inthis region. A recent analysis of the 2007 IPCC FourthAssessment global models by Hoerling and Eischeid (2006)indicates that the combination of increased temperatureand weak to no trend in snowfall will produce un-precedented drought conditions over the next 50 years in the

Atmospheric Research 97 (2010) 462–477

⁎ Corresponding author.E-mail address: [email protected] (R. Rasmussen).

0169-8095/$ – see front matter © 2010 Elsevier B.V. All rights reserved.doi:10.1016/j.atmosres.2010.04.010

Contents lists available at ScienceDirect

Atmospheric Research

j ourna l homepage: www.e lsev ie r.com/ locate /atmos


Southwest U.S. Seager et al. (2007) come to similar conclusionsregarding future runoff in the Southwest through an indepen-dent analysis of the IPCC Fourth Assessment global models.While the above predictions from the global model runs fromthe 4th IPCC Assessment indicate dire consequences for theSouthwest U.S., it also must be noted that the assessmentindicated that globalmodels typically performpoorly in regionsof complex terrain. The Upper Colorado Region is, in particular,driven by high altitude snow melt, so climate assessments inthis region using global models are uncertain. It is thereforecritical to examine climate impacts in this region using higherresolution models in order to more realistically simulateorographic precipitation and evaporation processes.

Key aspects of snowfall, snowpack, evapotransporationand runoff potentially improved by high resolution climateruns with adequately resolved topography are:

1. Proper depiction of vertical motions leading to increasedintensity of clouds and snowfall.

2. Formation of an isothermal layer at 0°C frommelting snowleading to additional snowpack.

3. Improved simulation of airflow blocking effects on theflow and associated snowfall.

4. Proper depiction of terrain-induced embedded convection.5. Improved spatial depiction of the local snow accumulation,

accounting for local ridge shadowing reduction of snow-melt and sublimation.

6. Improved depiction of evapotransporation and runoff.

In this study, we perform simulations of winter precipi-tation between 1 November and 1 May for four retrospectiveyears at various resolutions using various parameterizationsand compare the model results to SNOTEL (SNOwpackTELemetry) observations. The current paper will attempt toaddress the following questions:

1. Can a properly configured high resolution regional modeladequately simulate seasonal snowfall over the ColoradoHeadwaters region?

Fig. 1. Retrospectivemodel domain and location of SNOTEL sites (black dots). Shown in (a) is the full model domain, and (b) a sub-domain focused on the ColoradoHeadwaters region. Locations of some towns and cities in and near the sub-domain are indicated by stars.

Table 1List of simulations performed in the current study. The rightmostcolumn indicates the number of SNOTEL sites that were operationalduring the respective years in the sub-domain (Fig. 1b). The SNOTELdata were used for model verification.

Wateryear

Simulationperiod

Modelresolution

Number ofSNOTEL sites

2001–2002 1 Nov.–1 May 2 km Dry 952003–2004 1 Nov.–1 May 2 km Average 1022005–2006 1 Nov.–1 May 2 km Average 1082007–2008 1 Nov.–31 Oct. 2 km Wet 1122007–2008 1 Nov.–1 May 6, 18, and 36 km Wet 112

463K. Ikeda et al. / Atmospheric Research 97 (2010) 462–477


2. What is the sensitivity of the snowfall predictions to modelresolution?

Section 2 will present the model setup. Section 3 providesa description of SNOTEL data. Comparison of retrospectivemodel runs to SNOTEL observations will be presented inSection 4. A discussion of the role of improved verticalvelocity simulation on snowfall with different model resolu-tions is given in Section 5. Results and discussion are given inSection 6, and summary and conclusions in Section 7.

2. WRF model setup

The Weather and Research Forecast (WRF) regionalweather and climate model version 3.0 (Skamarock et al.,2008) was used to perform all the simulations presented in

this study. The model was configured for a single domain of1200×1000 km2 with 45 vertical levels (Fig. 1a). Horizontalgrid spacing in the domain was varied between 2, 6, 18and 36 km depending on the run. A large domain size waschosen in order to properly represent the influence ofupstream mountains on the flow and moisture depletion.The final model configuration includes the followingparameterizations:

• The Mellor–Yamada–Janjic (MYJ) Planetary Boundary Layerscheme (Janjic, 2002)

• Noah land-surface model with new enhanced snow albedo(Chen and Dudhia, 2001)

• The NCAR Community Atmosphere Model (CAM) longwaveand shortwave schemes (Collins et al., 2004)

• Thompson et al. (2008) cloud microphysics scheme

Fig. 3. Comparisons of the 2-km WRF to SNOTEL site average precipitation accumulation (mm) for the 6-month retrospective simulations from (a) 2001/2002(dry year), (b) 2003/2004 (average year), (c) 2005/2006 (average year), and (d) 2007/2008 (wet year).

Fig. 2. (a) A photograph of an all-weather precipitation gauge at the SNOTEL site in Brooklyn Lake, WY. (b) Snow water equivalence comparison of the BrooklynLake SNOTEL gauge (solid) to the NCAR VRG (long dash), and GEONOR (short dash) accumulations at the Glees site between 15 March and 8 April 2008. TheBrooklyn Lake SNOTEL site is located within 2 km of the Glees site.

464 K. Ikeda et al. / Atmospheric Research 97 (2010) 462–477


The choice of this configurationwas established through14-day simulations with various parameterization combinationsand comparison to SNOTEL data (results not shown in thisstudy). Precipitation processes over complex terrains arecomplex and thus the model can be sensitive to themicrophysical scheme used. Detail examination of sensitivityto microphysics parameterizations will be presented in aseparate paper.

The method to initialize and drive the model for theretrospective winter precipitation simulations was throughthe use of the North American Regional Reanalysis (NARR)data (Mesinger et al., 2006). The NARR dataset is availableevery 3 h over North America with a 32-km horizontal gridresolution. The three-hourly lateral boundary conditionsfrom the NARR were used to drive the WRF model duringthe 6-month simulation.

3. Verification data

Our analysis focuses over the Colorado Headwaters regionshown in Fig. 1b (referred to as the model sub-domain in thisstudy). SNOTEL observations (see Serreze et al., 1999 for arecent description) provide a long-term record of precipita-tion from weighing precipitation gauges at numerous sitesthroughout the Colorado Headwaters region. These sites are

owned and operated by the Department of AgricultureNatural Resource Conservation Service (NRCS), and aretypically located at elevations between 2400 and 3600 mabove mean sea level (MSL). The number of operational sitesin the sub-domain during each simulation year presented inthis study is given in Table 1.

Hourly and daily precipitation records from all of the pastwater years (1 November through 31 October) with 0.1 in(2.5 mm) resolution are available from the NRCS web site. Avariety of authors have described the use of these data,including known deficiencies (Serreze et al., 1999; Serreze etal., 2001; and Johnson and Marks, 2004; for example). Themain issuewithweighing type gauges for snowfall estimationis the undercatch due towind (Serreze et al., 2001; Yang et al.,1998; Rasmussen et al., 2001). Based on the location ofSNOTEL gauges in a forest clearing, most of the time the windspeed is less than 2 m s−1, for which an undercatch ofapproximately 10–15% is expected (Yang et al., 1998).Another possible limitation is the 0.1 in (2.5 mm) resolution.Over the simulation period of six months for the currentstudy, however, in which over a 200 mm of melted snow istypically collected, this is not an issue.

To further gain confidence in the performance of theSNOTEL precipitation gauge, an inter-comparison was madeof the snowfall accumulation from the Brooklyn Lake SNOTELsite located in the Medicine Bow mountain range inSoutheastern Wyoming with two nearly co-located indepen-dent weighing snow gauges. These independent snow gaugeswere installed and managed by NCAR for a WeatherModification evaluation project funded by the State ofWyoming and included a GEONOR weighing gauge (Rasmus-sen et al., 2001) and a Vaisala weighing Rain Gauge (VRG),both with a 0.25 mm resolution. Comparison was madeduring a 25-day period from 15 March 2008 to 8 April 2008,during which all three gauges were operating (Fig. 2). Theaccumulation curve for the SNOTEL precipitation gauge fallsin-between the accumulation curves for the two independentgauges, providing confidence in all three gauges, includingthe SNOTEL gauge. All three gauges were situated in forestclearings, and thus winds are expected to have a relativelysmall impact on gauge catch efficiency.

4. Model simulation result

4.1. Comparison of 2 km WRF model simulations to SNOTELobservations

A list of simulations conducted for this study is given inTable 1. Simulations over a 6-month period from 1 Novemberthrough 1 May at 2 km horizontal resolution were performedfor 2001/2002, 2003/2004, 2005/2006 and 2007/2008. Thesefour years were chosen to represent a low, two average, andone high snowfall years, respectively. In addition, a full yearsimulation was conducted for 2007/2008 in order to span thesnowmelt and summer convective season. In order to focusour discussion to cold season precipitation, results from the 6-month period simulations are discussed in this study.

Fig. 3 shows the precipitation accumulation at SNOTELsites from the four retrospective simulations with 2 kmresolution. The model runs are compared to the averageprecipitation accumulation from SNOTEL sites. The model

Table 2Monthly precipitation totals and 6-month total precipitation from modelruns and observations (SNOTEL) for four simulation years. Model values atSNOTEL sites were determined from the average of four closest grid points.The table compares average model values for all SNOTEL sites in the sub-domain (Fig. 1b) with average SNOTEL precipitation amounts. Difference andpercent difference (with respect to SNOTEL averages) between the modelruns and observations are also listed. The differences are computed as modelminus SNOTEL observation.

(a) 2001–2002

Nov. Dec. Jan. Feb. Mar. Apr. 6 mo.

Model (mm) 77.3 55.2 52.7 52.6 73.4 50.2 361.3SNOTEL (mm) 68.0 55.8 44.7 46.9 65.9 37.4 318.7Difference (mm) 9.3 −0.6 8.0 5.6 7.5 12.8 42.6Percent diff. (%) 13.7 −1.1 17.8 12.0 11.4 34.4 13.4

(b) 2003–2004


Model (mm) 139.2 78.6 70.6 88.1 43.9 114.6 535.0SNOTEL (mm) 118.6 75.2 68.2 79.9 45.6 106.0 493.5Difference (mm) 20.6 3.4 2.4 8.2 −1.7 8.6 41.5Percent diff. (%) 17.4 4.5 3.5 10.3 −3.7 8.2 8.4

(c) 2005–2006


Model (mm) 74.4 119.8 99.3 52.0 145.3 85.7 576.5SNOTEL (mm) 87.7 112.3 82.5 51.0 105.8 61.4 500.6Difference (mm) −13.2 7.5 16.8 1.0 39.4 24.4 75.8Percent diff. (%) −15.1 6.6 20.4 1.9 37.3 39.7 15.1

(d) 2007–2008


Model (mm) 41.2 157.5 156.7 115.5 73.9 54.0 598.8SNOTEL (mm) 38.2 149.7 136.3 117.4 69.7 71.8 583.1Difference (mm) 3.0 7.8 20.5 −1.9 4.2 −17.8 15.7Percent diff. (%) 7.8 5.2 15.0 −1.6 6.1 −24.8 2.7



Fig.

4.Mon

thly

timeseries

ofaccu

mulativeprecipitation(m

m)for(a)20

01/200

2,(b

)20

03/200

4,(c)20

05/200

6,an

d(d

)20

07/200

8from

theW

RFmod

elan

dSN

OTE

Lda

ta.T

hemod

elprecipitationat

theSN

OTE

Lsitesis

basedon

thesimplefour-point

averag

ingas

inFig.

3.Graysh

ades

represen

ton

estan

dard

deviationfrom

theav

erag

eda

ilyprecipitationtotals

atSN

OTE

Lsites.



precipitation accumulation at the SNOTEL sites was deter-mined in four different ways: (1) computing the averagevalue of the four closest grid points; (2) taking an inverse-distance weighted mean of the four closest grid points; (3)

performing a bilinear interpolation; and (4) performing acubic interpolation. All four techniques produced similarresults at this model resolution. The model precipitationshown in Fig. 3 is based on the simple four-point averaging.Table 2 summarizes Fig. 3 with a list of monthly and 6-monthtotal precipitation from the model simulations and SNOTELobservations and their differences. The results show that the6-month total precipitation from the model agrees to within10% for the 2003/2004 and 2007/2008 simulations, and 15%for the 2001/2002 and 2005/2006 simulations. The modelprecipitation is higher than the average SNOTEL precipitationfor all four years. Considering that the SNOTEL precipitationgauge likely undercatches up to 15% due to wind effects, theagreement is remarkable.

The monthly accumulation time series for the foursimulation years is shown in Fig. 4. The simulated precipita-tion curves follow the average SNOTEL observation well,

Fig. 5. Monthly total precipitation from the SNOTEL data and model simulations for (a) 2001/2003, (b) 2003/2004, (c) 2005/2006, and (d) 2007/2008. Percentdifferences with respect to observations are indicated at the top of each pair of bars.

Fig. 6. (a) Percent difference and (b) actual difference in monthly totalprecipitation for the four simulation years.

Fig. 7. Time history of average top 1% CAPE values within the sub-domainfrom 1 November 2005 to 1 May 2006.



Fig. 8. Spatial patterns of monthly total precipitation (mm) from the 2-kmWRF simulations (left panels) and SNOTEL observations (right panels) for (a) December2007, (b) January 2008, and (c) February 2008, (d)March 2008, and (e) April 2008. Precipitation amounts at SNOTEL sites are indicated using filled circles (legendsare in the bottom right).



indicating that individual weather events are well repre-sented by the 2-km resolution model.

Monthly total precipitation from the model and SNOTELdata and their relative differences are shown in Figs. 5–6,respectively. For a wide range of monthly totals, the modelagrees within ∼20% from November through March exceptfor March 2006. The percent difference increases in April inthe 2001/2002, 2005/2006, and 2007/2008 simulationsalthough the actual difference is only 20 mm or less. Thismay be partly related to the model's inability to properlyresolve small-scale convective motions as precipitationsgradually shift to convective regimes in April over the sub-domain. As evidence, Fig. 7 shows the temporal evolution ofthe average top 1% values of CAPE (Convective AvailablePotential Energy), a measure of potential strength ofconvection, in the sub-domain during the 2005–2006simulation. As expected, relatively large convective instabilityis only occasionally present in the deep winter months andgradually strengthens as the time progresses into the springseason. The enhanced instability in the spring would lead tothe development of vigorous small-scale convective elementsembedded within the large-scale synoptic and orographi-cally-generated precipitating systems. However, the convec-tive-scale motions cannot be adequately resolved at a ∼2 kmresolution and have a poor predictability, thereby leading tocomparatively large precipitation biases later in the simula-tion period. The enhanced convective activity in the earlyspring is also consistent with the increased portion ofparameterized rainfall in the coarse resolution simulations(not shown).

The spatial pattern of monthly accumulation from De-cember 2007 through April 2008 is shown in Fig. 8. The figurereveals that the model precipitation accumulation at specificSNOTEL locations also compares well to the observed SNOTELaccumulation. The SNOTEL sites are located at elevationstypically above 2400 m MSL, which are the locations of thehighest snowfall. The agreement between the model and theSNOTEL precipitation provides confidence that the highresolution model can properly simulate snowfall over theColorado Headwaters domain.

4.2. The impact of horizontal resolution on WRF model snowfallaccumulation

One of the goals of this study is to determine the modelresolution needed to make accurate snowfall predictions forboth weather forecast and climate models. For example, whatresolution do regional climate models need in order toadequately depict both snowfall amounts and spatial distribu-tion? Both of these aspects are important to hydrologiststrying to predict future streamflow and other hydrologicalquantities. To investigate this question, WRF model simula-tions were performed with 6, 18, and 36 km resolutions forthe 2007/2008 season over the model domain shown inFig. 1a. Resolutions of 18 and 36 km are typical of currentregional climate models (Leung et al., 2006), which typicallyunderestimate snowfall by 15–30% (Leung et al., 2003). Forthe 18 and 36 km resolutions, various cumulus parameteri-zation schemes currently available in version 3.0 of the WRFmodel [explicit (EX), Kain–Fritsch (KF, Kain and Fritsch, 1993,

Fig. 9. (a) Average precipitation accumulation from 112 SNOTEL sites and WRF simulations at 36, 18, 6, and 2 km horizontal resolutions. Results from simulationswith explicit (EX), Kain–Fritsch (KF), Grell–Devenyi ensemble (GD), and Betts–Miller–Janjic (BMJ) cumulus parameterization schemes are shown for the 18- and36-km resolution runs. The model values were interpolated at the individual SNOTEL sites using bilinear interpolations. Comparisons using three other ways (seetext) to obtain model precipitation values at corresponding SNOTEL sites showed similar results. (b) Percent difference in monthly total precipitation between themodel and SNOTEL data.



Kain, 2004), Grell–Devenyi ensemble (GD, Grell and Devenyi,2002), and Betts–Miller–Janjic (BMJ, Janjic, 1994) schemes]were tested. Although non-convective precipitation domi-nates inwintermonths, the intent of running themodels withvarious cumulus parameterization schemes was to examinemodel sensitivity to cumulus parameterization in late winteror early spring.

Fig. 9a compares SNOTEL average precipitation accumula-tions to model accumulations at four horizontal resolutions.The accumulations for the 2- and 6-km simulations are nearlyidentical and within 3% of the SNOTEL accumulation. However,the 18- and 36-km simulations are ∼15% and 23–31% lowerthan the 6-month SNOTEL total accumulation, respectively(Table 3). The 36-km resolution runs produced precipitationwell below the observations in all months, generally with a

negative bias on the order of 18% or greater (Fig. 9b). Thedifferences are relatively small among the various cumulusparameterization schemes. Because the differences withrespect to the observations are larger than those among thevarious schemes, results from the BMJ cumulus parameteriza-tion scheme will be shown hereafter.

The decreasing snowfall at high elevations as modelresolution becomes coarser is also evident at specific SNOTELlocations, as shown in Fig. 10. At a few sites, the coarserresolution models produced more precipitation than the 2-and 6-km simulations (e.g., Fig. 10j). Results such as shown forthe Nast Lake SNOTEL site (Fig. 10j) occurred typically in avalley or a region of rain/snow shadow on a lee side of a ridgewhere details in the local topography were not resolved in thecoarser models. Frequency plots of the absolute percent

Table 3(a) Monthly precipitation totals and 6-month total precipitation from model runs and observations (SNOTEL) for the 2007/2008 run at 2, 6, 18, and 36 kmresolutions. Model values at corresponding SNOTEL sites were determined from bilinear interpolations. The values are average of all SNOTEL sites in the sub-domain (Fig. 1). (b) Percent difference (with respect to SNOTEL averages) and (c) difference in precipitation between the model runs and observations are alsoindicated. The differences are computed as model minus SNOTEL as in Table 2.

(a) Monthly and 6-month total precipitation (mm)


36 km (EX) 29.7 112.0 110.9 77.0 47.1 33.8 410.436 km (BMJ) 29.4 114.6 112.6 77.2 48.8 36.5 419.136 km (KF) 32.7 122.6 118.6 81.7 53.7 37.2 446.536 km (GD) 28.8 112.7 109.5 74.8 45.0 30.6 401.518 km (EX) 34.6 130.4 135.9 95.2 58.6 41.9 496.418 km (BMJ) 34.7 133.0 139.3 95.3 59.0 42.4 503.718 km (KF) 36.1 132.1 136.6 95.2 59.6 42.6 502.218 km (GD) 33.6 129.9 134.7 91.5 53.7 39.2 482.66 km 41.2 153.7 155.6 111.8 73.1 53.0 588.42 km 41.5 158.2 157.6 116.0 74.1 54.1 601.6SNOTEL 38.2 149.7 136.3 117.4 69.7 71.8 583.1

(b) Percent difference (%)


36 km (EX) −22.2 −25.2 −18.6 −34.4 −32.5 −52.9 −29.636 km (BMJ) −23.1 −23.5 −17.4 −34.2 −30.0 −49.2 −28.136 km (KF) −14.4 −18.1 −12.9 −30.4 −23.0 −48.2 −23.436 km (GD) −24.7 −24.7 −19.6 −36.2 −35.5 −57.3 −31.118 km (EX) −9.5 −12.9 −0.3 −18.9 −16.0 −41.6 −14.918 km (BMJ) −9.1 −11.1 2.2 −18.8 −15.3 −41.0 −13.618 km (KF) −5.6 −11.8 0.3 −18.9 −14.5 −40.7 −13.918 km (GD) −11.9 −13.2 −1.2 −22.1 −23.0 −45.4 −17.26 km 7.9 2.7 14.2 −4.8 4.9 −26.3 0.92 km 8.7 5.7 15.7 −1.2 6.3 −24.6 3.2

(c) Difference (mm)


36 km (EX) −8.5 −37.8 −25.4 −40.4 −22.7 −38.0 −172.736 km (BMJ) −8.8 −35.1 −23.7 −40.2 −20.9 −35.3 −164.036 km (KF) −5.5 −27.1 −17.6 −35.7 −16.0 −34.6 −136.636 km (GD) −9.4 −37.0 −26.7 −42.6 −24.7 −41.2 −181.618 km (EX) −3.6 −19.4 −0.4 −22.2 −11.2 −29.9 −86.718 km (BMJ) −3.5 −16.7 3.0 −22.1 −10.7 −29.5 −79.418 km (KF) −2.1 −17.6 0.4 −22.2 −10.1 −29.2 −80.918 km (GD) −4.6 −19.8 −1.6 −25.9 −16.0 −32.6 −100.56 km 3.0 4.0 19.3 −5.6 3.4 −18.9 5.32 km 3.3 8.5 21.4 −1.4 4.4 −17.7 18.5

Fig. 10. Accumulative precipitations from select SNOTEL sites and corresponding model precipitation at 36, 18, 6, and 2 km horizontal resolutions. Bilinearinterpolation was used to obtain model data at SNOTEL sites.





difference in theWRFmodel accumulation from correspondingSNOTEL accumulation at 112 SNOTEL sites for the fourresolutions are shown in Fig. 11. The results show that the2 km simulation is within 20% of the SNOTEL amount 71% ofthe time, while the 18 and 36 km simulations are within 20%of the SNOTEL accumulation only 48 and 39% of the time,respectively. The results indicate the clear superiority of the2 and 6 km simulations over the 18 and 36 km resolutionsimulations typically used in current regional climatemodels.

5. Examination of the impact of higher resolution on localvertical motions and resulting snowfall

It was demonstrated in the previous section that theaccurate seasonal snowfall simulations were possible withhorizontal resolutions smaller than 6 km. In this section weinvestigate the impact of higher model resolution on localvertical motions and the resulting snowfall in order toexamine whether high resolution models allow improvedsimulation of mesoscale and cloud-scale processes importantto the accurate snowfall simulation for both weather andclimate time scales.

5.1. Proper depiction of vertical motions leading to increasedintensity of clouds and snowfall

A robust finding from this study is that vertical motionsassociated with winter storms over the Colorado Headwatersregion have updraft/downdraft magnitudes on the order of1 m s−1 or greater associatedwith gravitywave formation overterrain features, such as isolated peaks, complexes of highterrain, and ridges/valleys. Fig. 12 presents an example of the600 hPa vertical velocity at 0500 UTCon1December2007 fromthe2-kmsimulation.Note thatupdrafts are locatedupstreamofeach significant terrain feature withmagnitude of ∼1 m s−1 orgreater, with corresponding downdrafts of similar magnitudeon the lee side. These updrafts create locally large condensatesupply rates that lead to higher liquid water contents, snowmixing ratios, and enhanced graupel formations than wouldoccur in the typically weak updrafts associated with coarseresolution simulations (on the order of cm s−1). As a result,snow formation is focused close to the peaks of the terrainwhere updrafts are typically strongest, resulting in snowfallcontours that closely resemble topography contours in the 2and 6 km simulations (e.g., Fig. 8).

As an example, Fig. 13 shows a height-distance crosssection of vertical velocity between A and B (Fig. 13e) at 2, 6,

Fig. 11. Frequency distributions showing absolute percent difference in model simulated, 6-month total precipitation at (a) 2 km, (b) 6 km, (c) 18 km, and (d)36 km horizontal resolutions with respect to observations at each of the 112 SNOTEL sites. Values above bars indicate the number of SNOTEL sites in each of thepercent difference category.



18, and 36 km horizontal resolutions. Note the close corre-spondence of the 2- and 6-km resolution topographies toeach other, and the significantly smoother terrain features ofthe 18- and 36-km topographies. As a result, updraftsproduced by the 18- and 36-km topographies are broaderand weaker (by 50% or more) than the updrafts produced bythe 2 and 6 km topographies. In addition, the 2- and 6-kmtopographies induce updrafts close to the peaks, therebyproducing more precipitation near the peaks and overrelatively narrower horizontal regions. In contrast, since thelow level air in the 18 and 36 km simulations begins to glideup the windward side of the mountain hills generally earlieror at a more upwind location compared to the 2 and 6 kmruns due to the smoothed topography, precipitation amountsat lower elevations or in the valleys are larger in the 18 and36 km models. This effect is illustrated in Fig. 14, whichcompares the 3-hour precipitation total (snow plus rain) and3-hour snow and rain accumulations along the cross sectionin Fig. 13 from two different resolution runs. As shown, the 6-km resolution simulation focuses the precipitation at thetopographic peaks where relatively strong vertical motionsoccur. Thus, the higher resolution simulation moves theprecipitation towards an improved spatial distribution, withmore precipitation on the peaks and less in the valleys thanthe 18 and 36 km runs. Similar results were shown inprevious studies of orographic precipitation (for example,Garvert et al., 2007; Rasmussen et al., 1988; and Grubišić et al.2005).

Adequate depiction of topography is also important inproperly simulating precipitation phase (i.e., rain versus

snow). For example, Fig. 14 shows more snowfall at thepeaks in the 6-km run, where the temperature is colder thanthe 36-km run (Fig. 14e). The combination of higher snowfalland colder temperatures at these peaks impact the timing andamount of snowmelt.

Fig. 15 shows the spatial difference in 6-month totalprecipitation between the 2 km run and the 6,18 and 36 kmruns. The differences between the 2 and 6 km runs are small,as expected, but the 18 and 36 km differences from the 2 kmprecipitation total are significant. In general, the 2 kmsimulation has higher accumulation at topographic peaks orhigh elevations (e.g., see the accumulation differences atSNOTEL sites), and less at low elevations and in the valleys,consistent with the previous discussion.

6. Results and discussion

6.1. Comparison to observations

This study used SNOTEL observations of snowfall toevaluate the ability of a high resolution model to accuratelysimulate snowfall over the Colorado Headwaters region. TheSNOTEL snowfall observations were compared to co-locatedhigh resolution gauges and shown to perform well (Fig. 2b).Both the spatial and temporal distributions of SNOTELsnowfall observations for the four simulation years comparedwell to the WRF model simulations at 2 km grid resolution(Figs. 3–4, and 8). These results provide a high degree ofconfidence that the currently chosen configuration of theWRF model is capable of reliably simulating snowfall over afull winter season for a variety of conditions.

6.2. Model resolution

The comparison of precipitation amounts as a function ofmodel resolution reveals that a very good agreement withSNOTEL data is achieved if a model resolution of 6 km orsmaller is used. At this resolution, snowfall is predicted tomostly fall at the higher elevations, consistent with observa-tions. Snowfall is predicted to be as much as 57% less at theSNOTEL sites in the 36-km resolution runs, but with moredistributed in the valleys (Fig. 15). This in part results fromsmoother terrain features represented in the lower resolutionmodels and the consequent poor simulation of verticalmotions associated with local peaks and valleys (e.g.,Figs. 13–14). In addition, fine resolution models betterrepresent non-hydrostatic features of topographically-in-duced motions such as gravity waves (Smith, 1979), andthis would certainly contribute to improved snowfall simula-tions. Since the more-defined topographic peaks of the 2-kmmodel are higher and, therefore, colder, snow on the peakswill last longer in these simulations than in the 18 and 36 kmruns.

The spatially redistributed precipitation in the coarserresolution runs also impacts the rain–snow distribution overthe Colorado Headwaters domain. Fig. 16 shows differencesin 6-month total rain and snow between the 2- and 36-kmmodel runs. The results indicate more rain produced from the36-km simulation (warm colors) in the valleys or locations ofascent for westerly winter storms. This indicates that aportion of the higher precipitation amount in these areas

Fig. 12. The 600 hPa vertical wind speed at 0500 UTC on 1 December 2007from the 2-km simulation (color). Arrows indicate horizontal winds at600 hPa. Thin gray contours show the underlying topography.





comes from rain instead of snow. Additionally, the highestelevations have significantly more snowfall in the 2 kmsimulation as compared to the 36 km simulation (coldcolors). This would have a significant impact on the

subsequent melting and sublimation of snowpack andamount of runoff. When comparing model performance inan average sense (Fig. 17), the results show a similar rain–snow ratio over the domain. However, when consideringhydrological impact and changes to snowpack as climatewarms, our findings suggest that properly simulating thedistribution of snowfall with elevation is important.

Future studies will investigate the impact of climate changeon snowfall with the high resolution model used in this study.Model sensitivity to microphysical schemes will be presentedin future studies. In addition, hydrological impact from climatechange will be explored. Other mesoscale factors to beconsidered in future studies include the formation of additionalsnowpack resulting from the production of an isothermaltemperature layer at 0 °C due to melting snow, the role ofairflow blocking (see for example the equivalent potentialtemperature contours for the various model resolutions inFig. 13), embedded convection, local shading of radiation byridges, and the impact of locally resolved airflow and radiationon evapotransporation and subsequent runoff.

7. Summary and conclusions

The following bullets summarize this paper:

• Comparison of high resolution WRF simulations of seasonalsnowfall to SNOTEL observations over the Colorado Head-waters regions show very good agreement if a grid spacingof b6 km is used.

• The 18 and 36 km resolution runs underestimate 6-monthtotal SNOTEL snowfall by 13–31% as a result of:– Terrain smoothing and associated spreading of the

precipitation horizontally as a result of a broader andweaker updraft.

– Poor representation of mesoscale processes.• High resolution (2 km) simulations of annual snowfallsuggest that current global and regional model estimatesof snowfall (with a N18 km grid spacing) underestimatehigh elevation snow by 13–31%, and overestimate lowelevation snowfall by a similar amount.

• Future studies will focus on:– Impact of climate change on snowfall, snowpack, runoff,

and evapotransporation over the Colorado Headwatersregion using the high resolution model presented in thisstudy.

– Impact of mesoscale processes on snowfall and snowpackformation, including future climate simulation.

– Model sensitivity to microphysical scheme.

Acknowledgements

This research was made possible by an AcceleratedScience Discover computer time award by the NCARComputer Information Systems Laboratory. NCAR is spon-sored by the National Science Foundation. Roy Rasmussen is

Fig. 14. (a) 3-hour total precipitation, (b) 3-hour snow accumulation, (c) 3-hour rain accumulation, (d) model elevation, and (e) surface temperaturefrom the 6- and 36-km simulations for 1 December 2007 05 UTC between A(0 km) and B (500 km) in Fig. 13.

Fig. 13. (a)–(d) Height–distance cross sections of vertical velocity (color) at 1 December 2007 0500 UTC from model simulation at (a) 2, (b) 6, (c) 18, and (d)36 km horizontal resolutions. The cross section is from A (0 km) to B (500 km) in the terrain map shown in (e). Gray-filled area shows the terrain profile. The sliceplane is parallel to the mean upper level wind direction [black arrow in (c)]. Gray thin contours in (a)–(d) are equivalent potential temperature (K) in incrementsof 2 K. Thick gray lines indicate the 0 °C isotherm. Thick brown lines indicate where total condensate mixing ratio is 0.01 g kg−1.



Fig. 15. Spatial distribution of the difference in 6-month total precipitation between the 2 km simulation and (a) 6 km, (b) 18 km, and (c) 36 km from the 2007/2008 simulation. The differences are computed as the 2 km run minus the coarser resolution runs. Overlaid are SNOTEL sites (gray-filled circles) and elevation(gray contours).

Fig. 16. Spatial distribution of the difference in (a) 6-month total rain and (b) total snow between the 2 and 36-km simulations. Gray-filled circles indicate SNOTELsites. Gray contours are elevation in increment of 600 m.



grateful to Hans Pruppacher for his outstanding mentorshipand inspiration. This paper is dedicated to him.

References

Chen, F., Dudhia, J., 2001. Coupling an advanced land surface-hydrologymodel with the Penn State-NCAR MM5 modeling system. Part I: modelimplementation and sensitivity. Mon. Wea. Rev. 129, 569–585.

Collins, W.D., et al., 2004. Description of the NCAR Community AtmosphereModel (CAM 3.0). . (NCAR Technical Note NCAR/TN-464+STR, 226 pp).

Garvert, M., Smull, B., Mass, C., 2007. Multiscale mountain waves influencinga major orographic precipitation event. J. Atmos. Sci. 64, 711–737.

Grell, G.A., Devenyi, D., 2002. A generalized approach to parameterizingconvection combining ensemble and data assimilation techniques.Geophys. Res. Lett. 29. doi:10.1029/2002GL015311.

Grubišić, V., Vellore, R.K., Huggins, A.W., 2005. Quantitative precipitationforecasting of wintertime storms in the Sierra Nevada: sensitivity to themicrophysical parameterization and horizontal resolution. Mon. Wea.Rev. 133, 2834–2859.

Hoerling, M., Eischeid, J., 2006. Past peak water in the Southwest. SouthwestHydrol. 35, 18–19.

IPCC, 2001. Climate change 2001. In: Houghton, J.T., Ding, Y. (Eds.), TheScientific Basis. Cambridge University Press, Cambridge.

IPCC, 2007. Contribution of Working Group I to the Fourth AssessmentReport of the IPCC. . (Available online at http://ipcc-wg1.ucar.edu/index.html).

Janjic, Z.I., 1994. The step-mountain eta coordinate model: further develop-ments of the convection, viscous sublayer and turbulence closureschemes. Mon. Wea. Rev. 122, 927–945.

Janjic, Z.I., 2002. Nonsingular Implementation of the Mellor–Yamada Level2.5 Scheme in the NCEP Meso Model. . (NCEP Office Note, No. 437,61 pp).

Johnson, J.B., Marks, D., 2004. The detection and correction of snow waterequivalent pressure sensor errors. Hydrol. Process. 18, 3513–3525.

Kain, J.S., 2004. The Kain–Fritsch convective parameterization: an update. J.Appl. Meteor. 43, 170–181.

Kain, J.S., Fritsch, J.M., 1993. Convective parameterization for mesoscalemodels: the Kain–Fritcsh scheme. In: Emanuel, K.A., Raymond, D.J.

(Eds.), The Representation of Cumulus Convection in Numerical Models:Amer. Meteor. Soc. Monogr. 246 pp.

Leung, L.R., Qian, Y., Han, J., Roads, J., 2003. Intercomparison of globalreanalyses and regional simulations of cold season water budgets in theWestern United States. J. Hydromet. 4, 1067–1087.

Leung, L.R., Kuo, Y.-H., Tribbia, J., 2006. Research needs and directions ofregional climate modeling using WRF and CCSM. Bull. Amer. Meteorol.Soc. 87, 1747–1751.

Mesinger, F., et al., 2006. North American regional reanalysis. Bull. Amer.Meteor. Soc. 87, 343–360.

Rasmussen, et al., 1988. Comparison of orographic cloud and precipitationdevelopment of three different geographic locations using a 3-D nested-grid cloud model. Annalen der Meteorologie Vol. 1 (no. 25), 353–355(Offenbach a./M., n.s).

Rasmussen, R.M., et al., 2001. Weather support to deicing decision making(WSDDM): a winter weather nowcasting system. Bull. Amer. Meteor.Soc. 82, 579–595.

Seager, R., Ting, M., Held, I., Kushnir, Y., Lu, J., Vecchi, G., Huang, H.-P., Harnik,H., Leetmaa, A., Lau, N.-C., Li, C., Velez, J., Naik, N., 2007. Model projectionsof an imminent transition to a more arid climate in Southwestern NorthAmerica. Science 316, 1181–1184.

Serreze, M.C., Clark, M.P., Armstrong, R.L., McGinnis, D.A., Pulwarty, R.S.,1999. Characteristics of the western United Stated snowpack fromsnowpack telemetry (SNOTEL) data. Water Resour. Res. 35, 2145–2160.

Serreze, M.C., Clark, M.P., Frei, A., 2001. Characteristics of large snowfallevents in the montane western United States as examined usingsnowpack telemetry (SNOTEL) data. Water Resour. Res. 37, 675–688.

Skamarock, W.C., et al., 2008. A Description of Advanced Research WRFVersion 3. . (NCAR Technical Note NCAR/TN-475+STR, 113 pp).

Smith, R.B., 1979. The influence of mountains on the atmosphere. Advancesin Geophysics, Vol. 21. Academic Press, pp. 87–230.

Thompson, G., Field, P.R., Hall, W.R., Rasmussen, R.M., 2008. Explicit forecastsof winter precipitation using an improved bulk microphysics scheme.Part II: implementation of a new snow parameterization. Mon. Wea. Rev.136, 5095–5115.

Yang, D., Goodison, B.E., Metcalfe, J.R., Golubev, V.S., Bates, R., Pangburn, T.,Hanson, Cl.L., 1998. Accuracy of NWS 8" standard nonrecordingprecipitation gauge: results and application of WMO intercomparison.J. Atmos. Oceanic Technol. 15, 54–68.

Fig. 17. Domain average (a) total precipitation, (b) snow, (c) rain, and (d) rain–snow ratio from the 2, 6, 18, and 36 km model runs.


Simulation of seasonal snowfall over Colorado

Documents