1 Introduction

Numerical weather predictions (NWP) of the meteorological and environmental parameters of important weather events are strongly related to the time and space resolutions, initial conditions and data assimilation techniques. Over mountainous environments where model grids may not match the height changes because of quickly changing environmental conditions, measurements may include large uncertainties and variability; affecting the assessment of NWP model simulations.

The most difficult issue in assessing weather conditions over rough topography is to resolve the natural variability over short distances along the mountain slopes. It is well known that precipitation amount may increase or decrease with height, depending on how the thermodynamical conditions are distributed along mountain slopes (Isaac et al., 2012; Gultepe and Zhou, 2012; Mo et al., 2012). When the observations of precipitation type, amount, and phase changes along the mountain are limited, it is very difficult to assess model based predictions. Previous studies suggested that the lower the model resolution, the lower the precipitation amount, and the intensity of model based forecast precipitation rate decreases with increasing grid area size (Mailhot et al., 2012). They suggested that sampling strategies are important for model validation studies. This work showed that precipitation assessment using the model simulations should be done with the appropriate time and space scales, resolving the physical processes.

Uncertainty in the measurements of precipitation and visibility (Vis) is strongly dependant on particle density and definition of snow type as a function of temperature usually (Brandes et al., 2007; Gultepe et al., 2012). Although temperature (T) is an important factor in defining particle type and density, relative humidity with respect to water (RHw) and flow conditions also play an important role for particle shape and density estimation. Visibility estimated directly from particle spectra can include large uncertainties related to assumptions used for distinguishing particle shape. On the other hand, based on direct extinction of light (Gultepe and Milbrandt, 2011), Vis measurements can represent various particle shapes very well but a detailed particle spectra measurement is needed to relate Vis to particle shape and density. Although Vis measurements can be obtained accurately, precipitation rate (PR) measurements for snow may not be obtained accurately because of instrument issues. This suggests that selection of Vis and PR from various instruments should be done carefully. In the case of the Vaisala FD12P sensor, T between −3 and +6 °C was defined for the wet snow conditions that may not be valid for NWP model parameterizations for snow (Gultepe and Milbrandt, 2011).

The FD12P T sensor doesn’t represent outside air T but air T in the protected cylinder that results in a severe assessment issue for distinguishing snow type. The total precipitation sensor (TPS) PR may include large uncertainties when they are used in mountainous areas with strong turbulence conditions (Rasmussen et al., 2011). Gultepe et al. (2012) suggested Vis and precipitation measurements can be highly suspicious when T goes down below −30 °C because they are not calibrated for the cold temperatures. These studies suggest that measurements of meteorological parameters over the mountainous regions create challenging issues for meteorologists and researchers.

Rasmussen et al. (2011) provided a detailed summary of snow precipitation measurements. They stated that the estimation of snowfall rate remains one of the most challenging measurements to make because of the wide variety of snow types, shapes, size distributions, and particle densities. This is especially true for ground-based measurements using gauges (Goodison et al., 1989; WMO/CIMO, 1985; Yang et al., 1998, 2001). The weighing gauges weigh the accumulated snow in a bucket and use a glycol-based solution to prevent evaporation of water. Using gauges, PR is typically estimated using the amount of accumulation over a period of ~5–10 min. The Geonor weighing gauges, that use three transducers for measuring snow mass amount, are sensitive to vibration resulting from the wind effects (Thériault et al., 2012). Optical gauges (Gultepe and Milbrandt, 2010) that usually use a laser light, accurately measure particle volume, and perform well for rain. The Vaisala FD12P has combined an optical sensor with a simple measure of snow-water content obtained using a heated plate to provide improved snow estimates (Gultepe et al., 2012).

In the previous studies, T and RHw measurements were found to be highly correlated with wind speed (U h) and solar radiative fluxes (Huwald et al., 2009). Although manufacturers provide low uncertainties for T and RHw values, e.g. +1 and 5 %, respectively, these can be much larger in cold, windy, and changing moisture conditions (Gultepe and Starr, 1995). During the Fog Remote Sensing and Modeling-Ice Fog (FRAM-IF) project which took place at Yellowknife Airport NWT (North West Territories) for the winter of 2010–2011, RHw uncertainty reached up to 20 % (Gultepe et al., 2012). These uncertainties in the basic thermodynamic parameters can pose a serious challenge for assessing model simulations along short distances (<100 m) over mountainous regions (Bailey et al., 2012).

Variability of meteorological parameters observed over mountainous regions is usually very large (Gultepe et al., 2012). Using only surface in situ observations is not sufficient to represent the weather conditions along the horizontal and vertical dimensions; some remote sensing observations to capture environmental variability in the horizontal and vertical distances are also needed. For example, a profiling microwave radiometer (PMWR), microwave rain radar (MRR), and ceilometer as well as a scanning C-band radar were also help to better understand weather conditions in the horizontal and vertical scales (Gultepe et al., 2009; Joe et al., 2012). A PMWR provided T, q v (vapor mixing ratio), RHw, and liquid water content (LWC) in a profiling mode (Bianco et al., 2005) but it needs to be validated using in situ observations; this may be a challenge for stratiform clouds because of a weak signal compared to that of convective cloud systems. All these remote sensing platforms are needed to further assess the weather variability over the mountainous regions.

An overview of the measurements made during the Science of Nowcasting Winter Weather for the Vancouver 2010 Olympics and Paralympics (SNOW-V10) project was given by Joe et al. (2012) while the objectives of the overall project were summarized by Isaac et al. (2012). The present work focuses on the extensive measurements of fog, precipitation from weighing and optical gauges, particle spectra from 0.3 μm up to cm size, visibility, as well as other meteorological measurements such as 3D turbulence and solar radiation collected at the mountain top called the Roundhouse (RND at 1,856 m) and these measurements, representing extreme weather conditions, are given in next section.

The goals of this work are (1) summarize the instruments and their measurements for the Roundhouse (RND) site on Whistler Mountain, (2) show the variability and uncertainty in the precipitation type and amount, (3) show the response of the sensors to the temperature, wind, and solar radiation changes for various weather events (e.g. cases studies), and (4) emphasize the particle shape effect on precipitation rate estimation and discuss the variability of measurements over a fast changing altitude (e.g. ~500 m) and extreme weather conditions. There were no reference precipitation measurements as described in World Meteorological Organization (WMO) technical notes because of extreme weather conditions. The following sections will focus on RND instruments, analysis of the observations, results, discussion on uncertainty and variability, and conclusions.

2 Roundhouse (RND) Meteorological Observations

In this study, we will focus on the RND instruments and their observations collected during January–April 2010, representing extreme weather conditions such as high winds and turbulent conditions, blowing snow, and freezing precipitation. Figure 1 shows the locations of RND, VOA (Whistler Mountain high level), VOL (Whistler Mountain mid-level), VOT (Whistler Mountain timing flat), TFT (Whistler timing flat), and VOC (Nesters) stations. The symbols and acronyms used in the text are defined in the “Appendix”. To show the variability and scale dependency of PR and Vis, the measurements of the FD12P as well as T, RHw, and wind speed from the all mentioned sites are also used in the analysis. The heights of RND, VOA, VOL, VOT, and TFT are, respectively, 1,856, 1,640, 1,320, 805, and 776 m.

Fig. 1
figure 1

Shows the project sites over the Whistler Mountain: a topographic map (a) with stations along the mountain slope. The green line shows the path of the Whistler Gondola. The major instrumented sites were shown in (b), and instrumented towers are shown in (c). The FD12P is the Vaisala all weather present sensor and measured precipitation rate, amount, and visibility at 1 min sampling intervals. It was available over the all major stations. Pictures of PVI and GCIP are shown in (d) and (e), respectively

Full size image

Figure 1b shows the RND site and instrumented towers around the RND site. There were three towers, called tower 1 (T1), T2, and T3. Table 1 shows the list of the sensors and their definitions used in the towers. The main observations collected at RND were related to T, RHw, horizontal and vertical air speed (U h and w a), radiative fluxes, 8 Hz 3D wind speed and direction, particle spectra from disdrometers (LPM, OTT), Rosemount icing detector (RID) signal, precipitation type, amount, and rate (FD12P and HSS-VPF-730), particle video imaging (PVI), snow and droplet spectra, and particle shape (GCIP, FMD, snow photography), visibility (FD12P, HSS-VPF-730, Sentry), aerosol spectra (>0.3 μm at 8 channels, CAP) and short wave radiative fluxes (SWRF) (SPN1 and Eppley radiometers), snow amount and rate [Geonor and TPS (total precipitation sensor)], and radar reflectivity (MRR). The other measurements (Joe et al., 2012) were available from the remote sensing platforms (e.g. PMWR and MRR) located at the TFT (Timing Flats, 776 m), and Whistler Doppler Radar located at (VVO, 557 m). Radiosonde balloons were released from the VOC (Nesters, 651 m).

Table 1 The instrument list and their specific characteristics as well as their location at the site (Fig. 1) RND-L and RND-R represent the left and right directions, respectively, of the RND's north direction
Full size table

Because of the RND location, the measurements represent extreme weather conditions. There was no WMO suggested reference site for precipitation, as a double fenced intercomparison reference (DFIR) shield installation was not possible at any location along the Whistler mountain slope. Snow gauge shields were not used due to wet snow capping issues. The site was significantly affected by strong winds and turbulence, making measurements very difficult during extreme weather conditions such as blowing snow, freezing precipitation, and cold temperatures. The Geonor is an operationally used sensor. The TPS is a relatively new instrument and its response to strong wind environments was tested. Note that the TPS precipitation algorithm uses a collection efficiency correction as a function of wind speed. Many sensors had difficulty operating properly when blowing snow and freezing precipitation occurred. Solar radiation effects were large on T and RHw measurements and they are summarized in discussion section.

All the measurements were collected with an identical pair of sensors in case one of them failed or to assess their responses in various environmental conditions. The sensors were calibrated at RND and also at EC (Environment Canada) before the deployment. The visibility sensors were calibrated in the field using the calibration kits and procedures for that instrument. The precipitation sensors were tested using a known mass of water, for example, by pouring water into the weighing gauges. The sensors used for T and RHw were newly purchased and were factory calibrated. They were also checked using recommended procedures for possible issues after the project. If the sensors were not heated, errors due to freezing precipitation could affect measurements significantly such as wind measurement observed with the Young 3D wind sensor during wet snow and icing conditions. These conditions were identified using a RID sensor which responds to icing conditions.

3 Analysis and Instrumental Issues

Analysis of observations were performed by examining: (1) time series of various meteorological and thermodynamical measurements, (2) scatter plots of the same parameters from identical two sensors to assess the bias, (3) wind and SWRF effects on T, RHw, and precipitation rate, and (4) Vis versus PR plots to show the variability for the various particle shapes. Then, PR, Vis, T, RHw, and wind speed (U h) values from the three stations (RND, VOA, and VOL) were averaged to represent the 0.5 km horizontal scale (slope ~0.7 km).

In the analysis, the five cases with three major events, representing: (1) blowing snow (BSN) conditions, (2) snow (SN) conditions, and (3) freezing fog (FF) and icing (ICNG) conditions, were examined. These conditions are shown in Fig. 2. Note that all these conditions may result in very low Vis conditions. Therefore, in the analysis, their effect on Vis and PR calculations was considered carefully using the laser precipitation monitor (LPM) sensor and wind measurements because their integrated values can be used to assess specific weather events.

Fig. 2
figure 2

Severe weather events at the RND site: April 9 for blowing snow event (a), Jan 28 for heavy snow and low visibility event (b), and Jan 25 for heavy freezing fog event at VOA (c). Pictures of the various snow crystal types taken by the first author during snow storm conditions: splintering mechanism at −6 °C for Jan 17 2012 (d), dendrites and wet particles for Jan 21 (e), graupel for Jan 17 (f), rimed particles; including dendrites, sector crystals, and some aggregates for Jan 21 (g), and light snow with aggregates and stellar crystals for Jan 21 (h)

Full size image

The FD12P measurements may include large uncertainties when air T is between −3 and +6 °C during which FD12P based mixed precipitation information may not be true if the upper level conditions do not represent the mixed phase conditions or there is an inversion. Therefore, particle shape recognition at the low levels may not be representative of the true conditions at higher levels (Gultepe et al., 2012). It should also be emphasized that, at sizes less than 500 μm, disdrometers can include large uncertainties and should be used cautiously. During the project, snow particles were collected over plates covered with a piece of cloth (as in Fig. 2d–h) and their pictures were taken directly using a macro lens in the field (Gultepe et al., 2012). Various snow particle shapes and their distributions are shown in Fig. 2 and for splintered snow crystals, e.g. needles (−6 °C, January 28 2010) in box d, sector plates and stellar crystals, and needles in box e (January 28 2010), graupel with riming in box f (March 4 2010), various mixed snow crystals in box g (March 7 2010), and aggregates and stellar crystals in box h (March 9 2010). These images, collected over approximately 1 min time intervals, show that snow crystals had very variable particle shapes and distributions. Clearly, it is shown that particle shape at the surface may not be related to surface T. It should be noted that accurate PR from particle spectra can be obtained only when both snow and ice particle characteristics such as shape, concentration, cross-section area and fall velocity over entire size range are known.

Visibility and PR are obtained from either the present weather sensors or weighing gauges. Both parameters are strongly related to particle shape and density, and these parameters are also related to T and RHw, and dynamical activity within the clouds and boundary layer. FD12P PR measurements are related to optical characteristics of snow scattering and capacitance measurements of melted snow (SWE). When the melted water amount is used to estimate snow depth, usually a ratio of SWE ~1:10 is used, e.g. 2 mm converts to 20 mm snow physical depth. In this case, Vis versus PRSN for snow is plotted for various particle shapes to emphasize the particle shape effects. The LPM PRSN (mm h−1) value is obtained using the following equation:

$$ {\text{PR}} = 3,600\left( {\frac{\pi }{6}} \right)\sum\limits_{i = 1}^{m} {N_{i} V_{\text{f}} } D^{3} $$
(1)

where V f is the measured particle fall velocity (mm s−1), N i the ice crystal number concentration (cm−3), D the diameter (cm) or maximum size of ice snow particles. V f is determined directly from falling particles through the sampling volume over Δt time period. N i is calculated as:

$$ N_{i} = \frac{{C_{i} /V_{{{\text{f}},i}} }}{{{\text{SA}} \times \Updelta t}} , $$
(2)

where C i is the count of particles at ith bin, SA the sampling area (45.6 cm2), and Δt the sampling time period. The fall velocity (cm s−1) is directly obtained from the sensor algorithm. In order to get true snow water equivalent rate, PRSN at each bin needs to be multiplied with snow density (ρSN = αD β) described by Huang et al. (2010) where α = 0.178 and β = −0.922 for wet snow for D > 1 mm (Brandes et al., 2007). Huang et al. (2010) stated that if N i is reduced about 50 %, PRSN was underestimated at about 34 % with α = 0.21 and β = −0.80. Based on Eq. 2, it is recognized that particle shape effects are included in the PR through the V f. Note that the coefficients can change depending on snow type. It is clear that particle shape recognition from all disdrometers and FD12P are based on retrievals that may not represent true particle shape statistics. This suggests that without accurate particle shape recognition, water equivalent snow precipitation rate (PRSN) values may include large uncertainties. In this respect, FD12P measurements at least use the melted snow amount to estimate PRSN.

4 Results

The five cases related to FF and ICNG, BSN, and SN conditions (Table 2) were selected to accomplish the goals of this study. The Jan 17 2010 case is given in Sect. 5 to show the importance of variability in the model validations. The cases are summarized below.

Table 2 Cases used in the analysis: Jan 17 case includes observations from RND, VOA, and VOL that are used to emphasize variability in the measurements along the Whistler mountain slope and discussed in Discussion section. VOT observation were not available on this day
Full size table

4.1 Case Studies

4.1.1 Jan 25 Case (with FF, BSN, SN, ICNG)

Time series and cross correlation plots of the measurements, and wind effects on PR are shown in Fig. 3. Vis on this day went down to almost 1000 m for several hours (box a). PR from various sensors went up to 3 mm h−1 (box b) but significant differences occurred among them. The PRgeo from the Geonor sensor had significantly lower value compared to Vaisala FD12P (PRfdp) when turbulence and strong winds occurred (e.g. 1800 UTC). The PRtps from the Yankee TPS sensor was usually comparable to PRfdp. Maximum U h and w a reached 6 and 2 m s−1 (box c), respectively, but U h was usually less than 3 m s−1. Note that strong wind and turbulence significantly affected both PRgeo and PRtps measurements because of collection efficiency issues and significant dynamical effects on the plates of the TPS. The Jan 25 case had saturated conditions (RHw >97 %) for the entire day and this resulted in FF (see Rosemount Icing Detector relative tipping frequencies) over cold temperatures (−3 to −5 °C) (box d).

Fig. 3
figure 3

Time series of Vis from FD12P and RHw from HMP212 °C sensor (a), PR from FD12P, Geonor, and TPS (b), horizontal (U h) and vertical wind (w a) from ultrasonic anemometer (c), Rosemount icing detector relative frequency (red line) and T (blue line) (d), PR from FD12P versus PR from TPS (e) and versus PR from Geonor (f), PR from TPS versus horizontal wind (U h) (g), PR from Geonor versus U h (h), and PR from FD12P versus U h (i) for Jan 25 2010

Full size image

The PR comparisons shown on Fig. 3e, f are for PRfdp versus PRtps, and versus PRgeo, respectively. When PR <2 mm h−1; relative scattering is very large for both plots. For an example, when PRtps = 1 mm h−1, PRfdp can be between 0.2 and 2 mm h−1. For PRgeo, this difference can be between 0 and more than 2 mm h−1. Overall, PR varies from case to case. Figure 3g–i show PRtps, PRgeo, and PRfdp versus U h, respectively. PRtps decreases with increasing U h; when U h > 3 m s−1, PR decreases significantly (box g). PRgeo (box h) also decreases in a similar way. PRfdp (box i) measurements appear to be not affected as much as PR from other precipitation sensors.

The spectral precipitation sensors such as disdrometers can also be used for PR and Vis calculations. Figure 4a–d represent V f, Vis, and PRlpm from LPM, and V f from MRR, respectively. Note that the PR estimation obtained using Eq. 1 from LPM measurements does not use counts with sizes <0.1 mm (Fig. 4a). This figure shows V f versus particle diameter with counts on the color bar. Two areas of particle counts (C) on the plots are important to emphasize; first, C > 500 in counts at sizes <0.5 mm in diameter signifies the BSN particles and small snow particles. The green line is the theoretical curve for V f representing cloud liquid droplets/drops. Second, below the curve, usually snow particles exist. Large snow flakes (diameter >3 mm) are seen with V f = 0.5–4.2 m s−1. Note that the stronger the vertical air velocity (w a) snow crystal gets the higher Vf with increasing particle size range.

Fig. 4
figure 4

Particle fall velocity (V f) versus particle width (diameter) (a); the color bar shows the counts (C) of particles (e.g. snow or droplets). The LPM sensor used for observations is shown in the inlet. Time series of Vis is shown in (b). The vertical red colored line indicates where the counts in (a) is taken. The precipitation rate (PRlpm) time series is shown in (c). The MRR particle fall velocities time-height cross section is shown in (d). The results are shown for Jan 25 2010

Full size image

During fog and snow conditions, Vislpm went down below 1 km for almost 50 % of time (Fig. 4b). After 6 pm, Vislpm was below 500 m for about 5 h. The PRlpm was usually less than 5 mm h−1 (Fig. 4c) and found to be larger than PRfdp and PRgeo. Vf obtained from MRR was between 0.3 and 5 m s−1 (Fig. 4d) which was consistent with Vflpm (Fig. 4a).

Using the NASA PVI probe (Newman et al., 2009), time series of snow crystal width as a maximum dimension (W) is shown in Fig. 5a; the color bar shows the spectral number concentration density (N id, m−3 mm−1). Clearly, W < 1 mm, N id becomes about 5 m−3 mm−1. Variability in N id versus diameter (D) is shown in Fig. 5b over 10 min time intervals. For this case, the slope of the spectra did not change much but N i over 10 min intervals changed significantly for a given D. Figure 5c shows the snow crystal shapes from PVI; including mostly dendrites and aggregates with sizes up to 10 mm.

Fig. 5
figure 5

Time series of width (W) of the particles measured with PVI (a). The color bar indicates the particle number concentration density (N id). The spectra obtained over 10 min intervals are shown on (b) and ice crystal images collected during the snow storm at the RND station are shown in (c). The results are shown for Jan 25 2010

Full size image

4.1.2 March 17 Case (FF, Strong Gust, Light SN)

The time series plot in Fig. 6 is obtained in a similar way to that in Fig. 3 given for January 25 case study. Visfdp was less than 1 km for the first 6 h, and then, occasionally, snow showers occurred (box a). PR from various sensors was usually below 0.3 mm h−1 (box b) but significant differences occurred among them. The PRtps did not report the observations because of a malfunction. The PRgeo had significantly larger values as compared to PRfdp because of strong turbulence and winds. Therefore, PRgeo measurements cannot be reliable for PR <0.5 mm h−1. During freezing drizzle events (FDRZ), 3D wind components were not measured accurately before 1200 UTC because the Young ultrasonic sensor was iced up. During non-icing conditions, the U h and w a (vertical air velocity) reached to 8 m s−1 and 1 m s−1 (box c), respectively. Note that wind speeds, similar to the January 25 case, significantly affected PRgeo measurements because of collection efficiencies. In this case, saturated conditions (RHw >95 %) until 1800 UTC resulted in FDRZ (see RID tipping relative frequencies) over temperatures from −7 to −12 °C (box d).

Fig. 6
figure 6

Time series of various measured parameters are shown in box a–d as given in Fig. 3 but for the March 17 case. In this case, icing conditions are indicated by PMWR measurements (e) and shown by black dashed lines for LWC. LWC estimated using GCIP data is shown in (f). In (e), the blue line indicates RND level where LWC is measured by both GCIP and PMWR. RHw and T are shown by colored regions and solid black lines, respectively. At RND level, T was about −10 °C

Full size image

The PMWR retrievals (Bianco et al., 2005) for the March 17 case are shown in Fig. 6e where the color bar is for RHw. The T and LWC are shown with light colored solid lines and light colored dotted lines, respectively. Between 0600 and 1800 UTC, icing over sensors at RND (1856 m) was very severe and LWC ~0.16 g m−3 based on GCIP measurements was observed at the RND level. The 3D wind sensor during icing event was iced up, resulting in a malfunctioning (Fig. 6c). Just before 0600 UTC, icing was very severe; LWC reached up to 0.3 g m−3. The GCIP images (Fig. 6f) suggested that drop diameters were between 50 and 300 μm. Integration of measurements from various sensors on this day suggested that freezing drizzle and precipitation conditions can occur often at mountainous regions.

4.1.3 March 30 Case (Heavy SN Conditions)

Figure 7 is obtained similar to the previous cases. The Visfdp during saturated conditions (RHw >95 %) came usually down to a few hundred meters during very low PRfdp (<0.5 mm h−1) measurements (box a). The PR from various sensors reached 3 mm h−1 between 1600 and 1800 UTC (box b) but there were significant differences between them. The PRtps had significantly larger values (~2 mm h−1) compared to PRfdp and PRgeo values (<0.5 mm h−1) when U h were more than 4 m s−1 (before 1100 UTC). Clearly, TPS sensor was not working accurately on this day when strong winds were occurring. The U h and w a reached to 6 and ±1 m s−1 (box c), respectively, that significantly affected PRgeo and PRtps measurements when PR <0.5 mm h−1. Because of collection efficiency issues of small particles and dynamical effects, both PRtps and PRgeo include large uncertainties. On this day, RHw was usually >95 % for the entire day; FF (see RID tipping relative frequencies) was occurring occasionally over cold temperatures (−5; −8 °C) (box d).

Fig. 7
figure 7

Time series and scatter plots of various measured parameters as given in Fig. 3 but they are for March 30 2010

Full size image

The PR comparisons from Geonor, TPS, and FD12P measurements are shown in Fig. 7e, f. When PRtps <2 mm h−1, PRfdp was usually less than 0.3 mm h−1. For large PR values, PRfdp was larger than PRtps. For an example, when PRtps = 3 mm h−1 at 1300 UTC, PRfdp was usually less than 2 mm h−1. For PRgeo <0.5 mm h−1 (1100 UTC), PRfdp was less than 1 mm h−1; this likely reflects the Geonor’s slower response to low precipitation rate conditions. Figure 7g–i show PRtps, PRgeo, and PRfdp versus U h, respectively. The PRtps (box g) increases with increasing U h which is opposite to the PRgeo trend (box h) in general. The PRgeo decreases (~50 %) with increasing U h when U h > 2 m s−1. The PRtps measurements were much larger than both PRgeo and PRfdp. The PRfdp (box i) shows no wind dependency.

During the extreme weather conditions, thermodynamical parameters such as RHw and T can be affected by both SWRF and U h. Figure 8a–f were obtained for T and RHw along y axis, respectively. The T h45c from HMP45C versus T h212 from HMP45C212, T tps, and T sr50 are shown in Fig. 8a. With increasing T, differences between them increase. If we assume that T h45c is correct and its value is −4 °C, then, T tps ~−3 °C, T sr50 ~−6 °C, and T h212 ~−5.5 °C. Differences decrease when T goes down to −8 °C. Figure 8b shows T versus SWRF. When the sun comes out, T measurements range between −6 and −8 °C (difference ~2 °C). A sudden jump in T value about 2 °C at 160 W m−2 was likely due to direct solar radiation effect on the sensor shield. This difference rises to 5 °C when SWRF was about 500 W m−2. It seems that both T h212 and T sr50 were less responsive to SWRF changes, indicating that their shields were more effective for this case. Figure 8c (same parameters as in Fig. 8b) shows various T measurements versus U h. It seems that T tps, T h45c, and T h212 were significantly affected by increasing wind speed, resulting in lowering T from −2 to −6 °C. T tps measurements were affected severely with increasing wind speed compared to other measurements.

Fig. 8
figure 8

Scatter plots of temperature from HMP45C (T h45c) versus others (T sr50; T tps; and T h45c) (a), T h45c, T h212, and T tps versus SWRF (b), T from all sensors versus horizontal wind speed (U h) (c), RHwh212 versus RHw45c (d), RHwh45c versus short wave radiative flux (SWRF) (e), and RHw45c and RHwh212 versus U h (f) for March 30 2010

Full size image

The plot of RHwh212 versus RHwh45c is shown in Fig. 8d. In this figure, for a given value of RHwh45c ~90 %, RHwh212 changes from 87 to 92 %. A 5 % uncertainty in RHw measurements for this case was found to be reasonable as defined in the company’s manual. Figure 8e shows RHw versus SWRF. For a given SWRF, RHw from both H45C and H45C212 sensors usually varies between 85 and 93 %. An uncertainty in RHw measurements is found to be about 8 %. Increasing U h from 0 to 7 m s−1 also resulted in RHw values ranging between 85 and 95 %. This may be a result of evaporating effects of the wind speed.

4.1.4 April 5 Case (FF, SN)

Results for this case are shown in Figs. 9 and 10. The Visfdp during light SN and FF conditions was about a few hundred meters for PR <0.5 mm h−1 (box a). The PR from the various sensors was less than 3 mm h−1 for entire day (box b); the PRtps increased to 2 mm h−1 while PRfdp had values less than 0.5 mm h−1 between 1800 and 0700 UTC when U h was about 3–4 m s−1. Clearly, TPS measurements were strongly affected by the intense winds. On this day, the U h and w a reached 4 and ±1 m s−1 (box c), respectively. RHw was usually >95 % before 1800 UTC, and decreased to 75 % for very light SN PR. The FF (see tipping RID values) was indicated occasionally with cold T (−2; −5 °C) and saturated RHw (box d).

Fig. 9
figure 9

Time series and scatter plots of various measured parameters as given in Fig. 3 but they are for the April 5 2010 case

Full size image
Fig. 10
figure 10

Scatter plots of temperature from HMP45C (T h45c) versus others (T sr50; Ttps; and Th45c) (a), Th45c, Th212, and Ttps versus SWRF (b), T from all sensors versus horizontal wind speed (Uh) (c), RHwh212 versus RHw45c (d), RHwh45c versus short wave radiative flux (SWRF) (e), and RHw45c and RHwh212 versus U h (f) for April 5 2010

Full size image

The PR inter-comparisons from TPS, Geonor, and FDP measurements are shown in Fig. 9e, f. When PRtps <2 mm h−1, PRfdp was usually less than 0.5 mm h−1. For large PR values, PRfdp was usually less than PRtps but not always. For PRtps = 3 (or 1) mm h−1, PRfdp was usually ~1.8 (>2) mm h−1. For PRgeo <0.5 mm h−1, PRfdp was usually larger than PRgeo; this likely happened due to strong winds and turbulence. Figure 9g–i show PRtps, PRgeo, and PRfdp versus U h, respectively. The PRtps (box g) increases significantly with increasing U h. Also, PRtps was much larger than both PRgeo and PRfd12p during strong wind conditions. Even if a trend were taken from Fig. 9g, PRtps would be still larger than other PR values because it uses an internal algorithm to correct the PR. The PRgeo (box h) decreases with increasing U h. The PRfdp (box i) shows no strong wind dependency, and its maximum values were larger than PRgeo.

Figure 10a–f are obtained for T and RHw, given along y axis, respectively. The T h45c versus T h212, T tps, and T sr50 are shown in Fig. 10a. With increasing T, differences between any two increases. If we assume that T h45c is correct and its value is −4 °C; the values of T tps, T sr50, and T h212 range from −5.0 °C to almost 0 °C. When T goes down to −6 °C, T values from all sensors are in the range of ±0.5 °C of the 1:1 line. Figure 10b shows T versus SWRF. When sun comes out, T measurements ranged between −6 and −5 °C (difference ~1 °C). This difference approached 5–6 °C when SWRF was 600 W m−2. It seems that T h212 and T sr50 were less responsive to SWRF changes. Figure 10c shows T versus U h where T tps and T h45c were significantly affected with increasing wind, resulting in lowering T from 2 to −2 °C, and from −2 to −4 °C, respectively. T tps measurements were strongly affected by changing wind speed compared to other T measurements.

The RHwh212 versus RHwh45c is shown in Fig. 10d. In this figure, for a given RHwh45c ~90 %, RHwh212 varied by ~5 %. Uncertainty in RHw increased to 10 % when it was ~70 %. Also, changing sign of RHw difference between two measurements was likely due to solar radiation effects shown on Fig. 10e where RHw versus SWRF is plotted. For a given SWRF during sunrise, the difference between RHwh45c, and RHwh212 was about 5 %. This uncertainty in RHw during the afternoon was about 10–15 % when SWRF >350 W m−2. The RHw ranged between 70 and 100 % over U h range (Fig. 10f). More data points close to 100 % RHw suggest that saturation conditions exist usually when U h < 2 m s−1. RHw becomes smaller when stronger winds occurred likely resulting in evaporative cooling.

4.2 Variability in the Measurements along the Mountain Slope

Variability in the measurements is studied using the three stations shown in Fig. 1. The RND, VOA, and VOL based FD12P measurements of Vis and PR, T, RHw, and U h are used to show the variability over 500 m along the slope from the VOL to RND. The mean and standard deviation (SD) values, and variability versus height (at the station level) are shown in Table 3 for the entire Feb 2010 period. Mo et al. (2012) and Gultepe and Zhou (2012) also showed that clouds along the mountain slope were not always linearly distributed and mid-mountain clouds were usually present while low and high levels were cloud free. Their work clearly demonstrated the horizontal and vertical variability. These results are discussed in the following section in the context of model verifications.

Table 3 Mean, SD, variability (var), 10 %, 50 %, and 90 % values of FD12P PR, Vis, T, RHw, and Uh over entire Feb 2010 based on 1 min observations. FD12P PR threshold used was 0.05 mm hr−1 to account for light precipitation. Last two rows are for absolute MDM (maximum difference of means) and maximum relative variability (MRV = MDM/max mean) for all cases
Full size table

5 Discussions

5.1 Measurement Uncertainties

Uncertainty in the measurements at the RND site can be significant because of extreme weather conditions such as BSN, cold T, strong winds, and radiation effects. In this section, first we will summarize the comparisons between HMP45C212 and HMP45C sensors (Table 1) from 20 days of randomly selected data for T and RHw. The SWRF and wind effects on T and RHw, comparisons of precipitation and Vis from various sensors, and the impact of w a on PR are also given. In the end, the importance of averaging scales on the measured parameters, and their representativeness for model verification will be discussed.

5.1.1 T and RHw Comparisons

The T and RHw uncertainties for T down to −30 °C are usually given as +1 °C and 5 %, respectively, by the manufacturing companies. But these values can be much higher in extreme weather conditions (e.g. during mixed precipitation at ~0 °C; T < −30 °C; U h > 3 m s−1). Figure 11a, b show T and RHw, respectively, obtained from HMP45C and HMP45C212 sensors as well as T from SR50 and TPS. The T comparisons using a randomly selected 20 days of data from December 1 2009 to April 5 2010 indicated that T differences significantly increase from T = −13 to +1 °C, and T tps can be 2 °C larger than that of T h45c. On the other hand, T h212 and T sr50 can be 2 °C colder than that of T h45c at warm temperatures. Clearly, T uncertainty can be as high as 5 °C if we assume that T h45c measurements are correct. Both H45C and H45C212 were calibrated at Vaisala Inc. The RHwh45c and RHwh212 were plotted against each other in Fig. 11b. If RHwh212 is fixed at 95 %, RHwh45c changes from 80 % up to 97 %, and the net change is ~20 %. The percentiles and mean values are also shown on Fig. 11b. Overall, RHwh212 are overestimated as compared to RHwh45c.

Fig. 11
figure 11

Scatter plots of temperature from HMP45C (T h45c) versus others (T sr50; T tps; and T h45c) (a) where solid line is for 1:1 line, RHwh212 versus RHw45c (b) where yellow line is for 1:1 lines and mean with percentiles are also shown, T h45c, T h212, and T tps versus SWRF (c), RHwh45c versus short wave radiative flux (SWRF) (d) where solid line is a trend line, T from all sensors versus horizontal wind speed (U h) (e) where the solid line shows a possible trend on some data points, and PRgeo versus w a (f) for the randomly selected 20 cases from December 2009 to April 5 2010

Full size image

5.1.2 SWRF and Wind Effects on T and RHw

The SWRF can play an important role on T measurements because of the direct heating effect on the T sensor. Figure 11c shows T from various sensors (y axis) versus SWRF (x axis). It is obvious that SWRF may cause heating at the surfaces, resulting in an increase in T. When SWRF changes from 0 to 200 W m−2, T change is about 1 °C. After 200 W m−2, T values from all sensors became very noisy and with significant scatter being observed. Uncertainty in T was about ±3 °C because of changes in SWRF from 200 to 600 W m−2. All T sensors respond similarly to SWRF changes. Note that early in the morning, the SWRF effect was about 2 °C. Figure 11d shows the effect of SWRF on RHw measurements. In reality, SWRF can affect RHw directly because, increasing SWRF may result in increasing moisture on the ground by melting the snow. Therefore, it is not easy to make firm conclusions on this issue. Figure 11d suggests that some accumulation of data points below 80 % may be related to SWRF heating problem and this is needed to be further analyzed.

Wind effects on T sensors are shown in Fig. 11e. In general, wind speed effects could not be identified clearly on T sensors but both T h212 and T h45c were affected significantly with increasing wind speed when T is around 0 °C. This effect can be as high as 2 °C. Figure 11f shows w a effect on PRgeo and it is discussed in next subsection.

The SWRF and wind effects on T and RHw measurements are well known from many previous studies. A summary of radiation and wind effects on T and RHw is given by Huwald et al. (2009). They stated that over snow, 30 min mean T differences can be as large as 10 °C. Unshielded thermocouples were not affected as much as shielded ones for SWRF. They also stated that T errors decrease with decreasing solar radiative fluxes and increasing wind speed when a shield is used. The present work also suggested that SW heating effects can be as high 5 °C when T is close to 0 °C where diabatic effects can also be important. In our case, increasing wind speed resulted in an increasing error up to 3 °C in T measurements. Although it is not shown here, Huwald et al. (2009) stated that T measurement errors because of reflected SWRF increase much faster than incoming SWRF because of snow albedo effects. They used a sonic anemometer to improve the T measurements because its measurements are independent of SWRF. They also suggested a technique to improve the albedo effects and this will be considered for future studies.

5.1.3 3D Wind Effects on PR

Figure 11f shows the PRgeo against w a from the Young ultrasonic anemometer. This plot clearly shows that PRgeo reaches its maximum value when w a ~ 0 m s−1. Increasing +w a suggests that air parcel goes away from the Geonor inlet, resulting in decreasing PRgeo. Increasing −w a also results in decreasing PRgeo. This result cannot be explained easily because increasing downward motions (see Eq. 1) should increase PR but this is not usually the case. This means that wind effects on precipitation can only be resolved by considering winds in 3D. Thériault et al. (2012) stated that snow gauge collection efficiency is also strongly related to snowflake characteristics. The results from the current work suggest that a detailed flow model around the measuring gauges is needed to better assess the precipitation measurements quality.

5.1.4 PR Comparisons

The PR measurements during the project were collected at 1 min sampling rate. The PR measurements from the three transducers in the Geonor instrument are used to get instantaneous PR values after using 5 min running averages. Horizontal wind effects are not removed because of the lack of reference PR measurements. Also, it is shown that a correction cannot be made using only 2D winds (Fig. 11f). As shown during the case studies, increasing wind speed affects PR by decreasing its value. In ideal conditions, zero wind speed is perfect for accurate PR measurements. Therefore, horizontal wind effects can be taken away from the PR estimation algorithms. Figure 12a and 12b show PRfdp versus PRtps and PRgeo, respectively. Figure 12a shows that scattering increases with decreasing PR, and inconsistency is seen mostly for the low PR values. More scattering in PRfdp versus PRgeo plot in Fig. 12b exists when PR is less than 1 mm h−1. Clearly, when PR >2 mm h−1, scattering of the data points decreases. Note that for a given value of PRfdp, PRgeo and PRtps can be two times more or less than PRfdp.

Fig. 12
figure 12

Scatter plot of PRfdp versus PRtps (a) and versus PRgeo (b), Vis from sentry sensor (Vissen) versus Vis from FD12P (Visfdp) (c) where red circles are for human observations. The red solid line is a fit for all data points and yellow line is for 1:1 line. Scatter plot Visfdp versus PRfdp for various snow types is shown in (d) where red solid line is for the fit to all data points (equation is given on the plot). The mean (red filled circles) and percentiles (red thin lines) as well as SD bars (green lines) are also shown on (d) for all 20 cases. The labels of LSN, MSN, HSN, LIP, MIP, SG, ICE, MEAN, 10 and 90 % represent, respectively, light snow, medium snow, heavy snow, light ice pellet, medium ice pellet, snow grain, ice crystals, mean, 10 and 90 % percentiles. Time series of width (W) of the particles measured with a PVI are shown in (e) where the color bar indicates the particle number concentration density log(N id). The ice crystal images, showing the variability of the shapes, collected during the snow storm at the RND station are shown in (f) for February 16 2010

Full size image

5.2 Uncertainty and Variability in Vis

Visibility measurements are usually based on scattering and absorption properties of the particles in a small air volume nearby the transmitting and receiving sensors. Unfortunately, they don’t represent the larger scales and their measurements are usually extended over tens of kilometers that may not represent variability. Variability (Var) here is defined as the ratio of SD to the mean (Var = SD/mean) for a given time and space scale. Especially over mountainous regions, real Vis may include much larger variability compared to those over smooth surfaces. Figure 6 previously showed that Vis may change quickly over a few minutes. Figure 12c shows Vissen versus Visfdp plot where a fit to data suggests that Visfdp is 1.5 times larger than Vissen. Also, increasing Vis results in larger scatter of the observations. Time series of Visfdp shown earlier suggested that Vis during snow and fog events may change drastically. Figure 9a as an example shows that Vis at about 1200 UTC drastically decreased from tens of kilometers down to a few hundreds of meters in less than a few minutes, suggesting that variability of Vis can be significant over very short time periods. Time series of PR measurements from different instruments are shown for various cases in Figs. 4, 6, 7, and 9. These showed that PR also changes quickly over periods of a few minutes, indicating that NWP models should represent microphysical processes over time periods less than a few minutes.

5.2.1 Vis versus Snow Type and Intensity at RND

The FD12P’s particle shape recognition technique is a function of T, optical response to falling snow amount, and capacitance during the melting process. During the project, T between −3 and +6 °C was assigned to mixed phase precipitation based on the manufacturer's specifications. In that T range, as indicated earlier, T measurements may include large uncertainties that can affect the specification of snow type. Figure 12d shows the Vis versus PR obtained from the FD12P sensor as a function of snow type. The LSN, MSN, and HSN types are easily described using PR limits but other types of precipitation (see Fig. 12d) cannot be distinguished. Data points seen between Vis = 100 and 300 m for PR <1 mm h−1 may be related to light snow or ice crystals (e.g. snow grains, light ice pellets) but the same snow types are also seen between Vis = 2 and 10 km, suggesting that particle concentration was not determined accurately or snow type was not related to surface air T. A fit to all the data with 10 and 90 % lines is provided on the figure and its equation is obtained as

$$ {\text{Vis}} [{\rm km}] = 2.432{\text{PR}}^{ - 0.3684} [{\rm mm\, hr^{-1}}] - 1.16 .$$
(3)

This equation (with correlation coefficient R = −0.65) can be used for Vis estimation for snow PR but its percentile values should also be considered to get a probabilistic estimation of Vis (Gultepe and Milbrandt, 2010). Because of variability in Vis for a given PR value, we should not compare the fit values to mean values simulated with a forecasting model.

5.2.2 Variability in Snow Particle Type and Intensity at RND

Snow precipitation rate is directly related to snow particle types and their fall speed (which is also related to density). As shown in Fig. 2, snow crystal types can change significantly over 1 min time periods. Using the PVI sensor (Newman et al., 2009), time series of snow crystal width (W) are shown in Fig. 12e; the color bar shows the spectral concentration density (N id) of snow crystals. Clearly, for W < 1 mm, N id becomes about 5 m−3 mm−1. Figure 12f shows the snow crystal shapes from PVI; including columns/needles, dendrites, sector plates, and aggregates with sizes up to 5 mm. Equation 1 suggests that V f which is function of particle shape and density plays a significant role in the PR calculation.

5.2.3 Variability along the Slope: Daily 1-min Observations

The observations collected during SNOW-V10 created a unique data set to study variability over the Whistler Mountain slope. The Jan 17 2010 case is chosen as an example to show the variability along the Whistler Mountain. Figure 13 shows two pictures representing the weather conditions at 1000 LST (snowing 1800 UTC) and 1600 LST (fogy). The five stations were located along the slope but only three of them are used for discussions here. The observations of Vis and PR (from FD12P), T, RHw, and wind speed/direction were collected at each minute. Figure 14 shows the time series of Vis, PR, T, RHw, and wind speed from RND, VOA, and VOL stations to indicate variability over a distance about 500 m. Box (a) shows that Vis ranges from 100 m to 10 km depending on the stations at 1800 UTC. During snow precipitation (box b), Vis differences can be more than 1 km (1200 UTC). Snow PRfdp at 0200 UTC ranges from 0.5 mm h−1 (VOL) up to 3 mm h−1 (RND). The U h ranges from 1 m s−1 (VOL) to 7 m s−1 (VOA) at 1100 UTC (box c). During saturation (foggy) conditions, RHw becomes very close to 100 % at all stations except during clear air conditions (from 0000 to 0400 UTC), where differences between stations are about 40 % (box d). Absolute T differences become more than 3 °C during clear air conditions (box e) and ~1 °C when snowing at the all stations.

Fig. 13
figure 13

Variability in the weather conditions on Jan 17 2010: snowing at the RND site (1,853 m) at 1000 LST (a) and fogy/cloudy below the RND site at the VOA and VOL sites on 1600 LST (b). The fog/cloud filled the valley below the RND site on this day

Full size image
Fig. 14
figure 14

Time series of Vis (a), PR (b) from FD12P, U h from 3D ultrasonic anemometer (c), RHwh45c (d), and Th45c (e) for RND, VOA, and VOL sites to show the vertical variability over along a 500 m slope for Jan 17 2010

Full size image

5.2.4 Variability along the Slope: Monthly Averaged 1-min Observations

Variability in the measurements is studied using the data collected at the three stations shown in Fig. 1. The RND, VOA, and VOL based FD12P measurements of Vis and PR are used to show the variability over ~500 m height in the vertical (700 m along the slope) from the station VOL to RND. Table 3 shows the monthly mean, SD, and variability (SD/mean) of the 1 min observations in February 2010 from the individual stations along the slope. In fact, differences between two stations represent the scale of about 200 m. The results show that MRV (maximum relative variability) for PR, Vis, T, RHw, and U h over February 2010 can be up to 43 %, 73 %, 140 %, 10 %, and 76 % (Table 3), respectively, along the 700 m slope. This result can contribute to a significant error in comparisons with NWP models when a forecasting model has a resolution of about 1 km in the horizontal and 100 m in the vertical. Bailey et al. (2012) showed that both RND and VOA sites share the same GEM-REG-15 km and GEM-LAM-1 km model points; therefore, comparisons become difficult unless models resolved these scales. Note that individual variability of the parameters at each station can be up to 100 % per minute except for RHw (which is up to 30 %). The RHw measurements can be highly susceptible to measurement errors as described earlier. Further analysis of the observations on this issue will be made later using extended data sets from the stations along the slope (Fig. 1).

6 Conclusions

The main conclusions obtained from this work are important for nowcasting PR, Vis, T, RH, and U h using forecast models, and for their verifications using statistical methods and observations. The meteorological parameters can change quickly over short time and space scales as described. The models should have enough time and space resolutions to reduce the scale dependent uncertainties. For example, vertical (horizontal) space resolutions of the forecasting models should be better than 50 m. Otherwise, because of variability, comparisons using point observations may not be accurate, resulting in large uncertainty issues. The main points obtained from this work can be summarized as:

  1. 1.

    Time and space variability obtained from surface point observations can be significant and they may not be compared directly with NWP simulations.

  2. 2.

    Models need better subgrid scale representation of the processes to resolve smaller scales (e.g. <100 m).

  3. 3.

    Uncertainties in RHw and T because of wind and SWRF variations can be as high as 20 % and 2–3 °C, respectively.

  4. 4.

    Vertical air velocity (w a) can increase or decrease PR depending on U h direction, suggesting that 3D winds need to be used in improving snow PR measurements.

  5. 5.

    TPS measurements are strongly affected by the horizontal wind speed, and resulted in up to 100 % error in PR when a collection efficiency equation used in the embedded algorithm.

  6. 6.

    FD12P measurements were not affected by the wind as much as TPS and Geonor measurements, and it was the best sensor for mountain applications to record the precipitation rate (especially for rates <0.5 mm h−1). PR from Geonor and TPS was usually found to be strongly affected by wind speed.

  7. 7.

    Snow type changed drastically over 1 min time periods in many conditions, and the FD12P was a good sensor for distinguishing snow fall rate as light, medium, and heavy, but was not a good indicator for individual ice crystal shape recognition.

  8. 8.

    LPM sensor was overestimating PR but it was good for BSN detection and snow type.

  9. 9.

    Uncertainty in Vis from both FD12P and Sentry can be significant for both low and high Vis conditions, and also from human based observations.

  10. 10.

    Variability over 700 m scale along a mountain slope can be up to 100 % for many meteorological parameters; therefore, observations and models should have similar time and space scale resolutions for comparisons.

  11. 11.

    Based on observational uncertainties, additional methods and efficient project setups should be developed for model validations.

This work suggested that PR, Vis, T, and RHw measurements need to be improved over the complex terrain; also, other than point measurements, some areal based techniques should be developed for model validations. Future work on this issue using SNOW-V10 observations will be considered.