Journal of the Meteorological Society of Japan, Vol. 91, No. 6, pp. 763̶774, 2013 DOI:10.2151/jmsj.2013-603
763
Classification and Quantification of Snow Based on Spatial Variability of Radar Reflectivity Sanghun LIM Korea Institute of Construction Technology, Ilsan, Korea Colorado State University, Fort Collins, CO, USA
Dmitri MOISSEEV University of Helsinki, Helsinki, Finland
Venkatachalam CHANDRASEKAR Colorado State University, Fort Collins, CO, USA University of Helsinki, Helsinki, Finland
and Dong-Ryul LEE Korea Institute of Construction Technology, Ilsan, Korea (Manuscript received 23 November 2012, in final form 16 July 2013)
Abstract In this study, a classification methodology of snow particle types, i.e., crystals, aggregates, rimed snow, and graupel, by using spatial variability of the equivalent radar reflectivity factor is proposed. The methodology is formulated on the basis of the analysis of vertically pointing Doppler radar, scanning dual-polarization weather radar, and supporting surface observations. It is argued that by using the proposed snow-type identification methodology, it is possible to guide the choice of the particular parameters of power law relations of equivalent radar reflectivity factor̶liquid equivalent snowfall rate. The validity of the classification results are demonstrated by comparing the classification output to Vaisala WXT observations, which can be used to detect presence of high-density particles in snow. The performance of the proposed quantitative snowfall estimation algorithm is illustrated using an example of the data collected from the C-band operational Helsinki Vantaa radar and ground instruments (Vaisala PWD-11, Pluvio). Keywords weather radar; classification; quantification; snow
1.
Introduction Weather radar-based quantitative precipitation Corresponding author: Sanghun Lim, Water Resources Division, Korea Institute of Construction Technology, 283, Goyangdae-Ro, Ilsanseo-Gu, Goyang-Si, Gyeonggi-Do, 411-712, Korea E-mail:
[email protected] ©2013, Meteorological Society of Japan
estimation of snowfall is notoriously difficult. Radar observations of snow depend on size, orientation, and density of the snow particles. Variability of these physical properties is one of the major uncertainty sources in quantitative snowfall estimation with radar (Mitchell et al. 1990). The snowfall rate estimation using radar measurements has been studied for decades. Conventional radar-based snowfall estimation methods have used power law relations between
764
Journal of the Meteorological Society of Japan
the equivalent radar reflectivity (Z) and liquid equivalent precipitation rate (S) (Marshall and Gunn 1952; Sekhon and Srivastava 1970; Smith 1984; Fujiyoshi et al. 1990; Rasmussen et al. 2003; Matrosov et al. 2009; Huang et al. 2010; Zhang et al. 2011). These methods generally show wide variability owing to physical properties and behavior of snow. The variability of parameters in the Z̶S relations can cause a factor of two-order difference in snowfall estimation (Matrosov et al. 2009). The classification of winter precipitation according to hydrometeor classes such as aggregates, graupel, and rimed particles can give guidance for the refinement of snowfall estimation techniques. Classification of snowfall from weather radar measurements is difficult. Despite the successful application of fuzzylogic-based hydrometeor classification algorithms (Vivekanandan et al. 1999; Liu and Chandrasekar 2000; Straka et al. 2000; Lim et al. 2005; Ryzhkov et al. 2005; Keränen et al. 2007; Park et al. 2009; Chandrasekar et al. 2013) to summer precipitation, limited progress has been made in winter precipitation. As shown by experience, polarimetric radar signatures are not very different for many types of ice particles, i.e., aggregates and rimed ice particles. Because of the abovementioned reasons, the ability of radar observations to improve quantitative radar estimation of snowfall has been very limited (Brandes et al. 2007). In this paper, a snow-type identification technique that uses two-dimensional decision function is proposed. The decision function is constructed by a combination of reflectivity value and spatial variability of reflectivity. A classification-based snowfall rate estimation that can reduce the effect of variability owing to physical properties and behavior of snow is also suggested. The proposed algorithm is evaluated using the C-band operational Helsinki Vantaa radar (VAN) and ground instruments such as Vaisala PWD11 and Pluvio. Snowfall measurement errors from ground instruments can make it difficult to evaluate the Z̶S relations. Here the ground instrument error is not considered for evaluation. The paper is organized as follows: Section 2 introduces the measurement setup for this study. Section 3 describes a snow-type identification method using a combination of spatial variability and reflectivity value. Section 4 presents a snowfall rate estimation technique that is guided by snowfall types, whereas the proposed snowfall estimation algorithm is evaluated in Section 5. Finally, important results are summarized in Section 6.
2.
Vol. 91, No. 6
Measurement setup
In this study, measurements from the Helsinki Testbed (Koskinen et al. 2011) were used. These measurements include observations of the Finnish Meteorological Institute Vantaa dual-polarization Cband weather radar (Saltikoff and Nevvonen 2011), FMI surface sensors, University of Helsinki, Kumpula, and transportable Doppler weather radars. The Kumpula radar is a C-band polarimetric weather radar located atop the Department of Physics building (60° 12.26ʼ N, 24° 57.78ʼ E). The radar is positioned 59 m above the mean sea level and 30 m above the ground level. The transportable Doppler C-band weather radar used in this study is stationed 32 km to the north (azimuth 11.8° ) of the Kumpula radar in Järvenpää (60° 29.07ʼ N, 25° 4.91ʼ E). The radar is positioned 53 m above the mean sea level and approximately 3 m above the ground level. There is a clear line of sight between the radars. The layout of the instruments is shown in Fig. 1. The surface sensors used in this study are weighing precipitation gauges, snow depth sensors, present weather sensors (Vaisala PWD 11), and impact sensors (Vaisala WXT 510). In general, WXTs are not considered as snow measurement sensors. However, we believe that these sensors, because of their measurement principle, can provide an additional source of information on prevailing snow types. In WXT510, precipitation measurement is based on Vaisala RAINCAP sensor, which detects the acoustic impact of individual raindrops (Salmi and Ikonen 2005). The signals resulting from the impact are proportional to the diameter, and thus the volume of the drops, and therefore the signal of each drop can be directly converted to accumulated precipitation. The sensor can also distinguish hailstones from raindrops. During snowfall, this sensor should not produce signals since snowflakes are not sufficiently hard to cause a detectable impact. However, if a signal is detected, it can be attributed to dense snow particles, such as graupel or snow grains. It was observed, as will be shown later, that in some cases impacts are sufficiently strong to be detected by the standard WXT algorithms and identified as caused by hail. In several such cases, we could trace those observations to ground reports of large graupel particles. 3.
Snow-type identification
3.1 Setting the classification criteria To identify precipitation events with different snow types, measurements of snow depth and accumulated
December 2013
S. LIM et al.
765
Fig. 1. Layout of the measurement setup (Map. Google Maps. Google). The WXT location is marked by a star. PWD-11 and Pluvio sensors are located next to the radar in the Kumpula campus of University of Helsinki. The circles show the locations of precipitation measurement stations, measurements from which were used to create Table 1 and Fig. 2. The red balloons indicate locations of the radars used in this study. The yellow persons correspond to graupel observations on January 4, 2010 (S. Gr: Small graupel, L. Gr: Large graupel).
precipitation liquid equivalent were used. Snow growth processes influence the density of freshly fallen snow: aggregates are less dense than particles grown mainly by riming. Power et al. (1964) have found that −3 unrimed particles will have densities below 0.1 g cm , while rimed particles will have larger densities. Therefore, the snow ratio, ratio of snow depth change to liquid equivalent (Jodson and Doesken 2000), of less than 10: 1 would indicate rimed ice particles. Snowfall accumulations and snow depth changes from Finnish Meteorological Institute weather stations located within 60 km from the Järvenpää radar were used (Table 1, Fig. 1). The observations are conducted daily at 6 UTC and represent 24 h accumulations. The snow ratios calculated from these measurements are used to
estimate a predominant snow-type, i.e., rimed or unrimed, for each snowfall event. Snow depth was mainly measured manually, but at several stations, acoustic snow depth sensors were installed before March 2009. The liquid water equivalents (LWE) were manually measured by volumetric gauges. The measurements are presented in Table 1 and Fig. 2. The snow ratios, slopes of lines that fit the data and pass through the origin, are calculated using iteratively reweighted least-square linear regression with a bisquare weighting function (Holland and Welsch 1977). The method is less sensitive to outliers, and therefore is well suited for this analysis. It can be seen that observations of March 9, 2009, indicate the presence of dense ice particles. This inference is in line
766
Journal of the Meteorological Society of Japan
Vol. 91, No. 6
Table 1. Observed snow depths, liquid water equivalents, and calculated snow ratios. The observations were carried out at stations that were within 60 km from the location of the transportable radar. Superscript a denotes automatic snow depth observations. Station Name (Location) Helsinki Kaisaniemi Helsinki-Vantaa Vihti Hiiskula Espoo Nupuri Porvoo Harabacka Nurmijärvi Observatorio Hyvinkää Hyvinkäänkylä Mäntsälä Hirvihaara Hausjärvi Lavinto
Distance to Järvenpää (km) 35.5 19 31 37 31 23 18 18 38
Snow ratio
Fig. 2. Observed LWE and snow depths for the storms that occurred on March 9, 2009, and March 3, 2009. It can be seen that the event on March 9, 2009 is characterized by smaller snow ratios that indicates the presence of dense snow particles.
with witness reports. To further verify those observations and pinpoint the exact time of riming occurrence, the analysis of Doppler power spectra recorded by the transportable Doppler radar were used. During Doppler spectra recording, the Kumpula radar performed range̶height
03/03/09 LWE, mm
03/09/09
Depth, cm a
3.8 2.7 2.3 2.2 2.4 2.9 3.2 2.4 2.6
4 a 3 2 3 a 2 a 3 3 0 3 10.4
LWE, mm
Depth, cm
8.6 6.1 10 8.3 5.9 8.3 10.4 7.4 9.3
7 a 7 7 8 a 5 a 6 9 7 9
a
8.7
indicator (RHI) scans over the transportable radar. These scans were repeated every 2 min. Given the measurement geometry, the Kumpula radar beam is approximately 500 m wide at the location of the transportable radar and the lowest beam height is 350 m. The range resolution of the transportable radar is 100 m. In Fig. 3, the Kumpula radar RHI and corresponding transportable radar observations collected on March 3 and March 9, 2009, are shown. The aggregates of dendrites and densely rimed dendrites have fall velocities in the range of 0.7̶1.5 −1 m s (Locatelli and Hobbs 1974). Rimed aggregates −1 and lump graupel have velocities that exceed 1.5 m s (Locatelli and Hobbs 1974). Therefore, a transition −1 from 1 m s fall velocity to fall velocities exceeding −1 1.5̶2 m s indicates heavy riming (Locatelli and Hobbs 1974; Mosimann 1995). This signature can be seen in Fig. 3B3). From these observations, we can conclude that during observation, the prevailing particle type was rimed aggregates. Unfortunately, vertical Doppler measurements do not provide a clear distinction between aggregation and riming processes for all cases. It is difficult to separate aggregates from densely rimed crystals (small graupel) by using only the fall velocity information. For example, a comparison of Fig. 3A3 and Fig. 3C3 does not allow for the discrimination of prevailing particle types observed during those events based on Doppler velocity measurements alone. Surface observations at the radar site on 0900 UTC 3 March have shown that the prevailing precipitation type was small graupel of approximately 3 mm in size. A casual comparison of measured reflectivity fields
December 2013
S. LIM et al.
767
Fig. 3. Radar observations collected on (A) March 3, 2009, (B)̶(C) March 9, 2009. The two left figures indicate the Kumpula radar RHI reflectivity and differential reflectivity observations. The dashed lines show the location of the Järvepää radar. The right figures indicate the Doppler spectrum collected by the transportable radar at the Järvenpää measurement site.
in Fig. 3 shows that overall equivalent reflectivity factor exhibits stronger spatial variability for the riming cases, i.e., on March 9, 2009, than for the case where riming was not detected (March 3, 2009). We argue that the observed difference in the continuity of radar reflectivity fields is related to physical processes occurring and can be used as an indicator of riming. Wüest et al. (2000) have reported a correlation between variability of ice-particle vertical velocities and degree of riming (Mosimann et al. 1994). They have discussed two possible explanations of the observed correlation. The first explanation is the influence of turbulence on observed vertical velocities
of ice particles. Pinsky and Khain (1998) have demonstrated that turbulence increases collision probability between ice particles and water droplets, and therefore makes riming more efficient. An increase in riming efficiency would result in higher degrees of riming and the resulting superposition of turbulent motion and terminal fall velocity of ice particles would result in higher variance of the mean vertical velocity observed by radar. The other explanation is that riming will result in increased variation in terminal fall velocities of ice particles caused by changes in density. Wüest et al. (2000) have shown that even though this effect is
768
Journal of the Meteorological Society of Japan
Fig. 4. Variograms of effective radar reflectivity factor calculated for three events is shown. Based on ground and vertically pointing Doppler observations, March 3 is identified as the event where aggregation dominated the snow growth process. On March 9, rimed aggregates were observed on the ground at 10̶11 UTC and small graupel was observed at 09̶10 UTC.
present, it is not sufficiently strong to explain the observations. We believe that there is a third explanation that can explain the observations of Wüest et al. (2000) and the observed variability of the reflectivity field. Particle sorting due to different fall velocities of snowflakes of different densities could result in an increased variability of the Doppler velocities observed by a vertically pointing radar. Since terminal fall velocity as well as radar reflectivity depend on particle density, these two observations are related and should exhibit a similar behavior, i.e., increased variability of fall velocity corresponds to an increased variability of reflectivity field. To quantify the differences in the reflectivity field continuity, we have used a variogram-based analysis. Germann and Joss (2001) have used variograms to assess reflectivity field continuity and to study the impact of the extrapolation of radar derived precipitation intensity from the radar beam height to the ground. In a similar manner, we estimate a variogram, 2γ(h), of the observed reflectivity field as 2γ(h) =
2 1 ∑ Z(R)−Z(R + h) , N(h) ie N(h)
(1)
where Z is the radar reflectivity factor expressed in dB, R is the horizontal distance, h is the lag distance
Vol. 91, No. 6
between reflectivity measurements, and N(h) is the number of pairs used for averaging. The number of pairs depends on the lag distance. In this study, we have calculated variograms from RHI observations presented in Fig. 3. Prior to the calculations, the RHI measurements were converted to the Cartesian coordinate system, with vertical and horizontal spatial resolutions of 100 m. The variograms were calculated for each height for the lowest 1 km and a horizontal extent of 30 km. The resulting variograms were averaged; the final product is presented in Fig. 4. One can see that the variograms for these cases are very distinct. The rimed snowfall events exhibit higher spatial variability, especially the small graupel event. The aggregation event on March 3, 2009, on the other hand, shows a very homogeneous reflectivity field. 3.2 Classification methodology As shown above, the spatial variability of reflectivity factor exhibits some skill in discriminating between different snow types. Even though, aggregates and rimed snow cases have similar reflectivity values, spatial structures of reflectivity fields are rather different. The reflectivity field observed during riming exhibits more spatial variability than that observed during aggregation. In addition, graupel is also characterized by larger reflectivity values than other snow types (Straka et al. 2000). Therefore, by using a combination of spatial variability and reflectivity value, a snow-type classification can be established. To simplify the applicability of the method, a variogram analysis is substituted by the normalized standard deviation of reflectivity. Fig. 5 demonstrates the decision boundary for the identification between various snow types and decision boundary values are listed at Table 2. The dots and asterisks indicate data collected by the operational C-band Helsinki Vantaa radar on January 4 and 7, 2010, respectively. Note that the decision boundary of snow type can depend on the characteristics of radar system such as frequency and signal fluctuation. Therefore, the boundary may need adjustment for different radar systems. The boundary in Fig. 5 is adjusted by comparing the Helsinki Vantaa radar and ground instruments such as WXT. The normalized standard deviation of Z (NSD) is expressed as NSD(Zm,n) = M−1 N−1 2 1 2 2 ∑ i =− M−1 ∑ N−1 (Zm+i,n+j−Z m,n) j =− M×N 2 2 ; (2a) Zm,n
December 2013
769
S. LIM et al.
Table 2. Snow-type identification decision boundary in reflectivity versus normalized standard deviation of reflectivity. Decision boundary
Type Crystal Aggregate Rimed snow High-density ice (Graupel)
Zh(dBZ) 10 10 20 20
Zh < 10 < Zh < 35 < Zh < 20 < Zh < 35 < Zh < 35 35 < Zh
NSD(Zh) All NSD(Zh) < 0.85 NSD(Zh) > 0.85 0.85 < NSD(Zh) < 1.0 NSD(Zh) > 1.0 All
Fig. 5. Snow-type identification decision boundary in reflectivity versus normalized standard deviation of reflectivity. Asterisks indicate data on January 4, 2010, and dots on January 7, 2010, from the operational Helsinki Vantaa radar.
M−1 N−1 1 2 ∑ i =− M−1 ∑ 2 N−1 Zm+i,n+j , j =− M×N 2 2
Zm,n =
(2b)
where m and n indicate the azimuth and range of the gate, respectively, and M and N represent the number of gates at azimuth and range, respectively. By analyzing Figs. 3B and 3C, we can see that most core cell sizes are approximately 2̶3 km in depth. From this analysis, in this study nine gates with 500 m resolution and five azimuths are used (M = 4, N = 2). By using the normalized value, the effect of reflectivity can be reduced. For example, if the values of Z and NSD(Z) are 20 dBZ and 0.7, it is classified as aggregate. However, if the values of Z and NSD(Z) are 20 dBZ and 0.9, it is classified as rimed snow. 4.
Z = αS ,
The equivalent reflectivity factor for snowflakes can be written as (Smith 1984) 4
−5
2
2
2
Dmax
Dmin
〈σ b〉N(D)dD, (3)
where m is the complex refractive index of water and λ is wavelength. The liquid equivalent snowfall rate, S can be expressed as
−1
S = ρw
Dmax
Dmin
relation, and vt(D) is the fall velocity̶size relation. Reflectivity and snowfall relations are expressed in terms of a power law:
Snowfall rate estimation
Z = λ π (m + 2)／(m −1)
Fig. 6. Block diagram of the classification-based snowfall estimation system.
g(D)vt(D)N(D)dD,
(4)
where ρw is the density of water, g(D) mass̶size
(5)
where Z is the equivalent radar reflectivity factor in 6 −3 mm m and S is the snowfall rate (expressed as the −1 liquid equivalent per unit volume) in mmh . The coefficients α and β depend on environmental factors (temperature, humidity) and microphysical properties (size, fall velocity, phase, and density). Although there is significant variability of the coefficients of Z̶S relations, recent studies showed that the typical values of coefficients α and β were in the range 30̶140 and 1.2̶1.55 for X-band (Matrosov et al. 2009) and 100̶ 300 and 1.1̶2.0 for C-band (Huang et al. 2010), respectively.
770
Journal of the Meteorological Society of Japan
Fig. 7.
Snowflake bulk density as a function of size.
Table 3.
Parameters in Z
Type Crystal Aggregate Rimed snow High-density ice (Graupel)
̶
Vol. 91, No. 6
Fig. 8. Relation of reflectivity and liquid equivalent snowfall rate.
β
αS relations. Z = αS
β
α
β
170 70 140 202
1.5 1.3 1.6 1.6
The schematic diagram of the classification-based radar snowfall estimation system is demonstrated in Fig. 6. First, using the NSD versus Z space, the snow types are identified. Next the parameters of Z̶S relations are selected on the basis of the classification results. To understand the variability of parameters corresponding to different snow types, theoretical simulation has been conducted (Matrosov et al. 2009). For aggregates, relations of Locatelli and Hobbs (1974) and Heymsfield et al. (2004) were used, whereas Brandes at al. (2007) relation was used for rimed snow and Hogan et al. (2000) for crystals, respectively. Unlike raindrops, the relation between density and size of snowflakes varies significantly. Figure 7 depicts snowflake bulk density as a function of size for different relations, whereas relation of reflectivity and liquid equivalent snowfall rate is shown in Fig. 8. The coefficients of Z̶S relations used are shown in Table 3. The ranges of Z̶S power law coefficients agree reasonably well with results of Matrosov et al. (2009) and Huang et al. (2010). For high-density ice particles, the parameters were adopted from Gray and Male
Fig. 9. Comparison of classification from the Helsinki Vantaa radar and WXT located on Hietaniemi, Helsinki, for January 4, 2010 case. (a) Zh and normalized standard deviation of Zh, (b) classification results by the proposed method from the Vantaa radar, (c) temperature from WXT, (d) classification from WXT.
December 2013
S. LIM et al.
771
Fig. 10. Analysis of the January 4, 2010, event: (a) Reflectivity (blue solid) and normalized standard deviation of reflectivity (olive dotted) from the Vantaa radar at the University of Helsinki, Kumpula, (b) classification result (＊: High-density ice, ◇: Rimed snow, + Aggregate, o Crystal), and (c) comparison of snowfall rate from the proposed method (blue solid), PWD-11 (red dotted), and Pluvio (black solid-dotted). PPI plots of radar reflectivity at (d) 1900 UT 4 January 2010 and (e) 1940 UT 4 January 2010. Red circles indicate the location of the University of Helsinki, Kumpula.
(1981) and Rasmussen et al. (2003). By subsequently applying the selected parameters, the snowfall rate is estimated. 5.
Data Analysis
The proposed snowfall estimation algorithm is tested by the operational C-band Helsinki Vantaa radar and ground instruments (Vaisala PWD-11 and Pluvio, which are located at the University of Helsinki, Kumpula; WXT located at Hietaniemi cemetery). OTT-Pluvio is an all-weather precipitation gauge that uses superior weight-based technology to measure rainfall, snow or hail. PWD is the Vaisala present weather sensor that measures the intensity and accumulation of precipitation. The proposed classification and snow estimation
methodology was applied to two snowstorm events. One occurred on January 4, 2010. The event was characterized by several graupel reports, locations of which are shown in Fig. 1, and an exceptionally high radar reflectivity of 50 dBZ that is linked to 1 cm graupel reports. Another event used in this study is a large-scale snowstorm that arrived to Helsinki from the south-easterly direction at around 0400 UTC 7 January 2010. During the day, surface winds were −1 approximately 3 m s or less, which facilitated the comparison of radar and precipitation gauge observations. Figure 9 shows the comparison between classification results by Vantaa radar observation and WXT on the January 4, 2010, case. Fig. 9a indicates Z and NSD from Vantaa radar at the WXT site, whereas Fig. 9b is
772
Journal of the Meteorological Society of Japan
Vol. 91, No. 6
Fig. 11. Analysis of the January 7, 2010, event: (a) Reflectivity (blue solid) and normalized standard deviation of reflectivity (olive dotted) from the Vantaa radar at the University of Helsinki, Kumpula, (b) classification result (＊: High-density ice, ◇: Rimed snow, + Aggregate, o Crystal), and (c) comparison of snowfall rate from the proposed method (blue solid), PWD-11 (red dotted), and Pluvio (black solid-dotted). PPI plots of radar reflectivity at (e) 0810 UT 7 January 2010 and (f) 1030 UT 7 January 2010. Red circles indicate the location of the University of Helsinki, Kumpula.
the classification result. To evaluate the classification procedure, the classification results are compared to the WXT reports. Since snow is not sufficiently hard to cause a detectable impact, only heavily rimed particle such as graupel are registered by WXT. For this particular event, there were several ground reports of large; up to 1 cm; and small, of approximately 3 mm, graupel on the ground. These reports correspond well with the hail and rain classes of WXT. Therefore, if during winter at temperatures below 0° C, WXT reports hail, the report most probably corresponds to large graupel, while the report of rain corresponds to small graupel. By using this information and analyzing WXT observations in Fig. 9c, one can conclude that WXT reports large graupel at 1845 UTC and small graupel after that. The air temperature varyied from −4 to
−1° C during this event. From the comparison of results presented in Fig. 9, we can see that the classification results using spatial variability of radar reflectivity match well with the WXT reports, especially at high-density ice regions. It should be noted that rimed aggregates are not sufficiently hard to be detected by WXT, probably hence there were no reports before 1840 UTC and after 1910 UTC. The radar-based snowfall precipitation estimation, based on the proposed technique, has been compared with PWD-11 and Pluvio. The results are shown in Fig. 10. The blue and olive lines in Fig. 10a represent measured Z and NSD values, respectively. Data from Finnish Meteorological Vantaa radar is used in this study. The results of snow classification are presented in Fig. 10b. Figure 10c shows the comparison of
December 2013
snowfall rate from the PWD-11, Pluvio, and the radarbased estimate. Figures 10d and 10e show the plan position indicator (PPI) plots of radar reflectivity at 1900 UTC and 1940 UTC. Red circles indicate the location of the University of Helsinki, Kumpula where the surface observations were collected. Overall, we can conclude that radar-based snowfall estimate compares well with the surface observations. To check the performance of the method in different conditions, the method was used on the observations collected on January 7, 2010. Figure 11 is similar to Fig. 10, except that the data is from January 7, 2010. The PPI plots presented in Figs. 11d and 11e show radar reflectivity values observed at 0810 UTC and 1030 UTC, respectively. The classification indicates that the prevailing snow type during this day was snow aggregates. The radar-based precipitation estimates compare well with the PWD values and are somewhat higher than those of OTT Pluvio gauge. It is not easy to tell which sensor reports are more accurate in this situation. However, it is not uncommon to see differences in snowfall rate observations between different precipitation sensors. 6.
773
S. LIM et al.
Summary
A methodology for the classification and quantification of snowfall has been presented. The proposed classification methodology divides snow types into crystals, aggregates, rimed snow, and high-density snow (graupel). Compared with other classification schemes presented in literature, the proposed method is different in that it uses the spatial continuity of radar reflectivity field as well as reflectivity observations. It is shown that the spatial continuity of reflectivity field can be linked to the presence of riming, i.e., observed reflectivity shows more spatial variability in case of riming. This variability is possibly linked to particle sorting. The proposed method uses snow-type identification to guide the choice of the particular parameters of power law relations of equivalent radar reflectivity factor̶liquid equivalent snowfall rate. This technique can reduce the variation of quantitative snowfall estimation due to difference in physical properties of snow particles. The snow-type identification method compared favorably with WXT observations, which are sensitive to high-density particles (such as graupel). Furthermore, the proposed algorithm was applied to Cband radar measurements and resulting snowfall rates were compared to surface sensor observations. The results show that the selective choice of power law parameters corresponding to snow types can provide
accurate snowfall estimation. Acknowledgments This research is supported by the Vaisala Inc.ʼs wintertime weather precipitation program (Science and Nowcasting Olympic Weather for Vancouver 2010), by the Academy of Finland (grant 263333), and by a grant from Strategic Research Project (Development of Flood Warning and Snowfall Estimation Platform using Hydrological Radars) funded by the Korea Institute of Construction Technology. The authors acknowledge Minda Le for assistance with theoretical snowfall rate calculations. References Brandes, E. A., K. Ikeda, G. Zhang, M. Schönhuber, and R. M. Rasmussen, 2007: A statistical and physical description of hydrometeor distributions in Colorado snowstorms using a video disdrometer. J. Appl. Meteor. Climatol., 46, 634̶650. Chandrasekar, V., R. Keränen, S. Lim, and D. Moisseev, 2013: Recent advances in classification of observations from dual polarization weather radars. Atmos. Res., 119, 97̶ 111. Fujiyoshi, Y., T. Endoh, T. Yamada, K. Tsuboki, Y. Tachibana, and G. Wakahama, 1990: Determination of a Z̶R relationship for snowfall using a radar and high sensitivity snow gauges. J. Appl. Meteor., 29, 147̶152. Gray, D. M., and D. H. Male, 1981: Handbook of Snow: Principles, Processes, Management and Use. Pergamon Press, 776 pp. Germann, U., and J. Joss, 2001: Variograms of radar reflectivity to describe the spatial continuity of Alpine precipitation. J. Appl. Meteor., 40, 1042̶1059. Heymsfield, A. J., A. Bansemer, C. G. Schmitt, C. Twohy, and M. R. Poellet, 2004: Effective ice particle densities derived from aircraft data. J. Atmos. Sci., 61, 982̶1003. Hogan, R. J., A. J. Illingworth, and H. Sauvageot, 2000: Measuring crystal size in cirrus using 35- and 84-GHz radars. J. Atmos. Oceanic Technol., 17, 27̶37. Holland, P. W., and R. E. Welsch, 1977: Robust regression using iteratively reweighted least-squares. Communications in Statistics: Theory and Methods, 6, 813̶827. Huang, G.-J., V. N. Bringi, R. Cifelli, D. Hudak, and W. A. Petersen, 2010: A methodology to derive radar reflectivity̶liquid equivalent snow rate relations using C-band radar and a 2D video disdrometer. J. Atmos. Oceanic Technol., 27, 637̶651. Jodson, A., and M. Doesken, 2000: Density of freshly fallen snow in the central Rocky Mountains. Bull. Amer. Meteor. Soc., 81, 1577̶1587. Keränen, R., E. Saltikoff, V. Chandrasekar, S. Lim, J. Holmes, and J. Selzler, 2007: Real-time hydrometeor classification for the operational forecasting environment. Preprints, 33rd Conf. on Radar Meteorology, Amer.
774
Journal of the Meteorological Society of Japan
Meteor. Soc., Cairns, Australia, 11. Koskinen, J. T., and co-authors, 2011: The Helsinki testbed: A mesoscale measurement, research, and service platform. Bull. Amer. Meteor. Soc., 92, 325̶342. Lim, S., V. Chandrasekar, and V. N. Bringi, 2005: Hydrometeor classification system using dual-polarization radar measurements: Model improvements and in-situ verification. IEEE Trans. Geosci. Remote Sens., 43, 792̶801. Liu, H., and V. Chandrasekar, 2000: Classification of hydrometeors based on polarimetric radar measurements: development of fuzzy logic and neuro-fuzzy systems, and in situ verification. J. Atmos. Oceanic Technol., 17, 140̶164. Locatelli, J. D., and P. V. Hobbs, 1974: Fall speeds and masses of solid precipitation particles. J. Geophys. Res., 79, 2185̶2197. Marshall, J. S., and K. L. S. Gunn, 1952: Measurement of snow parameters by radar. J. Meteor., 9, 322̶327. Matrosov, S. Y., C. Campbell, D. Kingsmill, and E. Sukovich, 2009: Assessing snowfall rates from X-band radar reflectivity measurements. J. Atmos. Oceanic Technol., 26, 2324̶2339. Mitchell, D. L., R. Zhang, and R. L. Pitter, 1990: Massdimensional relationships for ice particles and the influence of riming on snowfall rates. J. Appl. Meteor., 29, 153̶163. Mosimann, L., E. Weingartner, and A. Waldvogel, 1994: An analysis of accreted drop sizes and mass on rimed snow crystals. J. Atmos. Sci., 51, 1548̶1558. Mosimann, L., 1995: An improvemed method for determining the degree of snow crystal riming by vertical Doppler radar. Atmos. Res., 37, 305̶323. Park, H., A. V. Ryzhkov, D. S. Zrnić, and K. Kim, 2009: The hydrometeor classification algorithm for the polarimetric WSR-88D: Description and application to an MCS. Wea. Forecasting, 24, 730̶748. Pinsky, M. B., and A. P. Khain, 1998: Some effects of cloud turbulence on water-ice and ice-ice collisions. Atmos. Res., 47̶48, 69̶86. Power, B., P. Summers, and J. DʼAvignon, 1964: Snow crystal forms and riming effects as related to snowfall density
Vol. 91, No. 6
and general storm conditions. J. Atmos. Sci., 21, 300̶ 305. Rasmussen, R., M. Dixon, S. Vasiloff, F. Hage, S. Knight, J. Vivekanandan, and M. Xu, 2003: Snow nowcasting using a real-time correlation of radar reflectivity with snow gauge accumulation. J. Appl. Meteor., 42, 20̶36. Ryzhkov, A. V., T. J. Schuur, B. W. Burgess, P. L. Heinselman, S. Giangrande, and D. S. Zrnić, 2005: The joint polarization experiment polarimetric rainfall measurements and hydrometeor classification. Bull. Amer. Meteor. Soc., 86, 809̶824. Salmi, A., and J. Ikonen, 2005: Piezoelectric precipitation sensor from Vaisala. WMO Technical Conference on Meteorological and Environmental Instruments and Methods of Obsevation (TECO 2005), Bucharest, Romania, 4̶7. Saltikoff, E., and L. Nevvonen, 2011: First experiences of the operational use of a dual-polarisation weather radar in Finland. Meteor. Z., 20, 323̶333. Sekhon, R. S., and R. C. Srivastava, 1970: Snow size spectra and radar reflectivity. J. Atmos. Sci., 27, 299̶307. Smith, P. L., 1984: Equivalent radar reflectivity factors for snow and ice particles. J. Climate Appl. Meteor., 23, 1258̶1260. Straka, J. M., D. S. Zrnić, and A. V. Ryzhkov, 2000: Bulk hydrometeor classification and quantification using polarimetric radar data: synthesis of relations. J. Appl. Meteor., 39, 1341̶1372. Vivekanandan J., D. S. Zrnić, S. M. Ellis, R. Oye, A. Ryzhkov, and J. M. Straka, 1999: Cloud microphysics retrieval using S-band dual-polarization radar measurements. Bull. Amer. Meteor. Soc., 80, 381̶388. Wüest, M., W. Schmid, and J. Joss, 2000: Coupling between riming and the dynamics of precipitating clouds. Proccedings of 13th International Conf. Clouds and Precipitation, 421̶424. Zhang, G., S. Luchs, A. Ryzhkov, M. Xue, L. Ryzhkova, and Q. Cao, 2011: Winter precipitation microphysics characterized by polarimetric radar and video disdrometer observations in central Oklahoma, J. Appl. Meteor. Climatol., 50, 1558̶1570.
