删除或更新信息,请邮件至freekaoyan#163.com(#换成@)

Evaluation of Unified Model Microphysics in High-resolution NWP Simulations Using Polarimetric Radar

本站小编 Free考研考试/2022-01-02

Marcus JOHNSON1,2,
Youngsun JUNG1,2,
Daniel DAWSON3,
Timothy SUPINIE1,2,
Ming XUE1,2,
Jongsook PARK4,
Yong-Hee LEE4

1.Center for Analysis and Prediction of Storms (CAPS), University of Oklahoma, Norman, OK 73072, USA
2.School of Meteorology, University of Oklahoma, Norman, OK 73072, USA
3.Purdue University, West Lafayette, IN 47907, USA
4.Korea Meteorological Administration, Seoul 07062, Korea
Manuscript received: 2017-07-20
Manuscript revised: 2017-12-20
Manuscript accepted: 2017-12-27
Abstract:The UK Met Office Unified Model (UM) is employed by many weather forecasting agencies around the globe. This model is designed to run across spatial and time scales and known to produce skillful predictions for large-scale weather systems. However, the model has only recently begun running operationally at horizontal grid spacings of ∼1.5 km [e.g., at the UK Met Office and the Korea Meteorological Administration (KMA)]. As its microphysics scheme was originally designed and tuned for large-scale precipitation systems, we investigate the performance of UM microphysics to determine potential inherent biases or weaknesses. Two rainfall cases from the KMA forecasting system are considered in this study: a Changma (quasi-stationary) front, and Typhoon Sanba (2012). The UM output is compared to polarimetric radar observations in terms of simulated polarimetric radar variables. Results show that the UM generally underpredicts median reflectivity in stratiform rain, producing high reflectivity cores and precipitation gaps between them. This is partially due to the diagnostic rain intercept parameter formulation used in the one-moment microphysics scheme. Model drop size is generally both under- and overpredicted compared to observations. UM frozen hydrometeors favor generic ice (crystals and snow) rather than graupel, which is reasonable for Changma and typhoon cases. The model performed best with the typhoon case in terms of simulated precipitation coverage.
Keywords: Unified Model,
microphysics,
polarimetric radar,
radar simulator,
numerical weather prediction
摘要:全球许多天气预报机构都在使用英国气象局天气与气候统一模式(Met Office Unified Model, 简称MetUM), 它以有效预测大尺度天气系统而闻名, 能够模拟各种时空尺度事件. 然而最近才有预报机构(如: 英国气象局, 韩国气象局)开展该模式在1.5公里水平网格中的业务运行工作. MetUM的微物理方案最早是针对大尺度降水系统设计和调试的, 为评估其在模拟对流性降水过程中固有的潜在偏差或缺点, 本文使用韩国气象局预报系统提供的一次韩国梅雨锋降水和台风“三巴”(2012)降水资料开展数值试验. 通过比较偏振雷达观测资料和MetUM模拟输出的一系列偏振变量值, 发现: MetUM常常低估层状云降水平均反射率, 模拟产生反射率虚假高值中心和破碎的雨带. 这一问题应部分归因于单参数微物理方案中雨粒子截距这一变量采用的诊断公式. 本研究的其他发现有: 模式估计的水滴大小通常都比观测值偏大或偏小; 模式更易模拟出冰晶和雪等常规冰粒子, 较少生成霰粒子, 这对于梅雨锋和台风降水过程模拟是合理的; 从模拟的降水范围看, MetUM在台风个例中效果最好.
关键词:天气与气候统一模式(MetUM),
微物理方案,
偏振雷达,
雷达模拟器,
数值天气预报





--> --> -->
1. Introduction
As computing power consistently increases, operational centers run numerical weather prediction (NWP) models with convective-scale (≤ 4 km; Weisman et al., 1997) grid spacing (e.g., Tang et al., 2013; Goldenberg et al., 2015; Kim, 2015; Park et al., 2015a; Ballard et al., 2016). Model microphysical processes become significant at this resolution and drive the evolution of the precipitating system. In order to gain understanding of microphysics complexity, behavior, and potential biases, recent research has focused on microphysics scheme performance (e.g., Cintineo et al., 2014; McMillen and Steenburgh, 2015; Morrison et al., 2015; Johnson et al., 2016) and sensitivity (e.g., Morrison and Milbrandt, 2011; Morrison et al., 2012; Van Weverberg et al., 2012). Specifically, polarimetric radar data is a powerful tool for comparing model output with observations (e.g., Jung et al., 2012; Dawson et al., 2013; Brown et al., 2016; Putnam et al., 2017) because radar polarimetry can provide observational microphysics information (i.e., differential reflectivity Z DR is related to hydrometeor shape).
Microphysics schemes in NWP models typically represent particle size distributions (PSDs) using a gamma distribution: \begin{equation} N(D)=N_0D^{\mu}\exp(-\Lambda D) ,\ \ (1) \end{equation} where N0, μ and \(\Lambda\) are the intercept, shape and slope parameters, respectively, and D is the particle diameter. For one-moment (1M) schemes in which one PSD moment (typically mixing ratio) is predicted, \(\Lambda\) typically varies freely while N0 is usually fixed (e.g., Lin et al., 1983; Rutledge and Hobbs, 1983; Tao and Simpson, 1993; Hong and Lim, 2006) or diagnosed (i.e., as a function of temperature or mixing ratio [see (Hong and Lim, 2006), (Thompson et al., 2008), and the Thompson graupel intercept parameter in (Morrison et al., 2015)]. Diagnosing N0 in 1M schemes allows N0 to vary, but not independently of its tied parameter (i.e., mass). Diagnostic N0 has shown improvement over fixed N0 (Zhang et al., 2008; Wainwright et al., 2014; Pan et al., 2016), while improvement was not as clear in other studies (Straka et al., 2005; Milbrandt and Yau, 2006; Van Weverberg et al., 2011). As previous studies have focused mainly on convective events, additional studies regarding the tuning of N0 to large-scale events would be helpful for regional forecasts.
Two-moment (2M) schemes add an additional predicted variable (typically number concentration) that allows N0 to vary independently of mass (e.g., Milbrandt and Yau, 2005; Morrison et al., 2009; Mansell et al., 2010). Several studies have compared 1M microphysics scheme performance with multi-moment (2M or higher) schemes. While 1M schemes are theoretically computationally faster than multi-moment schemes with additional predicted moments, one of the largest deterrents of employing 1M schemes are their inability to replicate size sorting (e.g., Dawson et al., 2010, 2014; Jung et al., 2010; Johnson et al., 2016), a fundamental microphysical process where larger hydrometeors fall faster than smaller ones. This mechanism is presumably more important in deep convective storms within directional wind shear environments, where a strong updraft allows larger hydrometeor growth (and subsequently, size difference) and transport sedimentation spatially distributes them.
It is desirable that microphysics schemes retain consistent performance over various precipitation modes. However, some microphysics schemes have been developed and tested for (and therefore, potentially biased toward) large-scale (e.g., Wilson and Ballard, 1999; Hong and Lim, 2006; Thompson et al., 2008) or storm-scale precipitation systems (e.g., Milbrandt and Yau, 2005; Morrison et al., 2009; Mansell et al., 2010). As horizontal grid resolution decreases, the microphysical processes and parameterizations in the scheme do not necessarily change. Ideally, microphysics performance remains unchanged with varying grid scale. In reality, microphysics schemes have the potential to grow model error when they run outside scales for which they are tuned. For example, microphysics schemes tuned for large-scale precipitation systems might favor small rain drops through aggressive breakup and could poorly simulate large raindrops often seen in supercell storms. While some studies have examined model performance sensitivity to horizontal grid resolution (e.g., Bryan and Morrison, 2012; Potvin and Flora, 2015; Verrelle et al., 2015), more rigorous study across several microphysics schemes is needed to provide guidance for microphysics scheme improvement by modelers and help users choose the best microphysics/resolution combination for their modeling purposes.
While some of the studies previously mentioned have shown the superior performance of multi-moment microphysics schemes over 1M schemes, 1M schemes are still popular for operational models primarily because of their low computational cost. Further, some studies provide optimism that 1M scheme performance could be improved. The Thompson microphysics scheme, whose snow processes use 1M parameterization (i.e., snow mixing ratio is predicted) with a combination of two PSDs, predicts accumulated snowfall closer to observed totals than other 1M schemes and performs similarly to the Morrison 2M scheme in (Liu et al., 2011). Further, (Bryan and Morrison, 2012) demonstrated that increasing grid resolution noticeably increased 1M performance in terms of surface precipitation and storm evolution for a simulated squall line, although the 2M scheme still outperformed the 1M scheme.
This current study examines the performance of the Unified Model (UM) microphysics scheme using two distinct cases: a Changma front, and Typhoon Sanba (2012) over the Korean peninsula. Polarimetric variables are computed from UM output using the radar simulator based on Jung et al. (2008, 2010) and compared to observations from the Biseul-san radar, which is an S-band polarimetric radar. It is one of six radars operated by the Ministry of Land, Infrastructure and Transport to improve flood forecasts. More details on the radar can be found in (Park et al., 2015b). This study aims to identify any biases and weaknesses present so that modelers can improve the scheme and help researchers and forecasters to interpret forecasts given microphysical biases. The paper is organized as follows: section 2 details the UM configuration and the polarimetric radar data simulator; section 3 compares the structure of polarimetric observations with UM simulated polarimetric variables; section 4 analyzes the polarimetric distributions; section 5 expands the comparisons to frozen hydrometeors; and section 6 provides conclusions.

2. Overview of UM microphysics and the polarimetric radar data simulator
2
2.1. UM microphysics
--> The UM microphysics scheme is rooted in (Wilson and Ballard, 1999) [itself derived from (Rutledge and Hobbs, 1983)], although modifications continue to update and improve the scheme (e.g., Wilkinson et al., 2013). The UM microphysics scheme is unique in that it contains many parameterization choices. Here, we list relevant parameterizations in Korea Meteorological Administration (KMA) model simulations. At the most basic level, UM microphysics contains three hydrometeor categories: cloud water, rain, and ice. The default UM microphysics scheme is unique in that cloud ice and snow are contained in the ice category, although an option exists to separate these into individual categories. In our simulations, the ice category is represented by a single generic ice distribution (Field et al., 2005, 2007). Further, the UM microphysics configuration employed in this study contains an additional graupel category, an essential hydrometeor category for deep convection. Briefly, the UM is centered at 37.57°N, 126.97°E over the Korean peninsula on 744× 928 grid points with a predominant horizontal grid spacing ? x of 1.5 km, and includes 21 grid points of varying grid zones with ? x increasing to 4 km at the lateral boundaries. The model contains 70 terrain-following vertical levels up to 39 km. The model is integrated using a semi-implicit, semi-Lagrangian method, with time step ? t=50 s and model output every hour. Forecasts are initialized at 0000 UTC using three-dimensional variational data assimilation every 6 h and integrated up to 36 h. More details can be found in Table 1.
In our configuration, the PSDs are exponential for rain, gamma for graupel, and a linear combination of exponential and gamma distributions for ice (Table 2). As the hydrometeor categories use 1M parameterization, N0 must be parameterized in the PSDs. Rain and graupel N0 is a power-law function of the slope parameter \(\Lambda\): \begin{equation} N_0=n_{{\rm a}x}\Lambda^{n_{{\rm b}x}} ,\ \ (2) \end{equation} where n ax and n bx are constants for hydrometeor x. The rain N0 relationship is from (Abel and Boutle, 2012). Ice N0 is a function of the second and third moments of the ice distribution M2 and M3 (Field et al., 2007). Rain and graupel densities are set to 1000 and 500 kg m-3, respectively. Ice assumes a power-law mass relationship that does not assume a spherical shape, resulting in varying bulk density.

2
2.2. Polarimetric radar simulator
--> The polarimetric radar simulator employed in this study to compute polarimetric variables is well-documented in the literature (Jung et al., 2008, 2010; Dawson et al., 2014). Briefly, the simulator constructs PSDs from model output consistent with UM microphysical assumptions. However, we acknowledge that the generic ice category is treated as snow (and will be referred to as snow for the remainder of the paper) with a constant density of 100 kg m-3 for scattering purposes. This should not be a large source of error, as high-density ice crystals are quite small in the PSD used in the scheme, and thus contribute little to polarimetric calculations. The water fraction is diagnosed using a linear relationship as a function of temperature. As temperature increases from -2.5°C to 2.5°C for snow and from -5°C to 0°C for graupel, the water fraction increases from 0 to 0.8 for snow and from 0 to 0.4 for graupel. These temperature and water fraction ranges are chosen to tune the simulated melting layer to match observations in terms of depth and intensity. Finally, scattering amplitudes are retrieved from precomputed T-matrix tables that vary with particle diameter and water fraction. Polarimetric variables reflectivity at horizontal polarization (Z H), differential reflectivity (Z DR), specific differential phase (K DP), and the correlation coefficient HV), are computed from these scattering amplitudes (Zhang et al., 2001; Jung et al., 2010). For more details on the polarimetric simulator, please refer to the above mentioned publications.

3. Polarimetric structure for the model and observations
In this section, the polarimetric simulator is applied to two distinct weather cases——a Changma front, and Typhoon Sanba (2012)——and compared to observed polarimetric variables measured by the Biseul-san radar. For forecast verification, it is typical to compare observations and model forecasts valid at the same time. However, position error that grows fast during the forward model integration makes it difficult to compare precipitation systems in the same domain, especially when the verification data have limited coverage (i.e., the Biseul-san radar utilized in this study only covers 150 km). As radar beam height increases with increasing distance, precipitation systems that are being compared should be at a similar distance from the radar. In addition, polarimetric measurements exhibit large dynamic ranges depending on widely varying PSDs within the precipitation system. It would be more practical to compare the variables in the radar echoes that show similar characteristics in both observations and model simulations. Therefore, we compare precipitation systems showing similar qualitative reflectivity structure (i.e., rainbands) and distance from the radar in the observed and model reflectivity to allow for temporal errors.
Because the prevailing geography of Korea is mountainous, many radars are placed on the tops of mountains and the lowest elevation angles are below 0° (e.g., -0.5°). The Biseul-san radar used in this study is located on the top of Biseul Mountain (1085 m above mean sea level). We analyze model and observational plots of polarimetric and microphysical variables at the 0.5° elevation angle to ensure the primary precipitating hydrometeor is rain, eliminating substantial ground clutter contamination or beam blockage. The beam height of this elevation angle (taking into account radar altitude and Earth's curvature) within the radar coverage is below 4000 m.

2
3.1. Changma front (10 July 2012)
--> At 0000 UTC 10 July 2012, an east-west Changma front attached to a low-pressure system was positioned west of the Korean peninsula (not shown). While the low gradually detached from the Changma front and moved northeast, the Changma front settled just over the southern part of the Korean peninsula near the coastline at 0000 UTC 11 July 2012 (Fig. 1a). The green line shows the 20°C isodrosotherm, and the Changma front is often located south of this line. Figure 1b depicts the accumulated rainfall from AWS rain gauges on the peninsula over 12 h, interpolated linearly over the domain. Accumulated rainfall exceeds 100 mm sparingly in the middle of the Korean southern coast, which is where the Changma front is positioned. Further north, accumulated rainfall continually decreases and even falls below 5 mm, notably in the central and eastern parts of South Korea. These regions possibly received little rainfall because the stationary front stayed to the south and the low-pressure system moved to the northeast.
Figure1. The (a) surface chart at 0000 UTC 11 July 2012 and (b) rain accumulation (units: mm) from AWS gauge data over 12 h ending at 2330 UTC 10 July 2012 for the Changma front case.


Figure2. The (a, b) Z H (units: dBZ), (c, d) Z DR (units: dB), (e, f) K DP (units: ° km-1) and (g, h) ρ HV at 2130 UTC 10 July 2012 in observations (left-hand panels) and the model (right-hand panels) for the Changma front valid as a 3-h forecast at 2100 UTC 10 July 2012 at the 0.5° elevation angle.


Figure 2 shows the radar observations of the Changma front at 2130 UTC 10 July 2012, and model output for the simulated Changma front, analyzed at the 3-h forecast valid at 2100 UTC. Z DR and K DP contain 0.3 dB and 0.02° km-1 thresholds, respectively, to suppress noise. Overall, Z H shows widespread precipitation up to 55 dBZ, with rather smooth gradients (Fig. 2a). Z DR (Fig. 2c) is generally below 2 dB, indicating the main precipitating hydrometeors are likely small to medium-sized. K DP (Fig. 2e) is noisy over the radar coverage domain, except for high reflectivity cores in the south. Small areas of K DP exceed 1° km-1 in the main precipitation cores (Z H>35 dBZ), indicating heavy precipitation. Observed ρ HV is very high (Fig. 2g) over the entirety of the precipitation area, except for sparse reduced values in low signal-to-noise ratio regions. The overall large ρ HV values indicate the precipitating system at low levels is dominated by the presence of pure rain.
The storm structure of the simulated Changma front (Fig. 2b) is more detached compared to observations. Isolated high Z H cores with narrow stratiform rain are scattered within the radar coverage. Model underprediction of precipitation coverage is evident, as observed precipitation coverage (defined as Z H≥ 5 dBZ) is 75% of the radar coverage area, compared to the model's 42% (Fig. 2b). From the microphysics perspective, the diagnostic intercept parameter is likely one of the main reasons for the fragmented storm organization. As rain mass decreases toward the storm edges, drop size decreases rapidly and inversely proportional to the mixing ratio, and becomes increasingly prone to evaporation which is proportional to the rain intercept parameter. Subsequent timesteps with reduced rain mass result in an increased diagnosed rain N0, which further increases evaporation and decreases drop size. Further, other studies using the UM with grid spacing smaller than a few km have identified the model's struggle to adequately resolve convection, both in size and intensity (e.g., Tang et al., 2013). There are other studies that have tried to attribute overly intense cores and a lack of precipitation coverage to the local non-conservation associated with semi-Lagrangian advection and/or deficits in the subgrid turbulence scheme (e.g., Hanley et al., 2015; Nicol et al., 2015). While those studies revealed sensitivity, the main issue of overly intense updrafts and too little light rain remained.
The model also produces higher Z DR values (>2 dB; Fig. 2d) in the precipitation cores compared to observations. Given the 1M nature of UM microphysics, D 0r is monotonically related to rain mass. Thus, increasing reflectivity (increasing rain rate) corresponds to increasing median drop size, resulting in larger Z DR in the reflectivity cores. Outside of the cores, Z DR is frequently below 1 dB, which is somewhat similar to observations. Similar to observed K DP, significant K DP values (Fig. 2f) are found only in the precipitation cores with large drops. Although the spatial coverage of simulated K DP is significantly underpredicted similar to Z H and Z DR, the range of simulated K DP (0.25-3° km-1) is very similar to that of observations. The model ρ HV (Fig. 2h) is generally near 1 over the analyzed domain, indicating the primary precipitating hydrometeor at low levels is rain, matching observations. The sparse model ρ HV reduction is caused by interpolation error of model polarimetric variables to the radar elevation angle.

2
3.2. Typhoon Sanba (17 September 2012)
--> Typhoon Sanba (2012) made landfall on the southern coastline of the Korean peninsula. Surface charts at 0000 UTC 17 September 2012 reveal an intense low of 955 hPa with maximum winds reaching 148 km h-1 as the typhoon's center was positioned just south of the Korean coastline (Fig. 3a). The UM forecast is able to simulate the large-scale structure of Typhoon Sanba (2012) reasonably well, capturing the location of the heavy precipitation in the eyewall and rainbands (Fig. 3b). The highest 12-h accumulated rainfall over the peninsula is concentrated over the south (Fig. 3c), where the eyewall made landfall. The maximum substantially exceeds that in the Changma front case. A few stations reported over 400 mm of rain accumulation, with one exceeding 500 mm. The typhoon weakened as it made landfall but heavy rain continued to cause substantial damage while the typhoon moved northeast.
The observations considered for this case are at 0100 UTC 17 September 2012. The eye of the typhoon is just south of the coastline, with the typhoon's rainbands in the north covering much of the radar coverage area (Fig. 4a). Z H reveals widespread moderate to heavy precipitation. Further from the eyewall, precipitation becomes lighter at the edge of the radar coverage. It appears drops are relatively larger close to the typhoon's eye, where high Z DR is found (Fig. 4c). Given the convective nature of the rainbands, drops have more potential to grow before falling out of the updrafts. Significant observed K DP is found in the inner part of the eyewall, with maximum K DP exceeding 2° km-1 (Fig. 4e). The ρ HV is generally near 1 over the entire radar coverage area, which makes sense given the warm-rain processes that dominate typhoons (Fig. 4g).
Figure3. The (a) surface chart at 0000 UTC 17 September 2012, (b) simulated reflectivity Z H for the UM 6-h forecast valid at 0000 UTC 17 September 2012 and z=~ 166 m, and (c) rain accumulation (units: mm) from AWS gauge data over 12 h ending at 0300 UTC 17 September 2012 for Typhoon Sanba (2012).


Figure4. The (a, b) Z H (units: dBZ), (c, d) Z DR (units: dB), (e, f) K DP (units: ° km-1) and (g, h) ρ HV at 0100 UTC 17 September 2012 in observations (left-hand panels) and the model (right-hand panels) for Typhoon Sanba (2012) valid as a 7-h forecast at 0100 UTC 17 September 2012 at the 0.5° elevation angle.


The isolated convective cores seem less problematic for the UM simulated typhoon compared to the previous case, suggesting that the performance of the microphysics scheme may depend on the scale of the precipitation system (Fig. 4b). In fact, the microphysics scheme was originally developed for large-scale systems. The simulated high reflectivity near the coastline is consistent with observations that show the northern part of the eyewall/rainbands. Still, precipitation coverage is underpredicted, as the model precipitation encompasses 76% of the radar coverage area compared to the observational 90%, possibly due to the N0 relationship and semi-Lagrangian/subgrid turbulence dynamic reasons previously mentioned. While larger observed Z DR appears in the inner side of the eyewall, where drops can grow large in strong convection, large model Z DR coincides with high reflectivity throughout the domain (Fig. 4d). Enhanced model K DP is also found further away from the eye, collocating with high reflectivity and Z DR (larger drops; Fig. 4f) because of their monotonic relationships with q r. The larger K DP found near the edges of the radar domain is due to snow. The model ρ HV is generally near 1, and only a minor reduction is found in sparse areas near the edges of the domain due to snow and graupel, and interpolation error (Fig. 4h). Thus, the primary model hydrometeor is pure rain, which matches observations and the expected hydrometeor behavior of a typhoon given its dominant warm-rain processes.

4. Polarimetric distributions
2
4.1. Changma front
--> In order to evaluate the ability of UM microphysics to capture the natural variation of PSDs, observed, model, and model rank histograms of polarimetric variables are constructed for the two test cases (Figs. 5 and 6). All of the histograms use orange, green, blue, purple and violet shaded areas to denote the 0.2 percentiles in the distribution. The observation and model rank plots denote the observation 0.2 percentiles (to facilitate comparison between the model and observations), while the model histograms denote the model 0.2 percentiles. The observations and UM contain different resolutions and thus a different number of data points; therefore, a raw comparison is not valid. The model rank histograms are constructed by sorting observed data, finding locations of percentiles spaced at 0.1, and then distributing the model data into these observed percentiles. By analyzing how the model data fills the observed percentiles, a direct comparison between the model and observed distributions is possible, and model biases are readily apparent. The black solid lines in the plots represent model uniform distributions (hereafter UDs), which would occur if the model data perfectly matched the observed distribution. The 5-dBZ threshold in all plots, and additional 0.3-dB and 0.25° km-1 thresholds, are included in the histograms to filter noisy data. Additionally, observations are removed when the ρ HV is less than 0.9, because precipitation is mostly pure rain at the 0.5° elevation angle. It is important to mention that comparisons among cases should be taken with caution because of the small coverage of the radar used in this study. It only captures part of the precipitation, and therefore the results may not be representative of the entire storm system.
Figure5. Observed (left-hand panels), model (middle panels), and model rank (right-hand panels) of (a-c) Z (units: dBZ), (d-f) Z DR (units: dB) and (g-i) K DP (units: ° km-1) for the Changma front case. Percentiles at 0.2 intervals are denoted by color shifts in the plots. The observation and model rank plots display observation percentiles, while the model plots show model percentiles. The black lines in the model rank column denote a theoretical UD in which model data are distributed in the same manner as observations.


Observation and model reflectivity histograms are binned at 1 dBZ (Figs. 5a and b). The model Changma front reflectivity distribution tends to contain a larger frequency of smaller reflectivity values (<15 dBZ) than observations, while missing the larger peak of observed reflectivity (~30 dBZ). This is also reflected in the model rank histogram, as the 0.0-0.5 percentile bins exceed the model UD and dip below this line between the 0.5-0.9 percentiles where the observation reflectivity peak is centered (Fig. 5c). The model rank histogram slightly rises above the UD line in the largest percentile bin, but there is a clear underprediction of overall reflectivity.
Observational Z DR produces a smooth distribution at 0.05-dB bin intervals, albeit with several missing bins (Fig. 5d). This is due to observational Z DR rounding, where observations are stored at 0.06-0.07-dB intervals. As a result, the missing bins repeat for bins ending at 0.25 dB intervals. One notable difference between the two histograms is that the model Z DR frequency continually decreases with increasing Z DR (Fig. 5e), while observed Z DR peaks in the 0.2-0.4 percentile area. Similar to reflectivity, the model is producing more small Z DR values compared to observations. This is reflected in the model rank Z DR histogram, where the 0.0-0.2 percentiles exceed the UD (Fig. 5f). The repeating low-high step shape of the histogram is thought to be due to the observational Z DR rounding previously mentioned. The model rank histogram also reveals a longer model Z DR tail, as the 0.9-1.0 percentile exceeds the UD. Much of the middle Z DR percentiles (0.2-0.8) are below the UD line, lending to the relatively higher number of low/high model Z DR frequency. The observation and model K DP histograms are similar in that small K DP dominates the frequency (Figs. 5g and h). For this reason, the histograms are displayed logarithmically. Compared to the model, observations contain a much longer K DP tail. This is reflected in the model rank histogram, where the 0.6-1.0 percentiles are below the UD (Fig. 5i). On the other hand, the 0.0-0.4 percentiles are at, or exceed, the UD. Combined with the Z and Z DR histograms, large model drop size is primarily responsible for high Z and Z DR percentiles, while the rainfall amount may be underestimated.

2
4.2. Typhoon Sanba ( 2012)
--> For Typhoon Sanba (2012), both the observation and model reflectivity distributions seem to be negatively skewed Gaussian (Figs. 6a and b), and have narrower distributions compared to the Changma case. However, the model Z contains higher frequencies of smaller and larger reflectivity compared to observations (Fig. 6c). A U-shaped model rank histogram is prominent in this case, in which the smallest and largest percentiles exceed the UD, while middle percentiles stay under the line. Unlike reflectivity, the shapes of the Z DR distributions are quite different. Observational Z DR produces a smooth normal distribution with a peak at around 1 dB, which is larger than the above case. Conversely, the model Z DR does not have a Gaussian distribution and peaks at smaller values (Figs. 6d and e). A U-shaped model rank histogram suggests that the model overpredicts the frequency of both the smallest and largest raindrops (Fig. 6f). The model Z DR also clearly has a longer tail than observations. Similar to the Changma case, the largest Z and Z DR values are rather small (Z<55 dBZ and Z DR<3 dB) (Figs. 6b and e). In this Z DR range, the size effect on K DP is not dominant, and thus the model K DP shows a shorter tail compared to observations (Figs. 6g and h). As a result, the model rank histogram shows a rather flat distribution (Fig. 6i).
Figure6. As in Fig. 5 but for the typhoon case.



5. Hydrometeor properties
2
5.1. Simulated vertical hydrometeor profiles
--> In this section, we expand the scope of the analysis to include upper levels, the aim being to examine the sensitivity of simulated frozen hydrometeors to precipitation systems. Vertical profiles of horizontally averaged hydrometeor water content (HWC) over grid points where HWC is greater than 0 g m-3 are plotted for each case in Fig. 7. UM's generic ice category, which contains both ice crystals and snow, is generally favored over graupel at each height for the Changma front case (Fig. 7a). Graupel water content extends up to about 12 km. The UM's propensity for ice/snow over graupel is reasonable, as Changma fronts are less convective than systems that favor rimed ice, such as supercells or squall lines. The melting level seems to be near z=~ 5 km, as ice water content quickly decreases near this level and rain water content increases. The model graupel did not reach the surface, suggesting complete melting before reaching the surface.
Ice/snow is similarly favored over graupel at each height for Typhoon Sanba (2012) (Fig. 7b), but by far more than the Changma case. The ice/snow peak is more than twice that of the Changma front, exceeding 0.5 g m-3. Graupel is found within a limited layer around the freezing level, where it can grow through riming. However, heavy graupel falls out quickly, while most ice produced above the freezing level comprises small ice particles and aggregates in the typhoon (Houze, 2010). The typhoon melting level also appears to be near z=~ 5 km, as frozen water content decreases and rain water increases below this height.
Figure7. Vertical plots of model rain, ice/snow and graupel horizontally averaged HWC (units: g m-3) for the (a) Changma front and (b) Typhoon Sanba cases.



2
5.2. Hydrometeor classifications
--> Hydrometeor classification algorithms (HCAs, Park et al., 2009) are applied to the Changma front and Typhoon Sanba (2012) observations at the 1.6° elevation angle (Fig. 8), which captures higher altitudes (i.e., more frozen hydrometeors) than the elevation angle used in sections 3 and 4. HCAs can identify the dominant hydrometeor types in the radar resolution volume, and therefore can be used to evaluate UM's ability to properly simulate hydrometeor fields. The model dominant hydrometeor type is defined as the hydrometeor type that contributes most to linear reflectivity. This allows for a direct comparison of hydrometeor fields between model and observations. The hydrometeor types included in this study are: ground clutter/anomalous propagation (GC/AP), biological scatterers (BS), dry snow/ice crystals (DS/CR), wet snow (WS), rain (RA), rain/rimed ice (RR), and rimed ice (RI). Readers are referred to (Putnam et al., 2017) for further details on each method.
Figure8. HCAs for the (a, c) observations and (b, d) model in the (a, b) Changma front case and (c, d) typhoon case at the 1.6° elevation angle. The hydrometeors considered are: ground clutter/anomalous propagation (GC/AP), biological scatterers (BS), dry snow/ice crystals (DS/CR), wet snow (WS), rain (RA), rain/rimed ice (RR), and rimed ice (RI).


Observationally, lower levels of the Changma front are typically composed of rain (Fig. 8a). A melting layer transition region is mostly composed of rain, wet snow, and dry snow to the east of the radar. To the south, the transition zone is deeper and consists of two layers: an upper layer with rain, dry snow, wet snow and rimed ice; and a lower layer with rain and rain/rimed ice. Sparse melting is found to the west. Finally, in the upper levels of the radar scan, crystals and dry snow are prominent. Similar to observations, lower levels in the UM are typically composed of rain (Fig. 8b). Model transition regions are primarily composed of rain/rimed ice and wet snow in all directions. Rimed ice is most prominent to the east of the radar domain, but also present to the north and south. In observations, it is typically contained to both the west and south of the radar. This suggests that the model tends to overpredict the presence of graupel compared to observations. The upper levels of the simulated Changma front in the radar domain are typically dry snow/ice crystals, which matches observations well. Still, the presence of rimed ice above the melting layer is greater than in the observations.
Typhoon Sanba (2012) is mostly rain at lower levels (Fig. 8c). This lower-level rain coverage is smaller than that in the Changma front case because of the autumn season. Distinct melting occurs to the west of the radar site, with rain, dry snow and wet snow populating these regions, along with rimed ice. Elsewhere, hydrometeors are typically rain and wet snow in the melting layer. Heights above the melting layer are primarily composed of ice crystals and dry snow. Simulated UM hydrometeors are typically rain at low levels (Fig. 8d), in agreement with the observed predominant hydrometeor type. The melting transition region between frozen and liquid hydrometeors is primarily composed of rimed ice and wet snow in all directions, which is similar to the model Changma case. Model levels above the transition region are primarily composed of dry snow/ice crystals, in agreement with observations. Similar to the Changma case, rimed ice populates upper levels more frequently than observations.

6. Summary and discussion
This study examines UM microphysics for two convective cases at 1.5-km grid spacing over the Korean peninsula: a Changma front, and Typhoon Sanba (2012). Simulated polarimetric radar variables are compared to observations from the S-band Biseul-san radar. Clearly, the model struggles with convection, as reflectivity Z H gaps are present in each case. The consequences of inadequately resolving convection are significant to the model's forecast, as the model is underpredicting precipitation coverage. The diagnostic relation between the rain intercept parameter and mixing ratio results in a rapidly increasing (decreasing) drop size for an increasing (decreasing) mixing ratio. This is partially responsible for the large Z H gradients present in the cases compared to observations. The ability to correctly predict PSDs in microphysics schemes is potentially important, as they have a significant impact on the evolution of precipitating systems through their feedback to thermodynamics and dynamics. Thus, it is imperative to understand microphysics biases and address them, especially for convective-scale modeling where the microphysics error dominates the forecast errors. Quantitatively, the model generally produces more small and large Z DR than observations, which is consistent with the diagnostic relationship discussed above. K DP is less affected by the diagnostic relationship, as it is proportional to a lower order moment than Z and Z DR. Thus, the model drop size bias should be taken into account when interpreting simulated radar variables. The UM's generic ice category, which contains both crystals and snow, overshadowed graupel as the dominant hydrometeor, with neither particle type appreciably reaching the surface. This is reasonable because neither storm system has a strong updraft. Still, graupel might be incorrectly parameterized, as the model graupel is more prominent above the melting level compared to observations.
Many UM microphysics shortcomings in this study stem from the 1M nature of the scheme. Schemes with 1M categories have the potential to perform reasonably well for large-scale storm systems (Liu et al., 2011). In that regard, the UM microphysics raindrop size distribution is tuned with aircraft observations collected from stratocumulus and trade-wind cumulus (Abel and Boutle, 2012), which may not be adequate for midlatitude deep convective systems where the Korean peninsula is located. As a result, the model struggles with typical mid-latitude weather systems such as the Changma front in this study. An incorrect N0 parameterization by many orders of magnitude can adversely affect the rain PSD and its moments, which is directly linked to many microphysical processes. Further, radar variables are sensitive to drop size distributions. For example, radar reflectivity is dependent on drop size to the sixth power, and differential reflectivity is related to the axis ratio of hydrometeors. As radar observations are typically used to monitor severe weather, large biases in simulated radar variables can misguide forecasters as well as introduce large errors in assimilation. A well-calibrated regional rain N0 relationship could result in improved model microphysics and forecasts. For improved flexibility and model performance, it is desirable that a 2M version of the UM microphysics scheme be developed. As an example, the 2M version would potentially be able to simulate both heavy rain dominated by small drops with relatively uniform size in warm-rain processes, and heavy rain with many large raindrops in cold-rain processes. While adjusting the rain N0 configuration for different weather systems could mitigate this problem to some degree, evolving the scheme to 2M would allow more freedom to simulate a wider range of weather systems.

濠电姷鏁告慨鐑藉极閹间礁纾婚柣鎰惈閸ㄥ倿鏌涢锝嗙缂佺姳鍗抽弻鐔虹磼閵忕姵鐏堢紒鐐劤椤兘寮婚悢鐓庣鐟滃繒鏁☉銏$厽闁规儳顕埥澶嬨亜椤撶偞鍋ラ柟铏矊椤曘儱螖婵犱線鍋楅梺璇″枟閿曘垽骞冮埡鍐<婵☆垳鍘х敮楣冩⒒娴gǹ顥忛柛瀣噽閹广垽宕熼姘К闂佹寧绻傞ˇ浼存偂濞嗘垟鍋撶憴鍕婵炲眰鍊濋崺銏ゅ醇閳垛晛浜鹃悷娆忓缁€鍐磼椤旇偐效妤犵偛绻樺畷銊╁级閹寸偛绁舵俊鐐€栭幐楣冨窗閹伴偊鏁婇煫鍥ㄧ⊕閳锋帡鏌涚仦鎹愬闁逞屽墮閸㈡煡婀侀梺鎼炲労閻忔稑鈽夐姀鐘殿槹濡炪倖鍔戦崐鏍р枔閹屾富闁靛牆妫楅崸濠囨煕鐎n偅灏伴柕鍥у椤㈡洟鏁愰崶鈺冩澖濠电姷顣介崜婵嬪箖閸岀偛鏄ラ柍鈺佸暞婵挳鏌ц箛鏇熷殌妤犵偐鍋撳┑鐘殿暜缁辨洟宕戦幋锕€纾归柡宥庡幗閸嬪淇婇妶鍛櫤闁稿绻濋弻鏇㈠醇濠靛洨鈹涙繝娈垮枟婵炲﹪寮婚埄鍐ㄧ窞濠电姴瀚惃鎴濃攽閳╁啫绲婚柣妤佹崌瀵鏁撻悩鑼槰闂佹寧绻傞幊宥嗙珶閺囩喓绡€闁汇垽娼цⅷ闂佹悶鍔嶅浠嬪极閸愵喖顫呴柣妯虹仛濞堥箖姊洪崨濠勭畵閻庢凹鍣e鎶藉幢濞戞瑧鍘遍梺鍝勬储閸斿本鏅堕鐣岀闁割偅绻勯悞鍛婃叏婵犲啯銇濈€规洏鍔嶇换婵嬪磼濮f寧娲樼换娑氣偓娑欋缚閻矂鏌涚€c劌鈧洟鎮惧畡鎳婃椽顢旈崟顓濈礈闂備礁鎼崐鍫曞磿閺屻儻缍栫€广儱顦伴埛鎴︽偡濞嗗繐顏╅柛鏂诲€濋弻锝嗗箠闁告柨瀛╃粋宥夊箹娓氬洦鏅濋梺闈涚墕濞层劑鏁嶅⿰鍐f斀閹烘娊宕愰弴銏犵柈妞ゆ劧绠戦崙鐘绘煛閸愩劎澧涢柣鎾寸懃椤啰鈧綆浜妤呮煃鐠囪尙澧涙い銊e劦閹瑩寮堕幋鐘辩礉婵°倗濮烽崑娑樏洪鈧偓浣糕枎閹惧厖绱堕梺鍛婃处娴滐綁宕洪崨瀛樷拻闁稿本鑹鹃埀顒勵棑缁牊绗熼埀顒勭嵁閺嶎収鏁冮柨鏇楀亾缁炬儳婀遍幉鎼佹偋閸繄鐟查梺绋款儜缁绘繂顕i崼鏇為唶婵﹩鍘介悵鏇烆渻閵堝骸浜濇繛鑼枛瀵濡搁埡鍌氫簽闂佺ǹ鏈粙鎴︻敂閿燂拷
2濠电姷鏁告慨鐑藉极閹间礁纾婚柣鎰惈閸ㄥ倿鏌涢锝嗙缂佺姵澹嗙槐鎺斺偓锝庡亾缁扁晜绻涘顔荤盎閹喖姊洪崘鍙夋儓妞ゆ垵娲ㄧ划娆掔疀濞戞瑢鎷洪梺闈╁瘜閸樺ジ宕濈€n偁浜滈柕濞垮劜椤ャ垻鈧娲滈弫濠氬春閳ь剚銇勯幒鎴濐仾闁抽攱鍨块弻娑樷槈濮楀牆浼愭繝娈垮櫙缁犳垿婀佸┑鐘诧工閹冲孩绂掓潏鈹惧亾鐟欏嫭绀冩俊鐐扮矙瀵偊骞樼紒妯轰汗閻庤娲栧ú銈夌嵁濡ゅ懏鈷掑〒姘e亾婵炰匠鍛床闁割偁鍎辩壕褰掓煛瀹擃喒鍋撴俊鎻掔墢閹叉悂寮崼婵婃憰闂佹寧绻傞ˇ顖炴倿濞差亝鐓曢柟鏉垮悁缁ㄥジ鏌涢敐搴″箻缂佽鲸鎸婚幏鍛村礈閹绘帒澹堥梻浣瑰濞诧附绂嶉鍕靛殨妞ゆ劧绠戠壕濂告煟閹邦厽缍戞繛鍫熷姍濮婃椽宕橀崣澶嬪創闂佸摜濮甸懝鎯у祫闂佸憡顨堥崑鎰板绩娴犲鐓冮柦妯侯槹椤ユ粌霉濠婂懎浠滄い顓″劵椤﹁櫕銇勯妸銉含鐎殿噮鍋嗛埀顒婄秵閸撴稓澹曢挊澹濆綊鏁愭径瀣敪婵犳鍠栭崐鎼佹箒濠电姴锕ゅΛ妤呮偂閹邦儮搴ㄥ炊瑜濋崝鐔兼煃瑜滈崜姘辩矙閹烘洘鎳屽┑鐘愁問閸ㄤ即顢氶鐘愁潟闁圭儤鍨熷Σ鍫熸叏濡も偓濡宕滄潏鈺冪=闁稿本姘ㄥ瓭闂佹寧娲忛崕鑼矚鏉堛劎绡€闁搞儴鍩栭弲婵嬫⒑闂堟稓澧曢柟宄邦儔瀵娊顢橀姀鈾€鎷洪梺鍛婃崄鐏忔瑩宕㈠☉銏$厱闁靛ǹ鍎抽崺锝団偓瑙勬礃濡炰粙宕洪埀顒併亜閹哄秹妾峰ù婊勭矒閺岀喐娼忛崜褏蓱缂佺虎鍙€閸╂牠濡甸崟顖涙櫆闁兼祴鏅濋弳銈夋⒑閸濆嫭婀扮紒瀣灴閸┿垺鎯旈妶鍥╂澑闂佸搫娲ㄦ刊顓㈠船閸︻厾纾介柛灞剧懅缁愭梻绱撻崒娑滃閾荤偤鏌涢弴銊ユ灓濞存粍鐟╁缁樻媴閸涘﹤鏆堝┑鐐额嚋缁犳挸鐣烽姀锝冧汗闁圭儤鍨归敍娑㈡⒑閸︻厼鍔嬫い銊ユ閸╂盯骞嬮敂鐣屽幈濠电娀娼уΛ妤咁敂閳哄懏鐓冪憸婊堝礈濞嗘垹绀婂┑鐘叉搐缁犳牠姊洪崹顕呭剱缂傚秴娲弻宥夊传閸曨偂绨藉┑鐐跺亹閸犲酣鍩為幋锔绘晩閻熸瑦甯為幊鎾诲煝閺傚簱妲堥柕蹇娾偓鍐插婵犲痉鏉库偓鎰板磻閹剧粯鐓冮悷娆忓閻忔挳鏌熼瑙勬珚妤犵偞鎹囬獮鎺楀幢濡炴儳顥氶梻浣哥秺濡法绮堟笟鈧弻銊╁Χ閸涱亝鏂€闂佺粯蓱瑜板啴寮搁妶鍡欑闁割偅绮庨惌娆撴煛瀹€瀣М妤犵偛娲、妤佹媴閸欏浜為梻鍌欑劍閹爼宕愬Δ鍛獥闁归偊鍠楀畷鍙夌節闂堟侗鍎忛柣鎺戠仛閵囧嫰骞掗幋婵愪患闂佺粯甯楀浠嬪蓟濞戙垹绠涙い鏍ㄧ〒閵嗗﹪姊哄ú璇插箺妞ゃ劌鎳橀崺鐐哄箣閿旂粯鏅╃紓浣圭☉椤戝棝鎮鹃崼鏇熲拺缂備焦锕╁▓鏃傜磼缂佹ê绗ч柛鎺撳浮瀹曞ジ鎮㈡搴g嵁闂佽鍑界紞鍡涘礈濞戙埄鏁婇柡鍥ュ灪閳锋垿鏌i悢鐓庝喊闁搞倗鍠庨埞鎴︻敊閻愵剚姣堥悗娈垮枟婵炲﹪宕洪敓鐘茬<婵犲﹤鎷嬮崯搴ㄦ⒑閼姐倕孝婵炲/鍥х妞ゆ劦鍋傜槐锟�547闂傚倸鍊搁崐鎼佸磹妞嬪海鐭嗗ù锝夋交閼板潡姊洪鈧粔鏌ュ焵椤掆偓閸婂湱绮嬮幒鏂哄亾閿濆簼绨介柨娑欑洴濮婃椽鎮烽弶搴撴寖缂備緡鍣崹鍫曞春濞戙垹绠虫俊銈勮兌閸橀亶姊洪崫鍕妞ゃ劌妫楅埢宥夊川鐎涙ḿ鍘介棅顐㈡祫缁插ジ鏌囬鐐寸厸鐎光偓鐎n剙鍩岄柧缁樼墵閹鏁愭惔鈥茬盎濡炪倕楠忛幏锟�4濠电姷鏁告慨鐑藉极閹间礁纾婚柣鎰惈閸ㄥ倿鏌涢锝嗙缂佺姵澹嗙槐鎺斺偓锝庡亾缁扁晜绻涘顔荤凹闁哄懏鐓¢弻娑㈠Ψ閵忊剝鐝栧銈忓瘜閸ㄨ泛顫忓ú顏呭仭闂侇叏绠戝▓鍫曟⒑缁嬫鍎戦柛鐘崇墵瀹曟椽濮€閵堝懐鐫勯梺閫炲苯澧村┑锛勬暬瀹曠喖顢欓崜褎婢戦梻浣筋潐閸庢娊顢氶鈶哄洭鏌嗗鍡忔嫼缂備礁顑嗛娆撳磿閹扮増鐓欓柣鐔哄閹兼劙鏌i敐鍛Щ妞ゎ偅绮撻崺鈧い鎺戝閳ь兛绶氬顕€宕煎┑鍡氣偓鍨攽鎺抽崐鏇㈠疮椤愶妇宓侀柟鎵閳锋帡鏌涚仦鍓ф噮妞わ讣绠撻弻娑橆潩椤掑鍓板銈庡幖閻忔繈锝炲⿰鍫濈劦妞ゆ巻鍋撻柣锝囧厴椤㈡盯鎮滈崱妯绘珖闂備線娼х换鍫ュ垂閸濆嫧鏋斿Δ锝呭暞閳锋垿姊婚崼鐔剁繁婵$嫏鍐f斀闁炽儴娅曢崰姗€鏌涢埞鍨伈鐎殿噮鍣e畷濂告偄閸濆嫬绠ラ梻鍌欒兌椤㈠﹪锝炴径鎰闁哄洢鍨洪崕宥嗙箾瀹割喕绨奸柣鎾跺枛閺岋綁寮崼鐔告殸闁荤姵鍔х槐鏇犳閹烘挻缍囬柕濞垮劤閻熸煡鎮楅崹顐g凡閻庢凹鍣i崺鈧い鎺戯功缁夐潧霉濠婂懎浠︾紒鍌涘浮閹剝鎯斿Ο缁樻澑闂備胶绮崝妯衡枖濞戞碍顫曢柨鏇炲€归悡鏇熶繆閵堝懎顏柣婵愪簻鑿愰柛銉戝秴濮涢梺閫炲苯澧紒瀣笩閹筋偅绻濆▓鍨仭闁瑰憡濞婇獮鍐ㄧ暋閹佃櫕鐎诲┑鐐叉閸ㄧ敻宕虹仦鍓х閻庢稒岣块惌鎺旂磼閻樺磭澧电€殿喛顕ч埥澶愬閻樼數鏉搁梻浣呵圭换鎰板箺濠婂牆鏋侀柡宥庡幗閳锋垹绱掗娑欑婵炲懏姊荤槐鎺旂磼濡偐鐤勯悗娈垮枦椤曆囧煡婢跺ň鍫柛娑卞灡濠㈡垿姊绘担鐟邦嚋缂佽鍊块獮濠冩償椤帞绋忛梺鍐叉惈閹冲繘鍩涢幋锔界厱婵炴垶锕崝鐔兼煙閾忣偅绀堢紒杈ㄥ笚濞煎繘濡搁敂缁㈡Ч婵°倗濮烽崑娑氭崲濮椻偓楠炲啴鍩¢崘鈺佺彴闂佽偐鈷堥崜锕傚疮鐎n喗鈷掑ù锝呮啞閸熺偛銆掑顓ф疁鐎规洖缍婇獮搴ㄥ礈閸喗鍠橀柛鈺嬬節瀹曘劑顢欑憴鍕伖闂備浇宕甸崑鐐电矙閸儱鐒垫い鎺嗗亾闁告ɑ鐗楃粩鐔煎即閵忊檧鎷绘繛杈剧到閹诧紕鎷归敓鐘插嚑妞ゅ繐妫涚壕濂告煏婵炲灝濡煎ù婊冩贡缁辨帡顢氶崨顓炵閻庡灚婢樼€氫即鐛崶顒夋晣闁绘ɑ褰冪粻濠氭⒒閸屾瑧顦﹂柟纰卞亞閳ь剚鍑归崜娑㈠箲閵忋倕绠抽柡鍐ㄦ搐灏忛梻浣告贡鏋紒銊у劋缁傚秴饪伴崼鐔哄幐闂佹悶鍎洪悡渚€顢旈崼鐔封偓鍫曟煠绾板崬鍘撮柛瀣尭閳绘捇宕归鐣屽蒋闂備胶枪椤戝懘鏁冮妶澶樻晪闁挎繂娲﹀畷澶愭偠濞戞帒澧查柣搴☆煼濮婅櫣鎷犻垾宕団偓濠氭煕韫囧骸瀚庨柛濠冪箓椤繒绱掑Ο璇差€撻梺鑽ゅ枛閸嬪﹪宕电€n剛纾藉ù锝呭閸庢劙鏌涢妸銊ュ姷婵☆偆鍠庨—鍐Χ閸℃ê钄奸梺鎼炲妼缂嶅﹪骞冮悙鍝勫瀭妞ゆ劗濮崇花濠氭⒑閸︻厼鍔嬮柛鈺侊躬瀵劍绻濆顓炩偓鍨叏濡厧浜鹃悗姘炬嫹40缂傚倸鍊搁崐鎼佸磹閹间礁纾瑰瀣捣閻棗銆掑锝呬壕闁芥ɑ绻冮妵鍕冀閵娧呯厒闂佹椿鍘介幐楣冨焵椤掑喚娼愭繛鍙夌墪鐓ら柕濞у懍绗夐梺鍝勫暙閻楀﹪鎮″▎鎾寸厵妞ゆ牕妫楅懟顖氣枔閸洘鈷戠€规洖娲ㄧ敮娑欐叏婵犲倻绉烘鐐茬墦婵℃悂濡烽钘夌紦闂備線鈧偛鑻晶鐗堢箾閹寸姵鏆鐐寸墬閹峰懘宕ㄦ繝鍕ㄥ亾椤掑嫭鐓熼幖鎼灣缁夐潧霉濠婂啰鍩i柟顔哄灲瀹曞崬鈽夊▎蹇庡寲闂備焦鎮堕崕鑽ゅ緤濞差亜纾婚柟鎹愵嚙缁€鍌炴煕濞戝崬寮炬俊顐g矌缁辨捇宕掑顑藉亾瀹勬噴褰掑炊閵婏絼绮撻梺褰掓?閻掞箓宕戦敓鐘崇厓闁告繂瀚崳褰掓煢閸愵亜鏋旈柍褜鍓欓崢婊堝磻閹剧粯鐓曢柡鍥ュ妼娴滅偞銇勯幘瀛樸仢婵﹥妞介獮鎰償閿濆洨鏆ゆ繝鐢靛仩鐏忔瑦绻涢埀顒傗偓瑙勬礃閸ㄥ潡鐛Ο鑲╃<婵☆垵顕ч崝鎺楁⒑閼姐倕鏋戦柣鐔村劤閳ь剚鍑归崜鐔风暦閵忥絻浜归柟鐑樻尨閹锋椽姊洪崨濠勭畵閻庢凹鍘奸蹇撯攽鐎n偆鍘遍柟鍏肩暘閸ㄥ綊鎮橀埡鍌欑箚闁告瑥顦慨鍥殰椤忓啫宓嗙€规洖銈搁幃銏ゅ传閸曨偄顩梻鍌氬€烽懗鍓佹兜閸洖绀堟繝闈涙灩濞差亜鍐€妞ゆ劑鍎卞皬缂傚倷绶¢崑鍕偓娈垮墴濮婂宕掑顑藉亾妞嬪孩顐芥慨姗嗗厳缂傛氨鎲稿鍫罕闂備礁鎼崯顐﹀磹婵犳碍鍎楅柛鈩冾樅瑜版帗鏅查柛顐亜濞堟瑩姊洪懡銈呮瀾閻庢艾鐗撳顕€宕煎┑鍡欑崺婵$偑鍊栧Λ渚€锝炴径灞稿亾閸偆澧垫慨濠勭帛閹峰懘宕ㄦ繝鍌涙畼濠电偞鎸荤喊宥夈€冩繝鍌滄殾闁靛繈鍊栫€电姴顭跨捄鐑橆棡闁诲孩妞介幃妤呭礂婢跺﹣澹曢梻浣告啞濞诧箓宕滃☉銏犲偍闂侇剙绉甸埛鎴︽煕濠靛棗顏╅柡鍡欏仱閺岀喓绮欓崹顔规寖婵犮垼顫夊ú鐔肩嵁閹邦厽鍎熸繛鎴烆殘閻╁酣姊绘笟鈧ḿ褎顨ヨ箛鏇燁潟闁哄洠鍋撻埀顒€鍊块幊鐘活敆閸屾粣绱查梻浣告惈閸燁偊宕愰幖浣稿嚑婵炴垶鐟f禍婊堟煏韫囧﹤澧茬紒鈧€n喗鐓欐い鏃囶潐濞呭﹥銇勯姀鈩冪闁挎繄鍋ら、姗€鎮滈崱姗嗘%婵犵數濮烽弫鎼佸磻閻樿绠垫い蹇撴缁€濠囨煃瑜滈崜姘跺Φ閸曨垼鏁冮柕蹇婃櫆閳诲牓姊虹拠鈥虫珯缂佺粯绻堝畷娲焵椤掍降浜滈柟鐑樺灥椤忣亪鏌嶉柨瀣诞闁哄本绋撴禒锕傚箲閹邦剦妫熼梻渚€鈧偛鑻崢鍝ョ磼椤旂晫鎳囬柕鍡曠閳诲酣骞囬鍓ф闂備礁鎲″ú锕傚礈閿曗偓宀e潡鎮㈤崗灏栨嫼闂佸憡鎸昏ぐ鍐╃濠靛洨绠鹃柛娆忣槺婢ц京绱掗鍨惞缂佽鲸甯掕灒闂傗偓閹邦喚娉块梻鍌欐祰椤鐣峰Ο琛℃灃婵炴垯鍩勯弫浣衡偓鍏夊亾闁告洦鍓涢崢鍛婄箾鏉堝墽鍒板鐟帮躬瀹曟洝绠涢弬璁崇盎濡炪倖鎸撮崜婵堟兜閸洘鐓欏瀣閳诲牓鏌涢妸鈺冪暫鐎规洘顨婂畷銊╊敍濞戞ḿ妯嗛梻鍌氬€搁崐椋庢濮樿泛鐒垫い鎺戝€告禒婊堟煠濞茶鐏︾€规洏鍨介獮鏍ㄦ媴閸︻厼骞橀梻浣告啞閸旀ḿ浜稿▎鎾虫槬闁挎繂鎳夐弨浠嬫煥濞戞ê顏柡鍡╁墴閺岀喖顢欓悾灞惧櫚閻庢鍠栭悥濂哥嵁鐎n噮鏁囬柣鎰儗閸熷本绻濋悽闈浶fい鏃€鐗犲畷鏉课旈崨顔芥珖闂佸啿鎼幊搴g矆閸屾稓绠鹃柟瀵稿仧椤e弶銇勯锝嗙闁哄被鍔岄埞鎴﹀幢濡桨鐥柣鐔哥矌婢ф鏁Δ鍛柧闁哄被鍎查悡鏇㈡煃閳轰礁鏆熼柟鍐叉嚇閺岋綁骞橀崘娴嬪亾閹间讲鈧棃宕橀鍢壯囨煕閹扳晛濡垮ù鐘插⒔缁辨帡鎮欓浣哄嚒缂備礁顦晶搴ㄥ礆閹烘鐓涢柛娑卞枛娴滄粎绱掗悙顒€顎滃瀛樻倐瀵彃鈹戠€n偀鎷洪梻鍌氱墛缁嬫挻鏅堕弴鐔虹閻犲泧鍛殼濡ょ姷鍋涘Λ婵嬪极閹邦厼绶為悗锛卞嫬顏归梻鍌欑濠€杈ㄧ仚濠电偛顕崗姗€宕洪妷锕€绶為悗锝冨妺缁ㄥ姊洪幐搴㈩梿妞ゆ泦鍐惧殨妞ゆ洍鍋撻柡灞剧洴閸╃偤骞嗚婢规洖鈹戦敍鍕杭闁稿﹥鐗滈弫顕€骞掑Δ浣规珖闂侀潧锛忛埀顒勫磻閹炬剚娼╅柣鎰靛墮椤忥拷28缂傚倸鍊搁崐鎼佸磹閹间礁纾瑰瀣椤愪粙鏌ㄩ悢鍝勑㈢痪鎹愵嚙椤潡鎳滈棃娑樞曢梺杞扮椤戝洭骞夐幖浣哥睄闁割偁鍨圭粊锕傛⒑閸涘﹤濮﹂柛鐘崇墱缁粯绻濆顓犲幈闂佽宕樼亸娆戠玻閺冨牊鐓冮柣鐔稿缁犺尙绱掔紒妯肩疄濠殿喒鍋撻梺鎸庣箓濡盯濡撮幇顑╂柨螖婵犱胶鍑归梺鍦归崯鍧楁偩瀹勬壋鏀介悗锝庝簻缁愭盯鏌f惔銏⑩姇瀹€锝呮健瀹曘垽鏌嗗鍡忔嫼闂佸憡绻傜€氼剟寮虫繝鍥ㄧ厱閻庯綆鍋呯亸鐢电磼鏉堛劌绗ч柍褜鍓ㄧ紞鍡涘磻閸涱厾鏆︾€光偓閳ь剟鍩€椤掍緡鍟忛柛锝庡櫍瀹曟垶绻濋崶褏鐣烘繛瀵稿Т椤戝懘宕归崒娑栦簻闁规壋鏅涢悘鈺傤殽閻愭潙鐏存慨濠勭帛閹峰懘宕ㄦ繝鍐ㄥ壍婵犵數鍋犻婊呯不閹达讣缍栨繝闈涱儏鎯熼梺鍐叉惈閸婂憡绂掗銏♀拺閻庡湱濮甸妴鍐偣娴g懓绲婚崡閬嶆煕椤愮姴鍔滈柣鎾寸懇閺岋綁骞囬棃娑橆潽缂傚倸绉甸崹鍧楀蓟閻旂厧绀傞柛蹇曞帶閳ь剚鍔欓弻锛勪沪閻e睗銉︺亜瑜岀欢姘跺蓟濞戙垹绠婚柛妤冨仜椤洤螖閻橀潧浠滅紒缁橈耿瀵偊骞樼紒妯绘闂佽法鍣﹂幏锟�1130缂傚倸鍊搁崐鎼佸磹閹间礁纾瑰瀣捣閻棗銆掑锝呬壕闁芥ɑ绻冮妵鍕冀閵娧呯厒闂佹椿鍘介幑鍥蓟濞戙垹绠婚柤纰卞墻濡差噣姊洪幖鐐插缂佽鐗撳濠氬Ω閳哄倸浜滈梺鍛婄箓鐎氬懘濮€閵忋垻锛滈梺閫炲苯澧寸€规洘甯¢幃娆戔偓鐢登归獮鍫熺節閻㈤潧浠﹂柛銊ョ埣閺佸啴顢曢敃鈧紒鈺冪磽娴h疮缂氱紒鐘荤畺閺屾盯顢曢敐鍥╃暭闂佺粯甯楅幃鍌炲蓟閿涘嫪娌紒瀣仢閳峰鎮楅崹顐g凡閻庢凹鍣i崺鈧い鎺戯功缁夐潧霉濠婂嫮鐭掗柨婵堝仱瀹曞爼顢楁担鍙夊闂傚倷绶¢崑鍡涘磻濞戙垺鍤愭い鏍ㄧ⊕濞呯姴螖閿濆懎鏆為柣鎾寸懇閺屾盯骞嬪▎蹇婂亾閺嶎偀鍋撳鐐
相关话题/Evaluation Unified Model

闂傚倸鍊搁崐鎼佸磹閻戣姤鍤勯柛鎾茬閸ㄦ繃銇勯弽顐粶缂佲偓婢跺绻嗛柕鍫濇噺閸e湱绱掗悩闈涒枅闁哄瞼鍠栭獮鎴﹀箛闂堟稒顔勯梻浣告啞娣囨椽锝炴径鎰﹂柛鏇ㄥ灠濡﹢鏌涢…鎴濇灀闁圭ǹ鍟村娲川婵犲孩鐣烽悗鍏夊亾闁归棿绀佺粻鏍ㄤ繆閵堝懏鍣洪柡鍛叀楠炴牜鈧稒岣跨粻姗€鏌i埡浣规崳缂佽鲸鎸婚幏鍛槹鎼淬倗鐛ラ梻渚€娼荤紞鍥╃礊娴e壊鍤曞┑鐘崇閸嬪嫰鏌i幘铏崳妞は佸洦鈷戦柛蹇氬亹閵堟挳鏌¢崨顔剧疄闁诡噯绻濆畷鎺楁倷缁瀚肩紓鍌欑椤戝牆鈻旈弴銏″€块柛褎顨嗛悡娆撴煕閹存瑥鈧牜鈧熬鎷�2濠电姷鏁告慨鐑藉极閹间礁纾婚柣鎰惈閸ㄥ倿鏌涢锝嗙缂佺姵澹嗙槐鎺斺偓锝庡亾缁扁晜绻涘顔荤盎閹喖姊洪崘鍙夋儓妞ゆ垵娲ㄧ划娆掔疀濞戞瑢鎷洪梺闈╁瘜閸樺ジ宕濈€n偁浜滈柕濞垮劜椤ャ垻鈧娲滈弫濠氬春閳ь剚銇勯幒鎴濐仾闁抽攱鍨块弻娑樷槈濮楀牆浼愭繝娈垮櫙缁犳垿婀佸┑鐘诧工閹冲孩绂掓潏鈹惧亾鐟欏嫭绀冩俊鐐扮矙瀵偊骞樼紒妯轰汗閻庤娲栧ú銈夌嵁濡ゅ懏鈷掑〒姘e亾婵炰匠鍛床闁割偁鍎辩壕褰掓煛瀹擃喒鍋撴俊鎻掔墢閹叉悂寮崼婵婃憰闂佹寧绻傞ˇ顖炴倿濞差亝鐓曢柟鏉垮悁缁ㄥジ鏌涢敐鍕祮婵﹨娅i幏鐘诲灳閾忣偅顔勯梻浣规偠閸庢粓宕惰閺嗩亪姊婚崒娆戝妽閻庣瑳鍛床闁稿本顕㈠ú顏勵潊闁靛牆鎳愰敍娑㈡⒑閸︻厼鍔嬫い銊ユ閸╂盯骞嬮敂鐣屽幈濠电娀娼уΛ妤咁敂閳哄懏鐓冪憸婊堝礈濞嗘垹绀婂┑鐘叉搐缁犳牠姊洪崹顕呭剱缂傚秴娲弻宥夊传閸曨偂绨藉┑鐐跺亹閸犳牕顫忛搹瑙勫磯闁靛ǹ鍎查悵銏ゆ⒑閻熸澘娈╅柟鍑ゆ嫹
濠电姷鏁告慨鐑藉极閸涘﹥鍙忓ù鍏兼綑閸ㄥ倻鎲搁悧鍫濈瑲闁稿顦甸弻鏇$疀鐎n亖鍋撻弴銏″€峰┑鐘插閸犳劗鈧箍鍎卞Λ娆撳矗韫囨稒鐓忛柛顐g箥濡插綊鏌嶉柨瀣伌闁哄本绋戦埥澶婎潨閸繀绱g紓鍌欑劍椤ㄥ棛鏁Δ浣衡攳濠电姴娴傞弫鍐煥濠靛棙澶勯柛鎺撶☉椤啴濡堕崘銊т痪濠碘槅鍋勯崯顖炲箞閵娾晛鐒垫い鎺戝閻撳繘鏌涢锝囩畺闁挎稑绉垫穱濠囶敃閵忕媭浼冮梺鍝勭焿缁查箖骞嗛弮鍫晬婵犲﹤鎲涢敐澶嬧拺闁告縿鍎辨牎闂佺粯顨堟慨鎾偩閻戣棄顫呴柕鍫濇噽椤旀帒顪冮妶鍡樷拻闁哄拋鍋婂畷銏ゅ箹娴e厜鎷洪梺鍛婄☉閿曘儳绮堢€n偆绠惧ù锝呭暱濞诧箓宕愰崼鏇熺叆婵犻潧妫欓ˉ鎾趁瑰⿰鍕煉闁哄瞼鍠撻埀顒佺⊕宀h法绮婚弽褜鐔嗛悹鍝勬惈椤忣偆绱掓潏銊ョ闁逞屽墾缂嶅棙绂嶇捄浣曠喖鍩€椤掑嫭鈷戠紒顖涙礃閺夊綊鏌涚€n偅灏い顏勫暣婵″爼宕卞Δ鈧ḿ鎴︽⒑缁嬫鍎愰柟鐟版喘瀵顓奸崶銊ョ彴闂佸搫琚崕鍗烆嚕閺夊簱鏀介柨鐔哄Х閻e搫霉濠婂啰鍩g€殿喛顕ч濂稿醇椤愶綆鈧洭姊绘担鍛婂暈闁圭ǹ顭烽幃鐑藉煛娴g儤娈惧銈嗙墬缁嬫垿顢氶柆宥嗗€垫繛鎴烆仾椤忓懐顩叉い鏍ㄥ焹閺€浠嬫煟閹邦剙绾ч柍缁樻礀闇夋繝濠傚缁犵偟鈧鍠楅悡锟犮€佸Δ鍛妞ゆ巻鍋撻柍褜鍓欓悥濂稿蓟閿濆绠涙い鏃囧Г濮e嫰姊虹涵鍛棄闁稿﹤娼″璇测槈閵忕姈褔鏌涢妷顔句虎闁靛繈鍊栭ˉ鍡楊熆鐠轰警鍎戠紒鈾€鍋撳┑鐘垫暩婵挳宕愰幖浣告辈闁挎繂妫庢禍婊堝箹濞n剙鐒烘繛鍫熸礋閺屾洟宕惰椤忣參鏌涢埡鍐ㄤ槐妞ゃ垺锕㈤幃娆忣啅椤旇崵妫繝鐢靛У椤旀牠宕归柆宥呯闁规儼妫勯拑鐔兼煥閻斿搫孝闁绘劕锕弻宥嗘姜閹殿喖濡介梺璇茬箣缁舵艾顫忓ú顏勫窛濠电姴瀚崰娑㈡⒑缁嬫鍎愰柟鐟版搐椤繒绱掑Ο璇差€撻梺鍛婄缚閸庤櫕顨欏┑鐘垫暩閸嬫﹢宕犻悩璇插耿闁归偊浜濋惈蹇涙⒒娴h櫣甯涢柛鏃€顨婂顐﹀传閵壯傜瑝闂佸搫鍟悧濠囨偂濞嗘挻鐓欐い鏍ф閼活垰鈻撻崼鏇熲拺鐎规洖娲ㄧ敮娑欐叏婵犲倻绉烘鐐茬墦婵℃悂濡锋惔锝呮灁闁归濞€楠炴捇骞掑┑鍥ㄧグ闂傚倸鍊烽悞锕傚箖閸洖纾圭憸蹇曞垝婵犳艾绠婚悹鍥蔼閹芥洟姊虹紒妯活梿婵炲拑缍侀幆灞解枎閹惧鍘电紓浣割儏閻忔繈顢楅姀銈嗙厵妞ゆ梻鏅幊鍥ㄦ叏婵犲嫬鍔嬮悗鐢靛帶閳诲酣骞嬮悩妯荤矌缁辨挻鎷呴崫鍕戯綁鏌eΔ浣圭妞ゃ垺宀搁弫鎰緞濡粯娅囬梻浣稿暱閻忓牓寮插⿰鍫熷€靛┑鐘崇閳锋垹鎲搁悧鍫濈瑨濞存粈鍗抽弻娑樜熼崫鍕ㄦ寖缂備緡鍠楅悷鈺佺暦閻旂⒈鏁嶆繛鎴炲笚鐎氬ジ姊绘担鍛婅础閺嬵亝绻涚€电ǹ鍘撮柛鈹垮劜瀵板嫰骞囬鐘插箰闂備礁澹婇崑鎺楀磻閸曨剚娅犻悗鐢电《閸嬫挾鎲撮崟顒傤槬缂傚倸绉撮敃銉︾┍婵犲偆娼扮€光偓婵犲唭顏勨攽閻樻剚鍟忛柛銊ゅ嵆婵″爼骞栨担姝屾憰濠电偞鍨惰彜婵℃彃鐗婇幈銊ノ旈埀顒勬偋婵犲洤鏋侀柛鎾楀懐锛濇繛杈剧到閹碱偅鐗庨梺姹囧焺閸ㄦ娊宕戦妶澶婃槬闁逞屽墯閵囧嫰骞掗崱妞惧闂備浇顕х换鎴︽嚌妤e啠鈧箓宕归鍛缓闂侀€炲苯澧存鐐插暢椤﹀湱鈧娲栧畷顒勬箒闂佸搫顦扮€笛囧窗濡皷鍋撶憴鍕閺嬵亪鎽堕弽顬″綊鏁愭径瀣彸闂佹眹鍎烘禍顏勵潖缂佹ɑ濯村〒姘煎灡閺侇垶姊虹憴鍕仧濞存粠浜滈~蹇旂鐎n亞顦板銈嗙墬缁嬫帒鈻嶉弽顓熲拺闁告繂瀚埢澶愭煕濡湱鐭欓柟顔欍倗鐤€婵炴垶鐟ч崢閬嶆⒑閺傘儲娅呴柛鐕佸灣缁牓鍩€椤掆偓椤啴濡惰箛鏇炵煗闂佸搫妫欑粩绯村┑鐘垫暩婵兘寮崨濠冨弿濞村吋娼欓崹鍌炴煕閿旇骞樼紒鈧繝鍌楁斀闁绘ê寮堕幖鎰版煟閹烘垹浠涢柕鍥у楠炴帒顓奸崼婵嗗腐闂備焦鍓氶崹鍗灻洪悢鐓庤摕闁哄洢鍨归獮銏′繆閵堝倸浜鹃梺鍝勬4缂嶄線寮婚敍鍕勃闁告挆鍕灡婵°倗濮烽崑鐐垫暜閿熺姷宓侀悗锝庡枟閸婂鏌涢埄鍐夸緵婵☆値鍐f斀闁挎稑瀚禍濂告煕婵犲啰澧遍柡渚囧櫍閹瑩宕崟顓犲炊闂備礁缍婇崑濠囧窗濮樿埖鍎楁繛鍡楃箚閺€浠嬫煟濡搫绾у璺哄閺屾稓鈧綆鍋勬慨宥夋煛瀹€瀣М濠殿喒鍋撻梺闈涚箚閸撴繂袙閸曨垱鐓涘ù锝呮憸婢э附鎱ㄦ繝鍕笡闁瑰嘲鎳愮划娆撳箰鎼粹檧鍋撻姘f斀闁绘﹩鍠栭悘顏堟煥閺囨ê鐏╅柣锝囧厴椤㈡稑鈽夊鍡楁闂佽瀛╃粙鎺楁晪婵炲瓨绮犻崹璺侯潖濞差亜宸濆┑鐘插閻e灚绻濆▓鍨仴濡炲瓨鎮傞獮鍡涘籍閸繍娼婇梺鎸庣☉鐎氼喛鍊存繝纰夌磿閸嬫垿宕愰弽顓炵婵°倕鎳庣粣妤呭箹濞n剙鐏い鈺傚絻铻栭柨婵嗘噹閺嗙偤鏌i幘瀵告创闁哄本鐩俊鐑芥晲閸涱収鐎撮梻浣圭湽閸斿秹宕归崸妤€钃熼柨婵嗩槹閸嬪嫰鏌涘▎蹇fЧ闁绘繃妫冨铏光偓鍦У椤ュ銇勯敂鐐毈闁绘侗鍠栬灒闁兼祴鏅濋ˇ鈺呮⒑缂佹◤顏勭暦椤掑嫷鏁嗛柕蹇娾偓鑼畾闂佺粯鍔︽禍婊堝焵椤掍胶澧悡銈嗙節闂堟稒顥戦柡瀣Ч閺岋繝宕堕埡浣锋喚缂傚倸鍊瑰畝鎼佹偂椤愶箑鐐婇柕濞р偓濡插牓鎮楅悷鐗堝暈缂佽鍟存俊鐢稿礋椤栨氨顔掑┑掳鍊愰崑鎾绘煕閻曚礁鐏︽い銏$懇閺佹捇鏁撻敓锟�20婵犵數濮撮惀澶愬级鎼存挸浜炬俊銈勭劍閸欏繘鏌i幋锝嗩棄缁炬儳顭烽弻锝呂熷▎鎯ф缂備胶濮撮悘姘跺Φ閸曨喚鐤€闁圭偓鎯屽Λ鈥愁渻閵堝骸浜濇繛鍙夅缚閹广垹鈹戠€n偒妫冨┑鐐村灥瀹曨剟宕滈幍顔剧=濞达絽鎼牎闂佹悶鍔屽ḿ鈥愁嚕婵犳艾围闁糕剝锚瀵潡姊鸿ぐ鎺戜喊闁稿繑锕㈠畷鎴﹀箻濠㈠嫭妫冮崺鈧い鎺戝閻撴繈鏌¢崘銊у妞ゎ偄鎳橀弻锝呂熼悜姗嗘¥闂佺娅曢幑鍥Χ椤忓懎顕遍柡澶嬪灩椤︺劑姊洪崘鍙夋儓闁挎洏鍎甸弫宥夊川椤栨粎锛濋梺绋挎湰閻熝囁囬敂濮愪簻闁挎棁顕ч悘锔姐亜閵忊€冲摵妞ゃ垺锕㈡慨鈧柣姗€娼ф慨锔戒繆閻愵亜鈧牕顔忔繝姘;闁规儳顕弧鈧梺閫炲苯澧撮柡灞芥椤撳ジ宕ㄩ銈囧耿闂傚倷鑳剁划顖氼潖婵犳艾鍌ㄧ憸鏂款嚕閸涘﹦鐟归柍褜鍓熷濠氬即閵忕娀鍞跺┑鐘茬仛閸旀牗鏅ラ梻鍌欒兌鏋Δ鐘叉憸缁棁銇愰幒鎴f憰濠电偞鍨崹褰掑础閹惰姤鐓忓┑鐐茬仢閸旀碍銇勯鐔告珚婵﹦鍎ょ€电厧鈻庨幋鐘虫缂傚倸鍊哥粔鎾晝椤忓牏宓侀柛鎰╁壆閺冨牆绀冮柍杞扮劍閻庮參姊绘担鍛婂暈婵炶绠撳畷锝嗘償閵娿儲杈堥梺璺ㄥ枔婵敻鍩涢幋锔界厱婵犻潧妫楅顏呫亜閵夛箑鐏撮柡灞剧〒閳ь剨缍嗛崑鍛暦鐏炵偓鍙忓┑鐘插暞閵囨繄鈧娲﹂崑濠傜暦閻旂厧鍨傛い鎰癁閸ャ劉鎷洪梺鍛婄☉閿曘儵鍩涢幇鐗堢厽婵°倕鍟埢鍫燁殽閻愭彃鏆i柡浣规崌閹晠鎼归锝囧建闂傚倷绀侀幉鈥趁洪敃鍌氱婵炲棙鎸婚崑鐔访归悡搴f憼闁抽攱鍨垮濠氬醇閻旀亽鈧帞绱掗悩鍐插摵闁哄本鐩獮妯尖偓闈涙憸閻ゅ嫰姊虹拠鈥虫灀闁逞屽墯閺嬪ジ寮告惔銊︾厵闂侇叏绠戦弸銈嗐亜閺冣偓濞叉ḿ鎹㈠┑瀣潊闁挎繂妫涢妴鎰渻閵堝棗鐏ユ俊顐g〒閸掓帡宕奸妷銉у姦濡炪倖甯掔€氼參宕愰崹顐ょ闁割偅绻勬禒銏$箾閸涱厾效闁哄矉绻濋崺鈧い鎺戝绾偓闂佺粯鍨靛Λ妤€鈻撻锔解拺闁告稑锕ユ径鍕煕鐎n偄娴€规洏鍎抽埀顒婄秵閸犳鎮¢弴銏$厸闁搞儯鍎辨俊鍏碱殽閻愮摲鎴炵┍婵犲洤鐭楀璺猴功娴煎苯鈹戦纭锋敾婵$偠妫勯悾鐑筋敃閿曗偓缁€瀣亜閹邦喖鏋庡ù婊勫劤闇夐柣妯烘▕閸庢粎绱撳鍡欏ⅹ妞ゎ叀娉曢幑鍕倻濡粯瀚抽梻浣呵圭换鎴犲垝閹捐钃熸繛鎴欏焺閺佸啴鏌ㄥ┑鍡橆棤妞わ负鍔戝娲传閸曨剙顎涢梺鍛婃尵閸犳牠鐛崘顭戞建闁逞屽墴楠炲啫鈻庨幘鎼濠电偞鍨堕〃鍛此夊杈╃=闁稿本鐟ㄩ崗灞解攽椤旂偓鏆╅柡渚囧櫍閸ㄩ箖骞囨担鍦▉濠电姷鏁告慨鐢告嚌妤e啯鍊峰┑鐘叉处閻撱儲绻濋棃娑欘棡闁革絿枪椤法鎲撮崟顒傤槹濠殿喖锕ュ浠嬪箠閿熺姴围闁告侗鍠氶埀顒佸劤閳规垿鎮欓幓鎺旈獓闂佹悶鍔屽ḿ锟犵嵁婵犲伣鏃堝礃閳轰胶锛忛梺鑽ゅ仦缁嬪牓宕滃┑瀣€跺〒姘e亾婵﹨娅e☉鐢稿川椤斿吋閿梻鍌氬€哥€氼剛鈧碍婢橀悾鐑藉即閵忕姷顓洪梺鎸庢濡嫰鍩€椤掑倹鏆柡灞诲妼閳规垿宕卞☉鎵佸亾濡や緡娈介柣鎰缂傛氨绱掓潏銊ユ诞闁诡喒鏅涢悾鐑藉炊瑜夐幏浼存⒒娴e憡鎯堝璺烘喘瀹曟粌鈹戦崱鈺佹闂佸憡娲﹂崑鈧俊鎻掔墛缁绘盯宕卞Δ浣侯洶婵炲銆嬮幏锟�