Developing Applications for Nowcasting Heavy Rainfall over Complex Terrain

Transcription

1 Developing Applications for Nowcasting Heavy Rainfall over Complex Terrain Rita Roberts 1, Juanzhen Sun 1, Eric Nelson 1, James Wilson 1, Fu Tien Tsai 2 1 National Center for Atmospheric Research, 3450 Mitchell Lane, Boulder Colorado, U.S.A, rroberts@ucar.edu 2 Central Weather Bureau, Forecast Meteorology Center, Taipei, Taiwan, fred@mfcsv.cwb.gov.tw Rita Roberts 1. Introduction Taiwan is extremely vulnerable to flash floods resulting from heavy rain produced by local thunderstorms and by a spectrum of large-scale synoptically forced weather. Taiwan Central Weather Bureau (CWB) forecasters are being tasked in the coming years to issue frequently updated, short-term forecasts of heavy precipitation for all of their 368 townships that range in size from km 2 along the coastal plains to >100 km 2 over their mountainous regions. A particular challenge in accurately predicting the initiation, precipitation intensity and duration of storms in these regions is correctly accounting for the impact of the terrain (both mountain and coastal regions) on storm formation and evolution. The Terrain-influenced Monsoon Rainfall Experiment (TiMREX) conducted in 2008 during Taiwan s warm season gathered high resolution dual-polarization radar, surface, sounding and special vertical profiling measurements prior to and during heavy rainfall events. These high resolution observations are being used to examine storm formation and evolution, identify critical nowcast predictor fields, and develop real-time applications for nowcasting heavy rainfall over complex terrain. In this paper we document the first steps in identification of robust predictor fields for Taiwan using a TiMREX heavy rainfall case from 31 May Understanding the forcing mechanisms 2.1 Overview of 31 May weather On 31 May, the Mei-Yu front moved from northern to southern Taiwan during the afternoon. This synoptic front typically occurs during Taiwan s warm season months of May and June and is characterized by mesoscale convective systems (MCS) that form along its ENE-WSW oriented convergence boundary. These MCSs tend to form over the Taiwan Strait and South China Sea and track eastward, bringing heavy and torrential rains to the plains and western slopes of Taiwan and can cause significant flooding. Figure 1 shows the track of the Mei-Yu front (black contour) on 31 May, locations of surface convergence boundaries (dashed black contours), the direction of the onshore low-level wind (yellow vector) in southern Taiwan and storm locations (red circles) overlaid onto the terrain. Evident in all panels is the ENE-WSE line of storms triggered by the Mei-Yu front. Ahead (south) of the front, locally-forced convection developed in southern Taiwan resulting in heavy precipitation along the west slopes of the Central Mountain Range (CMR).The first wave of locally-driven convection initiates well ahead of the front over a small 300 m ridge (Fig. 1b). These storms move eastward and intensify as they reach the slopes of the CMR (Fig. 1c). Outflows from these storms move to the SW creating a convergence zone with the onshore southwesterly flow and several new storms are initiated (Fig. 1d). One hour later (Fig. 1e), as the front approaches the 300 m ridge north of Kaoshiung, a NE-SW line of storms oriented parallel to the ridge rapidly develops. The interaction of very small scale outflows from the resulting storms collide eventually generating one large merged cell (Fig. 1f) that produced heavy rain in the area from UTC ( LT). 2.2 Numerical weather model prediction of precipitation and instability The CWB forecast office runs several different Numerical Weather Predication (NWP) models. The heavy rainfall events were well predicted in terms of the synoptic environment by the CWB high resolution global and mesoscale models (e.g, NFS-CWB, UK, JMA, and LAPS-WRF) several days in advance of 31 May. However the more specific prediction of rainfall location, timing and intensity left a lot of room for improvement. The LAPS- WRF and CWB-WRK, high spatial resolution, mesoscale models, showed some improvement over the other models but spatial offsets in location and timing were still observed. To meet the objective of providing very short-term forecasts of heavy rainfall for the townships, greater improvement will need to be realized in the mesoscale models; this challenge is not unique to Taiwan alone.

2 The NWP models can produce forecast output fields of Convective Available Potential Energy (CAPE) and Convection Inhibition (CIN). These fields give an indication of the instability of the atmosphere. A 30 hr forecast of CAPE valid the afternoon of 31 May showed the regions for highest convective instability were located in southwestern Taiwan. This is climatologically a favored region for heavy orographic precipitation to occur as a Low- Level Jet (LLJ) is frequently observed on the warm (south) side of the Mei-Yu front transporting moisture into the frontal zone and also acts to destabilize FIG. 1. See text for explanation. the atmosphere (Chen and Yu 1988; Chen et al. 2005). Regions with low CIN values indicate that there is little negative buoyancy to overcome for initiation of convection. In sub-tropical Taiwan, however, the atmospheric CIN is frequently low in magnitude and thus is not a useful discriminator or predictor of where new convection may occur. In addition, due to Taiwan s complex terrain and lack of surface station observations in those areas, it is difficult for the models to produce reasonable estimates of instability over the higher elevations. This is problematic, particularly if one wishes to make use of CAPE information with wind speed to categorize flow regimes associated with conditionally unstable flow over mountain tops. 2.4 Boundary layer winds and moisture Figure 2 shows the surface station networks and operational radar locations. There are 155 stations that report hourly and 41 stations that provide 10 min updates. During TiMREX at the S-Pol site, we were only able to view 25 of the stations in real-time. This was an insufficient number of observations for correlating wind direction and speed with storm initiation over higher terrain. Stations in southwestern Taiwan were spaced too far apart to track boundary layer features, such as convergence boundaries, although those stations located along the coast helped provide supplemental information on the development of sea breezes. Taiwan s four operational radars are similar to

3 the U.S. WSR-88D radars and collect data using VCP surveillance modes. These radars have been used primarily for rainfall estimation and producing radar reflectivity mosaics for tracking storms. These radars are not currently useful for tracking boundary layer features in the clear air, as the clear air return is significantly impacted by ground clutter contamination arising from the variable terrain. The current methods for filtering of the ground clutter are not as robust as the filtering done using the Clutter Mitigation Decision System (Dixon et al 2006) on the NCAR S-Pol radar, which not only removes ground clutter but retains as much of the clear air signal as possible. During TiMREX, the NCAR S-Pol radar was operated near Pingtung, and was sited about half way between RCCG and RCKT radars in Fig. 2. The radar was situated to collect high resolution, dual-polarization information over southwestern Taiwan and immediately to the north-northeast of the radar along the Gao-Ping Xi Valley and the foothills of the CMR. A mixture of surveillance, sector and RHI scanning was collected to document storm characteristics and evolution. The clear air radar return from S-Pol was extremely important for documenting boundary layer winds and convergence features and FIG. 2. Locations of Taiwan surface station networks and the four operational radars. Radar locations are labeled in black. Stations collecting data at hourly (blue cross-hairs) and 10 min (magenta open circle) update frequencies are shown. the role of these features in convection initiation. Reflectivity and velocity information from S-Pol are being assimilated into a 4DVAR system to retrieve high resolution, boundary layer wind fields; this will be discussed in more detail in section 3. Radar-derived refractivity fields were produced from S-Pol using an index of refraction (refractivity) technique developed by Fabry et al. (1997). This high-resolution moisture field will be studied to understand the near-surface variability of water vapor leading up to convection initiation. The retrieved refractivity fields only extend out to about 30 km in range from S-Pol, which will likely be a limiting factor in using this information in a Taiwan nowcasting system. 2.5 Storm attributes and evolution Figure 3 shows radar attributes of one of the storms on 3 May at 08:45 UTC. RHIs taken through the storm (Fig. 3) illustrate the deep (16 km) convection (Fig. 3b), strong implied vertical motion associated with the low-level convergence and upper level divergence (Fig. 3c), and the presence of large precipitation drops (Fig. 3d). The Particle ID field (PID; Fig. 3e) derived from the S-Pol dual-polarization fields indicate the presence of grauple above freezing level. The total accumulated rainfall over southern Taiwan during the afternoon ranged from 50 mm on the plains to > 100 mm along the mountain slopes, as recorded by surface station rain gauges. A comparison of these measurements with the KDP-derived precipitation rate field is shown in Fig. 4a. An RHI cross-section of KDP through the storm at the same azimuth as in Fig. 3 illustrates the 6 km deep coherent structure of the heaviest precipitation shaft (Fig. 4b). The dual-polarization fields will be used for discriminating between ordinary storms and storms likely to produce intense rainfall and addressing storm dissipation. It is important to be able to nowcast when a heavy rainstorm when end. Satellite visible imagery can provide useful short-term predictor information on storm growth through the use of algorithms which automatically classify cloud types and monitor of cloud top infrared temperatures with time

4 (Roberts and Rutledge 2003). These algorithms are particularly effective at detecting growing cumulus clouds if the satellite imagery is updated every 15 min. Taiwan has access to both the Japanese and Chinese (FY-2C and FY-2D) satellite data sets. The visible and infrared imagery are collected by the three satellites every 30 min with an added latency of 30 min before the imagery arrives at the CWB. It is unclear at this time how useful these fields will be in the nowcasting system with the data being almost an hour old before it can be ingested into nowcast algorithms. 3. Identifying predictor fields and developing nowcasting applications Until the NWP models can provide specific, highly reliable, accurate forecasts of precipitation occurrence on 1-5 km scale and on the 0-2 hr nowcast timescale, the approach taken here is to use an heuristic nowcast system that employs fuzzy logic to combine all available observations with selected numerical model output fields to provide location-specific forecasts of heavy rain. The NCAR Auto-Nowcaster system (Mueller et al 2003) is used as the framework and the fuzzy engine for developing a heavy rainfall nowcasting system for CWB. The challenge is to identify those predictor fields applicable to Taiwan s different weather regimes. In the following sections we discuss preliminary efforts to identify and produce necessary fields for the nowcasting system, along with making use of existing applications that are relevant to the Taiwan nowcasting challenge. a) b) c) d) e) FIG. 3. Storm attributes associated with heavy rainfall event on 31 May at 0845 UTC observed by the S-Pol radar. a) Radar reflectivity sector scan with RHI cross-section location shown by the yellow line. b) RHI of radar reflectivity; c) RHI of radial velocity; d) RHI of ZDR; e) RHI of dual-polarization derived particle identifications.

5 a) b) FIG. 4. a) S-Pol precipitation rate derived from KDP measurements on 31 May at 0845 UTC; b) RHI of KDP through the heavy rainfall event at the same time. 3.1 Documentation of weather regimes One of the first steps in developing a Taiwan nowcasting system is documenting the most important and frequent weather regimes that lead to heavy rainfall events over the island. Who is better suited to do this than the forecasters from the CWB who have developed conceptual models for each regime built on their scientific knowledge and operational experience. Checklists of attributes for each weather regime, similar to the checklist that the CWB forecasters use for determining if severe weather is going to occur, are now being documented by a couple of forecasters under the direction and guidance of one of the top forecasters in the CWB. Key attributes for each regime will likely include prevailing wind direction at low and mid-levels, significant thresholds for stability indices, synoptic and mesoscale characteristics of the NWP pressure and height fields, and characteristic moisture profiles, just to name a few. Once these regimes are defined and example cases for each regime are identified, then membership functions and weights can be assigned to create a unique fuzzy logic set of the predictor fields for each weather regime. 3.2 Importance of the 3-D wind fields and near-surface convergence as predictors The complex interaction of boundary layer and mid-level winds with the terrain is a determining factor on where new convection will occur. Several case studies (too numerous to cite here) have been conducted by Taiwan scientists on the relationship of wind speed and direction and location of heavy rainfall. A new study has examined 4 years of Taiwan radar and lightning data and compared these results with the hourly averages of surface wind direction on days of weak synoptic scale forcing (Lin et al 2010). However, none of this research has been translated into a real-time nowcasting application and specific predictor fields have not been identified. Real-time applications need to be developed based on the correlations of wind speed and direction (at various layers in the lower troposphere) with the spatial locations of rainfall, analogous to the recent study and prescription for the development of a heuristic nowcast system described by Panziera and Germann (2010) for nowcasting orographic precipitation. Surface station and twice-daily radiosonde wind profiles are not sufficient to represent the 3-D flow, however. Taiwan has one profiler located on the National Taiwan University campus, but this is not usually available for routine forecasting operations. CWB scientists are currently combining data from pairs of operational radars to produce real-time dual-doppler winds; this is a work in progress. One of the most promising methods being explored is the use of the NCAR 4-D Variational Doppler Radar Analysis System (VDRAS; Sun and Crook, 1997) for providing the 3-D wind fields needed. It assimilates radar reflectivity and radial velocity information into a numerical cloud model to produce high spatial (5 km) and temporal (10 min) resolution winds and temperature information over the depth of the boundary layer. The VDRAS system has been developed and used for radar data assimilation research and real-time applications for the past 20 years. VDRAS winds and temperature gradient fields have been produced for 31 May using S-Pol radar data and are shown in Fig. 5. Two time periods are shown prior to (0715 UTC) and during (0814 UTC) the heavy rainfall event that was highlighted in Fig. 4. The position of the Mei-Yu front during this period is located approximately in the SW corner

6 of the panels. South-southwesterly flow ahead of (south) of the front is evident, bringing moisture inland and causing an appreciable gradient in temperature along the western slopes of the mountains along with a pronounced convergence in the flow (Fig. 5a). One hour later (Fig. 5b), in the same vicinity of the large temperature gradient in Fig. 5a, a new storm has developed. As can be seen in Fig. 1, this storm produces an outflow that increases the convergence in the low levels and additional storm initiation and storm merger results, resulting in a heavy rainfall event for this area. A comparison of the VDRAS winds with the WRF winds at 0715 UTC in Fig. 5c illustrates a significant discrepancy in the direction of the winds. The WRF model fails to capture the southerly flow over southern Taiwan and therefore misses the strong convergence of flow that sets up between the Mei-Yu front, the onshore flow and local storm outflows. The predominantly westerly flow throughout the rest of the domain observed in the WRF winds would likely have caused a nowcast system to predict many more storms along the slopes of the mountains than actually existed. 3.3 Monitoring changes in atmospheric stability It is important to know that the atmosphere is conditionally unstable and general guidance can be provided by the NWP forecasts of CAPE. However these fields tend to cover very broad areas. Two applications are being tested for their applicability in detecting changes in atmospheric stability over time. First, CAPE and CIN fields are now being generated within VDRAS. The second application is a software algorithm called metarcapecin that uses hourly surface station data to update the surface temperature and dewpoint values in the sounding that were launched every 3 h during TiMREX. The algorithm then makes new calculations of CAPE and CIN and produces hourly, gridded fields of CAPE and CIN over the Taiwan domain. CAPE and CIN fields produced by VDRAS and metarcapecin are shown in Fig. 6 at 0300 UTC (1100 LT), at 0600 UTC (1400 LT) and at 0900 (1700 LT). The overall pattern of the instability is similar between the two sets of fields. The regions of larger instability are concentrated mostly in southern Taiwan along the west and east coasts where this is plenty of available moisture at all level (e.g. at Tainan). This is in agreement with what was observed in the CWB-WRF model output discussed earlier. The magnitudes of CAPE from the metarcapecin algorithm are higher than the VDRAS CAPE values and correspond to the magnitudes computed for the unmodified 3 hr soundings. The WRF model fields are used as background information in the VDRAS system and the CAPE fields are derived from that output. Observations are interpolated and smoothed to the coarser grid resolution of the model and thus these calculated values of instability tend to be lower in magnitude. Some of the most intense convection of the day occurs between 0600 and 0900 UTC in southwestern Taiwan. However the CAPE values all panels during this time period are modest in magnitude and generally decreasing some during the afternoon. The main concern noted by the TiMREX scientific steering committee on 31 May was the presence of a relative dry (subsidence) layer at 500 hpa in the 0300 and 0600 UTC soundings at Pingtung and Liou-Guei (located up the Gao-Ping Xi valley, right next to the mountains) and whether convection would break through this dry layer. As observed in the temperature gradient/wind fields from VDRAS in Fig. 5, the strong convergence and larger gradient in buoyancy that set up along the western slopes was sufficient to enable explosive growth of convection (up to 16 km AGL) to take off. The dry layer at 500 hpa is completely gone in the 0900 soundings taken at Pingtung and Liou-Guei. It is not clear yet if either of the two versions of CAPE shown in Fig. 6 provide enough additional information to be useful as stand-alone predictor fields for heavy rainfall as neither showed any significant changes that would indicate convection was more likely to occur in one area versus another. As observed with the Auto-Nowcaster system run in the U.S., the CAPE fields are best used in conjunction with other predictor fields. 3.4 Storm evolution and tracking Taiwan uses a program called QPESUMS to produce radar reflectivity, precipitation rate and accumulation mosaics and storm tracks. This product is updated every ten minutes and is important information to include in a heavy rain nowcasting system. We are also utilizing the Thunderstorm Identification Tracking, Analysis and Nowcasting (TITAN; Dixon and Weiner) algorithm that provides additional predictor information including storm initiation locations and diurnal occurrence (see Fig. 7 for 31 May), frequency of storm occurrence (climatology statistics), and tracks changes in storm area and growth of storms. In the absence or blockage of radar coverage over the mountains, the total lightning data collected during TiMREX will be explored for filling in the gaps in coverage. A nowcasting application (algorithm) already exists to intelligently that combines lightning and radar data together in a gridded field for this exact purpose.

7 a) b) c) FIG. 5. VDRAS winds (yellow vectors) at km AGL overlaid onto a) temperature gradient field at 0715 UTC, b) temperature gradient field at 0814 UTC, and c) terrain and WRF winds (magenta vectors) at 0715 UTC. Black contours represent the storms at 25 and 35 dbz increments. Outline of Taiwan is in gray and mountainous regions are outlined in cyan.

8 a) b) c) d) e) f) FIG. 6. CAPE fields for 31 May 2008 shown at 3 hr intervals. CAPE produced from the metarcapecin algorithm at a) 0300 UTC, b) 0600 UTC, and c) 0900 UTC. CAPE produced from VDRAS at d) 0300 UTC, e) 0600 UTC, and f) 0900 UTC. 4. Summary An initial start has been made to understand the different weather scenarios that result in heavy precipitation and to test the ability of existing capabilities to identify important precursor information for thunderstorm nowcasting in Taiwan. Collaborations are underway with Taiwan scientists and forecasters to transfer research results into the framework of an heuristic nowcasting system. The 31 May case was used here to illustrate some of the steps in the process of building such a nowcast system. Much research has been done on the Mei-Yu front and its convective triggering mechanisms are likely better understood than other synoptic regimes. The complicating factor for specific prediction of heavy rainfall particularly for the smaller townships arises when convective forcing is occurring on both the local and large scales. In the near future we will be concentrating efforts on documenting robust predictors of convection on TiMREX days with minimal to no synoptic-scale forcing. Acknowledgment We would like to thank the Taiwan Central Weather Bureau for this support of this work. We also acknowledge Tracy Emerson for data processing and creation of Figure 7.

9 FIG. 7. a) Storm initiation locations on 31 May from the TITAN algorithm overlaid onto the terrain and b) percentage of occurrence of storm initiation versus time of day for 31 May. References Chen, G. T.-J. and Yu, C.-C., 1988: Study of low-level jet and extremely heavy rainfall over northern Taiwan during the Mei-Yu season. Mon. Wea. Rev., 116, Chen, G. T.-J., Wang, C.-C., and Lin, D. T.-W., 2005: Characteristics of low-level jets over northern Taiwan in Mei- Yu season and their relationship to heavy rainfall events. Mon. Wea. Rev., 133, Dixon, M., Kessinger, C., Hubber, J., 2006: Echo classification and spectral processing for the discrimination of clutter from weather. Preprints, 32 nd Conf. on Radar Meteor., Amer. Meteor. Soc., Albuquerque, NM. Fabry, F., Frush, C., Zawadzki, I., and Kilambi, A., 1997: On the extraction of near-surface index of refraction using radar phase measurements from ground targets. J. Atmos. Oceanic Technol., 14, Lin, P.-F., Chang, P.-L., Jou, B. J.-D., Wilson, J., and Roberts, R., 2010: Warm season afternoon thunderstorm characteristics under weak synoptic-scale forcing over Taiwan. In review, Wea. Forecasting. Mueller, C. K., Saxen, T., Roberts, R., Wilson, J., Betancourt, T., Dettling, S., Oien, N., and Yee, J., 2003: NCAR Auto-Nowcast system. Wea. Forecasting, 18, Panziera, L., and Germann, U., 2010: The relation between airflow and orographic precipitation on the southern side of the Alps as revealed by weather radar. Quart. J. Roy. Meteor. Soc., 136, Roberts, R. D., and Rutledge, S., 2003: Nowcasting storm initiation and growth using GOES-8 and WSR-88D data. Wea. Forecasting, 18, Sun, J., and Crook, N. A., 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sci., 54,