Short-Term Explicit Prediction (STEP) Program

Background

The Short-Term Explicit Prediction (STEP) Program is a multi-NCAR Laboratory activity with the overarching goal to improve the short-term (0-36 hours) forecasting of high-impact weather events such as severe thunderstorms (heavy rain, tornados, downburst, flash flood, lightning and hail), winter storms (snow, freezing rain and drizzle), and hurricanes. The STEP program emphasizes several research areas that are crucial for advancing the science and application of the short-term prediction of high-impact weather, through collaborative effort incorporating national and international scientists, engineers, and operational personnel from universities, government institutions and the private sector. Most of the forecasting/nowcasting systems and analysis tools developed under STEP are available to the communities for the support of research and real-time operations.

In FY13, STEP focused on two major research themes: (1) the development of an end-to-end hydrometeorological system to improve flash flood prediction; and (2) observational field programs that enhance our understanding and prediction of convective precipitation.

RAL led the effort in the first theme and played important roles in the second theme. The development of the end-to-end hydrometeorological system required the collaboration of scientists and engineers with different skills in several areas including quantitative precipitation estimation (QPE), nowcasting, data assimilation and numerical weather prediction (NWP), verification, and hydrological prediction. Through an integrated effort of studying 10 historical Front Range flash flood cases, various components of the end-to-end system were improved and major accomplishments were achieved toward the ultimate goal of developing a fully coupled end-to-end hydrometeorological system. These accomplishments are summarized below.

Major accomplishments

Nowcasting Convective Precipitation over Complex Terrain

Figure 1.  Left panel: Radar reflectivity image from 8 August 2008 at 00:44 UTC, one hour prior to the flash flood over Denver Colorado, with TITAN automated storm detections (blue polygons) and 60 min storm extrapolation nowcasts (magenta polygons) overlaid.  Right panel: Radar reflectivity image at 01:44 UTC, at the time of the flood illustrating the storm initiation and intensification that occurred and was not nowcast by the TITAN algorithm.
Figure 1.  Left panel: Radar reflectivity image from 8 August 2008 at 00:44 UTC, one hour prior to the flash flood over Denver Colorado, with TITAN automated storm detections (blue polygons) and 60 min storm extrapolation nowcasts (magenta polygons) overlaid.  Right panel: Radar reflectivity image at 01:44 UTC, at the time of the flood illustrating the storm initiation and intensification that occurred and was not nowcast by the TITAN algorithm.

The over-arching goal of this STEP project is to provide accurate forecasts of precipitation amount and to improve flash flood prediction. Thus, under STEP funding RAL, MMM and EOL, have been conducting research to develop a QPE, heavy rainfall and streamflow prediction system for the Rocky Mountain Front Range using , nowcasting techniques and numerical forecast models.  Ten historical Colorado flash flood events from 2008-2012 are being analyzed and data run through the components of an end-to-end prediction system to assess the performance of these capabilities for predicting heavy rainfall, flash floods and stream flow. In partnership with Colorado State University’s CHILL radar staff, Front Range Observational Network Testbed (FRONT) data was collected on four additional heavy rainfall and flash flood events that occurred along the Front Range during August 2013.  These fourteen cases are being used to improve the performance of heavy rainfall and streamflow prediction algorithms.  A Denver flash flood event on 8 August 2008 was chosen as a focal point case for testing and evaluating all applications.  Research highlights and results are provided in the following paragraphs.

Given the typical small-scale and quickly-evolving nature of flash flood events particularly over complex terrain, it was no surprise that the NWP model forecasts for the ten historic cases were not able to predict flash flood occurrence and rainfall intensities on the spatio-temporal scales needed.  The NWP models were only able to provide reasonable forecasts of precipitation accumulation on the one synoptically-driven day. 

Predicting the precise location of convective heavy rainfall is, for the present, only possible on the nowcasting time scale. Nowcasting techniques that include the AutoNowcaster thunderstorm nowcasting (ANC) system, the 4-D Variational Doppler Radar Analysis System (VDRAS), and the Thunderstorm Identification, Tracking, Analysis and Nowcasting (TITAN) extrapolation software package were run on the 8 August 2008 case.  The TITAN extrapolations, shown in Figure 1, illustrate the challenges in accurately predicting flash floods even on the one-hour time scale because significant intensification and initiation of storms occur and are not captured by general extrapolation techniques. Thus prediction of heavy rainfall depends on observing and predicting mesoscale and storm-scale features. Figure 2 demonstrates how the ANC identified the collision of two convergence lines in the region of heavy rain in an area where VDRAS was indicating strong convergence.

Figure 2. White wind vectors show strong mesoscale convergence into the region of developing heavy rainfall. Wind vectors are based on VDRAS analysis. The red color area is a 60 min prediction by the AutoNowcaster where the red and purple convergence lines are nowcast to collide. The heaviest rain occurred in the western half of the collision zone.
Figure 2. White wind vectors show strong mesoscale convergence into the region of developing heavy rainfall. Wind vectors are based on VDRAS analysis. The red color area is a 60 min prediction by the AutoNowcaster where the red and purple convergence lines are nowcast to collide. The heaviest rain occurred in the western half of the collision zone.

In addition monitoring the characteristics and evolution of individual convective storms is necessary to determine which storms cause heavy local rainfall. Figure 3 shows that for the Aug 8 2008 case, monitoring not only surface convergence regions, but storm growth, storm mergers, and storm motion relative to a global mean motion are important in prediction of heavy rainfall events.

A number of preliminary studies have been initiated focusing on the unusual, extreme Front Range heavy rainfall event during September 2013; initial results were presented  at an NCAR seminar later that fall.  The critical nature of mesoscale features in determining where the heaviest rainfall occurred was again evident during this event, particularly in respect to low-level upslope winds relative to the terrain and mesoscale circulation and convergence features (see Fig. 4). Detailed examination of quantitative rainfall estimation for the 24 h period from 11-12 September 2013 heavy rainfall event has begun using different Z-R relationships and examination of the dual-polarization radar fields from the Denver operational NEXRAD radar.  A particularly interesting feature of the event was the small raindrops that occurred that indicated the event was more tropical in nature than is customary in Colorado. Note in Fig. 5 how a tropical Z-R relationship was much more accurate in estimating the rainfall compared to the typical Z-R relationship for Colorado.

High-resolution analyses based on radar observations

Figure 3. One of the predictor fields for nowcasting storms that may produce heavy rainfall.
Figure 3. One of the predictor fields for nowcasting storms that may produce heavy rainfall.

The success of nowcasting precipitation with ANC depends on the mesoscale predictor fields, which can be provided by the high-resolution analysis system, VDRAS. VDRAS is a 4D-Var (4-dimensional variational) based data assimilation system that aims at high-resolution analysis using Doppler radar observations. VDRAS was run for the 10 historical flash flood cases over a Front Range domain that is covered by the operational radars KFTG (Denver) and KCYS (Cheyenne) with a 2km horizontal resolution. The high-resolution analyses were used as predictors in ANC to demonstrate the importance of the mesoscale wind convergence obtained by VDRAS in prediction of convective initiation. Figure 6 shows the VDRAS vertical velocity at 1km above ground level (AGL) overlaid by the wind vectors at 150m AGL for the case of August 8-9, 2008. It is evident that the initiation of convection occurs in the region of high vertical velocities. The vertical cross-section clearly depicts the updraft resulting from the convergence of northerly flow associated with a cold front passage and the warm southerly flow.

VDRAS was also run for the Great Front Range Flood of September 2013. Figure 7 shows the VDRAS wind at 0130 UTC and 0300 UTC in one of the most intensive precipitation periods. It reveals the enhancement of southeasterly flow by a meso-cyclone circulation and the intensification of northeasterly flow near the Colorado-Wyoming border.

Figure 4.  Radar reflectivity and low-level surface station winds (barbs) at four different times during 11-12 September 2013 illustrating the evolution of the  low-level upslope winds relative to the terrain and mesoscale circulation and convergence features that played important roles in determining where the heaviest rainfall occurred.
Figure 4. Radar reflectivity and low-level surface station winds (barbs) at four different times during 11-12 September 2013 illustrating the evolution of the low-level upslope winds relative to the terrain and mesoscale circulation and convergence features that played important roles in determining where the heaviest rainfall occurred.

The merging of these winds in Boulder and Larimer Counties strengthens the upslope lifting. These detailed wind and convergence analyses with a frequent update cycle of ~10 minutes can provide mesoscale predictors for ANC nowcasting system and therefore have a potential to improve the quantitative precipitation nowcasting. Further in-depth studies for this flood case are underway.

Development of an Operational Flash Flood Prediction Capabilities

Operational flash flood watches and warnings today are issued based primarily on meteorological forecast criteria, rainfall rate and duration threshold exceedance or detection of rapidly rising waters.  The criteria for issuing flood watches and warnings rely on relationships between historical rainfall-runoff relationships.  While useful, such approaches can limit the lead time available to emergency responders, decision makers and affected citizenry.  Reliance on historical relationships can result in significant forecast uncertainty when watershed conditions have undergone significant change such as those induced by wildland fire or rapid urbanization.  Furthermore, reliance on past measurements of rainfall and runoff is highly problematic for areas without long records of precipitation and streamflow.  As such, flash flood forecasts in mountain front regions like the Colorado Front Range where long-term observations are sparse and where rapid landscape change is occurring possess a large degree of uncertainty.  These uncertainties stem from both the quality of the assumed underlying relationships between rainfall and streamflow as well as the lack of spatial detail from which such relationships were derived.  The end result is that many data poor regions in smaller tributary stream and river systems lack skillful flash flood forecasting capabilities. 

Figure 5. Plots of 24 h precipitation accumulation totals derived from radar reflectivity (Z) – rainfall (R) (Z-R) relationships.  Left panel:  Precipitation accumulations are underestimated using a typical NEXRAD Z-R relationship. Right panel:  Precipitation accumulations obtained from a “tropical” Z-R relationship represents more closely the actual rainfall that occurred.
Figure 5. Plots of 24 h precipitation accumulation totals derived from radar reflectivity (Z) – rainfall (R) (Z-R) relationships. Left panel: Precipitation accumulations are underestimated using a typical NEXRAD Z-R relationship. Right panel: Precipitation accumulations obtained from a “tropical” Z-R relationship represents more closely the actual rainfall that occurred.

The STEP program addresses this issue through the development and implementation of an end-to-end, physics-based (as opposed to empirically-based), flash flood prediction system.  Our work has three main components: 1) dataset collection and quality control, 2) operational system implementation and 3) product generation.  Within the first component we have developed methodologies to collect, in real-time, several different radar-based QPE products along with available surface rain gauge and stream gauge data.  Synthesis and quality-control of these products allows us to generate reasonable initial conditions for hydrological model forecasts.  Under the second component, we have implemented the WRF-Hydro modeling architecture to ingest both radar QPE products and quantitative precipitation forecast (QPF) output from numerical weather prediction models to make hydrological forecasts on a gridded, 100-meter resolution stream channel network across the Colorado Front Range.  Under the third component we have created an initial set of tools to generate forecast products from WRF-Hydro model output including predicted versus observed streamflow hydrographs and animations of streamflow forecasts for various Front Range river systems.  A prototype of this operational capability was working during the Great Flood of 2013 which dumped in excess of 15 inches of rainfall in 5 days along the Colorado Front Range.  Example forecast products generated from the WRF-Hydro system are shown in Figure 8 below.  While much work remains to assess and improve the quality of these streamflow forecasts, it is clear that the high spatial and temporal information produced by the WRF-Hydro forecast system captured several of the most severe flood waves that emerged from the Front Range foothills region and that the system has the capability to provide flood risk information with a level of spatial fidelity that is directly usable by emergency responders and the public. 

Development of an end-to-end verification capability

Figure 6.  VDRAS analyses valid at 0100UTC (upper left) and 0200UTC (upper right), showing the vertical velocity (color shade) and reflectivity greater than 30 dBZ (magenta contour), overlaid by wind vectors. The lower panel shows the vertical cross-section across the black line indicated on the upper right panel.
Figure 6. VDRAS analyses valid at 0100UTC (upper left) and 0200UTC (upper right), showing the vertical velocity (color shade) and reflectivity greater than 30 dBZ (magenta contour), overlaid by wind vectors. The lower panel shows the vertical cross-section across the black line indicated on the upper right panel.

Forecast evaluation is a critical element of the end-to-end hydrometeorological prediction system being developed within STEP.  The evaluation component of the project focuses on the assessment of the contributions of each component of the system, from QPE to quantitative precipitation nowcasting (QPN) and QPE, to forecasting streamflow.  By applying diagnostic approaches for evaluating the forecasts, it will be possible to assess the strengths and weaknesses of the particular components of the system.  In addition, overall performance information will provide guidance for applications of the system.  During 2013, initial evaluation approaches were applied to forecasts from the set of ten historical cases that were selected to use in developing the initial forecasting system, and the results were demonstrated to the project team for their assessment of which aspects of the verification were most informative.  The initial focus has been on model-based QPF from the Weather Research and Forecasting (WRF) model, for two of the extreme precipitation events included in the historical cases.

The project team selected both traditional and advanced spatial methods for the initial studies.  While traditional verification approaches provide some useful information for evaluating and comparing model versions, especially during periods of heavy precipitation, they generally suggest that the sample forecasts did not have any skill, even though they may contain some useful information.  Alternative spatial methods – the Fractions Skill Score (FSS; Roberts and Lean 2008) and the Method for Object-based Diagnostic Evaluation (MODE; Davis et al. 2009) – both provided more guidance regarding which aspects of the forecasts are providing meaningful information about performance.

Figure 7.  VDRAS wind analyses valid at 0130 UTC (upper left) and 0300 UTC (upper right) on 12 September 2013. The storms are indicated by red shades with reflectivity greater than 25 dBZ. The Front Range with topographical height greater than 2000m is shown by yellow shades.
Figure 7. VDRAS wind analyses valid at 0130 UTC (upper left) and 0300 UTC (upper right) on 12 September 2013. The storms are indicated by red shades with reflectivity greater than 25 dBZ. The Front Range with topographical height greater than 2000m is shown by yellow shades.

For example, Figure 9 shows distributions of forecast and observed precipitation for a flooding case from 2011, for two lead times (6 h and 23 h) from a baseline WRF forecast initialized at 0000 UTC.  It is apparent that the baseline WRF model forecasts had a different distribution of predicted precipitation than was observed.  In particular, the WRF predictions produced a large number of small precipitation values that were not observed.  For this case, the Stage IV precipitation analysis was used as the QPE field for the verification.  Figure 10 shows an example of a performance diagram for this case; this diagram (based on work of Roebber 2009) provides a way to visualize several different contingency table statistics simultaneously, including the Probability of Detection (POD), Success Ratio (SR, equal to one minus the False Alarm Ratio), the Critical Success Index (CSI), and frequency bias. This diagram suggests that the best performance is associated with low precipitation thresholds and depends on the lead time of the forecast; the subsets with the largest CSI values also have large frequency bias.

In contrast, Figure 11 shows results from an application of MODE, showing centroid (location) differences between matched precipitation regions in the forecast and observation fields for two lead times.  This diagnostic plot suggests that there were very different patterns of offsets between the forecast and observed storm systems at the two different lead times.  Other information that can be obtained from the MODE approach concern the forecast storm size (somewhat too small), intersection area between the predicted and observed systems (very small), and the precipitation intensity (somewhat too large on average and too small in the extreme).  Additional analyses of this case and the other historical cases are being undertaken to examine the performance of enhanced forecasts.

Improving QPF by assimilating radar observations

Figure 8.  a) Map of the Front Range flash flood prediction domain, b) Map of accumulated rainfall from 00z on 11 Sep. through 23z on Sep. 13 (colorscale ranges from 0-15 inches with warm colors having the most precipitation), c) Snapshot of streamflow values during peak runoff period around 1:15 a.m. local time on 12 Sep. (colorscale of streamflow values ranges from 0-50 cubic meters per second with warm colors having higher flow values)
Figure 8. a) Map of the Front Range flash flood prediction domain, b) Map of accumulated rainfall from 00z on 11 Sep. through 23z on Sep. 13 (colorscale ranges from 0-15 inches with warm colors having the most precipitation), c) Snapshot of streamflow values during peak runoff period around 1:15 a.m. local time on 12 Sep. (colorscale of streamflow values ranges from 0-50 cubic meters per second with warm colors having higher flow values)

One of the keys to improve qualitative precipitation forecast and hence streamflow prediction is to obtain better initial conditions by assimilating high-resolution observations such as those from Doppler radars. Considerable efforts were devoted to the improvement of WRF-based data assimilation systems so that radar observations can be effectively used in establishing the initial conditions of WRF. To support the collaborative effort of developing the end-to-end hydrometeorological system, WRF forecasts initialized by North American Model downscaling and by WRF 3DVar data assimilation without radars were produced for the 10 historical cases. A detailed case study was conducted for a Denver flash flood case that occurred on August 8-9, 2008. Radar observations were assimilated for this case with a 3km domain that covers eight NEXRADs. Figure 12 compares the 6-h (00-06 UTC) accumulated precipitations between the experiments without and with radar observations, verified by the Stage IV precipitation analysis.

One of the new developments that aimed to improve the convective-scale data assimilation and QPF was to change the momentum control variables from the original stream function and velocity potential in the WRF 3D/4DVar system to the direct west-east and south-north wind components. Experimental results with the multiple flash flood cases showed encouraging results from using the new control variables. Figure 13 compares the FSSs (Fractions Skill Scores) from two experiments that use the two sets of momentum control variables respectively. It is shown that the new control variables of u-wind and v-wind produce significantly improved precipitation forecast skills.

Improving WRF physics for improved prediction of high impact weather

Figure 9. Distributions of precipitation amounts for baseline WRF forecast grid and Stage IV observation grid, for two lead/valid time on 13 July 2011.  The center of the box represents the median value; bottom and top of the box are the 0.25th and 0.75th quantile values of the distributions, and lines inside an outside the boxes designate expected extremes (points beyond these “whiskers” are outliers).  The notches in the boxes represent 0.95th confidence intervals around the medians. Figure 9. Distributions of precipitation amounts for baseline WRF forecast grid and Stage IV observation grid, for two lead/valid time on 13 July 2011.  The center of the box represents the median value; bottom and top of the box are the 0.25th and 0.75th quantile values of the distributions, and lines inside an outside the boxes designate expected extremes (points beyond these “whiskers” are outliers).  The notches in the boxes represent 0.95th confidence intervals around the medians.
Figure 9. Distributions of precipitation amounts for baseline WRF forecast grid and Stage IV observation grid, for two lead/valid time on 13 July 2011. The center of the box represents the median value; bottom and top of the box are the 0.25th and 0.75th quantile values of the distributions, and lines inside an outside the boxes designate expected extremes (points beyond these “whiskers” are outliers). The notches in the boxes represent 0.95th confidence intervals around the medians.

a. Impact of microphysics and radar data assimilation on squall line simulation

A case study was conducted to investigate the impact of model microphysics parameterization on short-term forecasts of convective initiation, evolution, and quantitative precipitation forecasts (QPF).  In order to promote more accurate initiation of convective features in the simulations, RAL’s Real-Time Four-Dimensional Data Assimilation (RTFDDA) system that utilizes a latent heat nudging technique was used. Numerical experiments were conducted to test the radar data assimilation technique, as well as to evaluate the effect of changes in the microphysics scheme.

The event studied was a squall line observed on 20 June 2007 in central Oklahoma.  Convective activities initiated over the Great Plains in the afternoon of 19 June, leading to the development of separate mesoscale convective systems in the evening.  These systems merged, forming a squall line with a trailing stratiform region around 05 UTC of 20 June.  Idealized simulations of the case have previously shown the parameterization for raindrop breakup to have a strong influence on the evolution of organized convection via its impact on the cold pool (Morrison et al. 2012).

The RTFDDA was run on a 3-km domain, assimilated radar data of reflectivity mosaic at 00-05 UTC of 20 June, and then produced forecasts valid at 06-12 UTC.  The numerical experiments show that radar data assimilation improved the squall line forecasts consistently (Figure 14).  The simulated cold pool was better defined in the radar assimilation runs than in the no-radar assimilation runs.  However, even with radar data assimilation, the simulated cold pool was still too weak compared to surface station observations.  An examination of the raindrop sizes showed that the simulated median drop sizes were generally larger than what was observed by the disdrometer.  The microphysics scheme was then modified to increase the drop breakup efficiency.  Preliminary testing showed that the modified scheme impacted the simulated squall line in the right direction.

b. Improving the Thompson microphysical scheme

Figure 10. Performance diagram (Roebber 2009) showing multiple verification statistics for the 13 July 2011 case.  This diagram simultaneously shows POD, 1-FAR, CSI, and Frequency Bias values for various lead times and precipitation thresholds.  The different colors represent the precipitation thresholds used to define the precipitation events and individual points represent different forecast lead times.  The best scores are in the upper right corner; bias values close to one are typically considered optimal.
Figure 10. Performance diagram (Roebber 2009) showing multiple verification statistics for the 13 July 2011 case. This diagram simultaneously shows POD, 1-FAR, CSI, and Frequency Bias values for various lead times and precipitation thresholds. The different colors represent the precipitation thresholds used to define the precipitation events and individual points represent different forecast lead times. The best scores are in the upper right corner; bias values close to one are typically considered optimal.

During 2013, the Thompson et al (2008) microphysics scheme in WRF was upgraded to directly couple the radiative-effective radius of water drops and ice crystals computed directly within the microphysics scheme and passed to the radiation scheme.  This modification causes cloud characteristics to directly change radiation as occurs in the real atmosphere as compared to the inherent assumptions of particle size within the radiation scheme.  The newly coupled physics were used in one of the OU-CAPS Spring Experimental Forecast ensemble members during their real-time forecast exercise in May and early June 2013.  Analysis of 29 days of 48-hour forecasts is underway to compare the sensitivities to surface temperature and precipitation as well as cloud properties in conjunction with geostationary satellite data when using the uncoupled versus coupled physics.  A journal manuscript is being prepared to present the methodology and results.

FY2014 Plans

Examination of the historical flash flood cases will continue and each of the components in the end-to-end hydrometeorological system will be further developed in preparation for a real-time test that is planned for the mid-late summer in 2014 in the Front Range region. One of the important planned activities will be the engineering effort to couple components, creating an end-to-end system that can be operated in real time.  Beginning in the summer of FY2014, the FRONT (Front Range Observational Network Testbed) facility will begin to operate, collecting data from NCAR’s S-Pol and CSU’s CHILL dual-polarization radars, which will provide additional observations for use in this STEP project. Detailed plans for FY14 are summarized below:

Figure 11. Centroid differences from application of MODE to the 13 July 2011 precipitation forecasts, for lead times of 6 and 23 h.   The diagram depicts both the direction and magnitude of the forecast displacement from the observed storm area.  Individual points represent different definitions of objects by MODE, with point sizes representing different radius values used to define the smoothing to be applied, and colors representing the threshold value used to define the extent of the precipitation area included in the final MODE objects.
Figure 11. Centroid differences from application of MODE to the 13 July 2011 precipitation forecasts, for lead times of 6 and 23 h. The diagram depicts both the direction and magnitude of the forecast displacement from the observed storm area. Individual points represent different definitions of objects by MODE, with point sizes representing different radius values used to define the smoothing to be applied, and colors representing the threshold value used to define the extent of the precipitation area included in the final MODE objects.

QPE and nowcasting

  • Document and evaluate the different Z-R relationships for QPE and conduct research to automatically determine the appropriate Z-R relationship to use for the meteorological situation
  • Evaluate the performance of the nowcasting techniques for prediction of heavy rainfall for several of the 14 historic flood events
  • Evaluate the performance of the nowcasting techniques for prediction of heavy rainfall during the great Front Range floods (10-15 September)
  • Contribute to a BAMS article on the Front Range floods
  • Run the QPE and nowcasting techniques in real-time during July-August 2014 using the FRONT datasets and evaluate performance

VDRAS high-resolution analysis

  • Improve the analysis by considering the observation time at each radar beam
  • Test the new version of VDRAS with the terrain effect and evaluate its impact
  • Implement VDRAS on a new cluster machine
  • Test a configuration for a summer 2014 real-time implementation

Flash flood prediction

Figure 12. 6-h  accumulated precipitation observations (00-06 UTC, August 9, 2013) from the Stage IV analyses (left), WRF forecasts valid at  the same time period without radar data assimilation (middle), and WRF forecast with radar data assimilation (right).
Figure 12. 6-h accumulated precipitation observations (00-06 UTC, August 9, 2013) from the Stage IV analyses (left), WRF forecasts valid at the same time period without radar data assimilation (middle), and WRF forecast with radar data assimilation (right).
  • Improve the groundwater and channel flow physics components in the WRF-Hydro system to improve the recession characteristics of flood events
  • Increase the number and type of QPE and QPF products ingested into the WRF-Hydro system with the goal of developing more probabilistic forecasts of streamflow and flooding threat
  • Continue evaluation and use of different polarimetric QPE products in the creation of hydrologic forecasts
  • Implement a new methodology for incorporating high-time frequency nowcasts of precipitation into the WRF-Hydro system
  • Work with emergency managers and local forecasters to improve the quantity and quality of forecast products for dissemination

Verification

Figure 13. Comparison of Fractions Skill Scores (FSS) with a radius of influence of 10 km, between two forecast experiments with stream function and velocity potential (blue line) and u-wind and v-wind as momentum control variables. The FSSs are computed over 29 forecasts conducted for 7 of the 10 historical flash flood cases.
Figure 13. Comparison of Fractions Skill Scores (FSS) with a radius of influence of 10 km, between two forecast experiments with stream function and velocity potential (blue line) and u-wind and v-wind as momentum control variables. The FSSs are computed over 29 forecasts conducted for 7 of the 10 historical flash flood cases.
  • Continue to examine and enhance the spatial verification methods to be utilized in evaluations of forecasts produced in summer 2014
  • Implement a near-real-time evaluation system that will produce verification information on a regular basis, as well as for important events, during summer 2014
  • Consider the use of probabilistic QPE fields as a way to take into account important aspects of observation uncertainty in the verification. 
  • Include capabilities to evaluate the streamflow forecasts in the summer 2014 evaluations.

WRF-based Radar data assimilation

  • Continue improving the capability of WRF 3D/4DVar for radar data assimilation
  • Test an update cycle strategy that is suitable for the Front Range area
  • Configure and install 3D/4DVar on Yellowstone to get ready for real-time execution during the summer

WRF physics improvement

Figure 14.  Hourly rainfall from RTFDDA forecasts with no radar data assimilation (left column) and with radar data assimilation (middle column), and Stage II estimates (right column).
Figure 14. Hourly rainfall from RTFDDA forecasts with no radar data assimilation (left column) and with radar data assimilation (middle column), and Stage II estimates (right column).
  • Refine the hybrid radar data assimilation system with improved mixing ratio relationships to improve forecasted convective initiation and evolution
  • Configure and install RTFDDA on Yellowstone to get ready for real-time execetion during the summer
  • Develop a 3-moment graupel/hail hybrid category in the Thompson microphysics scheme to improve forecasted convective storm structure, evolution, and QPF.