Short-Term Explicit Prediction

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 efforts with national and international scientists and 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 FY17, RAL’s STEP effort continued to emphasize the research and real-time demonstration of the integrated Hydromet Prediction System (Figure 1) (STEP-Hydromet hereafter). The overarching objective of this effort is to advance the prediction of heavy rainfall, flash floods and streamflow through the integration of state-of-the-art rainfall estimation, precipitation forecasting/nowcasting, and hydrology modeling techniques into one seamless system. The major objectives this year were to conduct the analysis and evaluation of the real-time data collected during the 2016 STEP Hydromet Experiment and the third real-time Hydromet Experiment, conducted from 1 June – 15 August 2017 along the Colorado Front Range. Additionally, RAL continued to lead the STEP research theme on the improvement of the WRF microphysics parameterization scheme as in previous years, and STEP staff participated in preliminary analyses of the NSF-sponsored PECAN (see http://pecan15.org) experiment datasets and predictability of nocturnal, elevated convection initiation.

 

Figure 1.   Flowchart for the STEP Hydromet Prediction System.
Figure 1.   Flowchart for the STEP Hydromet Prediction System.

1. Development and Demonstration of STEP-Hydromet

The STEP-Hydromet program is focused on providing to predictions on the time scale of minutes out to one day, with particular emphasis on 0-12 hour forecasts and 0-1 hour nowcasts on very high-resolution spatial grids (from 100 m – 3 km in resolution). In addition to active research activities that support the development of the hydromet system, the integrated system has been demonstrated in real-time using the Colorado Front Range as a testbed since 2014. All components of STEP-Hydromet are run in real time in an integrated fashion. The components included in the fully integrated system are:  1) radar based quantitative precipitation estimation (QPE) and rain gauge QPE, 2) quantitative precipitation nowcasting (QPN) from 10 min to 1 h from the heuristic-based Autonowcaster/Trident system. High-resolution winds and buoyancy analyses from VDRAS are also produced. 3) quantitative precipitation forecasts (QPF) from the WRF 3DVar NWP models with radar data assimilation and frequent update cycles, 4) streamflow prediction on a spatially-continuous 100 m resolution grid, from the WRF-Hydro hydrology model, and 5) near-real time performance evaluation of the QPE and QPF fields using a set of statistical metrics and techniques.

The numerical weather prediction (NWP) forecast models and the WRF-Hydro model have been run on NCAR’s Yellowstone supercomputer, while the AutoNowcaster, Trident and VDRAS nowcasting systems, and the EOL QPE mosaics ran on workstations located in RAL and EOL.  A dedicated web page

(https://ral.ucar.edu/projects/step_hydromet)

was set up for real-time viewing of precipitation accumulation fields, forecast and nowcast products, and streamflow prediction.  The JAZZ interactive java-based display system was also set up for viewing all the real-time observations (radar, satellite, surface station data), STEP nowcasts and NWP model forecast fields and was easily accessible on any workstation by clicking on a link on the dedicated STEP web site. Real-time streamflow, soil moisture and overland flow depth information was displayed on a dynamic web mapping service called HydroInspector, also developed as part of the STEP project. Upgrades and enhancements were made to the components of the Hydromet system prior to the start of the demonstration based on research conducted throughout the year. Ongoing research efforts aimed at improving the different components of this system are discussed below.

Quantitative Precipitation Estimation (QPE)

Figure 2. One-hour radar rainfall accumulations derived from the NCAR EOL Hybrid PID algorithm.  The centers of the black circles indicate the location of surface station rain gauges and the corresponding 1 h rainfall accumulations measured by these gauges are overlaid onto the image.
Figure 2. One-hour radar rainfall accumulations derived from the NCAR EOL Hybrid PID algorithm.  The centers of the black circles indicate the location of surface station rain gauges and the corresponding 1 h rainfall accumulations measured by these gauges are overlaid onto the image. 

Previous analysis of the QPE accuracy in the Colorado Front range using the 2015 data reported that improved results were obtained by using radar polarimetric variables to identify precipitation type and then applying specific radar rainfall (Z-R) equations for each precipitation type. However, overestimates of precipitation were common in hailstorms. Based on results from 2015 it was decided to use the NCAR/EOL (dual polarimetric) Hybrid PID algorithm that includes a revised R-KDP rainfall relationship  developed by Alexander Ryzhkov of NSSL.  Presently rainfall estimates from the summers of 2016 and 2017 using this technique are being compared with rain gauge estimates from dense rain gauge networks along the Front Range. Initial results are promising as shown in Fig. 2.  There is good agreement between the EOL radar-based QPE of 1 h precipitation accumulations and surface station rain gauge 1 h accumulation measurements.

Quantitative Precipitation Nowcasting (QPN)

It has been repeatedly shown that nowcasting is greatly improved when knowledge of boundary layer convergence lines are part of the nowcasting technique. Automated methods to detect these convergence lines have been sufficiently unreliable to utilize in the AutoNowcaster (ANC)/Trident system. The highest priority activity this year has been the efforts toward producing accurate and reliable surface convergence fields. Two approaches were taken. First, the automated boundary detection algorithm called COLIDE was updated to make use of dual-polarization radar data and the EOL dual-polarization Particle Identification (PID) algorithm. The PID field was used as a first step in data processing to mask out regions of existing storms that can occasionally interfere with the detection of convergence boundaries which have a linear shape on radar. This step enabled COLIDE to be run on only clear air features evident in the PBL, and a much improved automated detection was immediately evident. An elliptical filter was then applied to the COLIDE detections to determine convergence boundary orientations which will then be used for tracking boundary motions and extrapolating their positions in time. In the past, COLIDE produced symbolic products. This year it was modified to produce gridded output fields which will be easy to ingest into the ANC/Trident system. 

COLIDE fields were compared to the VDRAS gridded convergence field throughout the summer.  It was found that while VDRAS does a reasonable job of detecting large-scale boundaries, like synoptic and mesoscale fronts, it generally has considerable less ability to detect boundaries on the scale of gust fronts or smaller; COLIDE has just the opposite capability. Methods are planned to further refine COLIDE and conduct tests to determine whether combining COLIDE with VDRDAS cam optimize the strengths of both algorithms.

Considerable effort was expended this year to further tune and optimize the ANC/Trident system for the Front Range.  Efforts are presently underway to utilize initiation, growth and decay information from the AutoNowcaster into Trident to improve nowcasts of one hour rainfall amounts.  Rainfall nowcasts based on Trident are being compared with rain gauge estimates over varying size areas. These results will then be compared with those from persistence and extrapolation.

Qualitative Precipitation Forecasting (QPF)

The model-based QPF effort had three focus areas in FY16. First was the evaluation and verification of the real-time results from the 2016 and 2017 STEP Hydromet Experiment; second was the continued development of convective-scale data assimilation using WRFDA; and third was the real-time demonstration of data assimilation and QPF systems along with other STEP-Hydromet components during the summer of 2016.

Figure 3. Comparison of forecasting skills using FSS among the three experiments CYCLE, RADAR, and HRRR for the hourly precipitation threshold of 1mm (upper) and 2.5mm (lower). The FSS statistics for July (left) and August (right) are separately shown.
Figure 3. Comparison of forecasting skills using FSS among the three experiments CYCLE, RADAR, and HRRR for the hourly precipitation threshold of 1mm (upper) and 2.5mm (lower). The FSS statistics for July (left) and August (right) are separately shown.
Figure 4. Comparison of hourly precipitation patterns between CYCLE (middle row) and RADAR (bottom row), verified by MRMS (upper row) for the first three hour forecasts initialized at 03 UTC on Aug. 7, 2016.

Figure 4. Comparison of hourly precipitation patterns between CYCLE (middle row) and RADAR (bottom row), verified by MRMS (upper row) for the first three hour forecasts initialized at 03 UTC on Aug. 7, 2016.

The evaluation and verification of the real-time results was performed on the entire 3km WRF model domain as well as on three smaller domains. The verification for the whole 3km domain was done for both July and August in 2015 summer by comparing WRF 0-12h QPF with/without radars against MRMS gauge-corrected precipitation analysis. The Fraction Skill Scores for 1mm and 2.5mm are shown in Figure 3; each compares the skills for the following QPF runs:

CYCLE: initialized by WRFDA 3DVar analysis with continuous 3-hourly update cycles, assimilating only conventional observations; RADAR: partial cycled hourly radar data assimilation with CYCLE as the first guess at the multiples of 3 hour; HRRR: operational HRRR mapped to the same domain as the other runs.

Figure 4 shows an example of how precipitation forecasts differ with and without radar data assimilation. The forecasts with radar (bottom row) produced mode details that agree with the observations.

Figure 5. FSS (left) and BIAS (right) of hourly precipitation forecasts for two experiments with (red curve) and without (blue curve) radar no-rain data assimilation for the 1mm (top) and 5mm (bottom) thresholds. The shading indicates 95% confidence interval.
Figure 5. FSS (left) and BIAS (right) of hourly precipitation forecasts for two experiments with (red curve) and without (blue curve) radar no-rain data assimilation for the 1mm (top) and 5mm (bottom) thresholds. The shading indicates 95% confidence interval.

The evaluation and verification of the performance of WRF 3DVar NWP forecasts with and without radar data assimilation and the HRRR NWP model is in progress for the summer 2017. The evaluation is been performed over the entire WRF 3km model domain as well as over the Front Range nowcast domain. Three verification regions are being used in this study: one domain within 60km of the Denver KFTG radar, the second centered over the eastern Denver region and the third over Boulder County, where a high density of surface rain gauges are being used as truth. These results will be compared to 2014 QPF verification statistics reported on previously.  Similar to previous years preliminary results show very low skill numbers with little difference in skill numbers between the 3DVAR with and without data assimilation or between 1 and 6 hr nowcasts

The effort to improve convective-scale data assimilation in FY16 focused on the improvement of water vapor analysis via a more objectively based cloud analysis, the assimilation of “no-rain” observations from radar, and surface data assimilation. Figure 5 shows the impact of the no-rain assimilation as demonstrated by the precipitation prediction skill computed over two summer months in 2016. It clearly indicates that the no-rain assimilation significantly reduced the bias of the precipitation without much reduction of the FSS statistics.

The results in Figure 5 indicate that the runs with radar data assimilation (HRRR and RADAR) improve the skill over CYCLE for the entire 12-hour forecast period. The WRFDA 3DVar-based radar data assimilation run RADAR has higher skill than HRRR at the greatest number of forecast hours.

During the STEP Hydromet Experiment, conducted in the summer of 2017, the QPF systems were run on the same 3km domain (Figure 4) as in 2016. In addition to the real-time evaluation of the radar data impact, other runs were added to evaluate the impact of a new variable (the ice Thompson scheme) and the impact of adding a large-scale constraint to the rapid update cycles. Results of these runs against the control run are being evaluated.

Streamflow prediction from WRF-Hydro

Real-time, high-resolution hydrologic prediction work was performed during STEP-2017 using the community WRF-Hydro modeling system.  The system was configured somewhat differently than in previous years in order to better align STEP hydrologic prediction research with ongoing hydrologic system forecast development being done at the national scale by the WRF-Hydro modeling team in collaboration with the NOAA National Water Center.  These changes included the following:

  1. Using s similar model physics configuration of WRF-Hydro as is used in the NOAA National Water Model
  2. Implementation of a real-time streamflow data assimilation system which ingests real-time streamflow data from the USGS and Col. Division of Water Resources
  3. Ingest of the EOL mosaic QPE product instead of the operational NOAA MRMS product.
  4. Implementation of the HydroInspector web-mapping display tool for visualization of real-time hydrologic forecasts
  5. Implementation of a comprehensive hydrometeorological model evaluation system using the open-source Rwrfhydro model evaluation software

Example outputs of heavy rainfall and streamflow forecast displays from the HydroInspector tool are shown in Figure 6.

Figure 6. An example of outputs of heavy rainfall and streamflow forecast displays from the HydroInspector tool.
Figure 6. An example of outputs of heavy rainfall and streamflow forecast displays from the HydroInspector tool.

The event shown occurred on July 19, 2016 in a fairly remote, mountainous region southwest of Denver.  The map display shows a basin-average re-mapping of the EOL QPE product in order to highlight the tributary watersheds where heavy rainfall fell. The two time-series plots on the right hand side show the time-series of EOL QPE precipitation within the basin colored in dark red and the resulting streamflow predicted by the WRF-Hydro system.  This particular event was not well measured by existing streamflow observations making validation problematic. As noted above 2017 witnessed very few heavy to extreme rainfall events. Correspondingly, there were not many significant flood or flash flood events to study from the hydrologic model.  Analysis is currently underway of the continuous streamflow prediction performance throughout the summer and the STEP team is collectively analyzing a selection of heavy rainfall events that did occur during 2017.

Figure 7. An example of entire summer streamflow analyses from observations (black), an open-loop (no data assimilation) version of the WRF-Hydro model (yellow) and the analyses from the assimilation system (blue)
Figure 7. An example of entire summer streamflow analyses from observations (black), an open-loop (no data assimilation) version of the WRF-Hydro model (yellow) and the analyses from the assimilation system (blue)

Real-time data assimilation of streamflow was implemented in 2017, which had a significant impact on streamflow initial state characterization, particularly below reservoirs which exert significant control on streamflow behavior in the Colorado Front Range region. Figure 7 provides an example of entire summer streamflow analyses from observations (black), an open-loop (no data assimilation) version of the WRF-Hydro model (yellow) and the analyses from the assimilation system (blue).  It can easily be seen how the streamflow data assimilation keeps the model analyzed streamflow state close to the observations while without the data assimilation a background model bias emanating from a poorly described reservoir in the model is not able to keep track of the actual streamflow behavior below the reservoir.

Improvement upon the data assimilation methods used in the WRF-Hydro modeling system will be a major focus of research and development work leading to real-time forecasting activities in 2017.

FY18 plans

The STEP Hydromet Experiment will be operated during the summer of 2018 with upgraded system components, based on research conducted during the winter/spring 2017 Other research plans for each of the STEP-Hydromet components are summarized below:

QPE

  • No improvements are planned for the QPE component of the system.

QPN

  • Detailed analyses will be conducted on the evolution of heavy rainfall and flash flood events from the 2016 and 2017 demonstration. A climatology of heavy rainfall events along the Colorado Front Range during the past 8 years will be compiled.
  • Analyses and evaluation of the performance of the Autonowcaster/Trident nowcasts with the new predictor field will be examined. The impact of 10 min – 1 h nowcasts on the WRF-Hydro streamflow prediction will also be assessed for selected heavy rainfall/flash flood events along the Front Range.
  • The effort on the development of the model-based nowcasting techniques will continue using the VDRAS cloud model. A simplified scheme to advect precipitation  in a linear manner will be developed. The scheme will be compared with the full cloud model integration and extrapolation to assess its merit.

QPF

  • New data assimilation capabilities that were developed in the past two years will be tested using selected 2017 cases. These new capabilities include the assimilation of “no rain” data from radar reflectivity, a large-scale analysis constraint to maintain the synoptic scale balance, and a divergence constraint.
  • A research version with these new capabilities will be tested in real-time during the STEP Hydromet 2018.
  • An ensemble hybrid 3DVAR technique will be developed aimed at improving the background error covariance of the model forecast.

WRF-Hydro

 The main hydrologic research and prediction goals are as follows:

  • Improve the process representation of flood inundation through improvements in the channel-land surface model coupling
  • Conduct rigorous hydrologic model parameter calibration and regionalization over the STEP domain using the new National Water Model configuration of WRF-Hydro
  • Expand domain eastward to capture more deep convection events
  • Implement an upgraded streamflow data assimilation method that utilizes a time-filtered bias propagation that improves the blending between streamflow observations and the model simulated streamflow.  Also, we will begin research on ENKF methods for join streamflow-land data assimilation
  • Coordinate the potential generation of streamflow prediction ensembles with the rest of the STEP team.
  • Modify HydroInspector web mapping service to:
    • Display retrospective model skill
    • Access and display real-time verification statistics
    • Display catchment averaged precipitation
    • Create time-lagged ensemble probabilities

In sum, these goals are aimed at significantly improving streamflow and flood inundation prediction skill throughout the STEP domain. We will have a significantly stronger emphasis in 2017 on real-time model evaluation and diagnosis to better communicate combined precipitation and streamflow forecast skill.

2. Improving WRF physics for improved prediction of high impact weather

Development of prototype multi-moment graupel/hail hybrid category in Thompson microphysics parameterization

Figure 8. Relationship of fall velocity and particle diameter for various graupel densities.  The original Thompson graupel parameterization relationship for graupel using a constant density of 500 g/kg is shown as a solid gray line.
Figure 8. Relationship of fall velocity and particle diameter for various graupel densities.  The original Thompson graupel parameterization relationship for graupel using a constant density of 500 g/kg is shown as a solid gray line.

Previously, modifications were made to the Thompson microphysics code to allow for a variable graupel density to be diagnosed at every grid point and time step, rather than having the prescribed graupel density remain constant throughout the simulation domain and time period.  This work also included implementing a relationship from Milbrandt and Morrison (2013) that allows the particle fall speed to vary with graupel density as well (similar to that shown in Figure 7). This prior STEP work was the foundation for developing a multi-moment graupel/hail hybrid category.  For this multi-moment graupel/hail category, number concentration and bulk volume mixing ratio are now predicted variables, in addition to the bulk mass mixing ratio. This allows for the graupel density to vary in space and time and be diagnosed from the new predicted variables.  The fall speed relationship was also updated based upon data from Andy Heymsfield (NCAR/MMM) (Figure 8). This new prototype scheme was developed in FY17 and has been preliminarily tested in idealized simulations, as well as in simulations of real convective cases including a PECAN case on 19-20 June 2015.

Evaluating performance of microphysical schemes

In order to evaluate impacts on storm structure, QPF, and, in particular, storm evolution due to updates made to the Thompson microphysics scheme, an object-based evaluation tool that tracks storms over time was needed.  The Method for Object-based Diagnostic Evaluation (MODE)-Time Domain (TD) is a tool developed at NCAR for such analyses, allowing users to set several parameters to evaluate storms with certain spatial scales and intensities.  In FY17, the MODE-TD matching technique was evaluated in a few cases from the FY16 STEP-Hydromet summer experiment to improve the settings used for matching observed and modeled objects.  These evaluation methods will be applied to test cases of the prototype microphysics scheme to quantify impacts of the new multi-moment graupel/hail category.

FY18 plans

  • Evaluate and improve the prototype multi-moment graupel/hail hybrid category in the Thompson microphysics scheme to improve forecasted convective storm structure, evolution, and QPF.
  • Run the new prototype scheme as its own member in ensemble forecast experiments (i.e. OU CAPS, STEP-Hydromet)
  • Run MODE-TD on the ensemble forecast experiment members utilizing different microphysics schemes (prototype scheme, original Thompson scheme, others as available) and compare results for the different microphysics schemes utilized. 
  • Use evaluation results to inform further microphysics parameterization improvements.

3. Study on convective initiation with PECAN data

One of the objectives of the Plains Elevated Convection at Night (PECAN) experiment was to advance knowledge of the processes and conditions leading to pristine nocturnal convection initiation (CI). An isolated hailstorm in July 2015 provided an exceptional case study for project scientists.  This nocturnal hailstorm developed more than 160 km from any other convective storms and in the absence of any surface fronts or bores. The storm initiated within 110 km of the S-Pol radar, directly over a vertically pointing Doppler lidar, within 25 km of the University of Wyoming King Air flight track, within a network of nine sounding sites taking 2-hourly soundings, and near a mobile mesonet track. Importantly, even beyond 100 km in range, S-Pol observed the pre-convection initiation cloud that was collocated with the satellite infrared cloud image, and provided information on the evolution of cloud growth. The multiple observations of cloud base, thermodynamic stability and direct updraft observations were used to determine that the updraft roots were elevated.  Diagnostic analysis presented in the paper suggests that CI was aided by lower-tropospheric gravity waves occurring in an environment of weak but persistent mesoscale lifting.  A paper on this work has been published in Monthly Weather Review.

Figure 9. Location of developing hailstorm in relation to PECAN observation systems. The bold yellow FP, MP and MS stars refer to the location of the fixed and mobile PISA’s and soundings. The smaller yellow stars are MP3 and TWOLF; they did not take soundings but had Doppler lidars. The red and pink tracks show the position of the UWKA 30 min prior (red) to the radar image and 30 min after (pink). SR1 and SR2 are the SMART-R C-band radars. The radar image is at 0420. A mobile mesonet track, not visible, is positioned directly under the UWKA east-bound leg. The yellow rectangle shows the PECAN domain.
Figure 9. Location of developing hailstorm in relation to PECAN observation systems. The bold yellow FP, MP and MS stars refer to the location of the fixed and mobile PISA’s and soundings. The smaller yellow stars are MP3 and TWOLF; they did not take soundings but had Doppler lidars. The red and pink tracks show the position of the UWKA 30 min prior (red) to the radar image and 30 min after (pink). SR1 and SR2 are the SMART-R C-band radars. The radar image is at 0420. A mobile mesonet track, not visible, is positioned directly under the UWKA east-bound leg. The yellow rectangle shows the PECAN domain.
Figure 10. Time series of vertical velocity from zenith pointing Doppler lidars (MP3 and TWOLF). The bottom panel is an enlargement of the TWOLF data shown inside the purple rectangle in the middle panel. Positive velocities (yellow) are updrafts and negative velocities (green) are downdrafts. The TWOLF data within the purple box has been edited at cloud base to remove unreliable data above cloud base. For MP3 the dark areas beginning above 3 km are cloud bases. MP3 is at an elevation of 700 m MSL and the first recorded gate is at 15 m AGL. TWOLF is at an elevation of 800 m MSL and the first recorded gate is at 400 m AGL.
Figure 10. Time series of vertical velocity from zenith pointing Doppler lidars (MP3 and TWOLF). The bottom panel is an enlargement of the TWOLF data shown inside the purple rectangle in the middle panel. Positive velocities (yellow) are updrafts and negative velocities (green) are downdrafts. The TWOLF data within the purple box has been edited at cloud base to remove unreliable data above cloud base. For MP3 the dark areas beginning above 3 km are cloud bases. MP3 is at an elevation of 700 m MSL and the first recorded gate is at 15 m AGL. TWOLF is at an elevation of 800 m MSL and the first recorded gate is at 400 m AGL.

Figure 9 shows the observation facilities that were available for this study and the location of the of the storm at initiation time prior to its development into a hail storm. Figure 10 shows the gravity waves that were observed by two vertical pointing Doppler lidars. These gravity waves were likely partially responsible for the storm’s initiation.

 

FY2018 Plans

  • Analysis of PECAN cases will continue in collaboration with scientists from EOL and MMM to understand the processes associated with nocturnal elevated convection initiation and the predictability of these events.
  • STEP scientists will contribute to a manuscript on PECAN ECI to be submitted to BAMS.
  • Detailed analyses will be performed for a few selected cases using VDRAS to examine the dynamical mechanism of elevated convection.