How NSSL research provides real-time precipitation estimations and flash flood prediction for high-impact events

Some of the costliest and deadliest weather events in the United States are flash floods. On average, more fatalities are attributed to flash flooding than other short-fused severe weather hazards, like tornadoes, hurricanes, and lightning.

Flash flooding — the rapid rise of water in a normally dry area — is mostly related to excessive rainfall resulting in significant groundwater runoff and quick rises in waterways. NOAA National Weather Service (NWS) forecasters rely on accurate quantitative precipitation estimations (QPEs). QPE are input into diagnostic tools and models to help NWS forecasters predict and warn on the potential for flash flooding, like flash floods that occurred in Tennessee on Aug. 21, 2021.

Areas west of Nashville, particularly in Humphreys County, received over 1 foot of rain in a matter of hours. This included a period where 3-4 inches of rain fell per hour over multiple consecutive hours. Approximately 17.02 inches of rain was recorded in McEwen located in Humphreys County. This preliminary total eclipses the state record for rainfall in a 24-hour period, which was 13.60 inches in 1982. Twenty people perished in this Tennessee flood event.

A gif loop of radar reflectivity over middle Tennessee showing the increase flash flood levels.
A Multi-Radar Multi-Sensor reflectivity loop covering the duration of the western Middle Tennessee flash flood event ton Aug. 21. (Gif provided by Randy Bowers.)

NWS forecasters can use a series of products to diagnose an ongoing weather event to determine what might be happening. Researchers at the NOAA National Severe Storms Laboratory (NSSL) and the Cooperative Institute for Mesoscale Meteorological Studies (CIMMS) at the University of Oklahoma developed two systems to help with forecaster analysis and warning decision making — the Multi-Radar Multi-Sensor (MRMS) system and the Flooded Locations and Simulated Hydrographs (FLASH) system.

The Systems

The MRMS system is a platform that combines various weather observations and model data to create a suite of products, including various QPE fields.

A key to the MRMS system is the quality control of radar data. Quality control algorithms remove radar artifacts from blockages, wind farms, biological scatter (like birds and bugs), and other data contaminations. The MRMS system then applies the latest scientific advancements in precipitation estimation using dual-polarization radar technology to provide accurate precipitation data in real-time every two minutes.

NSSL and CIMMS researchers regularly analyze MRMS QPE performance, including historic events like the Tennessee flash flooding. Product evaluations are conducted through internal web pages that allow for statistical comparisons of MRMS QPEs to independent gauge observations.

Using 24-hour analysis centered around 1200 UTC (7:00 AM local time) to collect both daily CoCoRaHS rain gauges along with hourly automated gauge observations, a few notable trends appear in the data. The overall analysis showed well correlated and clustered comparisons between the MRMS radar-based QPE and the gauge observations with rather small errors. The MRMS dual-polarization radar QPE had some overestimations with totals less than two inches, while some slight underestimation was observed with totals exceeding four inches. Yet, the nearly equivalent values between the gauges and MRMS in the area of greatest rainfall shows how well the system handled the event.

A screenshot of MRMS dual-polarization QPE data.
Analysis of MRMS dual-polarization QPE ending 1200 UTC on Aug. 21 (left column) and Aug. 22 (right column) with bubble plots (top row) and scatterplots with statistics (bottom row) using hourly and daily gauge observations. (Screenshot provided.)

The second application developed by NSSL and CIMMS researchers to help with flash flood prediction is the Flooded Locations and Simulated Hydrographs (FLASH) system. The FLASH system is the first system to generate hydrologic modeling products specific to flash flooding at the flash flood time scale — new model runs are generated every ten minutes — in real-time for the entire country.

The FLASH system also provides products that compare QPE values to flash flood guidance — a measure of how much rainfall is needed to flood small waterways — in addition to the average recurrence intervals — a measure to determine the rarity of the precipitation totals based on how frequently they occur. All products within the FLASH system use the MRMS dual-polarization radar QPE as their input.

Three separate screenshots of the FLASH model products showing QPE and flooding.
Analysis of the following FLASH products at 1300 UTC 21 August 2021: maximum QPE-to-FFG ratio (left), maximum QPE average recurrence interval (center), and CREST maximum unit streamflow (right). (Screenshot provided.)

At the peak of the rainfall over Humphreys County, Tennessee, the QPE comparison products were at the upper end of the plotted scales. The accumulated rainfall was at least four times that of the NWS flash flood guidance for the area, and the average recurrence interval of the rain was beyond the plotted scale in the system (at least 200 years — approximately 0.5% chance of occurring per year).

The product that best conveys the flash flood potential and its possible severity is the maximum unit streamflow product from the Coupled Routing and Excess Storage (CREST) hydrologic model. The maximum unit streamflow values — the amount of water runoff normalized by its basin area — have been shown to capture the spatial coverage of flash flooding and provide context to its potential severity.

The projected unit streamflow values based on MRMS precipitation rates during the Tennessee flash flood event on Aug. 21, 2021, showed three key features:

  • How quickly the flash flood threat escalated.
  • How the extreme values pointed to a potentially catastrophic event.
  • How the model routed the water to show the impacts on local rivers even after the rainfall ended.
A graphic of the CREST maximum unit streamflow from the FLASH system. The graphic shows flood waters maxing out over time.
CREST maximum unit streamflow from the FLASH system from 0600–2100 UTC 21 August 2021. (Graphic provided.)

Researchers at NSSL and CIMMS continuously work to enhance the performance of the MRMS and FLASH systems to improve precipitation estimations and flash flood predictions. Efforts with machine learning and artificial intelligence are paving the way for increased performance in areas where radars struggle to accurately capture precipitation. Probabilistic hydrologic modeling with the use of forecast precipitation with the FLASH system looks toward the future of warning for flash floods within the FACETs (Forecasting a Continuum of Environmental Threats) paradigm.

Share this:

Using a dual-pol radar feature to anticipate downburst development

Downbursts—an area of strong winds in a thunderstorm—can damage trees and buildings, disrupt air travel, and cause loss of life. Decades of work by scientists has revealed a lot of information about downbursts including certain features seen on radar, known as precursor signatures, that can help forecasters anticipate when a downburst might develop. However, downbursts are still quite challenging to predict—especially in low-shear, summer-time thunderstorms—perhaps because the downbursts and their precursor signatures can develop quickly and be difficult to observe. 

Therefore, researchers at the University of Oklahoma Cooperative Institute for Mesoscale Meteorological Studies with the NOAA National Severe Storms Laboratory decided to look into it further. They studied  a dual-pol radar feature, known as a specific differential phase (KDP) core, because it could provide information about important processes that cause a downdraft to develop and get stronger. Our research shows KDP cores could be a reliable and easily observable downburst precursor signature that can help NOAA National Weather Service forecasters predict where a downburst could develop next. 

Our research began one afternoon in the halls of the National Weather Center in Norman, Oklahoma, when Randy Bowers, a forecaster at the NOAA National Weather Service Norman Forecast Office (OUN) and I (Charles Kuster, a research scientist with the CIMMS working at NSSL), were having an informal discussion about a completely unrelated topic. At the end of our conversation, Randy mentioned seeing a consistent area of high KDP—the KDP core— while issuing severe thunderstorm warnings and wondered if we could do some research on it together. I had been looking for opportunities to work more with forecasters, so this topic sounded like an amazing opportunity. We jumped in and began identifying potential cases, collecting data with a research radar in Norman, and brainstorming how to best study KDP cores.

The KOUN radar with storm clouds behind it.
The NOAA National Severe Storms Laboratory research radar collects data on a downburst-producing thunderstorm in Norman, Oklahoma. (Photo by Charles Kuster, OU CIMMS/NSSL)

Ultimately, we selected 81 downbursts in 10 different states to analyze. I got to work on comparing the size, magnitude, and vertical changes in the KDP cores associated with strong and weak downbursts, while Randy examined atmospheric conditions and possible warning applications. We also began working with Jacob Carlin, a CIMMS research scientist also working at NSSL, who explored model simulations of downdrafts and important microphysical processes—such as melting and evaporation—that can result in a stronger downburst. Another research scientist, Terry Schuur (OU CIMMS/NSSL), also brought microphysics expertise and experience to the team while researchers Jeff Brogden (OU CIMMS/NSSL) and Robert Toomey (OU CIMMS/NSSL) provided extensive support and expertise in radar data analysis software. Andy Dean, a forecaster with the NOAA/NWS Storm Prediction Center, provided data about atmospheric conditions around the downbursts.

Together, we determined KDP cores were a reliable downburst precursor signature in the events we studied. All 81 downbursts were preceded by a KDP core by as much as 30 minutes. The KDP core also reached its maximum intensity, typically about 10–15 minutes before the downburst reached its maximum intensity. In addition, there were very few instances where a KDP core was observed and no downburst occurred (i.e., very few null events).

Anticipating downburst intensity using KDP cores was more challenging because there was overlap between the characteristics of KDP cores associated with strong and weak downbursts, but in general, a stronger KDP core was more likely to be associated with a stronger downburst.

We also found the atmospheric conditions during each event were very important. When atmospheric conditions were less favorable for downburst development, we observed stronger KDP cores, which likely means more melting, precipitation loading, and evaporation are needed for a downburst to develop in such conditions. Ultimately, our work showed KDP cores provide a good signal that a downburst is likely to develop soon, assuming atmospheric conditions are favorable for downburst development, and can help forecasters triage storms to determine which one has the highest chance of producing a downburst.

A screenshot of a graph showing the KDP Core Size Near Melting Layer for all Downbursts.
KDP core size (red line) generally increases in the 30-minute period prior to downburst development. (Photo provided)
A photo montage showing the development of the KDP core.
Example of a KDP core in mid-levels of a thunderstorm (i.e., near the melting layer) and the associated downburst near the ground seen in Doppler velocity data (V). In the velocity images, red colors show air moving away from the radar and green colors show air moving towards the radar. (Photo provided)

We would like to thank everyone who has made this research possible so far and those who continue to help push it forward. Everyone from the CIMMS and NSSL administrative staff, NSSL IT, radar engineers, NOAA National Weather Service Norman Forecast Office, and webinar organizers are very much appreciated. For any questions, please contact nssl.outreach@noaa.gov or see https://doi.org/10.1175/WAF-D-21-0005.1 for more detailed information.

Share this:

WoFS in the virtual NOAA Hazardous Weather Testbed

 

The first week of April concluded the end of the 2021 Warn-on-Forecast Testbed Experiment as part of the NOAA Hazardous Weather Testbed. In this experiment, a total of 16 forecasters from nine southern regions National Weather Service Forecast Offices (WFOs), the NOAA NWS Storm Prediction Center (SPC), and the NOAA NWS Weather Prediction Center (WPC) came together over four weeks to explore the use of Warn-on-Forecast System (WoFS) guidance in the watch-to-warning time frame.

Like many other scientific activities, this experiment was delayed and then moved virtually due to the ongoing COVID-19 pandemic. Despite the many challenges this unique situation presented, our research team is pleased to report the experiment was very successful. This success is attributable to the significant efforts of numerous University of Oklahoma Cooperative Institute for Mesoscale Meteorological Studies (CIMMS) and NOAA National Severe Storms Laboratory scientists, as well as participants and collaborators in the NWS

Together, Pat Skinner (with OU CIMMS/NSSL), Patrick Burke (with NSSL), Burkely Gallo (with OU CIMMS/SPC), and I (Katie Wilson with OU CIMMS/NSSL) designed and executed this experiment to examine how forecasters envision WoFS guidance fitting into both their existing and visionary forecast processes, and to explore the ways that WoFS guidance can be used most effectively given national center and local office forecasting responsibilities.

A screenshot showing an experimental National Weather Service severe weather graphic. The black graphic shows storm areas highlighted.
An example of an experimental decision support graphic influenced by WoFS and constructed during one of the 2021 WoFS experiment case studies. (Screenshot)

A Collaborative Undertaking

During the experiment, Skinner delivered an overview presentation to build familiarity with WoFS guidance prior to participants’ completion of two case studies. These cases formed the first of two major activities during the experiment, which was for participants to immerse themselves in simulated real-time events and use WoFS guidance to make forecast and communication decisions. 

To prepare for the case study activity, each week Burke provided participants with a hands-on AWIPS-2 demo. Furthermore, his work to design AWIPS-2 perspectives and procedures, which was conducted jointly with Gallo, enabled a simulation setup that was more familiar to participants, especially to those from national centers who do not use AWIPS-2 in the way local office forecasters do.

A screenshot of the experimental WoFS guidance in the AWIPS-2 viewer.
For the first time, experimental WoFS guidance was viewable in the AWIPS-2 interface.

The preparation of these case studies was a major task undertaken by Jonny Madden (OU CIMMS/NSSL), Justin Monroe (OU CIMMS/NSSL), Jorge Guerra (OU CIMMS/NSSL), and Dale Morris (OU CIMMS/NWS Warning Decision Training Division).

The case studies presented two notable firsts:

  • Running AWIPS-2 in-the-cloud such that participants could complete the case studies from their own homes, and;
  • Presenting WoFS guidance in AWIPS-2, including the development of a tool to visualize paintball plots. 

Madden, Monroe, Guerra, and Morris worked together to accomplish numerous tasks, including: aggregating and processing a full suite of observational and model datasets for both cases, setting up the WES- 2 Bridge and AWIPS-2 interfaces, and collaborating with federal partners to get datasets onto the cloud framework. Much of what was accomplished for the case study portion of this experiment has laid the AWIPS-2 in-the-cloud groundwork for future virtual experiments. 

The second major activity during the experiment was focus groups. Together, Wilson and Gallo led three semi-structured discussions each week to explore forecasters’ visions for how WoFS will impact the current and future forecast process. Additionally, the presence of both national center and local office forecasters meant that much was learned about each others’ workflows, how one another makes decisions, and where there is an opportunity to strengthen collaboration. In a post-experiment questionnaire, participants rated the focus groups as a highly effective activity for sharing thoughts and ideas, and was the most enjoyed activity of the week.

A graphic showing how the team used Google Meet Jam Board to spur discussion. The graphic has two circles, with forecast offices in one area and SPC and WPC in another.
The team used the Google Meet jam board to spur discussion in focus groups.

In addition to the efforts of scientists at OU CIMMS and NOAA NSSL, we were grateful for input from our collaborators at NWS Southern Region, including Chad Gravelle (SR HQ), Todd Lindley (OUN SOO), Stephen Bieda (AMA SOO), and Randy Bowers (OUN). Gravelle and Lindleyalso joined the experiment for multiple weeks, and Randy created two excellent weather briefing videos to prepare forecasters for the case studies. A big thank you also goes to our pilot participants, Laren Reynolds (El Paso, Texas) and Joseph Merchant (Lubbock, Texas), for volunteering their time to fulfill an important support role throughout the whole experiment. This support role emerged following findings from the pilot week, and made for a much stronger experiment.

We are extremely appreciative to the 16 NWS forecasters who participated in this experiment. We realize the stressful conditions many people continue to live and work with, and have done so over the past year. We also realize the disappointment from not being able to attend this experiment in the NOAA Hazardous Weather Testbed, in Norman, Oklahoma, as originally planned. However, participants showed up to our virtual experiment with enthusiasm and made meaningful contributions to the experiment. We collected an enormous amount of data, and we can’t wait to analyze it and share what was learned.

For questions on this or other WoFS-related research please contact WoFS Program Lead, Patrick Burke, nssl.outreach@noaa.gov.

Share this:

WoFS at the AMS, and the 2020 Iowa Derecho

The Annual Meeting of the American Meteorological Society is the largest recurring conference in our field. In January 2021, the 101st Annual Meeting of AMS took place in a virtual venue, but that didn’t stop the experimental Warn-on-Forecast System (WoFS) from taking center stage in a variety of ways. At least seven posters and a dozen oral presentations covered stories specific to WoFS. Many more covered closely related aspects of mesoscale modeling and forecast and warning operations. WoFS presenters included Norman-community researchers and students, but also National Weather Service forecasters from national centers as well as local offices. One group especially well represented were science operations officers from the group of nine Southern Plains NWS offices that have been evaluating WoFS as part of a two-year project. Many showed real-world examples of the ways in which WoFS is already influencing lead time and specificity of information shared with the public and other users.

Poster presentation of Warn-on-Forecast.
Poster presentation led by Todd Lindley, Science and Operations Officer at the NOAA NWS Weather Forecast Office in Norman, Oklahoma.

There was so much enthusiasm for developing WoFS-style probabilistic and rapidly updating guidance — with novel data assimilation for the watch to warning time scales — an entire conference session was dedicated to WoFS and included a panel discussion titled, “Utilization and Development of Rapidly Updating Mesoscale Models for IDSS (Incident Decision Support Services).”

A graphic saying Utilization and Development of Rapidly Updating Mesoscale Models for IDSS.
Image used to advertise and introduce a WoFS-related panel discussion at the 101st Annual Meeting of the American Meteorological Society. (Courtesy of AMS)

Perhaps no presentation spoke more to the potential utility of WoFS than Patrick Skinner’s talk, “Predictability of the 10 August 2020 Midwest Derecho.” The “Iowa Derecho” was one of the biggest weather stories of 2020. Occurring at the height of the growing season, the swath of destructive winds was not only life-threatening but also obliterated crops in its path, making this the costliest single thunderstorm event in United States history.

Presentation slide showing satellite imagery before and after the Iowa Derecho. Image courtesy of NASA.
Presentation slide showing satellite imagery before and after the Iowa Derecho. Image courtesy of NASA.

Predictability varies for thunderstorm events, and many numerical models did not do a particularly good job of helping forecasters to anticipate such a devastating event, even the day of the storm. To test whether the experimental WoFS could have contributed to an improved forecast of the event, researchers first had to expand the model domain to capture the evolution of such a fast-moving and long-lived storm. Once this had been accomplished, the results of the forecast runs proved very promising. A forecast based on data that was available 12 hours before the derecho correctly predicted a fast-moving, bowing thunderstorm system with significant severe winds (> 75 mph) near the ground. In the loop below, red shading represents the swath of WoFS-predicted significant severe winds, and the small blue squares and red triangles plot the locations where damaging winds and tornadoes, respectively, were observed on August 10, 2020.

A gray GIF loop of a WoFS retrospective forecast. The forecasted areas over Iowa change color from orange to red over time.
Loop of a WoFS retrospective forecast that was initialized with data from 03 UTC on August 10, 2020, approximately 12 hours prior to the onset of damaging winds from the derecho. Shown here are the ensemble 90th percentile of maximum 10-meter wind gusts, the probability matched mean of radar reflectivity, and (very faintly) paintball splatters representing reflectivity from individual ensemble members. Local storm reports are plotted over the data as squares (severe wind), circles (severe hail), and triangles (tornadoes). The model system predicts a derecho-like system, including a broad swath of winds greater than 75 mph. (Courtesy of NOAA NSSL)

In the future, when a fully developed WoFS becomes available for events such as these, this could lead to earlier anticipation of a high-end event. The initial WoFS forecasts were displaced a little to the north of the worst damage, but as the early stage of the storm development got underway, WoFS forecasts adjusted to the correct latitude/location, still with a few hours of lead time before the worst of the storm would have occurred.

A screenshot of the Warn-on-Forecast System with plots summarizing the probability of severe near-ground winds.
Plot summarizing the probability of severe near-ground winds from a WoFS run initialized with data from 11 UTC on August 10, 2020. As time drew nearer to the derecho event, WoFS forecasts adjusted southward, then overlapping more directly with the swath of very damaging winds that would occur a few hours later. (Courtesy of NOAA NSSL)

Preliminary results indicate the poor depiction of overnight thunderstorms in Nebraska and South Dakota led to large errors in the operational forecast models. The models generated too many thunderstorms early in the forecast period, thus limiting the energy available for daytime storms in Iowa. Employing rapid and high-resolution assimilation of radar and satellite data, the experimental WoFS forecasts better depicted the overnight storms, and therefore better reflected the large amounts of energy available for the damaging daytime storms in Iowa. The group led by Skinner plans to publish this research in the near future.

For questions on this or other WoFS-related research please contact WoFS Program Lead, Patrick Burke, nssl.noaa@outreach.gov.

Share this: