Tropical Cyclone

BACKGROUND

RAL’s Joint Numerical Testbed Program (JNTP) has a number of efforts related to tropical cyclone (TC) forecasting.  These will be described below as they relate to the JNTP’s Developmental Testbed Center (DTC); the Tropical Cyclone Modeling Team (TCMT); the Tropical Cyclone Guidance Project (TCGP); the Tropical Cyclone Data Project (TCDP); and a new initiative called the Hurricane Risk Calculator.

The Developmental Testbed Center works closely with NCEP’s Environmental Modeling Center (EMC) to support the Hurricane Weather Research and Forecasting (HWRF) system to the research community. The team also tests new capabilities coming from the research community to determine their potential for improving the forecast skill of HWRF.  The goals of this work are to accelerate the improvement in TC forecasts by providing a mechanism for efficiently transitioning research into operations, and through extensive testing of new capabilities to determine their impacts on operational predictions.

The focus of RAL’s Tropical Cyclone Modeling Team is the development of new diagnostics and evaluation tools for tropical cyclone forecasting. Currently, the primary sponsor of the TCMT is NOAA’s Hurricane Forecast Improvement Project (HFIP).  In coordination with the HFIP teams, the TCMT collects, evaluates, and provides results of tropical cyclone track forecasts to the broader HFIP community. Statistical approaches and new graphical displays are being developed by the TCMT. Current efforts are focused on developing new methods of evaluation for ensemble tropical cyclone forecasts of rapid intensification and development of a specialized display and diagnostic evaluation system for use at the National Hurricane Center (NHC).

RAL’s Tropical Cyclone Guidance Project (TCGP) serves as a platform for the collection and dissemination of real-time TC forecast data and products, for the development and testing of new prediction methods, and for the development and testing of new forms of risk communication. TCGP collects real-time tropical cyclone guidance data from numerical prediction centers around the world and collates these data into a publicly-available global repository of TC forecast aids. TCGP’s public-facing web page features the real-time data as well as visualizations of the forecast data. The site receives millions of hits each year from a wide range of users including forecasters, emergency managers, government agencies (e.g., NOAA, FEMA, DHS), private-sector firms (e.g., ship-routing, transportation and logistics, energy producers, energy and risk trading, media), weather enthusiasts, and the general public. TCGP is currently supporting several projects at NCAR funded by NOAA’s Hurricane Forecast Improvement Project, including TCMT’s HFIP Ensemble Rapid Intensification (RI) task and a separate HFIP-funded project to develop new frameworks for predicting extreme rapid intensification. TCGP is also the foundation on which the Hurricane Risk Calculator is being built.

RAL’s Tropical Cyclone Data Project (TCDP) develops and maintains datasets of tropical cyclone observational data for the research community. The extensive data processing, curation, and quality control make these data more accessible for a wide variety of research uses. Currently, TCDP hosts three research-grade datasets and a new historical database of TC wind structure parameters:

VDM+: The Enhanced Vortex Message Dataset: Structure, Intensity, and Environmental Parameters from Atlantic Tropical Cyclones

FLIGHT+: The Extended Flight Level Dataset for Tropical Cyclones

QCAR-R: The QuikSCAT Tropical Cyclone Radial Structure Dataset

TC-OBS: The Tropical Cyclone Observations-Based Structure Database.

These datasets were first released to the public in FY2016 and have undergone several updates. The TC-OBS database uses objective methods to optimally estimate the various database parameters, as well as to provide time-dependent error bounds on the estimated parameters. It is intended to provide the highest quality database possible for parametric wind modeling applications and model evaluation activities (e.g., verification), and to support basic and applied research on TC intensity and structure change.

The Hurricane Risk Calculator began as a small exploratory project funded by the RAL Opportunity Fund in FY2018. It has since been expanded by substantial funding from the UCAR President’s Strategic Initiative Fund. The goal of the project is to explore novel methods of calculating and disseminating hurricane risk information and demonstrate their deliver to mobile apps. The approach involves intersecting the location-specific wind hazard information with the structure-specific vulnerability at the user’s location. When both the wind hazard and vulnerability are described probabilistically, it is then possible to calculate the user’s risk due to hurricane winds. This risk can then be translated into various forms that are relevant and meaningful for the user. Examples could include the risk of incurring major structural damage to their dwelling or the likelihood that their dwelling will be habitable following the storm. That risk information can then be contextualized against the risk and costs of long-distance evacuation, providing the user with actionable guidance on whether they should shelter-in-place, evacuate to a local shelter, or evacuate to a more distant location. The ultimate goal of the project is to wrap these capabilities into a mobile app to provide life-saving alerts to residents in areas impacted by tropical cyclones. 

FY2019 ACCOMPLISHMENTS

Developmental Testbed Center

Advancing HWRF physics

During FY2019, the DTC’s Hurricane team partnered with DTC Visitor Program Principal Investigators and subject area experts to help coordinate and test the performance of alternate physics schemes and innovations relative to the current parameterizations within the HWRF physics suite. To determine which innovations should be considered for operational upgrades, tests are performed with alternate configurations to isolate which innovations provide improved tropical cyclone (TC) prediction. The Mellor-Yamada-Nakanishi-Niino (MYNN) planetary boundary layer (PBL) parameterization became available for testing within the HWRF system from a DTC Visitor Program project by Robert Fovell (SUNY-Albany). A full report of this project can be accessed on the DTC website:https://dtcenter.org/sites/default/files/visitor-projects/DTC_HWRF_proposal_2015_REPORT_v2.pdf

Figure 1. Mean track errors (a), absolute intensity errors (b), and mean intensity errors (c) in the AL basin with respect to lead time. The CTRL (operational) is in black, and MYNN is in red. Pairwise differences (experiment minus control) are shown in light shades with 95% confidence intervals.
Figure 1. Mean track errors (a), absolute intensity errors (b), and mean intensity errors (c) in the AL basin with respect to lead time. The CTRL (operational) is in black, and MYNN is in red. Pairwise differences (experiment minus control) are shown in light shades with 95% confidence intervals.

Additional testing and evaluation was conducted for real HWRF cases, in order to assess the performance of the MYNN PBL scheme relative to the current HWRF operational Global Forecast System (GFS) Eddy-diffusivity Mass-flux (EDMF) PBL scheme. The control configuration (CTRL) was run using the full operational HWRF physics suite (including the GFS-EDMF PBL), whereas the experimental configuration (MYNN) replaced the operational scheme with the MYNN PBL, holding all other physics parameterizations constant.

The track and intensity results indicate fairly neutral results (Figure 1). The differences in track error are statistically indistinguishable, with small SS differences favoring the CTRL configuration at early lead times for absolute intensity error (forecast hours 6, 12, 24, and 36). Although not SS, the mean differences in the intensity bias suggest the MYNN configuration produces slightly less over intensification relative to the CTRL configuration (Figure 1). Most notably, the MYNN configuration produces a slightly larger spin-down in the initial forecast hours, corresponding to the SS differences favoring the CTRL configuration in the absolute intensity errors.

Hurricane Michael was a high-impact 2018 Atlantic hurricane that made landfall in the panhandle of Florida as a category-five storm in October. Early in the lifecycle of the storm, the operational HWRF showed the potential for the storm to intensify rapidly and make landfall as a major hurricane. In fact, three consecutive HWRF cycles (2018100718–2018100806) forecasted Michael to be a category-four hurricane at landfall. When these cycles were initialized, Michael was still a tropical storm. Since Hurricane Michael was such an important storm to forecast accurately, a case study analysis using the HWRF configurations with the GFS EDMF and MYNN PBL schemes was conducted to quantify how the model forecast differed between the two configurations.

Figure 2. Eyewall-averaged vertical profiles of normalized a) tangential and b) radial wind speed in Hurricane Michael from the 36-hour forecast of the 20181009 00 UTC cycle of CTRL (red solid line) and MYNN (red dashed line). The mean air temperature profile from the eyewall dropsondes is shown in black. The error bars represent +/- one standard deviation.
Figure 2. Eyewall-averaged vertical profiles of normalized a) tangential and b) radial wind speed in Hurricane Michael from the 36-hour forecast of the 20181009 00 UTC cycle of CTRL (red solid line) and MYNN (red dashed line). The mean air temperature profile from the eyewall dropsondes is shown in black. The error bars represent +/- one standard deviation.

While the thermodynamic comparison between the two simulations implied similar storm structures (not shown), the kinematic comparison reveals much larger differences. Figure 2 shows eyewall-averaged vertical profiles of tangential and radial wind. For both components of the wind, the MYNN configuration provides a superior match to the dropsonde observations. The height of the peak tangential wind in MYNN (z=400 m) exactly matches observations, but is too close to the surface in the CTRL (z=200 m). Above the height of the peak tangential wind, the reduction with height is much slower in MYNN than in the CTRL and closely mirrors the observations. For radial wind, the inflow is much stronger throughout the boundary layer in MYNN than in the CTRL and is nearly identical to the observations from 0–900 m.

The operational HWRF did not undergo a full implementation during 2019, therefore testing and evaluation conducted by the DTC during this period was to inform HWRF developers of promising upgrades for future operational implementations. A description of the full evaluation can be found in the final report on the DTC website located at:https://dtcenter.org/eval/hwrf_MYNN_2018/

Tropical Cyclone Modeling Team

Development of a Tropical Cyclone Display and Diagnostic System

Fig. 3: Examples of the NHC-Display tool showing the F-Deck editor (top panel), B-deck editor (bottom panel).
Figure 3. Examples of the NHC-Display tool showing the F-Deck editor (top panel), B-deck editor (bottom panel).

A next-generation display and diagnostic system has been developed to support evaluation needs of the U.S. National Hurricane Center (NHC) and the broader tropical cyclone (TC) research community.  The new hurricane display and diagnostic capabilities allow forecasters and research scientists to more deeply examine the performance of operational and experimental models.  The system is built upon modern and flexible technology, including OpenLayers Mapping tools that are web-based and platform independent. The forecast track and intensity along with associated observed track information are stored in a MySQL database.  The system provides an easy-to-use interactive display system, and provides a variety of diagnostic tools to examine tropical model forecast skill.  Consensus forecasts can be computed, displayed and saved to a standard forecast output file.  The system is designed to display observed (best track) and model forecast track and intensity information for both real-time and historical TC cyclones.  The display is also capable of displaying both observed and forecast product model fields.  Display configurations are easily adaptable to meet the needs of the end-user preferences.

Fig. 4: Example of the wind radii display tool.
Figure 4. Example of the wind radii display tool.

New technologies implemented into the display system this year include an advanced tool for editing the hurricane fix-position database (F-deck) and the best-track database (B-deck).  The F-deck editing tool allows NHC staff to add or edit the estimated location of hurricanes using fixed-position information from aircraft analysis, radar, satellite, microwave, and scatterometer observations.  The application can be used to edit best-track points and fix data pointed that are stored in a separate database.  The tool functionality include visual editing of best-track points location and form based editing of best-track and fix points using map or plot.  Once the new best-track and fix points are finalized, they can be updated on the official database.  This information is used to improve the location of the hurricane in the B-deck database during post-hurricane season analysis.  An example of the editing tools is shown in Figure 3.

Fig. 5: Example of the gridded moisture field and wind shear vector forecast products.
Figure 5. Example of the gridded moisture field and wind shear vector forecast products.

A wind radii tool has been added to the available features this past year.  Users can view wind radii graphics for 34kt, 50kt, 64kt or all thresholds on the maps.  Wind radii display capability is available for both best track and/or model data.  Example of the wind radii display capabilities is shown in Figure 4.  Gridded forecast and observation fields have been added to the display system.  The new gridded products include: wind_shear, precipitable_water, moisture, and SST contour products.  The display system calculates derived fields using GFS model output and stores the output in netCDF files.  The fields are extracted from thehttps://nomads.ncep.noaa.gov and ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/gfs/prod/ servers.  Products are available for 120 hours into the future and the past 24 hours centered on the current time.  An example of the moisture product overlaid with the wind shear vectors is shown on Figure 5.  The display system has improved functionality including: pan zoom, configurable labeling, and meta data information pop-up windows.

The latest version of the display system is available at the web-link:http://www.hfip.org/nhc-display

Ensemble Rapid Intensification Products

Building on last year’s HFIP Demo Rapid Intensification (RI) activities, the RI probability for all available model configurations were computed and displayed for each initialization time in 2019. Ensemble RI products were generated and posted online for distribution to the community are available from website: https://www.ral.ucar.edu/projects/hfip/d2019/ensRI/.

New diagnostic products are being developed to improve understanding of ensemble rapid intensification (RI) forecasts uncertainty.  The project development focused on several new prototype visualization products colored by a selected diagnostic parameter.  Diagnostic products focused on environmental conditions (wind shear, maximum potential intensity, sea surface temperature, etc.) and inner core storm structure (precipitation symmetry, radius of maximum winds, inertial stability, etc.) were developed.  Prototype visualizations were developed using parameters available from the ATCF a-decks (intensity, minimum sea level pressure, forecast lead time).  Example diagnostic products are shown in Figs. 6 and 7.

Fig. 6: Example plot of the trajectory of each forecast model colored by forecast lead time.
Figure 6. Example plot of the trajectory of each forecast model colored by forecast lead time.
Fig. 7: Example plot of the trajectory of each forecast model colored by its predicted intensity converted to the Saffir-Simpson Hurricane scale.
Figure 7. Example plot of the trajectory of each forecast model colored by its predicted intensity converted to the Saffir-Simpson Hurricane scale.

Tropical Cyclone Guidance Project (TCGP)

In past years, TCGP was mainly a platform to collect and disseminate real-time TC forecast data and associated visualizations. In FY2019, support from multiple projects allowed the first steps of an expansion of TCGP into a platform that can support a wider variety of TC-related activities, such as probabilistic wind exceedance modeling and impact-based risk communication.

New visualization informed by social science

During FY2019, TCGP continued to provide reliable visualizations of the publicly available TC guidance products. While these products are aimed at expert users (e.g., forecasters), they are also used by people in many other disciplines, as well as general publics around the world. There is always the possibility that users may mis-interpret products in decision making. One person asked a question about a TCGP product on Twitter and it caught the attention of social scientists. The user had wondered whether the models were listed in order of accuracy. This query was forwarded to the TCGP developer, Dr. Jonathan Vigh. Then, based on suggestions from Melissa Bica, a PhD candidate in computer science at CU Boulder, a new prototype visualization was created that emphasizes the more accurate forecast aids (e.g., models) and official forecasts, while de-emphasizing forecast aids that have limited skill, but are technically useful to forecasters (e.g., climatology and persistence models which serve as baselines for skill, beta and advection models which are useful diagnostics, and global ensembles which may have poor intensity skill). Figure 8 shows a prototype example of the new visualization. Thick black solid and dashed lines represent the current and previous official forecasts, respectively. Prominent colors and thick lines are used to represent the forecasts of skillful regional models. Less prominent colors (e.g, light pastel colors) with thinner lines represent the forecasts of forecast aids that have less intensity skill, such as global ensembles and climatology and persistence aids. This feedback loop, in which social science informs the design of products, is an example of the type of integrated research needed for science to more effectively serve the public.

Fig. 8: Example of TCGP’s current intensity forecast product (top) and a prototype of the new visualization (bottom).
Figure 8. Example of TCGP’s current intensity forecast product (top) and a prototype of the new visualization (bottom).

Implementing a probabilistic prediction framework for intensity and wind

During FY2019, an HFIP-funded collaboration between the Massachusetts Institute of Technology and NCAR resulted in the addition of a probabilistic framework for predicting TC intensity and wind exceedances. Through supplemental funding from the NCAR Advanced Study Program’s Graduate Visitor Program, Dr. Kerry Emanuel’s PhD student, Jonathan Lin, visited NCAR for three months and implemented his “Forecasts of Hurricanes Using Large-Ensemble Output” (FHLO) framework on TCGP. FHLO uses the tropical cyclone tracks of global ensemble prediction systems to simulate 1000 synthetic tracks with similar characteristics, but which more fully sample the various track scenarios. Then it uses quantities derived from the fields of the global ensemble members to run an intensity emulator to develop physically-realistic intensity forecasts for each of the 1000 tracks. A wind model is then run, allowing wind exceedance probabilities to be computed. Figure 9 shows an example of the 1000 FHLO track forecasts for Hurricane Dorian, while figure 10 shows an example of the intensity forecasts. The effect of the timing of landfall can be seen in the individual ensemble intensity forecasts. The advantage of this framework is that the spatial variability and flow-dependent uncertainty of the large scale environment can be more fully sampled, leading to more realistic probabilistic forecasts than would be possible from a simpler Monte Carlo-type simulation.

Figure 9: Example of ensemble track forecasts for Hurricane Dorian. Red lines show the tracks from the underlying global ensemble. Thin grey lines show the 1000 synthetic tracks of the FHLO ensemble.
Figure 9. Example of ensemble track forecasts for Hurricane Dorian. Red lines show the tracks from the underlying global ensemble. Thin grey lines show the 1000 synthetic tracks of the FHLO ensemble.
Figure 10: Example of ensemble intensity forecasts for Hurricane Dorian. Thin grey lines show the intensity forecasts of the 1000 FHLO ensemble members. Red lines show the percentiles (5th, 25th, 50th, 75th, and 95th) of the FHLO intensity ensemble. Blue line shows the intensity forecast from HWRF.
Figure 10. Example of ensemble intensity forecasts for Hurricane Dorian. Thin grey lines show the intensity forecasts of the 1000 FHLO ensemble members. Red lines show the percentiles (5th, 25th, 50th, 75th, and 95th) of the FHLO intensity ensemble. Blue line shows the intensity forecast from HWRF.

TCGP website: http://hurricanes.ral.ucar.edu/

Tropical Cyclone Data Project (TCDP)

Research Use of TC Datasets of Aircraft and Satellite Observations

In FY2019, TCDP continued to support the research community by hosting and maintaining high-quality datasets of aircraft flight level data and a historical database of TC wind structure parameters derived from these. As shown by counts of registered users in Table 1, research uses of these datasets has continued to increase. Many of the users are graduate students who are using the datasets in their thesis or doctoral research. Each of these datasets and the historical database were first released in FY2016. Since then, a number of papers have been published using TCDP datasets.

Table 1: Registered user statistics for TCDP datasets as of 14 Nov 2019.
Table 1: Registered user statistics for TCDP datasets as of 14 Nov 2019.

FLIGHT+ dataset extension and development of a new data product

Through funding from HFIP, work commenced in FY2019 to extend FLIGHT+ to include the recent 2016 - 2018 seasons. These seasons, which included impactful TCs such as Hurricanes Harvey, Irma, Maria, and Michael, are of considerable interest to researchers. The codeset used to process the data was transitioned to a GitHub repository and updated to support a multi-developer environment. Also, a new Level 4 data product was added to provide users with the azimuthal means of flight level data. An example is shown in Figure 11, in which an azimuthal mean is computed from the various constituent radial legs in each of the four quadrants of Hurricane Maria. Beyond ~55 km, no azimuthal mean is computed because radial data were missing in at least one quadrant. Evidence of multiple wind maxima associated with an eyewall replacement cycle is evident in both the individual radial legs and the azimuthal mean. This new capability can help to understand inner core structure changes, further increasing the usefulness of FLIGHT+ for process studies and development of new prediction methods.

TCDP website: https://verif.rap.ucar.edu/tcdata/

Figure 11: Example showing azimuthal mean of tangential wind for Hurricane Maria. Individual flight legs are shown in red. Azimuthal mean for this group time is shown in blue.
Figure 11. Example showing azimuthal mean of tangential wind for Hurricane Maria. Individual flight legs are shown in red. Azimuthal mean for this group time is shown in blue.

The Hurricane Risk Calculator

Work commenced on the several components of this UCAR-funded project, including refactoring TCGP to support the probabilistic prediction framework (FHLO) that will be used to drive the wind hazard side of the Hurricane Risk Calculator. Also, a researcher collective was established to bring a wide range of experts together from various disciplines including meteorology and physical modeling, structural engineering, web and cloud engineering, social science, emergency management, and verification. The collective is organized into teams around their respective disciplines and will share results with the wider collective as the project moves along. We expect that the various teams will work to secure further funding to take each component of this vision forward.

FY2020 PLANS

Developmental Testbed Center

For FY2020, the DTC will continue its work aimed at improving the HWRF physics through partnerships with physics developers. The performance of alternate physics schemes and innovations to the current parameterizations within the HWRF physics suite will be investigated. Retrospective forecasts using the most recent HWRF model version will be conducted to evaluate the performance of each innovation. Upgrades to the atmospheric component of HWRF will be passed to EMC to be included in its pre-implementation testing for the HWRF 2020 implementation. Additionally, efforts will transition from the operational HWRF system to the Unified Forecast System (UFS) hurricane application, Hurricane Analysis and Forecasting System (HAFS).

Tropical Cyclone Modeling Team

For FY2020, the TCMT will continue to enhance graphical display diagnostic tools with additional analysis features and continue to include the additional gridded fields (satellite, forecast products, SST observations). New diagnostic tools are being developed using input from the National Hurricane Center (NHC) staff.  Additionally, ensemble rapid intensification (RI) forecasts for the 2020 hurricane season will be evaluated and made available through the TCMT website.  The TCMT is also developing new techniques to evaluate ensemble RI forecasts.

Tropical Cyclone Guidance Project

For FY2020, TCGP will continue to expand into a platform that support development and evaluation of additional TC-related data services, prediction tools, and risk communication methods. TCGP is collaborating with the Shanghai Typhoon Institute to set up an international multi-nodal network for sharing TC forecast and observational data. TCGP is also being refactored to run as a cloud service in the Amazon Web Service (AWS). This should open the door to commercialization possibilities. TCGP will begin providing probabilistic data products from FHLO. TCGP will also support real-time testing of forecast aids based on machine learning techniques, such as a logistic regression-based forecast aid (HLOG).

Tropical Cyclone Data Project

In FY2020, TCDP will release a major update to the FLIGHT+ Dataset, extending the dataset to cover the years 2016 - 2018. The release will also include the new azimuthal mean data product and fix several known issues with the data processing of certain storms and data sources. It is also a goal to begin processing the flight level data in real-time and providing this as a data service to support prediction tools in TCGP.

Hurricane Risk Calculator

In FY2020, the Hurricane Risk Calculator will demonstrate the new risk calculation and communication framework in a web application. A retrospective study will be conducted to determine the accuracy of the framework by comparing predicted damage states for retrospective forecasts of major past events with the actual assessed damage for approximately 1000 structures. A key goal is to find a commercial partner to demonstrate these capabilities in a consumer-facing mobile app.