[Background] [In-situ turbulence]
[Remote sensing] [Turbulence forecasting]
[Turbulence characterization]

1. Background

RAP has been involved in a number of research and development areas over the past several years aimed at better understanding turbulence as it relates to aviation safety. Federal Aviation Administration (FAA)-supported aviation turbulence forecasting continues to be a major work area. RAP turbulence research areas have traditionally included work aimed at improving and implementing methods for better measurements of turbulence, either in-situ based or using remote sensing devices such as radar and lidar (either ground-based or airborne). Post-911 research has also begun focusing on turbulence processes in the urban boundary layer. Due to our still limited understanding of turbulence processes in the boundary layer and free atmosphere, research is being conducted to better characterize turbulence based primarily on field measurement campaigns and high resolution numerical simulations. RAP turbulence research is currently sponsored by the FAA Aviation Weather Research Program (AWRP), the NASA Advanced Satellite Aviation Weather Products Program (ASAP), and DARPA.

2. In-situ turbulence measurement algorithm

Under the sponsorship of the FAA, work was begun in the early 1990’s at NCAR to develop and deploy an in situ turbulence measurement and reporting system for commercial aircraft. The concept was to use existing sensors, avionics and communication networks to produce and disseminate a state-of-the-atmosphere turbulence metric – the eddy dissipation rate (EDR). These data would then be used by a variety of users for operational and scientific purposes. Operational users include pilots, airline dispatch and meteorology personnel, aviation forecasters, and air-traffic personnel. Furthermore, these data would also be used by the turbulence research and development community for building and improving turbulence detection, nowcast and forecast products.

The EDR reports are intended to augment the existing turbulence pilot reporting data. As is well-known, these pilot reports (PIREPs) are subjective measures of the aircraft’s response to the turbulence, as opposed to quantitative, state-of-the-atmosphere measurements. Furthermore, PIREPs are sporadic in space and time, and very few null reports are made. The EDR reporting system was designed to address many of the deficiencies with pilot reports. That is, to provide routine and quantitative measurements of atmospheric turbulence intensity levels – including null reports.

A turbulence detection product is currently being developed by John Williams, Larry Cornman, and Danika Gilbert for use with ground-based Doppler radars. As part of the algorithm verification, a number of data sources are being used, such as the EDR reports. In the following figures, a case study is presented which demonstrates the utility of the EDR reports in the development and verification of the radar turbulence product. Figure D-1 is a composite reflectivity field for 31,000 ft using four nearby WSR-88D radars. The three turbulence encounters can be seen in this figure, in east-central and western Iowa, and in eastern Nebraska. Note that the reflectivity values are quite low, and in fact the westernmost encounter is in clear air. At these reflectivity levels, airborne radar would not suggest anything that would be of concern to a pilot.

Figure D-1. Reflectivity in dBZ for the 18 November 2003 flight.

Figure D-2. Comparison of EDR values from the aircraft reports and the WSR-88D algorithm, for the 18 November 2003 flight. Valid time for the radar data is 00:30 UTC. EDR values produced by the prototype radar detection algorithm, show good agreement for the two in-cloud encounters.

The Graphical Turbulence Guidance (GTG) product, developed in RAP by Robert Sharman, Gerry Wiener, Jamie Wolff and Rod Frehlich, is an automated turbulence nowcast and forecast diagnostic system that is used operationally by aviation forecasters. It currently uses pilot reports of turbulence to appropriately weight the forecast elements. There are a number of drawbacks to the use of pilot reports, and hence efforts have been initiated to use the EDR reports to augment the pilot reports within GTG. One advantage of the routine reporting of turbulence is the increased spatial and temporal coverage over event-based reporting – such as occurs with pilot reports. Routine reporting of turbulence will allow for localized weighting schemes in GTG, which should improve the overall performance.

One of the promising new diagnostics being developed for GTG is based on a structure function analysis of the model velocity grids (Frehlich and Sharman, 2004). This diagnostic produces an EDR field, albeit one that is averaged over the spatial scales of the model grid. An example of this EDR diagnostic – for the horizontal structure function – is shown in Figure D-3. Note that the enhanced levels of turbulence are diagnosed in the regions with large gradients in the horizontal wind. The EDR and pilot reports for this time period are overlaid. It can be seen that there are far more EDR reports than pilot reports.

Figure D-3. Model-based EDR diagnostic field for 31 kft, valid at 00 UTC. EDR reports are shown as open circles (EDR1/3= 0-0.1), orange (0.1-0.2), and red (0.2-0.3). Pilot reports for that flight level are also shown.

[Top]

3. Remote sensing of turbulence

Remote detection of turbulence has the potential to provide an important new source of information to the aviation community, supplementing turbulence forecasts and in situ reports with real-time information about in-cloud turbulence, including the dynamically-evolving turbulence associated with convection that is responsible for a majority of aircraft turbulence encounters. RAP scientists J. Williams, L. Cornman, and D. Gilbert, along with software engineers Dave Albo, Steve Carson, and Jaimi Yee, have been working to develop, implement, and verify radar-based turbulence detection algorithms that make use of existing airborne and ground-based Doppler radars. The ground-based detection work has been supported by the FAA's Aviation Weather Research program, while the NASA Aviation Safety and Security Program funded the airborne radar work.

In the area of airborne turbulence detection, the principal accomplishment this year was the completion of a MATLAB software package that provides automated scoring tools for evaluating airborne turbulence detection performance using either flight test or simulated data. This software was designed to mimic human subjective scoring that accounts for the operational implications of warning or failure to warn, yet perform the analysis quickly and objectively. The data visualization and scoring capabilities of the software were demonstrated using data from the NASA B-757's Spring 2002 flight tests that evaluated the NCAR/ATR airborne turbulence detection algorithm, and the scoring results reconfirmed its excellent performance. In addition, four simulated radar and aircraft data cases were produced and scored with the automated algorithm to demonstrate how radar manufacturers' algorithms could be objectively evaluated as part of the FAA certification process. Following a successful presentation and delivery to NASA, the automated scoring software became part of a toolset available to researchers and radar manufacturers for evaluating and tuning their airborne turbulence detection systems.

Efforts to develop a real-time, ground-based turbulence detection capability also made substantial headway this year. The NCAR Turbulence Detection Algorithm (NTDA), a new Doppler radar turbulence detection algorithm designed for use on the operational NEXRAD and TDWR radars, was further refined and tested. This fuzzy logic algorithm makes use of radar-measured reflectivity, radial velocity, and spectrum width to produce estimates of eddy dissipation rate (EDR), an aircraft-independent measure of turbulence intensity, along with an associated quality control confidence for each value. The NTDA was verified by running it on archived NEXRAD data obtained from the National Climatic Data Center and comparing the radar-derived EDR with in situ EDR values from field programs, commercial in situ reports, and Flight Data Recorder information provided by the National Transportation Safety Board. These comparisons suggest that the NTDA often detects hazardous in-cloud turbulence well before an aircraft encounter, though further refinements may be necessary to reduce over-warning. An example of the NTDA's potential utility was provided by an analysis of an Airbus A340 turbulence accident that occurred at 31,000 feet over northeast Arkansas on 6 August 2003, injuring forty-five passengers and flight attendants. In that case, the NTDA successfully detected severe turbulence at the location of the encounter 14 min in advance (see Figure D-4 ), despite radar reflectivity so low that it was invisible to the pilots on their airborne weather radar. An analysis of aircraft and NEXRAD data from the NASA B-757 Spring 2002 flight tests using overlay plots like the one depicted in Figure D-5 provided a more comprehensive evaluation. This analysis concluded that, among the forty flight test "events" for which NEXRAD data were available, the NTDA successfully detected moderate-or-greater turbulence 94% of the time, with a nuisance alarm rate of just 16%. However, it is not clear that these flight test cases are representative of those encountered by commercial aircraft; a more reliable measure of the NTDA's skill will be provided by ongoing comparisons of NTDA EDRs with commercial in situ data.

Figure D-4. Eddy dissipation rate (EDR) turbulence detection produced by the NTDA using data from a 2.4° elevation sweep from the KPAH (Paducah, KY) NEXRAD at 20, 14, 8, and 3 minutes before the severe turbulence encounter at the location denoted by the "X". EDR values are colored from 0 (blue) to 0.7 m^2/3s^-1(red) as indicated by the colorbar at right, and the plot axes display distances from the radar in km.

Figure D-5: In situ EDR values obtained from the 15 April 2002, NASA B-757 flight test data along a seven-min segment of the flight track, superimposed over NTDA EDR values from the KLTX (Wilmington, NC) 2.4 elevation sweep completed midway through this time interval. Both aircraft and radar EDR values are on the same scale as Figure D-4, ranging from 0 (blue) to 0.7 m^2/3s^-1 (red).

Finally, a major step towards an operational implementation of the NTDA was completed with the design and implementation of an algorithm for merging the EDR values from multiple radars onto a 3-D mosaic, providing a "map" of turbulence hazard at each flight level. The merging methodology works by centering a Gaussian distance weighting function at each point on a regular latitude-longitude-altitude grid having spacing of approximately 2 km x 2 km x 2,000 ft. A confidence- and distance-weighted median of the EDR values from all radar sweeps completed during a prescribed time window is then computed. This method automatically gives more weight to values from nearby radars, since they have a higher density of data points, and effectively interpolates through data-poor regions. Sample output from the merging/gridding method is illustrated in Figure D-6.

Figure D-6: Turbulence hazard map at altitude 25,000 ft over southern Virginia and eastern North and South Carolina valid at 20:27 on 15 April 2002. The map was produced by merging the NTDA EDRs from the KFCX, KAKQ, KRAX, KMHX, KCAE, KLTX, and KCLX NEXRADs, whose locations are labeled in red on the plot. Reflectivity contours at 15 dBZ are superimposed in light blue. A 10-min segment of the B-757's flight track is overlaid just below the position of the KMHX NEXRAD. It depicts a severe turbulence encounter in low reflectivity that was nevertheless correctly detected by the radars.

4. Turbulence forecasting

Over the last several years, the FAA has funded RAP scientists (R. Sharman, R. Frehlich, J. Wolff, G. Wiener) to develop a turbulence diagnosis and forecast system for upper-level turbulence over the continental U. S. The forecast system, Graphical Turbulence Guidance (GTG), provides contours of turbulence potential based on RUC model forecasts out to 12-hours lead time. The original system, since March 2003, is part of the NCEP operational suite (available through the ADDS web site), produces forecasts of clear-air turbulence > 20,000 ft MSL. But there is a need to extend the forecast altitude range down to mid-levels, as low as 10,000 ft MSL. In order to do this, other sources of turbulence besides traditional clear-air sources related to jet streams and upper-level fronts must be considered. Turbulence diagnostic algorithms are required that can diagnose turbulence potential from numerical weather prediction model output, regardless of the source of turbulence. R. Frehlich is developing a technique to estimate small-scale turbulence from local estimates of the spatial structure functions of model variables, such as the horizontal velocity, vertical velocity and temperature to address this issue. The key assumptions used are the existence of a universal statistical description of small-scale turbulence and a locally universal spatial filter for the model variables. Under these assumptions, spatial structure functions of the model variables can be related to the structure functions of the atmospheric variables.

The shape of the universal spatial filter is determined by comparisons with the spatial structure function from aircraft data collected at cruising altitudes. This universal filter is used together with isotropic turbulence theory to estimate the small-scale turbulence levels. A simple yet universal description of the basic statistics (such as the probability density function and the spatial correlation) of these small-scale turbulence levels in the upper troposphere is predicted. The critical input parameter is the dimensions of the averaging domain used for the estimates of turbulence. Simple scaling laws as a function of the averaging domain were produced, as well as the fundamental statistical description of the spatial variability of the turbulent field. Various applications of these new results include:

• predicting the statistics of turbulence experienced by commercial aircraft;
  
  
• diagnosing and forecasting turbulence for aviation safety;
  
  
• estimating the total observation error for improving data assimilation; and
  
  
• improved sub-grid parameterization for operational weather prediction models.

Based on these results, the total observation error for typical rawinsonde measurements of velocity were found to be dominated by the sampling error or "error of representativeness" resulting from the effects of small-scale turbulence. Therefore, increasing the accuracy of rawinsonde data provides marginal improvements in total measurement error.

Any comparison of turbulence diagnostic metrics at upper levels produces large spatial variations as well as differences between the different metrics. This may reflect the sensitivity of the different metrics to the different physical processes that produce turbulence. An example of predicted turbulence from the EDR extracted from the local structure functions of horizontal velocity and the predictions of rms vertical velocity extracted from the local structure functions of vertical velocity are shown in Figure D-7. The EDR appears to be more indicative of fronts and jets while the vertical velocity appears to indicate convective activity. The predictive skill of turbulence predictions for the two noticeably different metrics is similar, indicating the need to effectively combine all the turbulence metrics with an optimal algorithm (i.e., GTG).

Figure D-7. Example of EDR estimates (upper panel) and w estimates (lower panel) derived from a RUC analysis at 10 km elevation.

Using these new turbulence diagnostics, together with other new diagnostics and using better algorithm optimization methods has led to a considerable improvement in predictive skill as measured by the ability to correctly predict moderate-or-greater turbulence encounters (PODY) and smooth air encounters (PODN). The relative PODY-PODN performance is shown in Figure D-8 for the original GTG version, GTG1 and the new experimental version, GTG2, for both mid-levels and upper-levels. As illustrated, the improvement in performance is significant.

Figure D-8. PODY-PODN performance curves derived from comparisons of GTG 6-hr forecast predictions to pireps over a three-month winter period valid at 18Z. [Top]

5. Turbulence characterization

5.1 Characterization of mountain wave turbulence

In FY04, R. Sharman, Bill Hall, and Teddie Keller conducted a series of numerical experiments aimed at better understanding mountain wave-induced turbulence (MWT) with the goal of developing improved MWT diagnostics. The feasibility of using multi-nested, high-resolution, non-hydrostatic numerical simulation models in an operational mode was also investigated. An initial assessment of the feasibility of using the Clark-Hall model in a semi-operational mode with one of the higher resolution nests placed over the Colorado Rockies was performed. Sensitivity tests of model resolution demonstrated that a horizontal resolution of 1-4 km is adequate to resolve most gravity waves and correctly identify likely turbulence regions. As would be done in an operational setting, initial and boundary conditions to the outer domain of the Clark-Hall model were provided by an operational NWP model, in this case, the RUC. One intense evaluation focused on a commercial airliner turbulence encounter over the Rocky Mountains near Alamosa, Colorado.

This case was chosen because actual flight data recorder (FDR) information during the event was made available by the operating airline. Comparing with the FDR data eliminated the inherent difficulties involved with turbulence PIREPs, such as inaccuracies in position and turbulence intensity. The simulation results indicated the presence of mountain waves and wave breaking in this region, with flow reversal, large vertical velocities and accompanying turbulence, all of which were verified using the FDR information (see Figure D-9 a and b). The success of using the Clark-Hall model for this initial case based on comparisons to the high resolution FDR confirms the validity of this approach. Further tests are ongoing using PIREPs for turbulence observations, for both turbulent and non-turbulent cases.

(a)

(b)

Figure D-9. Color contours of the east-west wind velocity (m^s-1) on a vertical slice along the flight track of the aircraft. The position of turbulence encounter is marked by the X. Note the region of stagnant (white) and reversed flow (green) just upwind of incident site. (a) 4-km horizontal resolution, (b) 1-km horizontal resolution.

5.2 Characterization of convectively induced turbulence

Deep convective clouds generate turbulence both in-cloud and out-of-cloud. Out-of-cloud convectively-induced turbulence (CIT) is poorly understood. A recent study by Lane et al. (2003) examined a single case of above-cloud CIT, and found that it was possible for the turbulence to extend far above the cloud. It was also shown that the turbulence was due to breaking gravity waves aloft, and the breaking of these waves was controlled in part by the wind shear above cloud top. The results of the Lane et al. study also suggested that the current FAA guidelines for above cloud CIT avoidance may be insufficient to avoid turbulence in all cases. Todd Lane and R. Sharman began studies to expand the results of the Lane et al. simulations to provide more general conclusions. Specifically a series of high-resolution numerical simulations with background flow conditions, including wind speed, wind shear, and stability were systematically varied to evaluate their effect on the turbulence generation processes. Figure D-10 (a and b) shows a sample from the simulation results. Turbulence, as measured by the subgrid turbulent kinetic energy (TKE), obviously extends to high altitudes above the cloud. Preliminary results suggest that:

• The vertical extent of turbulence, and area of turbulence above convection may be approximately inversely proportional to the lower-stratospheric stability.
  
  
• The speed of the cloud-top jet does not necessarily affect the area or intensity of turbulence above the cloud, but controls the processes that generate the turbulence. For example, in low wind speed cases, the turbulence is derived from propagating waves, and in high-wind speed cases, the turbulence is derived from breaking waves. However, the contribution of the propagating waves may be different in a 3-D model.
  
  
• When the maximum wind speed is constant, high wind shear inhibits deep turbulent layers, and some (moderate) wind shear provides optimum conditions for turbulence to develop over deep layers above cloud top.

D-10 (a and b). Sample cloud simulation results after 60 min of model time for two model runs. Blue shading denotes clouds. Yellow, orange, and red shading denotes regions with out-of-cloud total TKE. Note both simulations show the regions of gravity wave propagation and turbulence to extend considerable distances above the cloud. [Top]

Further work is planned for 2005 to extend the parameter space of simulations to include cases of less intense and more shallow convection, and more highly sheared wind profiles. In addition, under sponsorship from the NASA ASAP (Advanced Satellite Aviation-weather Products) we have performed high-resolution numerical simulations of a severe turbulence encounter by NOAA's G-IV research aircraft above a convective storm over the North Atlantic as part of the NORPEX ATReC (Atlantic-THORPEX Regional Campaign). The analyses of these simulations is ongoing, but tentatively it seems that the turbulence in this case was related to two factors: convection created inertia-gravity waves (IGW) that propagated vertically above the storm; and the large-scale Richardson number (Ri), which is reduced in jet stream regions above the storm, leading to local IGW-induced reductions in Ri to unstable values.

5.3 Characterization of island wake turbulence

R. Sharman and T. Lane used high resolution numerical simulations to investigate the lee side flow dynamics of island wakes, in particular the wake of the Hawaiian island of Kauai. The investigation was motivated by the breakup of the Helios solar-powered unmanned vehicle in the wake of Kauai on 26 June 2003. The simulations showed the persistent presence of shear lines that were mechanically forced by flow around the Kauai topography. In the strong wind cases, the magnitude of the shear was large enough to create lateral shear instabilities leading to breakdowns in the structure of the shear lines and turbulence. The location and intensities of the shear lines are highly transient. The shear lines extend considerable distances downstream before loosing their coherence to turbulent eddies. Sensitivity studies indicate that the low-level vertical structure of the ambient flow, as opposed to vertical stratification (stability), is primarily responsible for strength of the shear lines. This is consistent with the concept that the island of Kauai is a blocking obstacle to the usual easterly trade winds, creating large velocity deficits in its wake. The stronger the incident trade winds, the larger the deficit and the stronger the shear transition zones. An example of the simulation results is shown in Figure D-11. As can be seen, the island wake is highly transient and turbulent. Relative to the Helios demise it is of course not possible to duplicate the turbulent conditions in the Kauai wake, however, the simulations do indicate that the turbulence in the vicinity of the shear zones, if encountered, could have been significant.

Figure D-11. Contours of wind speed and sgs tke (orange) at z=1000m for model times 310 min through 360 min derived from a 167 m resolution simulation of the wake of Kauai. The heavy line in the right half of the figure is the Kauai western coastal boundary.

5.4 Characterization of building wake turbulence

R. Sharman and Piotr Smolarkiewicz (MMM) have conducted numerical studies of urban boundary layer flows and pollutant transport and dispersion using the non-hydrostatic model EULAG. The particular objective of this research is to quantify the air flow past the Pentagon building under various meteorological conditions, with representative building surface heating/cooling and roughness. Particular emphasis is on predicting the concentration levels of hazardous contaminants from intentional or inadvertent releases upstream of the Pentagon as they disperse and collect in various recesses of the building. A series of simulations for both high and low Reynolds number flows are required, to complement and compare to field observations and wind tunnel measurements of a Pentagon model. So far, the simulations have focused on computing high Reynolds number flows for various choices of physical and numerical parameters, in order to assess the optimal model configuration for practical predictability of the near-field flow realizations. Among others, the simulations using standard terrain-following coordinates were compared with an immersed-boundary approach. Contrary to restrictions experienced with other numerical simulation models, EULAG is able to effectively represent steep urban structures as terrain-following orography (Figure D-12), and produces results qualitatively similar to those where embedded structures are represented by fictitious forces in the equations of motion. In general, this work will lead to better representations of internal boundaries in atmospheric and oceanic numerical simulation models.

[Top]

Figure D-12. Simulation of a flow past the Pentagon building, with 2 and 1 m resolution in the horizontal and vertical respectively, using Gal-Chen & Somerville coordinate transformation. Contours of the square root of turbulent kinetic energy are shown in the xy cross section at z=10 m (top) and in the central yz cross section (bottom).

5.5 Characterization of urban boundary layers and boundary layer turbulence

In an effort to better understand the structure of the urban boundary layers and boundary layer turbulence, R. Frehlich and R. Sharman have analyzed Doppler lidar measurements and high resolution tethered sonic instrumentation to evaluate the boundary layer structure in stable, neutral, and unstable conditions. The data was gathered during an extensive field campaign conducted in the Washington DC area to characterize the atmospheric processes in an urban environment. The key measurements were provided by a 32-m tower instrumented with 3-D sonic anemomemeters, a sodar, a tethered blimp with high-rate velocity and temperature sensors, and a high-resolution Doppler lidar. One of the main goals of the campaign was to evaluate the ability of the Doppler lidar to extract accurate profiles of wind speed, direction, and turbulence information. An example comparison is shown in Figure D-13 and excellent agreement is produced for wind speed and direction as well as the turbulence level for well behaved conditions. More work is required to correctly remove the effects of background wind shear from the estimation algorithm, especially at the top of the boundary layer. This is the first comparison of in situ profiles of small scale turbulence and Doppler lidar estimates which have the potential to provide operational data in the boundary layer, as well as accurate profiles of wind speed and direction.

Figure D-13. Comparisons of lidar-derived (green circles), sodar-derived (blue circles), and tethered blimp high rate data (black lines) for respectively from left to right, wind speed, wind direction, temperature, velocity turbulence, thermal turbulence, and lidar backscatter for a 15-min. period taken near the Pentagon building near Washington DC on 11 May 2004.

[Top]

RAP Achievements ... In-flight Icing Snowfall and Freezing Convective Weather Atmospheric Turbulence Numerical Weather Prediction Water Resources Land Surface Atmospheric Remote Sensing Precipitation Enhancement Ceiling and Visibility Oceanic Weather Surface Transportation Statistics Applications NCAR Iinitiatives