D. Atmospheric Turbulence

[Background] [In-situ measurement] [Remote Sensing of Turbulence] [Turbulence Forecasting]
[Turbulence characterization] [Juneau Terrain-Induced Turbulence Project]

1. Background

RAP has been involved in a number of research and development areas over the past several years aimed at minimizing the number and severity of aircraft encounters with turbulence.  The work areas are divided into two complementary directions: 1) to improve and implement methods for better diagnosis of turbulence, either in-situ based or using remote sensing devices such as radar and lidar; and 2) to develop, implement, and verify automated forecasts of upper level turbulence.  In support of both areas, research is being conducted to improve our understanding of the nature or character of aircraft scale turbulence, based primarily on field measurement campaigns and high-resolution numerical simulations to recreate and study in detail actual turbulence encounters.  This work is sponsored by the FAA Aviation Weather Research Program (AWRP) and the NASA Aviation Safety Program (AvSP).

2.  In-situ measurement and reporting system

The only method currently available to obtain actual observations of aircraft-scale turbulence is the manual pilot reporting system (PIREPs).  Typically each report contains time, position and intensity information.  But there are some problems inherent in this system.  First, because a specific action is required by the pilot to initiate the report, typically only a few observations, sometimes none at all, are reported per flight leg. Second, the data are only available in regions containing air traffic; therefore there are very few reports at night and in remote areas.  Third, not all the reports received are actually entered into the automated systems.   Consequently, even over an hour’s time, there may be few PIREPs in the automated system, denying a complete mapping of the turbulence.  Besides this underreporting problem, the PIREPs that do actually make it into the automated system are subject to human error.  In order to mitigate these problems, NCAR RAP has developed an automated turbulence reporting system that takes in-situ measures of aircraft acceleration, attitude, mass, etc., and backs out an aircraft-independent measure of atmospheric turbulence.  This metric, currently established as an eddy dissipation rate (edr), is automatically computed and recorded every minute of cruise flight, and the maximum and mean are automatically downlinked to the surface, along with meteorological information, every 4 minutes.  The system is a software package that is easily loaded onto ACARS-equipped aircraft.  The result is many, perhaps several hundred, turbulence measurements per flight accompanied by precise position and time information.

This year, RAP scientists L. Cornman, C. Morse, and G. Meymaris concentrated on tuning and verifying the algorithm.  The results of one sample verification exercise are shown in Figure D1.

3. Remote Sensing of Turbulence

In FY02, research related to the detection of atmospheric turbulence by remote sensing devices continued.  Specifically, this research has been focused on two tasks: (1) the development of an improved turbulence detection algorithm for both airborne Doppler radars and ground-based WSR-88D (NEXRAD) radars, and (2) the development of turbulence detection algorithms for Doppler lidars.  This work continues to be directed by RAP scientist L. Cornman, in collaboration with R. Frehlich, S. Dalton, G. Meymaris, and J. Williams.

The airborne turbulence detection activity is sponsored by NASA’s Aviation Safety Program. A research team comprised of NASA, NCAR, radar manufacturers and others has been working to develop reliable methods for using existing forward-looking airborne Doppler radars to detect approaching convective turbulence.  Research and development activities included cloud-scale modeling of actual convective turbulence encounters, radar and aircraft simulations using both cloud model and analytical turbulence wind fields, and the enhancement and testing of a prototype hazard detection algorithm. This spring, flight tests of the hazard detection algorithm were conducted using the NASA B-757 research aircraft.  A rigorous verification found that turbulence was accurately detected at least 30 sec in advance for 80% of the encounters; moreover, the detection was often successful even in regions returning only a weak radar signal. This verification was performed on 55 turbulence encounters using overlay plots such as the one shown in Figure D2.

The WSR-88D activities this year continued to focus on the development and verification of an improved turbulence detection algorithm, making use of WSR-88D and co-located in-situ aircraft data collected during the STEPS-2000 field experiment.  This work follows on similar efforts using Mile High radar data from RAPS-92 and CSU-CHILL radar data from TCAD-99 that produced very promising results.  Unfortunately, the quality of the WSR-88D spectrum width data has proved quite problematic, and because these data are a primary source of information for the turbulence algorithm, a substantial effort to develop appropriate quality control techniques was required.  Statistical comparisons of turbulence estimates derived from the radar data with CO-located in-situ values helped yield both quality control and turbulence algorithm improvements, while overlay plots such as the one shown in Figure D3 have been a principal tool in individual case studies.

Figure D3. Turbulence eddy dissipation rate (EDR) estimated from a Goodland, Kansas WSR-88D radar scan, overliad with in-situ EDR turbulence measurements derived from the South Dakota School of Mines and Technology's T-28 research aircraft. Axis labels indicate kilometers from the radar, and the white arrow shows the position of the aircraft at the time of the radar scan.

Lidar detection capabilities were analyzed by RAP scientists R. Frehlich and L. Cornman using a combination of idealized (von Karman) wind field simulations and lidar detection simulations.  In particular, the accuracy of lidar estimates of eddy dissipation rate (edr) was determined with computer simulations.  The accuracy depends of two components: sampling errors and estimation errors.  A new algorithm for correcting the estimation error was developed and tested on both simulated and actual field data from the Juneau field program.  The estimates of edr were shown to be very accurately determined by lidar measurements for boundary layer type turbulence, but the length scale determination is more difficult.  This work was published in

Frehlich, R. and L. Cornman, 2002: Estimating spatial velocity statistics with coherent Doppler lidar. Journal of Atmospheric and Oceanic Technology, 19, 355-366.

[Top]

4.  Turbulence forecasting

Over the past several years RAP scientists and engineers R. Sharman, G. Wiener, S. Dettling, and J. Wolff developed and tested an automated algorithm for forecasting upper level turbulence locations and intensities.  The product, ITFA, (Integrated Turbulence Forecasting Algorithm), is intended to provide end users (e.g., aviation forecasters, airline dispatchers, and pilots), with easy to understand, yet accurate, nowcasts and forecasts.  Forecast images of likely turbulence regions are produced automatically based on RUC2 NWP model output which updates every 3 h.  From the NWP output, various turbulence indices, empirically or theoretically derived, are computed and combined using fuzzy logic to give the best agreement with available PIREPS.  The resulting images are provided on two web sites, the NCAR site (http://www.rap.ucar.edu/projects/itfa) and the NWS Aviation Digital Data Service (ADDS) (http://adds.aviationweather.noaa.gov/).  A sample image is shown in Figure D4, along with available PIREPs.  Performance metrics for all algorithms were derived by the methods outlined in the next section.  At this point the performance of the algorithm has been evaluated by the FAA and NWS as being acceptable for “guidance” use by aviation meteorologists.

Figure D4.  Example of an ITFA nowcast image with available PIREPs overlaid.  The color code is white=smooth, blue=moderate, yellow=severe, red=extreme.  In this case the algorithm correctly diagnosed turbulence as moderate.

The derivation and testing of new diagnostics and optimal strategies for combining the various diagnostics are areas of continued research.  This year, B. Sharman, J. Wolff, and G. Wiener, concentrated on making the algorithm more robust and added an empirically-based mountain wave turbulence algorithm. The mountain wave turbulence algorithm is based in part on a climatology of mountain wave turbulence PIREPs. Figure D5 shows contours of mountain wave turbulence reports taken over a 10 year period.  Even though the entire western half of the continental U. S. is considered “mountainous”, only selected regions within the major mountain chains seem to be conducive to creating turbulence.  For forecasting purposes then, only those regions indicated by the contours are considered.

In conjunction with the Oceanic Weather Program sponsored by the FAA, the ITFA methodology is being expanded (by C. Wolff and B. Sharman) to include turbulence nowcasts and forecasts over the entire globe by diagnosing NOAA’s AVN model.  Preliminary results of this effort are shown in Figure D6. This methodology is also being transferred (by G. Wiener, B. Sharman S. Dettling, and C. Caravone) to the Korean Meteorological Agency to run with their MM5 forecast model.

Figure D6.  Example computation of a turbulence index (Brown’s index) based on input from NOAA’s AVN global model.  This is one of the algorithms in the ITFA suite of diagnostics.  In this example the altitude is 32,000 ft and the time is 6 Aug 2002 18Z.

RAP has also investigated alternatives to the fuzzy logic approach of merging the individual turbulence diagnostics. C. Tebaldi is examining logistic regression, in which diagnostics are thresholded and merged based on Probability of Detection (POD) considerations. In this method, two PODs are evaluated, the probability of moderate-or-greater (MOG) turbulence detection (PODY), and the probability of null or smooth turbulence detection (PODN). Results of this method show superior performance to the current fuzzy logic approach (Figure D7), based on PODY-PODN tradeoffs.

Another area of research is the assessment of turbulence encounters in clear-air versus cloud.  RAP scientist J. Wolff has looked at the percentages of turbulence encounters above 20,000 ft in and out of cloud by comparing turbulence PIREPs to humidity data extracted from radiosonde ascents over the continental U.S.  (Figure D8).  At those altitudes MOG turbulence encounters are about 2.5 times more frequent in clear air than in cloud, but it must be remembered that the volume occupied by cloud is on the average smaller than the volume occupied by clear air (~20% cloud based on these studies).

Figure D8.  Percentages of turbulence encounters above 20,000 ft (null=blue, tan=light, red=moderate or greater) that were in cloud (left) and out of cloud (right) as deduced from comparisons to PIREPs in the vicinity of 0Z and 12Z radiosonde ascents over the continental US.

5.  Turbulence characterization

In order to evaluate turbulence sensor performance and its relation to aircraft response, numerical simulations of “turbulence” have been developed for sensor performance assessments.  A major difficulty in establishing data sets by simulation is that the range of scales that effect aircraft motion encompass about two decades, from about 20m to 2 km.  With current computational resources it is not feasible to make a simulation that will produce valid results for this full range of scales.  As an alternative, under the sponsorship of the NASA AvSP, RAP scientists R. Frehlich, R. Sharman, and T. Keller developed a technique to merge the results of mesoscale simulation output (~100 m) with an isotropic von Karman turbulence analytic representation of the smaller scales (“subgrid”) to get a realistic (in terms of spatial statistics) wind field data set valid at all length scales of importance to aviation applications. Figure D9a is an example of the spatial statistics (structure functions) derived from the merger process and Figure D9b shows the resultant vertical velocity fields.  In this process anisotropies in spatial statistics at the larger scales resolved by the mesoscale model are transferred to the smaller scales.  This raises a question about the validity of the technique, since as the scales become small enough, the turbulence would be expected to become isotropic.  However, those scales may be smaller than the scales important for aircraft response to turbulence (~ 25 m).  To test this, high rate wind data derived from NASA flights was examined for both low-level continuous turbulence events and higher-level shorter-lived discrete events.  Spectral levels for each velocity component were computed and compared to derive the levels of anisotropy.  As expected, the quieter regions are nearly isotropic, but the discrete events tend to be anisotropic down to the smallest scales resolvable by the aircraft instrumentation (~50 m).  Figure D10 shows an example of a particularly strong event in which the ratio of the vertical to horizontal energy was about 2.  The implication for forward-looking sensors is that since they measure only the wind components along the flight track (longitudinal) this measurement may not be a good predictor of the vertical component to which the aircraft responds most readily.

Figure D9.  Example of subgrid model determination and application. (a) The solid line is the longitudinal structure function from the interpolated cloud model.  Note the deficit in energy at small separations (scales). The dotted line is the desired Kolmogorov 2/3 model that matches the resolved scales of the cloud model. The dashed line is the merged structure function, i.e., the sum of the best fit subgrid model (represented by dash-dot line) and the interpolated cloud model. In this case the best fit model is simply a von Karman.  (b) Trace of vertical velocity comparing the cloud model winds (dashed line) with the merged cloud model winds.

Figure D10.  (a) Aircraft data derived from a NASA B757 test flight during a strong convective encounter.  (b) Best fit –5/3 spectra levels for the longitudinal spectrum Su and the vertical spectrum SW  In this case the turbulence is highly anisotropic, 3/4Sw/Su=2.06.

[Top]

6.      Juneau Terrain-Induced Turbulence Project

RAP has studied terrain-induced turbulence in the vicinity of the Juneau International airport for the past four years.  The problem in Juneau is complicated by a complex terrain around the airport, the need for arrival and departure routes to avoid the terrain, and frequent episodes of low ceiling and visibility.  RAP has conducted two intensive field data collections in the area and has generally confirmed that the conclusions of the previous Hong Kong and Colorado Spring programs carry over into Juneau; however, the geographic, operational, and climatic differences in Juneau limit the applicability. An operational system for providing wind and turbulence information has already been deployed in Juneau and efforts are underway to upgrade the system to provide specific terrain-induced hazard warnings.

Several years of high-quality data from the three wind profilers (Lemon Creek, South Douglas, and North Douglas) and with three mountain-top anemometers (Sheep Mountain, Mount Roberts, and Eagle Crest) have been collected. Although there are many gaps (some understood and others being investigated), the data does provide a valuable sample of the weather that can occur in Juneau. The seasonal distribution of these data emphasizes the winter months.  The time period covered by this data is much too short to consider the data set as climatology, but it is sufficient to provide many examples of the Taku, Gap, and Southeast flows of interest to this project. 

Examination of the dominant spatial locations of turbulence revealed the following “hot spots” for the four regions examined:  Lemon Creek, Gastineau Channel, Northern Douglas Island, and Coghlan Island.  In Lemon Creek, data collection was predominantly along the Lemon Creek and Fox departure paths.  Stronger turbulence appears deeper within the valley, and near the northwest edge of Blackerby ridge (see Figure D11 for a spatial distribution).  Most turbulence was observed below about 700 m, but this observation may be biased by the density of measurements; there were relatively fewer measurements higher in the valley. Examination of the data collected in each region as a function of measured wind speed revealed relationships between the wind speed and the onset of turbulence.  Although the amount and range of data is not sufficient to show quantitative thresholds, this result suggests that regression analysis, considering several factors at once, could be successful.

Substantial progress was made this year on the analysis of the Doppler on Wheels (DOW) data collected during the FY2000 field program.  The DOW radar offers a spatially and temporally comprehensive, multidimensional characterization of wind dynamics that cannot be provided by any other meteorological instrument currently deployed in the Juneau field area, including the research aircraft. During FY02, emphasis shifted toward using DOW data to identify and quantify turbulence.

Several algorithms were developed using dimensional analysis to estimate the degree of turbulence on the basis of DOW velocity and spectral width fields. Figure D12 represents a typical result of this empirically-based approach. Aircraft tracks are plotted with the value of measured turbulence color-coded. The DOW-based estimates of turbulence are graphically represented in the background. It is immediately apparent that a linear band of turbulence predicted on the basis of DOW data, located almost precisely south-southeast of the radar location, intersects the aircraft flight track at a point where one of the highest turbulence values was recorded.