Developing a Global Turbulence and Convection Nowcast and Forecast System

Transcription

1 1st AIAA Atmospheric and Space Environments Conference June 2009, San Antonio, Texas AIAA Developing a Global Turbulence and Convection Nowcast and Forecast System John K. Williams 1, Robert Sharman 2, and Cathy Kessinger 2 National Center for Atmospheric Research, Research Applications Laboratory, Boulder, CO Wayne Feltz 3, Anthony Wimmers 3, and Kristopher Bedka 3 University of Wisconsin Madison, Space Science and Engineering Center, Madison, WI Turbulence is widely recognized as the leading cause of injuries to flight attendants and passengers on commercial air carriers. Turbulence encounters frequently occur in oceanic and remote regions where ground-based observations are sparse, making hazard characterization more difficult, and where current World Area Forecast System products provide only low temporal and spatial resolution depictions of potential hazards. This paper describes a new effort to develop a global diagnosis and forecast system that will augment and enhance international turbulence and convective SIGMETs and provide authoritative global turbulence data for the NextGen 4-D database. This fully automated system, modeled on the FAA s Graphical Turbulence Guidance (GTG) and GTG Nowcast systems, will provide 3-D probabilistic turbulence nowcasts and forecasts globally above 10,000 feet MSL for 0-36 hour lead times, comprehensively addressing clear-air turbulence (CAT), mountain wave turbulence (MWT), and convectively-induced turbulence (CIT). The system will employ NCEP Global Forecast System model output and data from NASA and other national and international satellite assets to produce the CAT and MWT diagnoses based on both model-based turbulence diagnostics and satellite-based turbulence detection algorithms. The convective nowcast methodology makes use of GFS data and operational satellite data from GOES, Meteosat and MTSAT, and will be tuned and verified using data from NASA s TRMM, Cloudsat and MODIS instruments. The convective nowcasts will be coupled with the GFS environmental information to assess the near-term likelihood of CIT. This paper presents an overview of the system elements and initial algorithm development results. Future work will perform additional development and verification using comparisons with automated quantitative in situ turbulence reports, AIREPs and AMDAR data. A real-time demonstration including a webbased display and cockpit uplinks is also planned. T I. Introduction he World Area Forecast System (WAFS) was established in 1982 by the International Civil Aviation Organization (ICAO) and the World Meteorological Organization (WMO) to provide weather guidance for international aircraft operations. In concert with the US Federal Aviation Administration (FAA), the National Weather Service (NWS) established the Washington World Area Forecast Center (WAFC) in The Washington WAFC is responsible for providing aviation weather guidance to the 1 Project Scientist, RAL Aviation Applications Program, P.O. Box 3000, Boulder, CO , AIAA Member. 2 Project Scientists, RAL Aviation Applications Program, P.O. Box 3000, Boulder, CO Researchers, SSEC Cooperative Institute for Meteorological Satellite Studies, 1225 W. Dayton Street, Madison, WI Copyright 2009 by John K. Williams. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. 1

2 Americas, Atlantic, Pacific, and Eastern Asia, while a second WAFC in London supplies data to Europe, Africa, and Western Asia. Washington WAFC global products include facsimile charts of significant weather (SIGWX) forecasts produced at the NWS Aviation Weather Center (AWC) and significant meteorological information (SIGMET) warnings of hazardous en-route weather. The SIGWX prognostic chart approximates a human-drawn weather chart and depicts the type, location, movement and intensity of weather features such as fronts, cumulonimbus clouds, and regions of turbulence; it is updated at 6- hour intervals. International SIGMETs provide information about the presence of convection, turbulence and icing, and are updated at 4 hour intervals. The SIGMETs describe each event by its horizontal and vertical location, its intensity and movement as well as assigning an expiration time (see Figure 1). WAFS global weather products are issued infrequently and have relatively low resolution. For instance, the area enclosed by a SIGMET is typically so large that aircraft have little option but to traverse through it. This is much like the situation over the conterminous United States (CONUS) until a few years ago when the Graphical Turbulence Guidance (GTG) product and the National Convective Weather Forecast (NCWF; Megenhardt et al. 2004) became operational at the AWC. GTG is a turbulence forecast system that produces 1, 2, 3, 6, 9 and 12 hour forecasts of turbulence intensity (Sharman et al. 2006). NCWF produced 0-2 hour forecasts of convection likelihood, and is now being supplanted by the Consolidated Storm Prediction for Aviation (CoSPA; Dupree et al. 2009). These products from these automated forecast systems are used to support the creation of turbulence and convective AIRMETs and SIGMETs over the CONUS; they are also used to provide a graphical display directly to general aviation and commercial users (see Clearly, global turbulence and convection decision support systems would benefit from an analogous global system, whose development is described in this paper. The Global GTG system will also address NextGen requirements for System Wide Information Management (SWIM) by populating a 4-D data cube with timely, accurate, authoritative, probabilistic global turbulence nowcast and forecast grids and convection nowcast grids. In particular, Global GTG will supply automated, comprehensive assessments of atmospheric turbulence from clear-air turbulence (CAT), mountain wave turbulence (MWT), and convectively-induced turbulence (CIT) sources for altitudes above 10,000 feet and at lead times of 0-36 hours. The development of Global GTG brings together multiple previous NASA and FAA-funded research and development efforts. It draws on CONUS GTG development; empirical and modeling studies of how near-thunderstorm CIT is related to thunderstorm characteristics and environment information; development of numerical weather prediction (NWP) model-based and satellite-based MWT forecasting and detection, and oceanic convective weather diagnosis and nowcasting efforts. Thus, this effort will integrate a broad collection of research results and products together to create a comprehensive system. Figure 1: WAFS decision support products for the North Pacific. (Left) Example graphical display of international SIGMETs, which are updated every 4 hours. (Right) Example of a SIGWX facsimile chart, updated every 6 hours. 2

3 II. Major Sources of Turbulence Much empirical evidence exists documenting higher occurrences of aircraft turbulence encounters in the vicinity of jet streams and upper-level fronts, near the tropopause, over mountainous regions, and in or near cloud, especially convective cloud (e.g., Hopkins 1977, Lester 1993, Chandler 1987, Sharman 2004, Wolff and Sharman 2007). Global GTG must accommodate these various clear-air and in-cloud sources. Most of the clear-air sources are longer lived (e.g., Vinnichenko et al. 1980), but still may be rapidly evolving, and turbulence related to convection is known to be highly transient and patchy (e.g., Lane et al. 2003). Thus, Global GTG is broken out into two components, one to diagnose the longer-lived turbulence, generally associated with clear-air turbulence sources, from NCEP s Global Forecast System (GFS) model output, and a nowcast component that utilizes current observations to capture the more rapidly evolving features associated with mountain waves or convective storms. A. Clear-air turbulence The Global GTG components for non-convective turbulence nowcasts and forecasts are modeled after the CONUS GTG (Sharman et al. 2006), which uses a suite of NWP model-derived diagnostics for CAT and MWT, and only indirectly addresses turbulence associated with model-resolved clouds and convection. These diagnostics are combined in a weighted regression to get the best agreement with available observations. It was shown in Sharman et al. (2006) that a GTG combination based on climatological (static) weights, with the weights derived from the overall statistical performance of the diagnostics, performs almost as well as a version where the weights are dynamically updated based on comparisons to recent PIREPs. Since the number of turbulence reports available at any one time and region in the global domain will usually be insufficient for dynamic weighting, a static empirical model is expected to be employed for Global GTG. Figure 2: Comparison of turbulence diagnostics from the GFS 6-hour forecast valid at 18 UTC 4 November 2008 (left column) and RUC forecast (right column). The diagnostics are the Ellrod Index for 35,000 ft (top), EDR Index for 35,000 ft (middle) and Richardson number from thermal wind at 20,000 feet (bottom). 3

4 Adapting GTG to global use has required replacing the CONUS NWP model (currently RUC) with the GFS model to generate the constituent turbulence diagnostics globally; the combination of these diagnostics will then be tuned by training against in situ turbulence reports and AIREPs to optimize statistical performance globally. Figure 2 compares a GFS-based diagnostic to that generated from the RUC model. The overall patterns over the CONUS are quite similar, indicating that the coarser resolution GFS model is also capable of capturing at least some the large-scale features conducive to turbulence formation. However, as the bottom set shows, some turbulence diagnostics appropriate for northern latitude break down at the equator. Thus, the selection of diagnostics and determination of appropriate combinations will have to be performed on a regional basis. CAT may also be associated with tropopause folds, layers of stratospheric air that penetrate into the troposphere near a front and frequently exhibit dynamical instability (Shapiro 1980). Studies have shown that such folds may be identified via GOES water vapor gradients in conjunction with NWP model data, which establishes their vertical extent. The mechanism and example output from a prototype tropopausefold detection algorithm (Wimmers and Moody 2004a, 2004b) are shown in Figure 3. Validation of the tropopause fold algorithm running on GOES-12 data was performed via comparisons with United Airlines eddy dissipation rate (EDR) turbulence reports over the eastern United States from 01 May April The algorithm achieved a probability of detection of approximately 20% for light-orgreater turbulence observations, 5% for the much less frequent moderate-or-greater turbulence cases, and little skill for the very infrequent severe-or-greater cases. The most robust prediction of turbulence occurred in the months of December February. These results underscore the importance of combining multiple predictors sensitive to different turbulence sources to generate a comprehensive product. 150 stratosphere 14 Pressure (hpa) subtropical air mass front tropopause polar air mass Height (km) (~100 km) Figure 3: (Top left) Illustration of the tropopause-fold mechanism. (Top right) GOES Layer Average Specific Humidity (GLASH) product, which is an adjustment to the GOES water vapor channel to show specific humidity variations, with identified tropopause folds indicated in grey. The bottom plots depict an estimate of tropopause fold lower altitude (lower left) and upper altitude (lower right) bounds. 4

5 B. Mountain-wave turbulence Severe turbulence can result from breaking gravity waves associated with flow over mountains or other terrain features (Hopkins 1977, Lester 1993, Wurtele et al. 1996). This MWT often occurs at scales too small to resolve in operational NWP models, but the wave structures often can be detected in satellite water vapor imagery (Uhlenbrock et al. 2005, 2007). Automated wave detection and characterization algorithms may provide an additional valuable contribution to the diagnosis of regions susceptible to MWT. Mountain waves appear to be a relatively common phenomenon. An analysis of daily NASA MODIS imagery over the central Rockies from October-April and reveals that mountain waves were present in the 6.7 µm water imagery for 68% of all days during these months. However, the occurrence of strong turbulence associated with these waves is evidently quite rare: moderate or greater intensity turbulence occurred on average for only 0.35% of all EDR observations over the Rocky Mountain Region throughout the duration of this 3.5 year EDR database. An effort is underway to detect mountain waves via piecewise pattern recognition, and to identify wave interference and other characteristics that may be associated with turbulence. Figure 4 shows an example of the preliminary results from this work. Figure 4: (Left) MODIS 1-km 6.7 um water vapor imagery of a highly turbulent case from 6 March (Right) An experimental derived quantity, the wave interference scale, which provides a normalized score for the estimated combined contribution of wave strength and interference toward turbulence. C. Convectively-induced turbulence Convective or convectively-induced turbulence (CIT) may account for over 60% of turbulence encounters over the CONUS (Cornman and Carmichael 1993; see also Kaplan et al. 2005); the fraction may be somewhat less in oceanic regions due to typically high cruise altitudes and fewer routing constraints, but remain a significant concern. Convection can develop and evolve rapidly, and it may not be well predicted or represented by NWP models. For this reason, it is necessary to augment the GFSbased turbulence forecasts with timely assessments of CIT derived from observation-based convection analyses and nowcasts. An algorithm for diagnosing CIT over the CONUS based on a combination of environmental conditions obtained from the RUC model, thunderstorm features identified from satellite IR values and trends, radar reflectivity, radar in-cloud turbulence detection (Williams et al. 2006), and lightning data was described in Williams et al. (2007). Developing an analogous global CIT diagnosis system requires utilizing GFS model data in place of RUC and making modifications to accommodate 5

6 diverse geographical domains and the more limited storm characterization data available in oceanic and remote regions. A combination of the GFS model plus satellite data has been developed to characterize the presence of convection and environmental conditions likely to support convective development (Kessinger et al. 2009). Storms are identified and characterized based on a cloud-top height derived from satellite longwave infrared data and GFS model temperature profiles. The Global Convective Diagnosis (GCD; Mosher 2002) utilizes the difference between the satellite-measured longwave and water-vapor channels to identify deep convective clouds that have reached the tropopause. The cloud top and GCD are combined to obtain a scalar metric of thunderstorm intensity (see Figure 5). Thunderstorm nowcasting may be achieved via storm extrapolation, or a technique that involves data fusion of observation and model a) data Convection (Cai et al. Diagnosis 2009). Convective b) 1-hr nowcasting Convective products Nowcast originally c) 2-hr developed Convective for Nowcast use with the US GOES and Japanese MTSAT-1R satellites are being modified for use with Meteosat data to obtain coverage of Africa, Europe and Asia. NASA TRMM, CloudSat, and CALIPSO data, along with CONUS radar reflectivity and lightning data, are being used to evaluate the convective products performance. Convection locations and intensity will be used in combination with GFS model fields to diagnose regions of potential CIT. Cloud Top Height Global Convective Diagnosis Convective Diagnosis Oceanic Interest Figure 5: Satellite-derived convection information include cloud-top height (top), Global Convective Diagnosis (middle), and a convective intensity metric called the Convective Diagnosis Oceanic (bottom). 6

7 In addition to the location and intensity of convection, additional storm features may be useful for diagnosing CIT. For instance, using satellite longwave infrared data, locations of overshooting tops where a developing storm penetrates the stratosphere may be identified by looking for pixels that are cold and substantially colder than their surroundings; these overshooting tops show promise for more precisely diagnosing the location and severity of turbulence. An objective method has been developed for identification of overshooting convective cloud tops by utilizing infrared window channel observations to identify clusters of very cold pixels relative to the surrounding thunderstorm anvil cloud. Comparisons to visible channel imagery, CloudSat and CALIPSO profiles, and synthetic GOES-R ABI imagery have shown that the method has good accuracy and offers an improvement over existing overshooting top capabilities described in the literature. Figure 6 shows that over four convective seasons ( ), aircraft flying very near to overshooting tops experienced turbulence at a higher frequency and stronger intensity than non-overshooting cold cloud pixels. Figure 6: A comparison of overshooting tops detected in GOES-12 imagery and EDR turbulence observations over the convective seasons (April-September). The frequency of severe turbulence encounters has been multiplied by 10 to better illustrate the difference in frequency between overshooting tops and non-overshooting cold pixels. III Creating the Integrated Turbulence Nowcast and Forecast Products Developing an effective combination of model-based turbulence diagnostics, satellite detection products and convective nowcasts and CIT diagnostics requires case studies and empirical data mining over a variety of conditions and locations. Over the CONUS, Hawaii, and Central and northern South America, automated in situ reports from United and Delta Airlines aircraft provide routine measurements of EDR, a quantitative turbulence metric (Cornman et al. 1994, 2004) that may be used to support development and tuning a data fusion methodology. However, these data are not available globally. Sources of oceanic and international turbulence information include qualitative AIREPs and quantitative measurements from automated Aircraft Meteorological Data Relay (AMDAR) systems, including Aircraft to Satellite Data Relay (ASDAR) data from select British, Continental and Qantas Airlines aircraft. In particular, AMDAR measurements of U de (an automated turbulence intensity measure based on true air speed fluctuations) may supply useful quantitative turbulence information. Figure 7 illustrates the geographical coverage of these various turbulence measurements. 7

8 United EDR above 10,000 ft MSL to Delta EDR above 10,000 ft MSL to U de, various airlines to AIREPs, various airlines Figure 7: Coverage and density of various sources of in situ turbulence information for global turbulence development and verification. United Airlines EDR reports (top left), Delta Airlines EDR reports (top right), U de measurements from ACARS (lower left), and AIREPS (lower right). A database of global turbulence reports and associated environmental and convection characteristics is currently being developed. It will be used to identify case studies, perform data mining to establish predictive relationships, and train an empirical data fusion model. A machine learning technique called random forests (RFs; Breiman, 2001) has previously proved useful in developing empirical models for diagnosing regions of CIT (Williams et al. 2007) and for developing thunderstorm nowcasts (Williams et al. 2008a, Cai et al. 2009). Essentially, an RF is a group of decision trees which collectively form an ensemble of experts that vote on the correct classification of an input. In the course of training an RF, information is obtained on the relative importance of various candidate predictor variables; this new information may in turn be useful in understanding the underlying physical process or identifying distinct phenomenological regimes (Williams et al. 2008b, Abernethy 2008). The RF approach has proven to work well even when several hundred predictors require analysis. The final result of this algorithm development approach will be a combination of empirical models that provide integrated 3-D forecasts and nowcasts of atmospheric turbulence based on a combination of GFS model data and satellite data (see Figure 8). These will include 0-3 hour nowcasts that comprehensively address CAT, MWT and CIT and 3-36 hour NWP model-based forecasts. Separate predictive models will be created for each forecast lead time. 8

9 Figure 8: GFS-GTG turbulence forecast output, valid at 15 December UTC, at a 35,000 ft flight level. Overlaid are CDO (magenta shapes) showing locations of diagnosed convective storms, AIREPs and in situ turbulence measurement data from Delta, United and Qantas aircraft, valid between 90 minutes before and 90 minutes after the forecast time. AIREPs and in situ turbulence data are coded blue for light, orange for moderate and red for severe turbulence. The Global GTG product will combine the GFS forecast with satellite and convective nowcast information to create comprehensive turbulence grid, which may be tuned and verified based on the collocated turbulence observations. IV. Performance Evaluation Performance evaluation and tuning of the turbulence forecasts and nowcasts will use both qualitative measures and quantitative techniques. Qualitative assessments will be performed by soliciting feedback from demonstration users of the global turbulence and convection nowcasts and forecasts, who will obtain Global GTG products via a web-based interactive graphical display and customized conditions ahead text-based maps uplinked via ACARS to the cockpits of en-route aircraft. Participating United Airlines pilots will respond to a web questionnaire that asks them to describe the flight conditions and the accuracy and utility of each experimental cockpit uplink message received relative to standard weather information sources. A similar webpage will be created to allow airline meteorologists, dispatchers, or air traffic controllers viewing the web-based display to provide feedback. This demonstration will be similar to the one previously performed for a CONUS turbulence product (Craig et al. 2008). The development of a data fusion model for generating comprehensive turbulence nowcasts is currently underway. However, the convective diagnosis component of the system is already running in real-time. Sample displays of convective cloud top information near the location of the Air France 447 accident on June 1, 2009 is shown in Figure 9; this uplink capability was originally developed by the FAA Aviation Weather Research Program s Oceanic Weather Product Development Team (Kessinger et al. 2006). Although the cause of this accident is still unknown at the time of this writing, it seems likely that the information provided by a real-time uplink of weather conditions ahead would have improved the pilots situational awareness, providing valuable augmentation of information available from SIGMETs and SigWx charts and from the airborne radar. The real-time global turbulence product demonstration will confirm the value of this sort of real-time weather information. 9

10 B B A A /EXP CLOUD TOP FI AF447/AN NXXXAF 01 Jun '/' Cloud tops 30,000 to 40,000 ft//////ccc///// 'C' Cloud tops above 40,000 ft///////////cc///// *4.0N,30.0W///////// *////////C////////// //*//CC///CCC///////// ///*CCCC/C/CC////////// ////*CCCCCCC//C///////// ///CC*CCCCC///CC//////// ///CCC*CCCCC///////////// //CCCCC*CCCC////////////// //CCCCCCC*CCC////////////// /CCCCCCCCC*CCC///// / //CCCCCCCCCC*CC///// // /// //CCCCCCCCCCC*C///////// ////// /////CCCCCCCCCC*C// // // ////// //////CCCCCCCC//*/ // / ////// CC//////CCCCCCC//* ///////// CCC////////////// * ///////// /C//////////// *1.3N,31.4W //// /////// */ *// / /*/// // /*/// / /*// /*// * // * /// * // // * //// * ///// * ////// * ///// /// * /// Pos Rpt / // * / 0133 // X 1.4S,32.8W // Valid for // / z // Pilot feedback at url: deleted] Figure 9: Convective cloud tops computed from GOES-12 and GFS model data. (Upper left) Display valid at 23:45 UTC 31 May 2009; A marks the approximate location of the last verbal contact with Air France Flight 447 at 1:33 UTC 1 June 2009, and B marks the approximate location of its last automated ACARS reports at 2:14 UTC. The two points are separated by about 320 nautical miles. (Upper right) Display valid at 2:45 UTC 1 June (Lower left) Sample customized graphical display of cloud tops ahead that would have been produced based on the INTOL (1.4º S, 32.8º W) position report at 1:33 UTC 1 June 2009 and the filed flight plan. Cloud tops above 30,000 ft are represented by green polygons, while those above 40,000 ft are red. (Lower right) An uplink message generated for display on a cockpit ACARS printer representing the same information in a text graphics map. Quantitative assessments of the turbulence forecasts and nowcasts will be performed using two primary assessment metrics. The first performance measure is based on receiver operating characteristic (ROC) curves (e.g., Marzban 2004, Sharman et al. 2006). Sharman et al. (2006) compared ROC curves for individual turbulence diagnostics and the GTG combination against AIRMETs and showed that GTG performed better than any individual turbulence diagnostic, and the improvement in GTG s accuracy over the operational AIRMETs was substantial (see Figure 10). Similar accuracy improvements of Global GTG over international SIGMETs are expected. 10

11 A second performance measure used in evaluations of probabilistic forecasts is the reliability diagram (e.g., Wilks 1995, Hamill 1997), in which the predicted probabilities of turbulence over some threshold (light, moderate or severe) are compared with the measured relative frequency of turbulence encounters given that prediction. A well-calibrated forecast or nowcast will produce a reliability curve close to the 1:1 line. This approach also accounts for forecast volumes, which should be as small as possible. Both the reliability diagram and ROC curves will be computed from all available data in the global domain for as long a period as is practical, and the turbulence product will be tuned to optimize its performance. Figure 10: ROC curves from Sharman et al. (2006) for 6-hr CONUS GTG forecast performance evaluated in comparison to PIREPs over a one-year period. The gray curves are those of individual diagnostics, the black curve is the GTG combination, and the dot is the AIRMET PoDYes vs. PoDNo performance. This analysis demonstrates that GTG outperforms all of the individual diagnostics and significantly outperforms the operational AIRMET. V. Conclusion This paper has described the motivation and some of the technical elements involved in the development of Global GTG, a software system that will ingest GFS model data and real-time measurements from several geosynchronous satellites and generate turbulence nowcasts and forecasts that comprehensively address the major known sources of upper-level atmospheric turbulence. The gridded, probabilistic turbulence forecasts and nowcasts and convection nowcasts will be evaluated via a user demonstration, including a web-based display and uplinks to en-route aircraft, as well as statistical comparisons to in situ turbulence measurements. Global GTG is intended to become part of the NextGen 4-D weather data cube and to supply decision support information for the WAFS. It is expected that Global GTG will significantly assist in both strategic planning and tactical turbulence avoidance. Acknowledgments This research is supported by NASA under Grant No. NNX08AL89G. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Aeronautics and Space Administration. Some of this research was performed in response to requirements and funding by the Federal Aviation Administration (FAA). The views expressed are those of the authors and do not necessarily represent the official policy or position of the FAA. Several software engineers and scientists at NCAR/RAL and UW SSEC/CIMSS have contributed significantly to the work described here. They include Gary Blackburn, Huaqing Cai, Jason Craig, Greg Meymaris, and Nancy Rehak. We very much appreciate their contributions. 11

12 References Abernethy, J., 2008: A domain analysis approach to clear-air turbulence forecasting using high-density in-situ measurements. Ph.D. Dissertation, Department of Computer Science, University of Colorado. Breiman, L., 2001: Random forests. Machine Learning, 45, Cai, H., C. Kessinger, D. Ahijevich, J. Williams, N., Rehak, D. Megenhardt, R. Bankert, J. Hawkins, M. Donovan, and E. Williams, 2009: Nowcasting oceanic convection using random forest classification. AMS 16th Conference on Satellite Meteorology and Oceanography, Fifth Annual Symposium on Future Operational Environmental Satellite Systems-NPOESS and GOES-R, and Fourth Symposium on Policy and Socio-Economic Research, paper J8.4. Chandler, C. L., 1987: Turbulence forecasting. In Atmospheric Turbulence Relative to Aviation, Missile, and Space Programs, NASA Conference Publication 2468, Cornman, L. B. and B. Carmichael, 1993: Varied research efforts are under way to find means of avoiding air turbulence. ICAO Journal, 48, Cornman, L. B., C. S. Morse, and G. Cunning, 1995: Real-time estimation of atmospheric turbulence severity from in-situ aircraft measurements. J. Aircraft, 32, Cornman, L. B., G. Meymaris, and M. Limber, 2004: An update on the FAA Aviation Weather Research Program s in situ turbulence measurement and reporting system. AMS Eleventh Conf. on Aviation, Range, and Aerospace Meteorology, P4.3. Craig, J. A., J. K. Williams, G. Blackburn, S. Linden and R. Stone, 2008: Remote detection and real-time alerting for in-cloud turbulence. AMS 13th Conference on Aviation, Range and Aerospace Meteorology, Donovan, M. F., E. R. Williams, C. Kessinger, G. Blackburn, R. Bankert, and F. R. Mosher, 2006: A satellite intercomparison of aviation convective diagnosis products for oceanic regions. Preprints-CD, 12th Conference on Aviation, Range and Aerospace Meteorology, American Meteorological Society, Atlanta, January Dupree, W. J., D. Morse, M. Chan, X. Tao, M. Wolfson, J. O. Pinto, J. K. Williams, D. Albo, S. Dettling, M. Steiner, S. G. Benjamin, and S. S. Weygandt, 2009: The 2008 CoSPA Forecast Demonstration. AMS Aviation, Range and Aerospace Meteorology Special Symposium on Weather-Air Traffic Management Integration, Paper P1.1. Hamill, T. M., 1997: Reliability diagrams for multicategory probabilistic forecasts. Wea. Forecasting, 12, Hopkins, R. H., 1977: Forecasting techniques of clear-air turbulence including that associated with mountain waves. WMO Technical Note No. 155, 31 pp. Kaplan, M. L., A. W. Huffman, K. M. Lux, J. J. Charney, A. J. Riordan, and Y.-L. Lin, 2005: Characterizing the severe turbulence environments associated with commercial aviation accidents. Part 1: A 44-case study synoptic observational analysis. Meteor. Atmos. Phys., 88, Kessinger, C., P. Herzegh, G. Blackburn, R. Sharman, G. Wiener, B. Hendrickson, K. Levesque, J. Craig, T. Tsui, J. Hawkins, R. Bankert, E. Williams, M. Donovan, G. Ellrod, R. Kistler, and D. Fleming, 2006: The FAA AWRP Oceanic Weather Program Development Team. Preprints-CD, 12 Conference on Aviation, Range and Aerospace Meteorology, American Meteorological Society, Atlanta, January Kessinger, C., H. Cai, N. Rehak, D. Megenhardt, M. Steiner, R. Bankert, J. Hawkins, M. Donovan, and E. Williams, 2009: The oceanic convection diagnosis and nowcasting system. AMS 16th Conference on Satellite Meteorology and Oceanography, Fifth Annual Symposium on Future Operational Environmental Satellite Systems-NPOESS and GOES-R, and Fourth Symposium on Policy and Socio-Economic Research, paper J8.3. Lane, T. P., R. D. Sharman, T. L. Clark, and H.-M. Hsu, 2003: An investigation of turbulence generation mechanisms above deep convection. J. Atmos. Sci., 60, Lester, P. F., 1993: Turbulence: A new perspective for pilots. Jeppesen Sanderson, Inc., 212 pp. Marzban, C., 2004: The ROC curve and the area under it as performance measures. Wea. Forecasting, 19,

13 Megenhardt, D., C. Mueller, S. Trier, D. Ahijevych, and N. Rehak, 2004: NCWF-2 probabilistic nowcasts, Preprints-CD, 11 Conference on Aviation, Range and Aerospace Meteorology, American Meteorological Society, Hyannis, 4-8 October Mosher, F., 2002: Detection of deep convection around the globe. Preprints, 10th Conf. Aviation, Range, and Aerospace Meteor., Amer. Meteor. Soc., Portland, OR, Shapiro, M. A., 1980: Turbulent mixing within tropopause folds as a mechanism for the exchange of chemical constituents between the stratosphere and troposphere. J. Atmos. Sci, 37, Sharman, R.D, 2004: Clear-air Turbulence (CAT). Encyclopedia of Nonlinear Science, Alwyn Scott, Ed., 1104 pp. Sharman, R., C. Tebaldi, G. Wiener, and J. Wolff, 2006: An integrated approach to mid- and upper-level turbulence forecasting. Wea. Forecasting, 21, Uhlenbrock, N., S. Ackerman, W. Feltz, T. Whittaker, B. Sharman, K. M. Bedka, and L. Gumley, 2005: The use of satellite water vapor imagery and model data to diagnose and forecast turbulent mountain waves. 21 st International Conf. on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography and Hydrology, American Meteorological Society, P2.40. Uhlenbrock, N. L., K. M. Bedka, W. F. Feltz, and S. A. Ackerman, 2007: Mountain wave signatures in MODIS 6.7 um imagery and their relation to pilot reports of turbulence. Wea. Forecasting, 22, Vinnichenko, N. K., N. Z. Pinus, S. M. Shmeter, and G. N. Shur, 1980: Turbulence in the Free Atmosphere. Plenum, 310 pp. Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An introduction. Academic Press, 467 pp. Williams, J. K., L. B. Cornman, J. Yee, S. G. Carson, G. Blackburn, and J. Craig, 2006: NEXRAD detection of hazardous turbulence, AIAA 44th Annual Aerospace Sciences Meeting and Exhibit, AIAA Williams, J. K., J. Craig, A. Cotter, and J. K. Wolff, 2007: A hybrid machine learning and fuzzy logic approach to CIT diagnostic development. AMS Fifth Conf. on AI Applications to Env. Science, 1.2. Williams, J. K., D. Ahijevych, S. Dettling and M. Steiner, 2008a: Combining observations and model data for shortterm storm forecasting. In W. Feltz and J. Murray, Eds., Remote Sensing Applications for Aviation Weather Hazard Detection and Decision Support. Proceedings of SPIE, 7088, paper Williams, J. K., D. A. Ahijevych, C. J. Kessinger, T. R. Saxen, M. Steiner and S. Dettling, 2008b: A machinelearning approach to finding weather regimes and skillful predictor combinations for short-term storm forecasting. AMS 6th Conference on Artificial Intelligence Applications to Environmental Science and 13th Conference on Aviation, Range and Aerospace Meteorology, paper J1.4. Wimmers, A. J. and J. L. Moody, 2004a: Tropopause folding at satellite-observed spatial gradients: 1. Verification of an empirical relationship, J. Geophysical Research Atmospheres, 109, art. no. D Wimmers, A. J. and J. L. Moody, 2004b: Tropopause folding at satellite-observed spatial gradients: 2. Development of an empirical model, J. Geophysical Research Atmospheres, 109, art. no. D Wolff, J. K., and R. D. Sharman, 2007: Climatology of upper-level turbulence over the continental United States. Submitted to the J. of Appl. Meteor. Climo. Wurtele, M. G., R. D. Sharman, and A. Datta, 1996: Atmospheric lee waves. Annu. Rev. Fluid Mech., 28,