The Weather and Climate Impact Assessment Science Initiative
    Home Current Projects Extremes in Aviation
  About the Program
Overview
People
Contact Information
Program Documents
2004 Review Documents
  Research
Current Projects
Publications
Presentations
  Links of Interest
Supporting Institutions
Other NCAR Initiatives
  For Program Staff
Upcoming Events
Mailing List
Website Statistics Initiative Staff Only
  Search This Website

...more search features  
Extremes in Aviation
 

Project Abstract

The aviation community faces weather threats from a unique perspective. A large scale forecast can be largely correct and verify well using existing measures, but within this forecast area, extreme weather conditions may exist that pose an imminent threat to the safety of the plane and the people. These threats include icing, turbulence and convective events. These events are difficult to model because they are extreme in magnitude, relatively rare in occurrence or short in duration. Overly cautious forecasts restrict airspace and impede the flow of traffic. Extreme value theory focuses on properties of events that occur rarely or have an unusual magnitude. Therefore, extreme value theory seems to provide a potentially useful alternative approach in forecasting aviation hazards.

As part of the Aassessment Iinitiative, extreme value theory is or will be applied to four forecasting problems that are currently under investigation in the Research Applications Program (RAP). These problems are (1) In-flight icing, (2) Turbulence, (3) Convection, and (4) Public Weather forecasts. Although research and development in three of the four forecasting areas is supported through the Federal Aviation Administration's Aviation Weather Research Program, this funding is for directed research, and not intended for exploratory research such as that proposed here. The following is a brief description of work that is underway.

 

Icing Hazards

Icing HazardsIn order to forecast the threat of icing to the aviation community, NCAR has developed a Current Icing Potential (CIP) algorithm. This is an expert model that uses observations and data from numerical weather prediction (NWP) models to diagnose the icing potential for aircraft. An ensemble of statistical models has been built using the the output from the CIP and environmental variables from the NWP models to create estimates of the probability of icing conditions. Creating statistical models as well as verifying and icing models is difficult because icing is an extreme weather condition. Unlike features like temperature, wind and relative humidity, it doesn't occur often. Initially it was intended that extreme value theory would be used to model icing. These statistics have not been used yet, because pilot reports (PIREPs), the most common measure of icing intensities, are discrete, ordered categories.

For a 9 week period in the winter of 2002, statistical models were built using 2 weeks of PIREPs, CIP data and NWP model data. These models were in turn verified using data collected the following week. The statistical model offered a slight improvement over the direct CIP output. Using the area under the receiver operating curve (ROC) as a criteria1, the statistical post-processing offered on average a 3.1% improvement. On a weekly basis, this improvement ranged from 0.3% to 4.8%.

With the CIP, NCAR has gone to great lengths to stress that the forecasted potential is not a probability. While the data needed to verify that the statistical forecast is a true probability do not exist, its possible that it is closer to a probability than the existing potential. The properties of the post-processed forecasts and their interpretability are being examined.

 
ROC Plot The figure to the left shows a typical ROC plot comparing the original CIP output compared with the statistically post-processed output.
 

PIREPs rely on a pilots judgment to describe the nature and intensity of the icing. This information along with the plane's location are transmitted by radio to where is entered into a database. For these reasons, the data is subjective and subject to transcription error. Furthermore, planes typically avoid areas where icing potential is thought to be great. To address deficiencies associated with the PIREPs, an effort has been made to create a statistical model using data collected by specially equipped research aircraft. The presence of icing was determined using measurements made on the aircraft. These observations are thought to be higher quality than the pilot reports. Unfortunately, models can't be concurrently made using both pilot reports and the research aircraft data since they aren't often found in the same vicinity.

Using the aircraft data, statistical models were built by holding out data from one of the six flights of the research aircraft. These statistical models were verified with data from the hold-out flight. Using the CIP data, models were created using data from the 2 weeks preceding a flight, then used to forecast icing potential for the verification flight. In these experiments, both models received an identical area under the ROC curve value of 0.79. Additionally, the significant variables used in the models were noted. Variables consistently used by the models included the CIP forecast, temperature and relative humidity.

 

Convective Hazards

Convective HazardsThe extreme nature of convective events is being explored in two ways. First, working with data provided by RAP's C. Mueller, conditions present during the initiation of convective storms are being used to predict the duration of convective storms. This is a challenging problem that has been approached in many ways. Our initial attempts have been to used a Random Forest ( a randomly generated set of regression trees ) to predict the duration of such events. Very preliminary work suggests that predicting an exact duration may not be possible, but predicting a minimum duration may be achievable. Currently we are addressing the issue that geographic features such as the ocean may prematurely end convective events - or our ability to track them.

H. Brooks from NOAA/NSSL is in the process of creating a sounding re-analysis for at least the last 45 years. As opposed to actual soundings from radiosonde balloons, these re-analysis soundings incorporate other sources of information such as ground weather station data. Furthermore, re-analysis data is presented on an evenly space grid. This data is being linked to observations of extreme convective events - as observed by radar coverage or other independent observations of storms. While this project is still in its early stages, it is hoped that a better understanding of the relation between the conditions at the sounding points and the intensity of the storms will allow for a better prediction of convective events.

1. The ROC curve plots the relation between the hit rate and the false alarm rate for a model across a range of thresholds. A perfect model would have an area under the curve value of 1. A completely random model would have a value of 0.5.

 

Project PI, Leads, and Staff

  • Barbara Brown, PI, Project Lead
    Research Applications Program, NCAR
  • Marcia Politovich, Project Lead
    Research Applications Program, NCAR
  • Harold Brooks
    National Severe Storms Laboratory, NOAA
  • Eric Gilleland
    Research Applications Program, NCAR
  • Matthew Pocernich
    Research Applications Program, NCAR

For more information about this project, please contact Barbara Brown at: bgb@ucar.edu, or Marcia Politovich at: marcia@ucar.edu

  Initiative Staff Only Denotes Initiative Staff Only ©2007 UCAR   |   Privacy Policy   |   Terms of Use   |   Top of Page  
NCAR Weather and Climate Impact Assessment Science Initiative