A New Algorithm to Identify Errors Caused by Migrating Birds in Profiler Winds

On August 15, 1996, a new Quality Control (QC) algorithm was implemented on the hourly winds from the NOAA Profiler Network (formerly called the Wind Profiler Demonstration Network). The algorithm is designed to flag winds that are suspect of being biased in wind velocity or direction by migrating birds. The algorithm is applied only at night, only below 4.5 km above mean sea level (below about 570 mb or 15,000 feet), and only during the spring and fall migration periods. No change to any existing profiler display software is expected. Winds suspected of being biased by migrating birds are identified as failing QC. A more detailed description follows, covering bird migration patterns, the new QC algorithm, and the results of an experiment conducted to determine the impact of birds to the wind profiler data, and to evaluate the performance of the algorithm (Miller et al., 1997).

Bird Migration Patterns

Different species of birds migrate at different times of the year. Definite peak migration periods occur at night during the spring and fall when low-level winds are favorable for their direction of travel. Specifically, they migrate northward during the spring using the low-level southerly jet, and southward in the fall. Large migrating geese are not a major problem to the profiler, as might be expected. The birds that cause the greatest impact to the profilers are small songbirds (passerines) that fly in a dispersed pattern rather than a tight flock. They typically start flying about 30 minutes after sunset and continue until about 30 minutes before sunrise, flying at an altitude of most favorable tailwinds (or very weak headwinds), generally < 2-4 km above the ground. Their maximum height appears to be limited by the ambient air temperature (> -2^oC). They typically fly at a lower maximum altitude in the spring (using the nocturnal low-level southerly jet) compared to the fall (when they fly at greater heights, apparently in the warmer, deeper mixed layer). The birds typically fly with an airspeed of about 8-15 ms^-1(15-30 knots) (Wilczak et al., 1995).

Contamination of Wind Profiler Data Caused by Birds

Wind profiling radars are designed to measure the extremely weak signals from the clear air throughout the troposphere and lower stratosphere. These weak signals require a long averaging time (1-minute) in each of the three fixed beam positions used by the profilers (generally pointing toward the east, north, and vertical). The reflected energy from a single bird flying through the radar beam even for just a few seconds is much greater than the clear-air signal averaged over the 1-minute dwell time. Therefore, the profiler reports a velocity equal to the flying speed of the bird(s) toward or away from the profiler (most noted in the northward pointing beam). The existing hourly consensus averaging algorithm (used for the past 10 + years) effectively removes isolated (a few per hour) erroneous velocity measurements before averaging the remaining "good" points. Problems occur when consistent "bad" data (in time) are competing with, or overwhelming, the "good" wind data. In addition, because the birds produce higher reflectivity than clear-air, they can also be detected as they fly through the profiler's antenna sidelobes. This causes differing velocity and spectral width components to be measured and averaged together. However, the erroneous positive and negative velocity measurements may average out to nearly the correct background wind, while the wider than normal spectral widths remain (a key element in the QC algorithm). In order to measure the extent of the contamination, data from four profilers in Kansas and Oklahoma were compared with data from collocated rawinsondes that were launched every 3 hours during the spring and fall of 1994. The mean (additional) error attributed to the migrating birds was found to be only about 2 ms^-1(biased toward higher wind velocity), with larger errors possible up to the airspeed of the birds (15 ms^-1), although errors this large were rare. The likelihood of bird caused errors > 8 ms^-1was only 5% during migration periods. In total, it was found that during the spring and fall migration periods, only about 2% of all the profiler wind data were being contaminated by birds (when considering all wind measurements in height and time). Although a small amount, the impact of these biased (high) wind velocities were significant at times to numerical model output and forecasts in the central plains of the U.S.

The Bird Detection Algorithm

A check for bird contamination in the hourly winds was devised by two meteorologists (F.M. Ralph and D.W. van de Kamp), based on work reported in Wilczak et al. (1995), and their personal observations. It is a simple six-step test that is primarily based on identifying the larger velocity variance signature thought to accompany the backscattered radar returns from birds. The other steps in the test are meant to reduce the likelihood of false alarms. The check flags an hourly observation as contaminated if all the following criteria are met:

• The time of year is appropriate for bird migration ("spring": 15 February - 15 June or "fall": 9 August - 30 November).
  
  
• The time of day is appropriate ("night": 0200 - 1200 UTC during the spring, 0000 - 1300 UTC during the fall, i.e., local civil twilight following sunset to local civil twilight preceding sunrise).
  
  
• The height is appropriate (< 4500 m MSL).
  
  
• The wind direction is favorable for migration (90º- 270º in the spring, 0º- 90º or 270º-360º in the fall).
  
  
• The velocity variance signature is appropriate (average of the north, east and vertical beams > 1.4 m²/s²).
  
  
• The increased velocity variance is not a result of precipitation (magnitude of downward vertical velocity < 0.8 ms^-1(wider spectral widths naturally occur during precipitation)).

The rawinsonde data were also used to statistically evaluate the algorithm. The Probability of Detection (POD) was generally good, successfully flagging over 60% of the "identified" bird contaminated wind data. (Bird contamination was defined by computing the standard deviation of wind component velocity differences (approximately 2 ms^-1) during non-bird migration periods, and using two times this standard deviation as the threshold for identifying velocity contamination in the spring and fall.) A rather high False Alarm Rate (FAR) of 44% was noted. This appears to be due to the positive and negative velocity errors (caused by birds in the sidelobes) averaging out to near the true wind velocity, while the spectral width average remains high and trips the bird detection algorithm. Therefore, the FAR appears to be somewhat fictitiously high. Further evaluation and adjustments to the algorithm are expected.

An obvious change after the algorithm became operational on August 15, was an increase in the number of winds below 4.5 km being flagged as failing QC. Use your best judgment to interpret these winds. For further information, contact Doug van de Kamp at (303) 497-6309 or vandekamp@fsl.noaa.gov, or the Profiler Control Center at (303) 497-6033.

References

Miller, P.A., M.F. Barth, J.R. Smart, L.A. Benjamin, 1997: The Extent of Bird Contamination in the Hourly Winds Measured by the NOAA Profiler Network: Results Before and After Implementation of the New Bird Contamination Quality Control Check. Preprints, 1st Symposium on Integrated Observing Systems, Long Beach, CA, Amer. Meteor. Soc., 138-144.

Wilczak, J.M., R.G. Strauch, F.M. Ralph, B.L. Weber, D.A. Merritt, J.R. Jordan, D.E. Wolfe, L.K. Lewis, D.B. Wuertz, J.E. Gaynor, S.A. McLaughlin, R.R. Rogers, A.C. Riddle, and T.S. Dye, 1995: Contamination of Wind Profiler Data by Migrating Birds: Characteristics of Corrupted Data and Potential Solutions. J. Atmos. Oceanic Technol., 12, 449-467.