Evaluation of ultrasonic snow depth sensors for automated surface observing systems (ASOS)
Date
2005
Authors
Brazenec, Wendy Ann, author
Fassnacht, Steven R., advisor
Doesken, Nolan, committee member
Kelly, Gene, committee member
Stednick, John, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
In the 1990's the National Weather Service deployed automated surface observing systems at hundreds of airport locations across the country. Prior to the automation, human observers made snow observations every six hours. Once the automated systems were deployed, snow measurements ceased due to the lack of an automated sensor to measure snow. This study explored how well ultrasonic snow depth sensors compared to manual snow observations at nine sites across the country. This study had four objectives: 1.) Develop a method of quality assurance and quality control 2.) Identify factors which affect sensor performance 3.) Compare automated sensors to manual observations of snow depth 4.) Derive an algorithm to estimate six hour snowfall from automated sensor snow depth. A reliable data smoothing/processing technique was achieved using filtering of large variability and smoothing with a moving average to smooth small variations in snow depth. Factors found to affect sensor performance included: snow crystal type, wind speed, blowing/drifting snow, uneven snow surface, extremely low temperatures, and intense snowfall. The Judd and Campbell sensors both did a satisfactory job measuring snow beneath the sensor within ±0.4 inches. Two separate algorithms were created due to differing degrees of precision between the two sensors. It was found that the Campbell sensor did a better job at estimating six hour snowfall than the Judd using an algorithm that calculated snowfall over 5 minute periods and applying a temperature based compaction model to the estimated snowfall. The Campbell agreed with the manual data with an average mean absolute error between measurements of 0.23 inches. The Judd sensor results improved by using an algorithm which calculated snowfall using the change in snow depth over sixty minutes, however, the Campbell results were better using the five minute snowfall algorithm. Overall, both sensors accurately depicted the snow depth on the ground, however the Campbell sensor was more accurate at predicting six hour snowfall using the algorithms presented in this research.