Frequency and magnitude biases in the 'Fryberger' model, with implications for characterizing geomorphically effective winds

The 'Fryberger model' uses standardized wind data to estimate regional aeolian sediment drift potential (DP), which allows for interpretation and classification of aeolian dune landscapes. This paper identifies systematic frequency and magnitude biases in the model that result from: (i) variations in wind direction sector range, which affects frequency bin size for determining percent occurrence of transporting winds, and (ii) use of wind speed class mid-point values over more statistically representative values (e.g., mean, median) or minimally classified (whole knot) wind speeds in the magnitude weighting factor for DP calculations. The significance of these biases is tested using unclassified data from Stanhope, Prince Edward Island, Canada. Frequency bias resulting from wind data aggregation into direction sectors of varying size produces statistically significant discrepancies in DP and RDP estimates for this dataset. When 36-point (10 s of degree) data are reduced to 16 sectors, cardinal directions receive one extra direction class and the magnitude of DP and RDP values decreases by 1.4% and 2.8%, respectively, compared to 16 equal (22.5°) sectors derived from original data. This results from under-representation of intermediate direction classes for this dataset. The effects on RDD (less than 1°) and directional variability ratio (RDP/DP) are insignificant. Magnitude biases result from the use of the wind speed class mid-point values instead of other statistically representative measures of wind speed in DP calculations. In that wind speed distributions are often positively skewed, mid-point values yield an over-estimate of total DP by as much as 34% and RDP by up to 22% for this dataset over those derived using mean, median and whole knot (minimally classified) values. This difference is statistically significantly at both the individual wind speed-direction category level and at the aggregate (directionally summed) level and causes RDP vectors to shift 3-5° to the south. Though the impacts of these frequency-magnitude biases are site-specific and modest, the effects on RDP estimation may be more significant in complex (i.e., multi-modal) wind regimes and/or in environments with more frequent high-magnitude winds from the cardinal directions. Given the increased availability of more precise, unclassified wind data, most of these systematic frequency-magnitude biases can be avoided. Recommendations on reducing inaccuracies imposed by these biases include: (i) using to-the-degree wind data where available and categorizing into 16 equal 22.5° direction sectors, (ii) using either wind speed class statistical mean values or minimally classified whole knot values in DP calculations, and (iii) recognizing that converting 36-point (10 s of degrees) data to 16 direction classes may introduce a frequency bias toward the cardinal directions and will cause inaccuracies in DP and RDP estimates of an amount that depends on the wind regime. Further research is needed to assess the implications of such biases in different wind regimes and into the influence of localized supply- and transport-limiting factors on regional-scale assessments of dune morphodynamics and mobility. These findings are also relevant for other applications that use categorized wind speed-direction data such as dust or contaminant plume dispersion modelling.