Measuring and testing aeolian sand transport models across a spatial grid network

Models relating wind forcing to resulting sand transport flux have been at the core of the discipline from its conception and remain an important central focus. The desired capability to accurately predict the amounts of sand that are moved in response to any given wind regime is one of the key practical applications of our scientific efforts. A plethora of predictive equations have been developed and tested over the course of many decades, with a great variety of parameter values and options, and in recent years the ever-advancing progress in deploying high-frequency wind and sand transport sensors in high-density measurement arrays has yielded a deeper understanding of the strong spatio-temporal variability in windblown sand transport. Most field experiments involve measuring time-series of wind forcing and the accompanying sand transport flux in one location or along one compact array, to then analyse for statistical relationships. In this contribution we present both an alternative measurement strategy as well as sophisticated (and unusual) statistical data analyses techniques that have not been applied to this research problem before. A field experiment on a wide sandy beach involved the deployment of 20 pairs of 50Hz sonic anemometer and 25Hz electronic sand trap over a grid network covering a distance of 90m alongshore and 30m cross-shore. Data were collected during an event where sand transport was driven by wind blowing roughly parallel to the beach, with well-developed and typical boundary layer flow conditions. Each station of co-located sonic and sand trap thus provides the opportunity to measure and develop a transport relationship that can be independently tested on the other stations in the grid. The grid as a whole, meanwhile, provides the opportunity to quantify spatio-temporal variability in a variety of ways. In addition to this alternative measurement strategy, a taught MSc course for physics and maths students was used to elicit the application of advanced and unusual statistical techniques to the analysis of potential relationships between the synchronous time-series from these stations. Techniques that students applied included sophisticated data-cleaning and detrending methods, stationarity testing, autoregressive-moving-average models, various types of non-linear regression, Gaussian process learning, stochastic models, k-means clustering, and Bayesian modelling. While these unusual statistical techniques by no means yield any specific superior transport model, they do introduce some interesting and useful new perspectives on our traditional lines of inquiry.