Predicting infrasonic wind noise levels using local topographic features

This presentation describes a method for predicting RMS wind noise levels within a user-specified infrasonic frequency band (e.g., 1–10 Hz) using local topographic features. This capability is especially valuable when performing site selection for infrasound sensor array deployment. The method is based on building models using measured infrasound data from known sensor site locations and the local topographic features corresponding to the site. The presented results correspond to using features such relative terrain elevation and relative vegetation height obtained from publicly available sources. Mean wind speed and direction are also included as predictive parameters and these can be from data measured at the infrasound sensor site or from estimates obtained from a tool such as the Weather Research Forecast (WRF) model as was utilized for this investigation. A novel feature of the approach is that the terrain specification is not uniquely site specific but also depends on wind direction. Essentially, the topographic features are specified relative to wind direction and not in absolute coordinates. This enables a much richer set of samples from which to build the predictive models. While many options are potentially available for model building, this work focused on the use of multiple-layer artificial neural networks as a basis for regression. The results presented correspond to data from several infrasound sensor sites in the U.S Array Project.