A climatology of trapped lee waves over Britain and Ireland, derived using machine learning

Abstract

<p class="paragraph">Trapped lee waves and turbulent rotor activity are a hazard for aviation, so forecasters must take into account the likelihood of lee wave activity when advising aviation personnel. The UK Met Office’s operational high resolution Numerical Weather Prediction model, UKV, resolves lee wave activity but there is currently no automated operational way to recognise such regions directly from the high resolution UKV data, short of a forecaster looking at the data themselves. If lee waves can be detected from the UKV data, we can use this output to gain a better understanding of lee waves and their impact over Britain and Ireland. For example, over which regions are lee waves likely to form? Which atmospheric conditions are the most conducive to lee wave propagation?<span data-ccp-props="{"> </span></p> <p class="paragraph">A machine learning (ML) model, a U-Net, was trained to identify and segment lee waves, and then retrained to derive characteristics about the waves (orientation, wavelength) from the model data. The wave amplitudes are also extracted using the wave mask from the segmentation model. These ML models were then applied to a 30 year dataset (1982 – 2012) of high resolution NWP output, driven by ERA-Interim data. From this, the location and characteristics of lee waves over Britain and Ireland from within this time period were extracted. In this presentation, we explore the climatology of lee waves in the data over this period, including the relationship between waves and the orography; frequency of waves; seasonal effects; the effects of weather patterns on waves; and diurnal effects. </p>