Integrating Scanning Lidar with LES for Wake Characterization in Complex Terrain

Abstract

Flows over complex terrains present uncertainties in wind farm power production and lifespan. Core to these uncertainties are wake dynamics, which lack adequate models and need further validation. In lieu of conducting field experiments, this study integrates scanning lidar measurements and large eddy simulation (LES) to characterize wakes in complex terrain. LES data of flow over complex terrain near Perdigao, Portugal has been generated prior to this study. In the current study a `synthetic’ lidar is placed inside the LES to sample the wind field in space and time as a commercial scanning lidar would. From the synthetic scan fields, the wake is located and characterized using metrics of wake center location and maximum velocity deficit. These metrics are evaluated against the same metrics as determined by LES fields. By treating the LES metrics as the ``target”, the relative accuracy of the metrics can be determined. This accuracy is a function of the scan geometry, allowing for the geometry to be optimized for wake characterization. Since a lidar scan represents neither an instantaneous field nor a mean field, it is not clear whether the target should be the LES instantaneous or mean fields, therefore the metric accuracy is assessed using different LES fields in the form of instantaneous snapshots of the wake, fields averaged over the time of a scan cycle, and long-term averaged fields.

A set of scan geometries was tested which vary in measurement point density, scan direction, and scan area to find that accuracy depends on a trade-off in the temporal and spatial resolution of the geometry. The results highlight not only the dependency of accuracy on the geometry parameters, but also on which type of LES field is used as the target. The long-term averaged errors were found to be unsuitable for lidars in this terrain, where the longest time scales exceed the simulation time, therefore accuracy was assessed using two types of fields, instantaneous fields taken at the start of the scan cycle, and the cycle-averaged fields. From these results two geometries were identified as performing the best.

Improvements are still needed in the wake detection algorithm, both in distinguishing points in the wake from terrain-generated turbulence and the inclusion of more wake characteristics such as size, shape, and orientation in order to develop more accurate scan geometries, hence improving predictions of wind farm production and longevity. Other scan geometry parameters also need to be explored, including scan paths and non-uniform measurement point densities.