On the lidar-turbulence paradox and possible countermeasures

Abstract. We describe the major difficulties in establishing a physics-based method that corrects lidar-based turbulence measures so that they become equivalent to standard turbulence measures. The difficulties encompass the so-called lidar-turbulence paradox, which we circumvent in two ways. The first uses a physics-based lidar-turbulence model and the second directly uses lidar measurements, both approaches aiming at training neural networks. The measurements are from continuous-wave Doppler lidar wind profilers deployed besides a tall 250-m meteorological mast at the Østerild test station in Denmark. Sonic anemometers on the mast match four lidar measurement levels, from 37 up to 241-m height. The physics-based lidar-turbulence model predicts well the behavior of the ratio of the lidar-to-sonic along- and cross-wind velocity variance up to 103 m. However, it predicts further decreasing ratios at 175 and 241 m, while the observations show increasing ratios for a number of stability conditions and length-scale ranges. The physics-based lidar-turbulence model is used to produce physics-based datasets, which are utilized to train neural networks. Compared to turbulence intensity measurements from a first lidar, the predictions of these neural networks are in better agreement with the sonic-based measures for most mean wind speed bins at 37 and 103 m. At 175 and 241 m, the predictions' accuracy reduces and better agreement is achieved within the highest mean wind speed ranges only. Measurements from a second lidar are used to generate predictions of turbulence intensity with neural networks trained with measurements from the first lidar. At 37 and 103 m, these predictions are also in better agreement with the sonic-based measures than those of the second lidar for most mean wind speed ranges. However, at 175 and 241 m, turbulence measures derived from the second lidar are generally close to the sonic-based values, while the predictions overestimate them. We speculate either that the assumption of turbulence homogeneity within the lidar scanning pattern might not hold at the site and/or that the Doppler radial velocity spectra of the lidars might be contaminated, thus impacting the radial velocity estimates particularly with increasing focus distance.