Estimating the benefit of Doppler wind lidars for short‐term low‐level wind ensemble forecasts

Abstract

This work focuses on the potential of a network of Doppler lidars for the improvement of short‐term forecasts of low‐level wind. For the impact assessment, we developed a new methodology that is based on ensemble sensitivity analysis (ESA). In contrast to preceding network design studies using ESA, we calculate the explicit sensitivity including the inverse of the background covariance boldB$$ \mathbf{B} $$ matrix to account directly for the localization scale of the assimilation system. The new method is applied to a pre‐existing convective‐scale 1,000‐member ensemble simulation to mitigate effects of spurious correlations. We evaluate relative changes in the variance of a forecast metric, that is, the low‐level wind components averaged over the Rhein–Ruhr metropolitan area in Germany. This setup allows us to compare the relative variance change associated with the assimilation of hypothetical observations from a Doppler wind lidar with respect to the assimilation of surface‐wind observations only. Furthermore, we assess sensitivities of derived variance changes to a number of settings, namely observation errors, localization length scale, regularization factor, number of instruments in the network, and their location, as well as data availability of the lidar measurements. Our results demonstrate that a network of 20–30 Doppler lidars leads to a considerable variance reduction of the forecast metric chosen. On average, an additional network of 25 Doppler lidars can reduce the 1–3 hr forecast error by a factor of 1.6–3.3 with respect to 10‐m wind observations only. The results provide the basis for designing an operational network of Doppler lidars for the improvement of short‐term low‐level wind forecasts that could be especially valuable for the renewable energy sector.

Ensemble‐based estimates of the impact of potential observations

“…In that case, we recommend choosing the strongest regularization that does not substantially decrease the mean variance reduction, as overregularization is less prone to cause high errors than underregularization is. We used this approach to determine the $$$ \alpha $$$ value for our study of the potential of wind lidar observations for improving wind forecasts (Nomokonova et al, 2022).…”

“…This increase in efficiency allows evaluation of many more possible observational networks under a wider range of meteorological conditions. In our companion paper, we achieved robust results by applying the explicit global method to a 1,000‐member forecast to estimate the benefit of wind lidar observations on short‐term wind forecasts (Nomokonova et al, 2022).…”

“…There is a good reason why the explicit methods shown in the last subsections were not used in any study before Nomokonova et al (2022), and that is because, without analysis localization, the same result can be achieved without explicitly calculating the sensitivity. Instead, the variance reduction can be computed much more cheaply directly in observation space without requiring an inversion of the background covariance matrix.…”

“…In this study, we revisit the methods used in the network design studies of Mauger et al (2013), Hakim et al (2020), and Tardif et al (2021) and present three different methods to account for analysis localization in full detail. The first method has not been applied in any study before our companion paper (Nomokonova et al, 2022) and requires inverting the background ensemble covariance matrix, which is computationally rather expensive and introduces a free parameter needed to regularize the matrix. The second method avoids the inversion but needs to be provided with estimates of signal propagation over time.…”

Guidance on how to improve vertical covariance localization based on a 1000-member ensemble

“…Due to these severe weather events, several studies focused on this exceptional period (e.g., Piper et al, 2016). Necker et al (2020a, b), Nomokonova et al (2022), andCraig et al (2022) provide further details on the weather situation in this period as these studies also explore the 1000-member ensemble simulation with a different purpose.…”