Investigating predictability of offshore winds using a mesoscale model driven by forecast and reanalysis data

Abstract

The atmosphere is inherently unpredictable by deterministic Numerical Weather Prediction models at both small and large temporal and spatial scales with some intermediate regime where predictability has been demonstrated; this study deals with time scales only. The chaotic nature at the smaller time scales is predominantly caused by turbulence and at the large scales by non-linearity of the Navier-Stokes equations. We investigate, based on observations carried out with a wind-lidar at the FINO3 research platform in the North Sea, the ability of the Weather Research and Forecasting model (WRF) to simulate the changes in the observations ahead of time. The simulations are performed in two ways. In one type the model uses boundary conditions from a reanalysis data-set (WRF‑ERA). Alternatively, the simulations are carried out using boundary conditions from a forecast (WRF‑GFS). In this study focus is on the predictability of changes in the wind speed and direction. A metric is suggested that chiefly accounts for point-wise changes in the wind speed and direction including turbulent structures. However, for completeness, a traditional metric that compared predicted and observed wind speed and direction directly is also applied. This metric does not reflect the turbulent structures of the flow for small lead times, as the new metric does. The traditional metric reveals very good skills (Fig. 2) up to a lead time of 4 days for simulations in forecast mode (WRF‑GFS). By applying the new metric and a correlation coefficient of 0.6 as the lower limit for the skill in the simulations at a height of 126 m, corresponds to a lead time of ≈4 hours (reanalysis) and ≈3 hours (forecast) for both wind speed and direction for turbulence limited lead times. This value is larger than typically found over land – being ≈2 hours. The difference likely relates to the marine conditions of the measurement site. For large lead times, when the simulations are nudged towards the reanalysis the forecast skill does not deteriorate for increasing lead times. This is in contrast to simulations nudged towards meteorological forecasts where the predictability is limited by the non-linearity of the Navier-Stokes equations and a correlation coefficient less than 0.6 was found for lead times larger than ≈6 days for wind speed and somewhat smaller – ≈4 days for the wind direction when applying the new metric. Thus, the window of predictability of the WRF simulations nudged towards a forecast is found to be in the interval ≈4 hours up to ≈6 days (wind speed) and ≈3 hours to ≈4 days (wind direction). These numbers refer to a height of 126 m. The predictive skill is found to be a function of height; at 626 m it is better than at 126 m for both wind speed and direction. For the traditional metric a correlation of less than 0.6 was realized for a lead time larger than ≈4 days for both wind speed and direction.