24-hour ahead 10-meter wind Forecasts

Is there any chance that observations have to be stopped in the course of the night because the wind is reaching the maximum operating velocity (20 m/s)?

This topic did not meet a satisfying answer until now. The results of Table 4 by [CRS4 97] show large errors and poor correlation of 10-meter wind forecasts and analyses with data measured at Paranal or La Silla.

As for 2-meter temperature, much of ECMWF forecast error is due to the lack of observational data in the Antofagasta area could not be corrected by implementing a LAM (20 km mesh) without injection of new observational data. However unlike 2-meter temperature not only the forecasts, but also the analyses are poorly correlated. Obviously a global model such as ECMWF and even the LAM have a mesh much too large for the steep terrain surrounding astronomical sites. One has probably to go down to a mesoscale model such as the one described in Section 4.2 (ca. 500 m mesh). A sample output on Fig. 4 shows that local overspeed and shears are then clearly visible. Other more straightforward methods such a modified mass consistency models could be investigated to improve the case of 10-meter wind forecast.

$$\vert\epsilon^{Kal}\vert \vert\epsilon^{per}\vert$$ site N r^Kal r^per r^for r^ana

La Silla 1242 2.5 0.64 2.9 0.59 0.57 0.43

Paranal 608 3.0 0.59 3.4 0.56 0.50 0.64

caption

[10-meter wind speed Kalman Filter Postprocessing] Absolute mean errors ($\vert \epsilon \vert$) and correlations (r) between observed 10-meter wind speed (m/s) and estimations by Kalman filter post-processing ()^Kal compared to persistence ()^per, ECMWF 24-hour forecast ()^for, and ECMWF analysis ()^ana. N is number of data used for comparison during the period from 1989 to 1993.