|
Titel |
Testing the data assimilation technique for short-term wind forecast in the PBL: a case study |
VerfasserIn |
E. Avolio, S. Federico, A. M. Sempreviva, C. R. Calidonna, M. Courtney |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250063395
|
|
|
|
Zusammenfassung |
In this contribution we show the results of using a data assimilation technique to improve the
short-term wind forecast at a site in northern Europe. The assimilation technique is a simple
four-dimensional nudging and, for this purpose, we set-up a version of the Regional
Atmospheric Modelling System.
The nudging technique consists of adding an extra-tendency term, to the prognostic
equations of the zonal and meridional wind components, which forces the variable toward the
observations.
-φm- (φobs –φm-)
-t = Ï f(r) (1)
where φmis model variable (zonal or meridional wind component), φobs is the observation, Ï
is relaxation time scale (900 s), f(r) is a Gaussian function f(r) = e0-(r-r)-§2 , and r0=50
km.
The method was applied in Denmark where suitable observations were available at the
Danish National Test Station for Large Wind Turbines, located at Høvsøre (Western Jutland,
Denmark), and refer to the measurements of vertical wind profiles; the instrument is
the WINDCUBE-¢ Doppler LIDAR. Data were available every 10 minutes at the
following levels: 40 m, 60 m, 80 m, 100 m, 116 m, 130 m, 160 m, 200 m, 250 m
and 300 m. The data represent the average of the measurement for the previous 10
minutes. Only data available at the 00 minutes of each hour were considered in this
study.
The RAMS model is set-up with four nested grids. The fourth grid has 1 km
horizontal resolution and is centred over the site. Model levels do not coincide with the
measurement levels, and, to assimilate and to verify the forecast, the observations
were linearly interpolated to the model levels. The physical configuration of the
model is the one adopted for operational forecast over the Calabria Region in South
Italy.
In order to show the potential impact of the nudging technique, we run the model in two
different configurations: (a) a simple forecast and (b) an analysis-forecast run. The runs
duration is twenty-four hours for both configurations. For each configuration, simulations
were performed for a one-month period from 21 April 2010 to 20 May 2010 (one simulation
per day, starting at 12 UTC).
In the simple forecast, RAMS uses the ECMWF (European Centre For Medium Weather
Range Forecast) gridded analysis and forecast data as initial and dynamic boundary
conditions available every 6 hours at 0.25 degrees horizontal resolution.
In the analysis-forecast run, in addition to the ECMWF initial and boundary conditions,
measurements at three-hour time interval are nudged into the model for the first 12 h of
simulation. For the second half of the period, the model is driven only by the ECMWF
forecast, as for the simple forecast run.
To compare the simple forecast and the analysis-forecast runs, we computed the
Mean Absolute Error (MAE) and the Root Means Square Error (RMSE) for one,
three, and six hours following the end of the nudging time, i.e. after the first 12 h of
simulation.
The method was verified we used hourly values at the 00 minutes. So, for each day, one,
three, and six model-observation pairs are available for the one, three, and six hours forecast
verification.
In this work, we show the results of those statistics. There are days in which the forecast
is improved by the nudging technique and days in which the nudging technique does not
improve or even worsen the forecast. Work is in progress to characterise each day in terms of
synoptical versus local situation in order to associate errors at each run at all heights.
Further, we aim to repeat the analyses for very-short term forecasts (up to 1 h). |
|
|
|
|
|