Back to the programme printer.gif Print

Tuesday, 11 March 2014
14:15 - 15:45 Aspects for offshore and complex terrain
Science & Research  

Room: Llevant
Session description

Siting in complex terrain is still a challenge for wind energy developers. This session will shed light on questions including:

What is the best way to estimate the energy resource over hilly and forested terrains? Are linearized models still useful and are computational fluid dynamics (CFD) models mature enough? What is the best way to use meso-scale models for wind energy resource estimation offshore? How does atmospheric stability affect turbulence?

Learning objectives:

  • Judge the performance of linearized flow models in comparison with CFD model for wind resource estimation in complex terrain
  • Understand how meso-scale models are best used for offshore annual energy production (AEP) estimation
  • Appreciate how atmospheric stability affects turbulence over forests and how to use standard measurements to estimate the stability
  • Get an insight into state-of-the-art meso-scale modelling for wind energy resource estimation
Lead Session Chair:
Jakob Mann, DTU Wind Energy

Joachim Peinke, Uni Oldenburg, Germany
Ib Troen DTU, Denmark
Ib Troen (1) F P
(1) DTU, Roskilde, Denmark

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

Dr Troen is co-author of the European Wind Atlas. He was the main responsible for the
development of the Wind Atlas Analysis and Application Program (WAsP),
and has 18 years of experience in wind power climatology.
He was principal scientific officer at the EU Commission in charge of climate change research of the EU framework programmes.
He is currently Special Consultant at the Wind Energy department at DTU
mainly working with improvement of the WAsP model core.


Complex terrain wind resource estimation with the wind-atlas method: Prediction errors using linearized and nonlinear CFD microscale models


In hilly or mountainous terrain with large terrain slopes, one cannot expect linearized flow
models to perform well. In recent years the use of more complete physical microscale CFD
models, such as those based on RANS (Reynolds-averaged Navier-Stokes equations) and
including nonlinear terms, have become more mainstream.Here we present a statistical
comparison of predictive skills of model systems based on linearized flow modeling and on the
2-equation closure RANS model “Ellipsys” [1].


The model setup is based on the Wind Atlas methodology as described in [2] as implemented
in WAsP for the linear model setup (“WAsP-IBZ”) and equally for the nonlinear version except
that here flow corrections due to orography and roughness inhomogeneities are provided by
the Ellipsys model (“WAsP-CFD”). The comparison is intended to document the skill of these
model setups for use in real world wind resource estimation. We have used data from 9 sites
with a total of 26 mast locations, each mast with several levels instrumented in general. The
data are mainly provided by wind power developers, and the masts and sites were therefore
chosen at or near potential wind energy installations. This means well exposed hills and ridges
in general windy settings. The sites are located in Europe (5), Americas (2), Southeast
Asia/Australasia (2). The data and site descriptions are covered by Non-Disclosure-Agreements,
so the individual cases cannot be discussed. We present a purely statistical analysis of the
errors in cross predictions between wind observation locations based on the digitized maps,
mast locations, anemometer/windvane heights and winddata (frequency distributions)
provided to us (thanks to EMD-International for allowing use of 6 of the datasets). Site
description maps and winddata are treated in WAsP in the usual way for the built-in linear
models; for WAsP-CFD the maps are sent with site coordinates to the CFD server (cluster),
where the grid generation is done automatically and the Ellipsys model is run for 36 equally
spaced upwind directions. The upwind (log) profile roughnesses are returned to WAsP together
with a 3D matrix of flow corrections (speedup and directional turning) covering a 2 by 2km
horizontal area centered at the specified site coordinates.

Main body of abstract

The wind atlas methodology [2] was designed for horizontal and vertical extrapolation of wind
conditions. That is, if one has long term measured wind data (speed and direction) from some
point (met mast location) at some height above ground, the method is used to estimate the
wind conditions (wind speed frequency distributions per direction sector) at some other point
of interest (hub height of wind turbine). The method assumes that winds in the points
considered are governed by the same large (meso-) scale wind forcing. In practice this means
that the horizontal distance over which the method can be meaningfully applied depend on the
scales of the overall climate and of the scales of flow modifications introduced by surface
inhomogeneities (roughness differences, thermal differences, hills and mountains, …). As
described in [2] the method further builds on the assumption that (simple) models can be used
to model the local effect of these inhomogeneities. The methodology then in principle consist
of modeling the local influences at the measuring point and subtracting these to give a
“generalized” wind climate and adding the modeled influences at the target point [2]. This is
the basis of the WAsP model. Here we compare the skill of this model setup using two different
configurations. One is the traditional WAsP, denoted WAsP-IBZ here (for IB: “internal-boundary-layers”
, the model used for terrain roughness changes, and “BZ” the built-in linearized flow model of WAsP);
the second configuration WAsP-CFD, where the built-in models are replaced with flow corrections calculated
by the Ellipsys CFD model [1]. The calculations require detailed height and roughness maps together with
the observed wind data in the form of sector wise histograms of wind speed frequencies. Site descriptions (maps),
mast locations and wind data from a total of 55 instrument locations all covering approximately 1 year or more
of concurrent data per site were used as input. This allows a total of 370 cross predictions of the wind climate.
Mast distances varied between approximately 1 km to 15 km, mast levels between 10m and 100m.
The exact same maps and wind data was used for the two configurations and we can therefore directly
compare the results case by case. Taking e.g. the purely horizontal extrapolations, that is the total
of 90 combinations of wind data observation points and validation points (other wind mast data location
within same site) where the two anemometer heights are equal. The heights available depend on the site
as the met masts are of different heights. We divide the heights in 4 roughly equal sized groups
(10-20m, 30-40m, 50-60m and 80-100m). For each we compare error statistics (percent error of the
predicted mean wind speed) for the model configurations plotted in the form of “risk” graphs:
All cases sorted in ascending order of the error for each configuration. We find that for these cases,
chosen specifically as very complex sites with very steep slopes near the points (met masts) considered
and with a mixture of scales of orographic features, thus site conditions where the linear models should fail,
that indeed the CFD configuration is clearly superior. At the same time a simple “land-filling” method
(omnidirectionally modification of the terrain surface to limit radial slopes to a fixed value (0.125))
before using the standard linear model gives comparable errors. Also the delta-RIX correction is competitive,
but it must be noted that the conversion factor needed (factor between delta-RIX and correction applied) appear
to depend rather strongly on anemometer height. Figures 1-4 show the prediction errors in percent (y-axis) for
each case with the standard WAsP-IBZ in blue, WAsP-CFD in red and the Delta-RIX corrected standard WAsP-IBZ in green.
Fig 1 is for the 10-20m heights, Fig 2, for 30-40m, fig 3 for 50-60m and fig 4 for 80-100m agl:
The delta-RIX correction factors (pct correction of mean per pct in dRIX) are 1.2, 1.2, 0.9 and 0.5 respectively.
Figures 5-8 are the same, but with the green curve from WAsP-IBZ after simple slope reduction performed on
the terrain grid.


The study shows as expected the superior performance of the nonlinear flow model over the
linear model system. However, it also shows, that the RIX (Ruggedness Index) correction [3} can
correct the linear results to error levels comparable to those of CFD. The RIX correction is
found, however, to be rather sensitive to the height above ground often making its application
difficult in practice. Also, half of the sites in this study have only two mast locations, making a
site specific delta-RIX calibration impossible. The linear model setup with the addition of a
simple terrain slope limiting method (“land-filling”) was found to statistically give results at
levels comparable to the CFD for heights above approximately 30m, and without much
sensitivity to measuring/prediction height. This setup gives promise for a fast and accurate
resource estimation, to be used alongside CFD calculations, at site settings most relevant for
wind energy development in very complex high-relief terrain. Further work is needed to refine
the slope limiting method. For the latter method, while it is appearing to be very promising, it
must be kept in mind that the cases considered here are exclusively very well exposed ridges
and summits and that the nonlinear CFD model has a wider range of applicability and may
provide more relevant siting parameters than the simplified models. We must also keep in mind
that a (possibly not insignificant) part of the apparent prediction errors may be due to e.g.
speed measuring errors, direction alignment errors, inaccuracies in maps, heights and mast
locations etc. In addition some of the locations are forested and neither of the models include
at present specific automatic provisions for this (e.g displacement), this makes the model errors
especially at the lower levels larger and more uncertain.

Learning objectives
Real world error estimates of wind resource assessment on the microscale by extrapolation from observed data
in very complex high-relief terrain.
Relative skill of simple linearized models with models with more complete physics included.

[1] Bechmann, A., J: Johansen and N.N. Sørensen, 2007. “ The Bolund Experiment – Design
of Measurement Campaign using CFD” Risø-R-1623(EN). ISBN 978-87-550-3638-3. Risø National Laboratory, Roskilde. 19pp.
[2] Troen, I. and E.L.Petersen, 1989. “European Wind Atlas”, ISBN 87-550-1482-8, Risø National
Laboratory, Roskilde. 656 pp.
[3] A.J.Bowen, A.J. and N.G. Mortensen, 2004. “ WAsP prediction errors due to site orography”.
Risø-R-995(EN). Risø National Laboratory, Roskilde. 65 pp.