1 Introduction In Flanders there are currently about 200 large windturbines (between 0.5 and 5 MW) installed. The number of small wind turbines (less than 100 kW) is roughly four times less, despite the much lower investment cost. It appears that the market of small wind turbines is deadlocked: small turbines still need technological improvement to close the performance gap with large wind turbines, but because there is no market, there is very little investment in research and development. Still, the public is supportive of wind turbines, and they think it important to invest in sustainable energy produc- tion (Van Hamme and Loix, 2011). In this paper we present the results of our feasibility study about small wind turbines in Flanders (Runacres et al., 2012). There are a number of small turbines on the market with adequate efficiency. Admittedly, many of the small wind turbines on the market are below par, but some are good enough to represent a viable alternative to other small-scale sources of renewable energy, such as photovoltaic panels. We have found that siting and resource assessment are an important impediment to the further development of small wind turbines. In particular, the classical methods for wind resource assessment are often not sufficient to obtain a reliable prediction of the annual energy production at a site suitable for small wind turbines (Willemsen and Wisse, 2002). The paper is organized as follows: we first present the general results of the feasibility study, focussing on the sensitivity of the approaches to predict annual yield. We then identify three key factors and propose alternatives that are scientifically more justified. 2 Feasibility study Are small wind turbines suitable to provide sustainable energy for SME’s in Flanders? To answer this question, we performed wind speed measurements at various sites and conducted an exhaustive review of small wind turbines to compare the different characteristics. Combining these two aspects, we can make predictions of the annual energy production of different turbines at different sites. We performed wind measurements at 10.5 and 15 m, in built-up as well as rural areas. Our measurements were supplemented by measurements from the Royal Meteorological Institutes of Belgium (KMI) and the Netherlands (KNMI) and wind data available from project partners. For every site we constructed wind roses and energy diagrams and fitted different types of probability density functions to the data. Although these estimated distri- butions represent the data quite well (in terms of mean speed and overall shape), we found that energy yield predictions are very sensitive for the way in which the wind data are combined with the power curve. In our review of wind turbines, characteristics such as cut-in and cut- out speeds, rated power, the power curve, rotor diameter, and cost were collected for over 700 small wind turbines. To predict the annual yield at a certain site we only used turbines for which independently measured power curves were available. We also installed and tested two of our own small wind turbines. We found that rated power is a misleading indicator of the actual annual yield of a small wind turbine. To systematize the resource assessment we analysed a tool developed by Bruno Claessens and his collab- orators at Apère. This tool was used to predict the return on investment for different turbines at 18 locations in or close to Flanders, taking into account local wind speed, independently measured power curves, subsidies and fiscal policies. We extended the tool with our own measurements, added more wind turbines and improved the description of the atmospheric boundary layer. The tool is available for download from www.microwindturbine.be. The tool allows to use actual wind measurements as input data, or one can select a nearby measurement mast and a correction factor to take into account the surroundings (Wieringa, 1992). Often though the turbine will be installed at a height that differs from the height of the wind measurement. The wind speed then has to be extrapolated using the so-called log laws representing the atmospheric boundary layer. We compared different approaches using wind measurements at different heights and CFD simulations. Here again it turns out that the particular procedure used to correlate one wind speed to a different location can change the annual yield by a sig- nificant margin. To summarize, we have identified three key factors that determine the reliability of the resource assessment: * the detailed shape of the power curve at the wind speeds of interest, * the manner in which wind speed measurements are extrapolated from one height to another, * the method used to extract the distribution from the wind measurements. 3 Shape of the power curve We have performed resource assessment studies for ten different turbines using different sets of wind measurements and different density functions of the wind distribution. We found that rated power is a misleading indicator of the actual annual yield of a small wind turbine. The wind speed at limited heights (typically in the order of 15 m) and typical operating conditions of small wind turbines are invariably much lower than the rated speed. The cut-in speed and the steepness of the power curve at these low speeds will ultimately determine the annual energy production in a medium-to-low wind speeds area such as Flanders. This statement is illustrated by an example. The annual energy production of two different small wind turbines are compared for a specific site. Figure 1 shows the power curves of both turbines. A higher rated power and lower cut-in speed for turbine 1 can be determined from the figure. The wind data from a specific site is used to estimate the annual energy production. The results of the measurement campaign for a site in Flanders are shown. The figure shows the probabilty distribution of the measurements and the weibull distribution. As we now predict the annual energy production by the product of the probality distribution and the power curve, a higher energy production for turbine 2 is determined. The estimated annual energy production for turbine 1 is 2251 kWh/year and 2626 kWh/year for turbine 2. This implies also that what constitutes a ‘most suited wind turbine’ will be site dependent. In order to systematically compare different wind turbines for a given site one needs to compare the product of the wind distribution and the power curves. 4 Representation of the wind distribution In particular for the built environment and close to obstacles, the use of the Weibull distribution can give rise to substantial errors. Errors in ex- cess of 10% were found when e.g. a Weibull distribution was used, as is customary in most resource analyses. This is mainly due to its inability to cope with the non-zero probability of very low wind speeds. This inability skews the distribution at higher wind speeds and translates into an error of the estimated annual yield. We compare different methods to alleviate this problem (Carta et al., 2009) and we compare the effect on the accu- racy of the predicted annual yield. We also investigate the applicability of such distribution for wind speed measurements at the rooftop of buildings. More specifically we investigate the method by Takle and Brown (1977) that combines the Weibull distribution with a probability for null wind speeds, and the maximum entropy probability density function (Ram ??rez and Carta, 2006) that is specifically suited to match frequency histograms of the cube of the wind speed, and thus a useful tool to evaluate wind resources at a given site. Takle and Brown (1977) define a hybrid probability density function P H : PH(U) = F0?(U) + (1 ? F0)PW (U) (1) where * F0 is the probability of observing zero wind speeds, * ?(U) is the Dirac delta function, * P W (U ) is the classical Weibull distribution, * and U the wind speed. Ram ??rez and Carta (2006) rather define their probability density function as M PM(U,?1,…,?M)=exp ?0+?iUi (2) i?1 The ?‘s are the distribution parameters, and ?0 in particular describes the non-null probability of the very small wind speeds. We show that we increase the accuracy of the annual yield production by using the maximum entropy distribution to represent the wind data. 5 Extrapolation of wind speeds at different heights Different approaches exist to represent the wind speed as a function of height above ground level. Often a logarithmic profile is assumed, such as (Manwell et al., 2009): with U(z)= ? ln z U the wind speed [m/s], U? the shear speed [m/s], (3) ? the von Karman constant (0.41), and U ? z + z0 z the height [m]. There are however different variations of these laws, and the differences between the different logarithmic approximations are substantial. In the paper we systematically compare different log-law approximations and their effect on resource assessment using CFD simulations. We demonstrate the results using annual yield predictions for a given site in Ranst, Flanders. In figure 3 a CFD simulation of the wind flow over the terrain is shown. To perform a numerical simulation of the site, a wind profile must be applied at the inlet boundary. This profile is shown in figure 4. 6 Conclusions In this paper we have discussed three key factors that greatly influence the accuracy of the annual yield prediction for small turbines: * the shape of the power curve at low to moderate wind speeds, * the choice of the probability density function to represent the wind distribution, * and the extrapolation of wind speeds to different heights. Based on our feasibility study for small wind turbines in Flanders, we used wind measurements, power curve measurements and CFD simulations to predict the energy production for different turbines at a given site. We find that the rated power is not a good predictor of annual energy yield in the typical conditions under which a small wind turbine operates in Flanders. We find that the standard use of the Weibull probability density function can lead to substantial errors. We were able to improve the accuracy of the yield prediction by tuning the probability density function and the height variation of the wind. References Carta, J., Ram ??rez, P., and Vel ?azquez, S. (2009). A review of wind speed probability distributions used in wind energy analysis: Case studies in the canary islands. Renewable and Sustainable Energy Reviews, 13:933-955. Manwell, F., McGowan, J., and Rogers, A. (2009). Wind Energy Explained. John Wiley & Sons, Chichester, 2nd edition. Ram ??rez, P. and Carta, J. (2006). The use of wind probability distributions derived from the maximum entropy principle in the analysis of wind en- ergy. a case study. Energy Conversion and Management, 47(1516):2564 - 2577. Runacres, M., Vermeir, J., and De Troyer, T. (2012). IWT 090192 Final Report–On the use of small wind turbines for sustainable energy produc- tion for SME’s and individuals. Erasmushogeschool Brussel. Takle, E. and Brown, J. (1977). Note on the use of weibull statistics to char- acterize wind-speed data. Journal Paper No. J-8941 of the Iowa Agricul- ture and Home Economics Experiment Station. 7 Van Hamme, E. and Loix, E. (2011). VEA–Survey of the Support for Wind Energy in Flanders. Vlaams Energieagenschap (Flemish Energy Agency) and GfK Significant. http://www.energiesparen.be/node/880. Wieringa, J. (1992). Updating the davenport roughness classification. Jour- nal of Wind Engineering and Industrial Aerodynamics, 41-44:357-368. Willemsen, E. and Wisse, J. (2002). Accuracy of assessment of wind speed in the built environment. Journal of Wind Engineering and Industrial Aerodynamics, 90:1183-1190.