Wind Speed Probability Density Research Articles

Estimation of Weibull Parameters for Wind Energy Application in Al Shihabi City

Wind energy is environmentally friendly and low-cost energy compared to other renewable energy technologies and is easily exploited by different users. Al-Shihabi ia a city with high potential wind resources. The aim of the research is to assess the mean winds and Weibull parameters such as shape parameter and scale parameter. The assessment helps in understanding the wind behavior for the application of the wind energy project. Field data for year 2019 were collected at an interval of 10 minutes at altitudes of 50 and 100 meters. Then the data was analyzed using the statistical Weibull distribution function. The annual average of monthly means at 50 m and 100 m heights are 5.04 m/s and 5.56 m/s, respectively. While the annual average of monthly maximum winds at 50 m and 100 m heights were 13.65 m/s and 15.05 m/s. The larger values were in summer months. The mean wind speed and maximum wind speed values suggest that Al-Shihabi City is a promising site to install a wind farm. Estimating Weibull distribution parameters showed for 50 m and 100 heights that the annual average of monthly values of shape parameter and scale parameter were 5.6840, and 2.4035, respectively. At 100 m height, It was found that the annual average of monthly values of shape parameter, scale parameter were 6.2634 and 2.3215, respectively. It can also be noted that the values of parameters are higher in summer than in winter. It was found that the Weibull distribution is applicable to describe the probability density of wind speed.

An Improved Mixture Density Network Via Wasserstein Distance Based Adversarial Learning for Probabilistic Wind Speed Predictions

This paper develops a novel improved mixture density network via Wasserstein distance-based adversarial learning (WA-IMDN) for achieving more accurate probabilistic wind speed predictions (PWSP). The proposed method utilizes the historical supervisory control and data acquisition (SCADA) system data collected from multiple wind turbines (WTs) in different wind farms to predict the wind speed probability density function (PDF) of a targeted WT at the next timestamp. To better capture the fluctuation pattern of historical wind speed sequences and estimate parameters of the probability mixture model for approximating the wind speed PDF, an improved mixture density network (IMDN) is proposed. To address drawbacks of the traditional maximum likelihood estimation (MLE) on training the mixture density network, a Wasserstein distance (WD)-based adversarial learning is developed and the reparameterization trick is employed for the gradient delivery. The effectiveness of the proposed WA-IMDN is validated based on SCADA data (One dataset is publicly accessible) by benchmarking against a set of the commonly considered and recently reported PWSP methods, such as the mixture density network (MDN), maximum likelihood estimation (MLE) based mixture density attention network (MLE-IMDN), recent DMDNN and Improved Deep Mixture Density Network (IDMDN). Results demonstrate the superior performance of the proposed WA-IMDN on the PWSP. To demonstrate the repeatability of the presented research, we release our code at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/IkeYang/WA-IMDN-</uri> .

A novel hybrid forecasting system based on data augmentation and deep learning neural network for short-term wind speed forecasting

As a clean, economical, and renewable energy source, wind energy plays a very important role in easing the shortage of fossil energy, environmental population, and climate change. However, due to the strong intermittency, volatility, and randomness of wind speed, the large-scale connection of wind energy into the power grid is restricted. Therefore, constructing a reliable prediction model to achieve high-accuracy wind speed prediction is necessary. For this purpose, a novel hybrid model for short-term wind speed prediction is proposed in this paper. First, empirical mode decomposition is employed to decompose the raw wind speed time series into a set of subseries. Then, a data augmentation technique is first used to generate more training data to avoid overfitting of the prediction model. Furthermore, a new predictor based on a convolutional neural network (CNN), a long short-term memory (LSTM) network, and an extreme learning machine (ELM) is proposed for deterministic wind speed prediction, where a fuzzy entropy-based partition strategy is implemented to assign subseries to the CNN-LSTM and ELM. To improve the prediction performance, a synchronous optimization method based on an improved hybrid particle swarm optimization/gray wolf optimizer is proposed for feature selection and parameter optimization. Afterward, kernel density estimation is used to estimate the wind speed probability density function for probabilistic prediction. Finally, the performance of the proposed model is compared with seven other models by using three wind speed datasets from four aspects: point prediction, interval prediction, probability prediction comprehensive performance, and prediction reliability. The experimental results show that the proposed method achieves excellent performance on wind speed time series prediction.

Observation analysis of wind climate in China for 1971–2017 under the demand of wind energy evaluation and utilization

To meet the requirement of the wind energy resource assessment and utilization, the statistical characteristics of wind climate in the China mainland were analyzed in this study by using observation wind dataset for 1971–2017 from the China Meteorological Administration (CMA). A series of key factors closely related to wind power development were analyzed, systematically including the spatial and temporal characteristic of wind speed, the distribution of wind speed at different density grades, the density of the cumulative distribution, strong wind days and weak wind days. The characteristics of wind patterns in different regions are also compares. Results indicate that: (1) the regions with the highest wind speed in China centrally locate in Northwest China, North China and Northeast China with the highest wind speed occurs in spring. The annual variation of wind speed in northern China is bimodal while in southern China it is unimodal. (2) The wind speed probability density distribution shows a significant spatial–temporal difference. The wind speed probability distribution is the most concentrated in spring for almost all regions. For each season, the concentrated and stable wind speed occurs in Xinjiang region in spring. While it appears in the region from Northwest China to the northern part of North China for summer and autumn, and in the southern part of North China and Central China for winter. In the case of equal cumulative probability, the highest wind speed locates in Northeast China and the lowest in Sichuan Basin. (3) For 1971–2017, the wind speed in all regions of China showed a decreasing trend on the whole. But this trend slowed down since 1990s. It is​ worth noting that the wind speed in some regions even increased since 2000. (4) The numbers of strong wind day have been decreasing in all regions, but the numbers of weak wind day demonstrate an evolutionary feature of a growth followed by decrease. In most regions, the numbers of weak wind day reached the maximum in the vicinity of 2000 and then decreased rapidly.

Identification of optimum wind turbine parameters for varying wind climates using a novel month-based turbine performance index

The capacity factor (CF) and power coefficient (Cp) are two important wind turbine characteristics. CF describes the power generation capacity during a given period, and Cp describes the efficiency of the wind turbine. Both quantities depend on the rated wind speed. Determining the optimal rated wind speed that maximizes a function of CF and Cp that is directly related to a wind turbine’s output wind power density thus is of utmost importance as it leads to a maximum energy output. This paper proposes a novel Month-based Turbine Performance Index (MTPI) that considers the hourly mean wind speed data month-wise and enables the evaluation of this desired optimum rated turbine speed (Vr,opt) for a given site. Here, the 2-parameter Weibull distribution is employed as a single tool to parameterize the wind speed data and determine the wind speed probability density function, wind power density, vertical wind shear, CF, and Cp of the wind turbine. The examined stations taken for the analysis are from Trivandrum, Ahmedabad, and Calcutta in India. Our index is especially important in regions with intra annular variability, since it is the first to consider monthly instead of annual data.

LES study of topographical effects of simplified 3D hills with different slopes on ABL flows considering terrain exposure conditions

The oncoming turbulence is of particular importance with respect to the investigation of the turbulent flow fields over hilly terrain, which plays an essential role regarding the optimization of micro-siting of the wind farm and reasonable structural design of wind turbines. In this study, large-eddy simulations (LES) are performed to elucidate the effects of wind shear coefficients (WSCs) and upstream turbulence intensities (TIs) on the complex wind flow fields over the simplified 3-D hills. Systematical analysis of the wind turbulence properties related to the micro-siting and fatigue loads of wind turbines, including fractional speed-up ratio (SUR), terrain-induced TI, velocity spectrum and wind speed probability density function, is performed in a quantitative manner. Stronger non-Gaussianity is identified in the wake of gentle-sloped and moderate-sloped hills, which is weakened by increased WSC and enhanced TI. In addition, remarkable SURs are observed mostly at the summit of a steep-sloped hill near the hill surface, and the fractional speed-up factor is found to increase significantly under higher WSC and larger TI. Furthermore, the influence of topography on the wind direction field is also examined, which shows little dependence on the WSCs and TIs.

A correction to the unimodal and bimodal truncated normal distributions for a more accurate representation of extreme and calm wind speeds

The use of wind speed probability density functions is a standard practice to represent different wind regimes. Generally, these regimes are distinguished by the following three characteristics: the shape of the distribution in the central wind speeds, amount of the calm wind speeds (CWS), and extreme wind speeds (EWS). An in-depth review has highlighted that none of the parametric distributions available is suitable to represent the three main characteristics at the same time. To overcome this gap, the use of the corrected mixture of two truncated normal distributions (CMTTND) and corrected single truncated normal distribution (CTND) are proposed to represent, respectively, bimodal and unimodal wind speed distribution shapes. The CMTTND and CTND are obtained by introducing a correction, respectively, to the mixture of two truncated normal distributions (MTTND) and to the single truncated normal distribution (TND). The MTTND and TND permit an accurate representation of distributions with high levels of CWS. The CMTTND and CTND employ a new parameter, to accurately quantifying also the relative frequencies associated with EWS. The performance of the CMTTND and CTND was assessed using a goodness-of-fit (GOF) test and statistical measures of error in the evaluation of the characteristic mean wind speeds. The analytical expressions of these mean wind speeds are obtained and validated by a numerical integration method for the first time in this work. The accuracy of these distributions is compared with that of other conventional probability distribution models, of which three are unimodal and six bimodal, in four Italian locations and three American locations. The analysis of the results showed that the CTND and CMTTND allow obtaining high GOF of the experimental distributions with R2 and RMSE higher and lower than, respectively, 0.977 and 0.054. Moreover, the CTND results in the most accurate distribution in the estimation of the characteristic mean wind speeds in the case of localities with unimodal experimental distributions and the CMTTND in the case of localities with bimodal experimental distributions. Contrary to other distribution, CTND and CMTTND accuracies grow by increasing the grade of the characteristic mean wind speed by reaching also estimation values lower than 2% of the real ones. This is a great advantage in the wind energy source determination in a location since the available energy depends on the mean cubic wind speed.

Wind speed probability density estimation using root-transformed local linear regression

Despite the many attractive features of wind-based renewables, the intermittency of wind generation stands out as a critical challenge. An accurate and reliable nonparametric wind speed probability density model is proposed in this paper based on the application of local linear regression in tandem with a root transformation method, introduced here for the first time. The proposed root-transformed local linear regression approach provides a robust estimate of wind speed distribution and produces more accurate results than kernel density estimation models. The performance of the root-transformed local linear regression estimator is assessed via comparisons with three popular parametric models (Rayleigh, Weibull and Gaussian distributions), two additional parametric models (Birnbaum–Saunders and Nakagami distributions) recently suggested for wind speed probability density estimation, and two nonparametric kernel density estimation models using two standard goodness-of-fit hypothesis tests (Chi-squared and Kolmogorov–Smirnov), coefficient of determination and two error metrics (root mean square error and mean absolute error). Results confirm the accuracy of the proposed root-transformed local linear regression estimator for modelling wind speed probability density, with substantial improvements in all error metrics over parametric and kernel density estimation models.

Combinatorial method for bandwidth selection in wind speed kernel density estimation

Accurate estimation of wind speed probability density at a given site is crucial in maximising the yield of a wind farm. This goal calls for devising probabilistic models with adaptive algorithms that accurately fit wind speed distributions. In this study, a non-parametric combinatorial method is implemented for obtaining an accurate non-parametric kernel density estimation (KDE)-based statistical model of wind speed, in which the selection of the bandwidth parameter is optimised concerning mean integrated absolute error ( L 1 error) between the true and hypothesised densities. The proposed model is compared with three popular parametric models and Rule of Thumb-based KDE model using standard goodness-of-fit and statistical tests. Results confirm the suitability of KDE methods for wind speed modelling and the accuracy of the proposed implemented combinatorial method. It is worthwhile mentioning that the implemented procedure is adaptive (i.e. data driven) and robust.

Optimization method for multi-access capacity of wind power based on copula

With the increasing penetration of distributed wind power, the problem of wind power curtailment is becoming more and more serious due to the unreasonable installed capacity. A method is proposed to optimize the wind power capacity allocation for regional power grids. The main factors that affect wind power accommodation are analyzed according to the wind speed probability density function and the wind power output model. The correlation model of different wind farms is established by the Frank-Copula function. Aiming at reducing the total wind power curtailment, a distributed wind power installed capacity optimization method is proposed with the correlation model. The optimal capacity allocation of the system is calculated by genetic algorithm and the effectiveness of the proposed method is verified by the case of IEEE RTS-79.

Evaluation of rare velocity at a pedestrian level due to turbulence in a neutrally stable shear flow over simplified urban arrays

The geometric dependency of the wind environment at a pedestrian level is an important issue that influences human comfort and safety in urban outdoor spaces. As such, this paper proposes to investigate the statistical features of wind speeds at the pedestrian level by calculating wind speed probability density functions based on flow field data from large-eddy simulations of simplified urban arrays, aiming to clarify the effects of urban geometry on rare velocity events such as strong gusts or extremely weak air flow. Though strong wind events occur infrequently, a positive correlation was demonstrated between percentile and mean wind speeds, indicating that the risk of gusty events increases with the increase of mean wind speeds. Conversely, the frequency of weak wind events shows an inverse correlation with mean wind speeds, showing that better ventilated urban arrays will retain higher wind speeds. Furthermore, these percentiles and occurrence frequencies are clearly expressed by the frontal area indices of urban block arrays. These results imply a trade-off between the following two objectives for urban area wind environments characterized by the urban geometry: enhancing air ventilation in urban areas and preventing strong wind gust events at a pedestrian level.

Estimation of wind speed probability density function using a mixture of two truncated normal distributions

Probability density functions (PDFs) are normally used to describe wind speed distribution for the proper selection of wind turbines in a given location. The identification of a suitable PDF is fundamental for accurately assessing the wind energy potential and designing the wind farms.To achieve this objective, the use of a mixture of two truncated normal distributions (MTTND), defined for v ≥ 0 and obtained by linearly combining two normal distributions with different means and variances, is proposed in this work for the representation of the wind speed PDF. The distribution is a function of five parameters, does not require a high computational burden and allows the representation of wind calm hours (v = 0). The use of the MTTND allows an accurate estimation to be obtained of the experimental discrete distribution of the probability density and cumulative probability, and the characteristic statistical quantities used to estimate the available energy and the performance indicators in the selection of both the site and wind turbine.The validity of the use of the MTTND was verified by comparison with the most widespread PDFs in the scientific literature: Weibull, Rayleigh, lognormal, gamma, inverse Gaussian and Burr. This comparison was developed using experimental wind speed data relating to five Italian locations and a location in Colorado (USA) belonging to the National Renewable Energy Laboratory. For each location, the parameters of each PDF were obtained with the least squares non-linear regression method. The results of the comparisons, in terms of the coefficient of determination R2 and root mean square error (RMSE) for goodness of fit and in terms of relative error in the calculation of the statistical quantities, show that the use of the MTTND gives rise to greater accuracy than a conventional wind speed PDF.

Development of a statistical bivariate wind speed-wind shear model (WSWS) to quantify the height-dependent wind resource

The goal of this study was to develop a statistical bivariate wind speed-wind shear model (WSWS). The development of WSWS is based on near surface wind speed data available from 397 measurement stations distributed over Germany, as well as on ERA-Interim reanalysis wind speed data available in 1000m above ground level (a.g.l.). These data were used (1) to calculate empirical distributions of wind speed in 1000ma.g.l., (2) empirical distributions of the wind shear exponent, and (3) to fit theoretical distributions to the empirical wind speed and wind shear exponent distributions. It was found that the four parameter Johnson SB distribution reproduces the shape of the wind speed in 1000ma.g.l. empirical distributions best. The four parameter Dagum distribution provided good fits to the empirical wind shear distributions. The parameterized wind speed and wind shear marginal distributions were then linked by 16 joint copulas. Goodness-of-fit evaluation of the joint copulas demonstrates that the Gaussian-Gaussian copula reproduces the empirical bivariate wind speed-wind shear distribution most accurately. By using WSWS it is possible to continuously calculate the wind speed probability density function in hub heights between 10ma.g.l. and 200ma.g.l. This allows WSWS to be applied to virtually any power curve for computing the wind energy yield and capacity factor in the analyzed hub height range. A one-time site-specific parametrization of WSWS is sufficient for a comprehensive height-dependent exploitation of the available wind resource.

Response of wind speed probability distribution parameters to urban expansion

Recent studies have indicated that wind speed probability distribution can be described closely using three-param- eters Weibull distribution, which has come to be widely used in the description of wind speed probability distribution over the past several years. The fitting effect of probability density curve would be better than the universal three-parameter Weibull distribution in calm wind by changing the universal three-parameter Weibull distribution function definition domain condition [ r , ∞) to a limited three-parameter Weibull distribution function definition domain [0, ∞), which would be more applicable to actual wind speed distribution. Timely observation wind speed of Beijing, Nanjing, Guangzhou, and other 5 large urban stations and their comparative county stations, and corresponding NCEP/NCAR Reanalysis-1 wind speed data collected from 1985–2014 were divided into sets, each of which had an average of six periods. The three parameters ( k , c , r ) of Weibull distribution of wind speed probability distribution are fitted using numerical iteration to analyze the changes in wind speed probability distribution parameters with respect to differences in wind speed section, and synthetic variations of those parameters can reflect changes in different wind speed sections, such as calm wind, light wind, and strong wind probability under urbanization, which can provide thorough understanding of the relationship to city size indexes and wind speed probability distribution parameters. Data analysis and mathematical derivation were here used to draw the following conclusions: (1) the relationship of urban-rural wind speed difference D and rural wind speed V x can be expressed as D = b 1 V x + i . Under these conditions, city wind speed parameters k , c , and r and the corresponding rural wind speed parameters c x , k x , and r x have the following relationship: c =( b 1 + 1) c x , r =( b 1+1) r x + i , k = k x . (2) As shown, parameter c is affected by unitary surface friction function, parameter r by the combination of various kinds of factors, and the change in parameter k is only an expression of climatic random fluctuation produced by constructing relative variables of c / c x = b 1 + 1, r / r x = b 1+1+ ir x , k / k x =1. (3) It has a correlation with the degree of variation in wind speed probability distribution parameters and the speed of city expansion. Urbanization resulted in significant variations in location parameter r , scale parameter c comes second and shape parameter k did not show any significant changes. (4) Parameter c decreased with city build-up area enlarges. Seven stations’ regression equations pass the significant tests and the variation rates are between −0.0203–−0.0016 km−2, which means that the expansion of city scale will narrow the distribution range of the wind speed probability density curve. Parameter r presents a rising trend (the value changes from negative to positive) along with the extension of city build-up area. Six stations’ regression equations showed significant results, and the rates of variation were between 0.0014 and 0.0171 km−2, which indicated that the likelihood of calm wind decreased as the urban area expanded. Analysis also showed a close relationship between parameter r and latitude. (5) Using reanalysis data to fit the comparison parameters c x and r x produced the relative variable c / c x , and r / r x , analyses showed results similar to the consequences of c and r . Shape parameter k showed a non-significant relationship with the city size index in comparison.

Protection of surface assets on Mars from wind blown jettisoned spacecraft components

Jettisoned Entry, Descent and Landing System (EDLS) hardware from landing spacecraft have been observed by orbiting spacecraft, strewn over the Martian surface. Future Mars missions that land spacecraft close to prelanded assets will have to use a landing architecture that somehow minimises the possibility of impacts from these jettisoned EDLS components. Computer modelling is used here to investigate the influence of wind speed and direction on the distribution of EDLS components on the surface. Typical wind speeds encountered in the Martian Planetary Boundary Layer (PBL) were found to be of sufficient strength to blow items having a low ballistic coefficient, i.e. Hypersonic Inflatable Aerodynamic Decelerators (HIADs) or parachutes, onto prelanded assets even when the lander itself touches down several kilometres away. Employing meteorological measurements and careful characterisation of the Martian PBL, e.g. appropriate wind speed probability density functions, may then benefit future spacecraft landings, increase safety and possibly help reduce the delta v budget for Mars landers that rely on aerodynamic decelerators.

An SST‐dependent Ku‐band geophysical model function for RapidScat

AbstractA new Ku‐band geophysical model function (GMF), which includes sea surface temperature (SST) dependence, named NSCAT‐5, is developed for improved RapidScat wind retrieval. The RapidScat scatterometer instrument mounted on the International Space Station (ISS) provides near‐global wind data over the oceans. Starting from the existing NSCAT‐4 GMF, the variation of is approximated as a second‐order Taylor expansion in sea surface temperature for each given wind speed , and the fitting coefficients are obtained from both observed and simulated radar cross sections, using ASCAT winds, of either vertical or horizontal polarizations. Furthermore, an intercalibration is performed which aligns the distribution of RapidScat wind speeds to that of ASCAT. NSCAT‐5 is obtained by correcting NSCAT‐4 with SST dependencies and intercalibration information. Validation of the new RapidScat wind products retrieved using NSCAT‐5 shows clear improvements over those obtained with NSCAT‐4: Wind inversion residuals no longer depend on SST, and the wind speed Probability Density Functions (PDFs) are closely overlapping with those of ASCAT. Also, RapidScat NSCAT‐5 minus ASCAT wind speed differences show no SST dependence. The work presented here opens a door for further improving the quality of wind products from Ku‐band backscatter measurements, and helps to build a long‐term and consistent essential Climate Data Record (CDR) of scatterometer winds.

Global comparison of the goodness-of-fit of wind speed distributions

The aim of this study was evaluating the goodness-of-fit of 24 one-component probability density functions and 21 mixture probability density functions to empirical wind speed probability density functions on a global scale. Era-Interim reanalysis wind speed data for the period 2011-01-01 to 2015-12-31 with a spatial resolution of 1°×1° were used to compare the goodness-of-fit of 69 combinations of probability density functions and mixture probability density functions fitted with four different parameter estimation methods. The distribution parameters were obtained by applying the moment method, the L-moment method, the maximum likelihood estimation method and the least-squares estimation method. Four goodness-of-fit metrics related to the probability-probability plot, three goodness-of-fit metrics related to the quantile-quantile plot and one goodness-of-fit metric related to the average wind power density were calculated to assess the suitability of distributions. One important result of this study is that mixture probability density functions like the seven-parameter Burr-Generalized Extreme Value, the seven-parameter Dagum-Generalized Extreme Value, the six-parameter Dagum-Weibull and the six-parameter Generalized Extreme Value-Weibull generally provide a superior fit to one-component probability density functions according to goodness-of-fit metrics related to the probability-probability plot. Another important result is that based on the evaluation of goodness-of-fit metrics related to the quantile-quantile plot, the five-parameter Wakeby probability density function is a suitable choice for onshore and the four-parameter Kappa probability density function for offshore wind speed regimes. The four-parameter Johnson system of distributions and Wakeby probability density functions provided the overall best fit for average wind power density. Only for few wind speed regimes, the often used two-parameter Weibull probability density function was identified as the most appropriate distribution. Maps were produced that country-by-country show the most appropriate on- and offshore distributions on a global scale.

ENGLISH

The objective of this study consists to the modeling and prediction of the regime of windin the intention to estimate the potential of wind energy at Douala in Cameroon. The probability densityof windspeed is modeled through an analysis using the Weibull distribution function. The model has been obtained using wind speed data collected from a metrological station at the Airport of Douala. Four methods for calculating the shape and scale parameters of the Weibull wind speed distribution for wind energy analysis are presented: The Maximum likelihood method, the Graphical method, the Empirical method and the proposed Energy pattern factor method. The application of these four methods is effective using a sample wind speed data set. With some statistical analysis, a comparison of the accuracy of each method is also performed. The study helps to determine that Empirical method is the most effective (K= 2.80 and C= 1.97).

Research on Wind-Induced Reliability of Double Column Suspended Guyed Tower in Strong Wind Area

The ultra-high voltage transmission line is long, which needs to pass through the various complex terrains. In order to ensure the safe structural system, understanding and grasping the wind-induced response and reliability of double column suspended guyed tower (DCSGT) in strong wind area of Gobi Desert are basic requirements. Through combining the conditional probability density of wind speed and frequency function of wind direction, the successive-type joint distribution model of wind speed and direction is obtained by multiplication theorem. In accordance with the principle of equivalent extreme value event, through utilizing the statistical moment point estimation method, the first four-order statistical moment information of equivalent extreme value variables is solved. The probability density function (PDF) of DCSGT response is calculated in accordance with these statistical moments. The dynamic reliability index and failure probability of DCSGT are solved by the moment method.

The importance of rare, high-wind events for dust uplift in northern Africa.

Dust uplift is a nonlinear thresholded function of wind speed and therefore particularly sensitive to the long tails of observed wind speed probability density functions. This suggests that a few rare high‐wind events can contribute substantially to annual dust emission. Here we quantify the relative roles of different wind speeds to dust‐generating winds using surface synoptic observations of dust emission and wind from northern Africa. The results show that winds between 2 and 5 m s−1 above the threshold cause the most emission. Of the dust‐generating winds, 25% is produced by very rare events occurring only at 0.1 to 1.4% of the time, depending on the region. Dust‐producing winds are underestimated in ERA‐I, since it misses the long tail found in observations. ERA‐I overpredicts (underpredicts) the frequency of emission strength winds in the southern (northern) regions. These problems cannot be solved by simple tunings. Finally, we show that rare events make the largest contribution to interannual variability in dust‐generating winds and that ERA severely underestimates this interannual variability.