Assessment of wind resources in two parts of Northeast Brazil with the use of numerical models

Abstract

A study was conducted to quantify the wind resources in two locations (municipalities of Paracuru and Triunfo) with different topographical conditions (flat and complex) in the Northeast Region of Brazil (NEB). To this end, data collected in situ with anemometer towers and a simulation of the mesoscale numerical Weather Research and Forecasting (WRF) model were used. These served as initial conditions for simulations of the microscale numerical model from the Wind Atlas Analysis and Application Program (WAsP). The WAsP model enabled estimation of the annual wind potential and its power density, both through measured and simulated input data (WRF). Wind speed fields and the average direction of the wind, parameters of the Weibull probability density function (an estimate of the mean annual power density and wind potential), were used to compare the simulations and the observations. The results show that both locations, Paracuru and Triunfo, have a favourable annual wind potential for the implementation of wind farm projects. The power density estimated through measured and simulated data exceeded 400 W m−2 at heights of 60 and 50 m above the ground. The demand for electricity is growing in Brazil as a result of the country's social and economic development. The high costs of energy sources such as oil and hydroelectric power are yet another encouragement to look for other sources of power generation. Among the various sources of renewable energy in the world, wind power has been one of the fastest growing sectors in recent years, 2010–2013 (Zhao et al., 2012; GWEC, 2014). Currently, Brazil occupies the 7th place in the world in terms of installed wind capacity between January and December 2013, with 953 MW (top six: (1) China, 16 088 MW; (2) Germany, 3238 MW; (3) United Kingdom, 1883 MW; (4) India, 1729 MW; (5) Canada, 1599 MW; (6) United States, 1084 MW (GWEC, 2014)). This document shows that Brazil has been one of the most promising countries for the generation of onshore wind power over the past 5 years, producing 4.7 GW of power in 2013 through three public bids and with a capacity of 7.0 GW to be installed by the end of 2022. The government projects 17.5 GW of installed wind capacity by the end of 2022 (GWEC, 2014). For onshore wind power, the geographical distribution of wind farms in operation and under construction in Brazil until the end of 2013 is highest in the Northeast Region of Brazil (NEB), with a total of 3213.4 MW installed and of 8035.3 MW under construction, especially in the states of Rio Grande do Norte, Ceará (CE) and Bahia (GWEC, 2014). Some regions in Brazil are characterized by their high wind energy potential. Amarante et al. (2001) have shown that the coastal (Ceará and Rio Grande do Norte states) and semiarid (with complex topography, more specifically Alagoas, Bahia, Paraíba, Pernambuco states) areas of NEB stand out as having the highest wind power potential. Rocha et al. (2012) state that the constant winds throughout the year in the coastal region, particularly in the city of Paracuru in Ceará state, favour wind energy generation. Lima and Filho (2010) used results of a microscale simulation using the software WAsP (Wind Atlas Analysis and Application Program; Mortensen et al., 2006) to show that the city of Triunfo (Pernambuco state) is a promising location for a more detailed study of wind power. The studies by Rocha et al. (2012) and Lima and Filho (2010) could encourage more investments, creating conditions for the large-scale generation of renewable wind power in these locations. In order to evaluate the wind energy potential of a region, long-term meteorological observations are necessary. Santos and Santos e Silva (2013) evaluated wind speed trends using measurements from meteorological stations over a period of 26 years in NEB. The results reveal areas with a decreasing wind speed trend in continental states, but no such trends for the coastal states. These authors concluded that new investments in coastal wind farms are viable. However, studies with more precise data, collected at smaller time intervals from anemometer towers in NEB, are still scarce. Carvalho et al. (2012) described the restrictions of wind measurements for new wind farm projects, such as the high cost, quality and availability of data and long-term measurements (over 1 year). To overcome the high cost and low quantity of anemometer towers, microscale and mesoscale numerical models are increasingly being used. Mesoscale numerical models for weather forecasting, such as the WRF (Weather Research and Forecast) model, are constantly being used in research to simulate the behaviour of wind, and have shown high sensitivity for various parts of the globe. Byrkjedal and Berge (2008) evaluated the wind resources in Norway using the mesoscale WRF and microscale WAsP models. The results of the WRF simulations were compared with observations, obtaining an R2 between 0.8 and 0.9. Carvalho et al. (2013) evaluated the wind resources in two sites located in Portugal using the same models as Byrkjedal and Berge (2008). The results proved satisfactory according to three coupling techniques using the simulated data from WRF as input condition for the WAsP model. Oliveira (2013) performed numerical simulations with the WRF model in order to analyse the wind energy potential of Paraíba state, Brazil. Their results showed that the highest wind potential densities can be found in coastal and mountainous areas. Carvalho et al. (2014) showed the sensitivity of the wind regime obtained with the WRF model through six different sources of initial and boundary conditions for Portugal, which has a high wind power potential. Carvalho et al. (2014) concluded that data from the National Centres for Environmental Prediction (NCEP) (Global Forecast System and Final Analyses) seem to be the best alternative compared with the results of ERA-interim (best result in relation to other reanalyses). Other studies using the mesoscale WRF and microscale WAsP models to characterize regions with wind energy potential can also be cited, including studies in Peru (González-Mingueza and Muñoz-Gutiérrez, 2014), the United States (for Colorado, Cheng et al., 2013), Brazil (Lima and Filho, 2010, 2012; Ramos et al., 2013) and China (Zhao et al., 2012). The objective of the present study was to quantify and compare the wind energy resources of two different areas in the NEB using simulated data from the WRF model and measured data from anemometer towers for the cities of Paracuru (coastal area in Ceará state; CE) for 2005, and for Triunfo (heartland of Pernambuco state; PE) for 2006. The results should provide support for the scientific community with new research, and assist companies and government agencies at the federal and state levels in attracting investments for the large-scale production of wind power, in addition to increasing the production of wind energy in NEB. The data from anemometer towers (TAs) were made available by projects managed by the Department of Infrastructure of Ceará (SEINFRA/CE; http://www.seinfra.ce.gov.br) and the network of the Sistema de Organização Nacional de Dados Ambientais (SONDA, National Organization System of Environmental Data; http://sonda.ccst.inpe.br/). The SONDA network is managed by the Instituto Nacional de Pesquisa Espaciais (INPE, National Institute for Space Research) with the support of the Ministry of Science, Technology and Innovation. Its purpose is to collect and improve the database on solar and wind energy resources in Brazil (Martins et al., 2007). The first TA is located in the semiarid sector of the NEB in an area with complex topography, the municipality of Triunfo, PE. The horizontal speed and wind direction sensor is a propeller anemometer manufactured by the R.M. Young Company, Wind Monitor-MA Model 05106. The second TA is located in the coastal region, in the municipality of Paracuru, CE. The speed and wind direction sensor of the TA in Paracuru is the NRG 9200Plus model manufactured by NRG Sytems Inc. Hereafter, the TAs in Triunfo and Paracuru will be referred to as TA-TRF and TA-PAR, respectively. The temporal sampling interval for the data is 10 min. These measurements were subsequently converted into hourly averages. Figure 1 shows the location of each TA used in this study, and Table 1 shows the local characteristics, such as geographic co-ordinates, terrain, vegetation and the period of data used. According to Alvares et al. (2013), the climate in Paracuru is of the type Aw (tropical with less rainfall in summer) in the Köppen classification. The maximum temperatures vary from 29.4 °C (March) to 30.7 °C (November) and the minimum values range from 21.2 to 23.7 °C, with an annual average temperature of 26.6 °C according to the Fundação Cearense de Meteorologia e Recursos Hídricos (FUNCEME, State of Ceará Foundation of Weather and Water Resources). The rainy season is concentrated between January and April, with a mean annual rainfall of 1238 mm. The prevailing wind direction according to the data supplied by Companhia Energética do Ceará (State of Ceará Energy Company, COELCE) is from the east (76%) and southeast (13.10%). The winds reach their maximum speed in August and December with 10.8 m s−1 and their minimum of 1.3 m s−1 during the rainy months. The municipality of Triunfo-PE has a mountainous Cw'a type climate according to the Köppen classification (Alvares et al., 2013), characterized as a mesothermic climate with a dry winter and rainy summer. The mean annual rainfall is 1222 mm with seven dry months and the highest rainfall in March and April. The mean annual temperature is 25 °C, with a range of 26.3 to 16.7 °C, according to data from the Instituto Nacional de Meteorologia (INMET, National Institute of Meteorology). The mean monthly climatological wind speed is 10 m s−1 with a range of 2.65–3.80 m s−1, with higher speeds during May to October and a prevailing southeasterly direction (INMET). The mesoscale model used was the Advanced Research Weather Research and Forecasting System (WRF-ARW), version 3.5, developed by the National Centre for Atmospheric Research (NCAR) (Shamarock et al., 2008). The simulations were performed with two nested horizontal domains, with a spacing of 15 and 5 km (Figure 2). The first area (A1) was used to evaluate the performance of the WRF model at Triunfo-PE (Figure 2(a)). The second area (A2) was used to evaluate the performance of the WRF model at Paracuru-CE (Figure 2(b)). The configuration of the number of grid points in the model in their respective areas can be seen in Table 2. The WRF-ARW simulations are carried out through a set of parameters based on microphysics, convection, turbulence, surface processes, boundary layer processes and radiation. In the present study, the employed parameters (Table 3) were based on other studies that showed satisfactory results for wind speeds and direction (Miglietta et al., 2013; Ramos et al., 2013; González-Mingueza and Muñoz-Gutiérrez, 2014) Iacono et al. (2008) Iacono et al. (2008) Hong and Lim (2006) Kain (2004) Monin and Obukhov (1954) Tewari et al. (2004) Janjic (1994) Fifty vertical levels were used in all domains. The initial and boundary conditions for the simulations of the WRF model were obtained through the Global Final (FNL) data from NCEP, with a horizontal spacing of 1.0° × 1.0° (ca 110 km × 110 km) latitude–longitude and an interval of 6 h. The information on topography was taken from the digital global elevation model developed by the US Geological Survey (USGS), which covers the entire continental portion of the planet. The soil moisture data (16 categories) came from the USGS and data on vegetation cover (20 categories) came from the Moderate-Resolution Imaging Spectroradiometer (MODIS) satellite. These three categories of information (topography, type of vegetation and soil moisture) were made available in a regular grid with a spacing of approximately 925 m. In the present study, simulations of 84 h were carried out until completing the period for each TA measurement (1 January to 31 December 2005, A1/TA-PAR; and 1 January to 31 December 2006, A2/TA-TRF). The first 12 h (spin-up) were discarded and the following 72 h were used in the analyses. In addition, as suggested by Carvalho et al. (2012), the nudging-by-grid option of the Four-Dimensional Data Assimilation (FDDA) system was used in all simulations of the WRF-ARW model. A more detailed description of this technique can be found in Shamarock et al. (2008). The values for wind speed and direction for every hour on the nearest point of each TA at a height of 50 m in the d02 domain were extracted from the simulations of the WRF-ARW model. d02 was chosen because some studies have shown that the more refined grid of the WRF-ARW model is better at detailing the behaviour of wind flows in areas with complex topographies (Carvalho et al., 2012, 2014). The simulation to estimate wind resources at the microscale was performed using the WAsP model, version 9.1, licensed by the Centro de Tecnologias do Gás e Energias Renováveis (Centre of Gas and Renewable Energies Technologies, CTGAS-ER). WAsP was developed by the Wind Energy and Atmospheric Physics Department at Risø National Laboratory. The model is used to estimate wind resources based on the horizontal and vertical extrapolation of the wind data for a given region (Bower and Mortensen, 2004). According to Bower and Mortensen (2004), the model is based on predominantly neutral atmospheric conditions. The corrections for a non-neutral atmosphere are obtained through changes in the parameters of the surface heat flux components. According to Mortensen et al. (2006), the model extrapolates the data measured by a TA on different types of topography in order to achieve the statistical description of winds within a grid for a given elevation above the surface of the analysed site, while correcting for the effects caused by the presence of obstacles. WAsP is composed of physical models that describe the flow of the wind under the effect of different topographies and roughness, and the shadow effects caused by obstacles. The physical components of WAsP contain four main modules: topography; obstacle; roughness; and atmospheric stability. The topography, roughness and obstacle modules are input information in WAsP. This study only used topography and roughness in the simulations performed. Based on a statistical series of wind measurements from a TA, the effects of the obstacles and roughness of the region were removed, resulting in a general climatological field for the region. At a close distance to the tower, a location was chosen for a wind turbine where the winds are estimated considering the general field, using the geostrophic drag law and obstacles of the wind turbine's location. WAsP generates statistical information about the winds: the frequency of its direction; Weibull probability distribution graphs with the parameters of form and scale; and power density and annual wind power potential for the chosen wind turbine at any given time. More details on the WAsP model can be found in Mortensen et al. (2006). WAsP requires data on initial and boundary conditions. This study used the wind speed and direction data for the two locations, both measured and simulated with WRF-ARW, as input. Based on these data, power density and annual wind power potential maps were developed for the locations close to TA-TRF and TA-PAR. The methodology for running WAsP with data simulated by numerical time models on the mesoscale has been employed extensively in recent studies and is well established. Some authors have used meteorological fields of wind direction and speeds simulated with the WRF model as initial conditions for WAsP (Carvalho et al., 2013). The horizontal grid spacing used for the calculation of wind resources in WAsP was 50 m × 50 m. Table 4 shows the number of grid points used for TA-TRF and TA-PAR. Its integration period was 1 year: 2005 for Paracuru and 2006 for Triunfo. In addition, the heights at which the data were measured and simulated are indicated at each site. The coupling technique between the mesoscale (WRF-ARW) and microscale (WAsP) models suggested by Carvalho et al. (2013) was used to analyse and compare the wind potential. This method consists of using the simulated wind speed and direction as input for the microscale model. This permits a direct comparison between the simulated and measured wind. Here, the data obtained with the WRF-ARW model are represented as a fictitious TA, with the speed and direction of the temporal series being calculated for both locations. The calculation of the wind atlas realized by WAsP consists in removing and reinserting the effects of the terrain on the wind data. To this end, the data on topography and roughness for each region under study were supplied to WAsP for the calculations needed to put together the wind, annual energy production (AEP) and power density atlas. The high resolution topography was obtained from the TOPODATA project data (Valeriano and Rossetti, 2012), which have a grid with a spatial resolution of 30 m × 30 m, developed by INPE (http://www.dsr.inpe.br/topodata/acesso.php). The roughness was developed for each given location through the satellite images obtained from Google Earth. The fields created by WAsP with simulated and observed data were compared in order to evaluate the performance of the model WRF-ARW as an additional source to be used in wind resource calculations. In wind energy applications, statistical methods are used for the random variations in wind speed. The simulated and observed time series are presented in terms of the Weibull probability density function (PDF), from which the following parameters are obtained: scale (A); shape (k); and mean speed (V). The Weibull PDF has been widely used to represent wind speed distributions in wind energy applications because it is able to provide a better fit with the experimental data, despite its simple form (Carvalho et al., 2013). According to the value of k, the Weibull distribution can be similar to other types of statistical distributions (k = 1.0, exponential; k = 2.0, Rayleigh; k = 3.5, normal) (Lima and Filho, 2010). Once the mean wind speed V and standard deviation are known, the Weibull parameters can be calculated. The amount of potential that can be obtained from the wind depends on the available wind energy and on the operating characteristics of the wind energy extraction device. However, due to the Betz limit, the turbines cannot use 100% of this potential. Wind power density is a useful way to assess the available wind resources in a location without taking any given wind turbine into account. The wind potential density, measured in watts (W) per square metre, indicates the amount of available energy at a location for conversion by a wind turbine. The WAsP computational model uses the input parameters described above and generates a ‘virtual wind atlas’ for the given map area. The wind atlas is created through simulations of the influences on wind distributions caused by the topography, roughness and obstacles present in the maps, determining the behaviour of the wind above the atmospheric boundary layer. This wind is known as geostrophic wind. The wind modelling is completed by including the topographic and the roughness effects on the geostrophic wind for each position of the turbine layout of the wind farm under study. From the estimated wind energy resources for each area of the WAsP based on the employed time series (simulated and observed), estimates are obtained for AEP and power density. For the AEP estimation, a wind turbine manufactured by ENERCON (model E-53) was considered, with a hub height of 50 and 60 m, and 810 kW of rated power, positioned at the same geographical co-ordinates as TA-TRF and TA-PAR and their respective grid points in the WRF model. The power curve of this wind turbine is shown in Figure 3. Table 5 presents the results for the parameters of the Weibull PDF calculated by WAsP for both the data simulated by WRF-ARW and those measured by the TA-TRF and TA-PAR. In the employed methodology, the WRF-ARW simulation underestimated mean annual wind speed by −3.77% for Paracuru and by −36.32% for Triunfo compared with the measured data. The simulations produced mean annual wind speed values (V) below the measured annual mean. For the Weibull parameters, the differences relative to the measured data were larger for k in Paracuru and for A in Triunfo. When both locations are compared, the WRF model has a poorer than expected performance for the location of Triunfo. Figures 4(a) and (b) show the Weibull curves for the simulated (WRF-ARW) and measured (TA) data. If Figure 4(a) is analysed for distribution at 60 m (TA-PAR) and 50 m (WRF-ARW-PAR), it can be seen that the most frequent speed range 8–10 m s−1 is the same for the measured and simulated wind. For the values measured at a height of 60 m, however, there is a difference of −6.5% in the speed range of 9 m s−1 for the simulated value (WRF-ARW-PAR). The curves with the WRF-ARW data are straighter for both locations compared with the measured values. Figure 4(b) shows the simulated Weibull curve for Triunfo at 50 m with a deviation to the left-hand side of the wind speed axis, indicating that the estimates made by the model are larger for frequencies of low wind speeds and, consequently, lower for frequencies of higher speeds (Figure 4(b)). The WRF-ARW model simulated the higher frequencies of wind speeds at around 6–10 m s−1, whereas the measurements showed more frequent speeds at around 12–16 m s−1. The differences between the simulated and measured Weibull functions are larger for Triunfo than Paracuru. Figure 5 shows the frequency histograms of wind direction for the measured (TA) and simulated (WRF-ARW) data. For Paracuru, the model was able to determine the prevailing wind direction (east) efficiently, although it could not show the relative percentage of occurrences in other sectors in a reasonable way (Figure 5(a)). Just as with the observed data, the simulation showed east prevailing wind direction. For Triunfo, the simulations and measurements showed southeast prevailing wind (Figure 5(b)). Although the measurements show that the locations have different wind regimes, generally the model considers both locations to have similar patterns of atmospheric circulation. One of the reasons for these discrepancies may be attributed to the fact that the model assimilates only information from the large-scale meteorological fields provided by NCEP-FNL for its edges. Therefore, the result might have been better if the WRF-ARW had assimilated the observed data for the execution of its simulations. Table 6 presents the results for the AEP estimated by WAsP for the wind turbine ENERCON-53 in the specified locations. Using the wind data simulated by WRF-ARW as input in WAsP produces lower output estimates than the measured data. The AEP estimate reveals a lower performance at Triunfo than at Paracuru (Table 6), which is to be expected considering the wind speed distribution at this site (Table 5). For mean wind speed, the location of Paracuru showed less difference. The WAsP model initialized with the data measured by TA-PAR at 60 m presents mean wind speed estimates of 8.98 m s−1 and a power density of 558 W m−2. When the initial conditions simulated with the WRF-ARW were used, however, the mean speed was 8.68 m s−1 with a power density 462 W m−2. Migoya et al. (2007) state that for an area to become suitable for the generation of electricity, a mean wind speed of 7.5 m s−1 with a power density greater than 300 W m−2 is recommended. Figures 6(a) and (b) show the field area determined in WAsP for the estimation of the power density for Paracuru. Both the measured (TA-PAR) and simulated (WRF-ARW-PAR) data are shown. When mean power density maps estimated by WAsP for Paracuru are compared through the coupling of the microscale and mesoscale models, it can be seen that the values are similar (Figures 6(a) and (b)). In the north and northeast sectors, the region of the Atlantic Ocean, the power density values were higher. The other sectors showed similar behaviour, although the map in Figure 6(b) shows some localized regions where power density was underestimated compared with Figure 6(a). The maps (Figures 6(a) and (b)) also reveal locations with a high potential density exceeding 400 W m−2. Figures 7(a) and (b) show the power densities at 50 m above ground estimated by WAsP for Triunfo. They show the results of similar WAsP estimates, but with lower power densities as a result of the simulated WRF-ARW-TRF data compared with the measured TA-TRF data. The coupling technique of WRF with WAsP for the complex terrain showed poor performance relative to the results for the flat terrain. A possible source of error was discussed by Carvalho et al. (2012), who showed that worse results could be produced for complex terrains because the terrain files imputed in the WRF model, such as roughness and topography, are of low resolution, representing smoother conditions than the actual conditions of the region. In TA-TRF, the mean speed estimated by WAsP was 13.41 m s−1 with a power density of 2058 W m−2, and for the corresponding modelled data the values were 8.50 m s−1 and 537 W m−2, respectively. The underestimation presented in the WRF-ARW data for Triunfo generated a lower power density compared with the measured data, which showed high wind speeds throughout 2006 (data not shown). These values reveal that the performance of the simulated data for Triunfo remained below expectation in comparison with the measured data, although the values were still suitable for power generation. Considering only one wind turbine in Triunfo gives an AEP of 5.83 GWh (Table 6), which is a satisfactory yield at a height of 50 m, while the simulated WRF-ARW-TRF data show an AEP of 3.67 GWh (Table 6). However, considering a single wind turbine does not take into account the losses caused by the wake effect (Pinto, 2013). The density maps estimated by WAsP make it possible to see that TA-TRF was not placed at the most favourable location. According to Figures 7(a) and (b), there are other grid points with greater wind resources. Oliveira (2013) examined wind resources in Paraíba state in the Northeast Region of Brazil (NEB) based on observed and simulated data using two numerical time models (Brazilian Developments on the Regional Atmospheric Modelling System (BRAMS) and Weather Research and Forecasting (WRF)). For the Borborema mesoregion, the results for the Weibull parameters k and A for measured and WRF simulated data were 5.07 and 8.16, and 6.77 and 9.97 m s−1, respectively. The characteristics presented in that mesoregion are similar to the location of Triunfo, Pernambuco (PE) used in the present study. Oliveira (2013) observed that the Weibull curves simulated by WRF generally were straighter, showing a smaller variation of the wind, just as the present study observed for the locations of Paracuru, Ceará (CE) and Triunfo. Rocha et al. (2012) analysed seven numerical methods to evaluate their efficiency in determining the Weibull parameters using wind speed data from Camocim and Paracuru in Ceará state, in the period from August 2004 to April 2006. The results for A and k in their study for 2005 in Paracuru were represented by the ranges 9.35–9.92 m s−1 and 2.49–4.21, respectively. The simulated (Advanced Research Weather Research and Forecasting System for Paracuru (WRF-ARW-PAR)) and measured (anemometer tower (TA)-PAR) Weibull A and k values found in the present study were close to those obtained by Rocha et al. (2012). The main source of error for the underestimation of wind speeds in the simulation for the Triunfo location is the low resolution of the terrain in the WRF-ARW model. This is in agreement with the results of Carvalho et al. (2013), who found major flaws in the WRF-ARW models in areas with complex topography in Portugal. Lima and Filho (2010) also assessed the wind energy potential of the city of Triunfo and performed simulations with Wind Atlas Analysis and Application Program (WAsP) for wind farms based on data measured over 30 months (2004–2007) by anemometer station. The characterization of the wind at Triunfo and the assessment of the wind energy potential show that the region has a mean wind speed of 11.27 m s−1 and a mean power density of 1672 W m−2 for a Vestas V52 wind turbine rated at 850 kW and with total capacity of 20 MW. Because of those results, the authors of that study believed that the region should be considered as an area with potential for future commercial wind power exploitation. The difference between the values obtained by those authors and those given in the present study could be related to the higher number of months used in the former and the fact that 2006 had atypically high wind speeds during the rainy season in the region, contrary to several studies of the NEB region that have shown that winds in this season are usually weaker (Oliveira, 2013; Ramos et al., 2013; Santos and Santos e Silva, 2013). The authors would like to thank REUNI and CAPES for the doctorate grant awarded to Alexandre Torres Silva dos Santos. The authors are grateful to two anonymous reviewers for their constructive comments, which helped improve the presentation of the results obtained. Thanks are also given to CTGAS-ER for support in the integration of the WAsP and WRF-ARW.