Sage Journals: Discover world-class research

Abstract

This study presents an analysis of the proposed Weather Research and Forecasting–Wind Atlas Analysis and Application Program (WRF–WAsP) models coupling methodology and evaluates the improvement in the accuracy of the wind predictions for the small island developing state (SIDS) of Fiji in the tropical Southwest Pacific region. The results revealed that the proposed WRF–WAsP coupling methodology can be used as a wind resource assessment methodology provided very-high resolution dynamically downscaled wind resource data is available in the order of 10 years for utility-scale wind power applications. The analysis also revealed that the 1 km × 1 km WRF model data from the finest (high-resolution) domain (d03) is best for coupling WRF–WAsP for wind resource parameters evaluation. The coupling methodology of WRF–WAsP improved the accuracy of the wind prediction by 0.2–6% for the wind resource parameters evaluated for Fiji.

Keywords

Fiji wind resource parameters mesoscale–microscale coupling coupling accuracy evaluation

Introduction

The demand for electricity generation from clean energy sources is increasing rapidly in developed and developing countries around the globe (GWEC, Global Wind Energy Council, 2022; REN21, 2023). This is done in an attempt to reduce carbon emissions from the use of fossil-fuel-based power stations and has led individual countries to set renewable energy targets for electricity generation in their national electricity grid networks. Among these countries is the small island developing state (SIDS) of Fiji, which aims to achieve 100% electricity generation from renewables by 2036 (Government of Fiji, 2021). The current quota of fossil-fuel-based power generation amounts to approximately 40–50% (EFL, Energy Fiji Limited, 2022), which must be replaced with renewable energy-based power stations to achieve this target. This is ambitious for the SIDS of Fiji in terms of identification of potential renewable energy resource sites that can be utilized for electricity generation, and the funding required to execute the development of large-scale renewable energy projects.

Fiji is a SIDS located in the tropical southwest pacific region. The 332 islands of Fiji fall between the latitudes of 12 °S to 22 °S and longitudes of 177 °E to 178 °W. These islands are mountainous and of volcanic origin with a maximum peak of 1300 m. The Fijian climate can be classified as a tropical marine climate, with austral summer (wet season) from November to April and austral winter (dry season) from May to October. The dominant wind direction is southeasterly as Fiji falls in the Southeast Trade Wind Zone. The Southeast Trade Winds are mostly dominant during austral winter, while during austral summer the winds are mostly dominated by other climate systems such as low pressures systems and tropical cyclones (Dayal, 2021).

A number of wind resource assessment studies have been conducted in different areas of Fiji using measured wind data at heights of 10–34 m. These study areas include Kadavu and Suva Peninsula (Sharma and Ahmed, 2016), Bligh Waters (Dayal, 2015), Rakiraki, Nabouwalu and Udu (Dayal et al., 2021b), Butoni Wind Farm (Dayal et al., 2022), Suva (Kutty et al., 2019), and Nasawana (Nair and Kumar, 2021). Most of these studies aimed to assist rural electrification in the interiors of larger islands and in the smaller outer islands, while relatively few studies have examined utility-scale wind farms for onshore and offshore applications (Dayal, 2015, 2021; Dayal et al., 2021b). The measured wind speed and the power density at studied sites within Fiji varies from 1.94 to 6.31 m/s and 11 to 241 W/m² at heights of 10–34 m, respectively. The wind direction at studied sites is dominantly southeasterly but in some cases it's in other directions because of local and island-scale circulations. Apart from the microscale studies, a few mesoscale studies including a Weather Research and Forecasting (WRF) model grid sensitivity study (Dayal et al., 2020a), a WRF model evaluation study (Dayal et al., 2020b) and a decade long mesoscale wind resource assessment study have been conducted for Fiji at a high-resolution of 1 km × 1 km at applicable wind turbine hub-heights to identify potential wind resource areas (Dayal et al., 2021a). A recent study also analyzed the wind resources at the existing 10 MW Butoni wind farm and reported that it is located on a weak wind speed regime of Wind Power Class 1 and has a capacity factor of 5.4% (Dayal et al., 2022). Wind resource data is scare, and the wind energy potential is unknown in most locations around Fiji.

To aid Fiji in the area of resource identification, a decade long (2009–2018) mesoscale wind resource assessment has been conducted at a grid resolution of 1 km × 1 km using the WRF model via dynamical downscaling and the two-way nesting approach for all of the Fijian Islands (Dayal et al., 2021a). This mesoscale wind resource mapping exercise can be used for wind resource site identification and to provide high-resolution wind-resource data at any specific location where wind data was previously not available and the potential for wind was unknown. However, this wind resource data must be used with a microscale model to carry out micrositing of wind turbines for energy calculations. It was decided to couple WRF model results with the linear wind flow model, Wind Atlas Analysis and Application Program (WAsP) to investigate whether better predictions on wind resource parameters is achieved with the coupled approach.

WRF–WAsP is coupling the mesoscale WRF model with the microscale WAsP model. The input data used for mesoscale modeling is at a very coarse (approximately 100 km × 100 km) resolution (Dayal et al., 2021a) and it is dynamically downscaled using a mesoscale model to generate high-resolution wind resource data from 100 km × 100 km to 15 km × 15 km, 5 km × 5 km, and 1 km × 1 km (Dayal et al., 2021a). Wind data results are made available at these higher resolutions, but these mesoscale data cannot be used for siting directly as it will represent the same wind speed and wind direction over the whole 1 km × 1 km grid cell. Here the microscale model becomes handy in micrositing at an even higher grid resolution of 50 m × 50 m. By doing so, this could improve the accuracy of the wind speed predictions by taking into account high-resolution microscale features of the topography and vegetation characteristics. The proposed method could assist in estimating the wind energy that can be harnessed for utility-scale electricity generation (Dayal et al., 2021a).

The accuracy of the wind speed predictions is of utmost importance when it comes to modeling wind resources for the purpose of utility-scale wind power applications (Al-Yahyai et al., 2010). Even with the modern technological advancements in wind speed measurements, numerous challenges are faced by wind engineers and developers when it comes to providing an accurate assessment of the wind resource especially in developing countries and with areas where no standard wind mast-based measurements are available. Numerous research methods have been developed over time for wind resource assessments from microscale to mesoscale and to the coupling of mesoscale and microscale approaches for wind resource assessment (Watson, 2023). The most recent area of research interest among scientists and engineers is integrating mesoscale and microscale models in attempts to provide an accurate industry standard wind-resource assessment methodology (Watson, 2023). In relation to the different scales of the atmosphere, mesoscale has a length scale of few kilometers to tens of kilometers, time scale of hours to days, features include land and sea breezes, tornadoes, weather fronts, low-level jets, Gap winds and Mesoscale models are used as the modeling approach. While microscale has a length scale of few meters to a few kilometers, time scale of seconds to minutes, features include small eddies, turbulence, tip vortices and computational fluid dynamics (CFD) models and linear wind-flow models are used as the modeling approach.

Mesoscale weather forecasting models have been used as a source of wind data for wind resource assessment and for wind resource mapping (Al-Yahyai et al., 2010; Watson, 2023). The modeling results have either been used directly in microscale models or for resource mapping of larger areas for the development of regional wind resource maps. These studies include the wind resource mapping for Norway using WRF–WAsP, with results showing deviations between 3% and 25% depending on the complexity of the terrain (Byrkjedal and Berge, 2008); the wind atlas for Egypt using the Karlsruhe Atmospheric Mesoscale Model (KAMM–WAsP) via statistical–dynamical downscaling, with 5% and 10% deviations in the simulated wind speed in simple and complex terrains (Mortensen et al., 2006); the wind atlas for Spain using the Skiron mesoscale model, with annual wind speed bias of 1.87 m/s over simple terrain and 2.5 m/s over complex terrain (Gastion et al., 2008); wind resource modeling in complex terrain using WRF–WAsP in Portugal for two sites, with wind speed deviations between −36.3% to 7.3% (Carvalho et al., 2013); assessment of wind resources in two parts of Northeast Brazil with WRF–WAsP, with −3.77% and −36.32% deviations in the annual mean wind speed for simple and complex terrains (Silva dos Santos et al., 2016); and building the wind atlas for Finland using the mesoscale model AROME and WAsP, with wind speed deviations of −9.23% to 1.64% (Tammelin et al., 2013). The deviations are between the observed (measured) wind resource parameters and the simulated coupled model wind resource parameters.

The major limitations of the previous studies are that mesoscale model results were obtained by running simulations at a single domain resolution greater than 2 km × 2 km for 1–5 years from coarse resolution general circulation model data as initial and boundary conditions. Because of the low grid resolution of the model simulation domain, one of the studies (Carvalho et al., 2013) suggests not to directly compare winds based on mesoscale models with measurements for the purpose of verification, nor to directly use winds from mesoscale models to a specific turbine site for power production calculations or assessment of wind conditions. This is because the important features of the topography in the vicinity of the measurement station or turbine site are not resolved by the mesoscale model at grid resolutions of 2 km × 2 km and above (Badger et al., 2010). The domain resolution plays a significant role in determining the prediction accuracy of simple (flat to little complex) and complex terrains. Coarse resolution may have little impact on the prediction accuracy in simple terrains, but it will have a higher impact on the prediction accuracy in complex terrains as the exact topography may not be accurately resolved in the model domain. The solution to fixing this issue is by using a high-resolution domain.

The microscale RANS-CFD model WindSim has also been tested for predicting wind speed and wind direction alongside WRF and via coupling over a complex terrain wind park in Nygårdsfjell in Norway for three high-wind events (Bilal et al., 2016). There was no significant improvement revealed in the WRF–WindSim coupling, suggesting that further testing over a large number of test cases is required. The study recommended further simulations over a larger period of time and with different combinations of WRF and WindSim settings (Bilal et al., 2016).

A similar methodology in terms of different model coupling (mesoscale and microscale) has been applied in another study on mesoscale–microscale coupling for wind resource assessment using averaged atmospheric stability conditions at the Honkajoki wind farm in Finland for 1 year (Durán et al., 2019). The results of the study revealed that coupled simulations reproduce a more realistic shear. For estimated energy production, there is no significant difference between coupled and standalone (no-coupling) models. However, a considerable difference in horizontal wind speed patterns was observed between the coupled and non-coupled approaches and this shows that the WRF model resolution has only a small influence on the coupled CFD results for this particular study (Durán et al., 2019). Background information about WRF and WAsP can be found in (Al-Yahyai et al., 2010; Dayal et al., 2021a; Mortensen et al., 2014; Watson, 2023).

Looking at the mesoscale–microscale coupling studies (Badger et al., 2010; Bilal et al., 2016; Byrkjedal and Berge, 2008; Carvalho et al., 2013; Durán et al., 2019; Gastion et al., 2008; Mortensen et al., 2006; Silva dos Santos et al., 2016; Tammelin et al., 2013), it can be inferred that it is still challenging to achieve the desired accuracy of wind speed predictions. There is still no acceptable industry standard wind-resource assessment methodology when it comes to coupling a mesoscale model with a microscale model. This provides an opportunity to investigate approaches in mesoscale–microscale coupling studies.

In this study, we propose an approach to directly couple WRF decade long dynamically downscaled time-series medium resolution simulation data (in the order of 15 km × 15 km and 5 km × 5 km) to higher resolution (in the order of 1 km × 1 km) mesoscale wind resource data as input boundary conditions to run the linear wind flow model, WAsP. This approach can be compared to ground-based automatic weather stations (AWSs) data to assess and evaluate the accuracy of the wind predictions. The best-coupled results can be used at potential wind resource sites to evaluate the potential of wind energy that can be harnessed for utility-scale wind power applications. To the knowledge of the author, this approach has not been tested previously with any study applying decade-long high-resolution (1 km × 1 km) dynamically downscaled WRF data to run the microscale model WAsP for micrositing. The WRF mesoscale model has been selected because it is the most commonly used mesoscale numerical modeling tool for atmospheric/meteorological/wind energy research while WAsP is the most commonly used/standard wind-industry microscale modeling tool for wind energy research.

In this article, we aim to answer two research questions:

Does a coupled WRF–WAsP methodology increase the accuracy of wind prediction for the SIDS of Fiji in the tropical Southwest Pacific?

Can the WRF–WAsP coupling wind resource assessment methodology be used as a standard mesoscale–microscale wind resource assessment methodology for the SIDSs?

The second section of this article provides background material, the third section describes the methodology, the results are presented in the fourth section, then discussed in the fifth section, and the conclusions are presented in the sixth section.

Background

The key wind resource parameters computed by WAsP (Mortensen et al., 2014) include the Weibull distribution, the Weibull A and k parameters of the wind speed alongside the mean wind speed and wind power. Information about these parameters can be found in literature (Manwell et al., 2009; Mortensen et al., 2014; Sathyajith, 2006).

Weibull distribution

In most locations, surface wind speed distribution is well approximated by a Weibull distribution (Byrkjedal and Berge, 2008; Gastion et al., 2008; Mortensen et al., 2006). Fitting the Weibull distribution (a probability distribution function) is therefore the method most frequently used in wind energy studies to obtain a smooth distribution despite under-sampling of wind speed due to limited observation periods. The Weibull distribution is a two-parameter function, mathematically represented by $f (u)$ while the cumulative distribution function is represented by $F (u)$ : $f (u) = \frac{k}{A} {(\frac{u}{A})}^{k - 1} e^{- {(\frac{u}{A})}^{k}}$ (1) $F (u) = 1 - e^{- {(\frac{u}{A})}^{k}}$ (2)where u is the reference wind speed at the height of measurement, k the nondimensional shape parameter, and A the scale parameter whose dimensions coincide with that of $u (m / s)$ . $k, A, u > 0$ . In the observed wind climate analysis of WAsP, the Weibull distribution function of wind speed is represented with an estimated Weibull A parameter which indicates on average how windy the site is, the Weibull-shape parameter k which indicates how peaked the distribution is, Weibull mean U which indicates the long-term mean wind speed and P which indicates the available power density at the site. A higher Weibull A parameter $(> 4 m / s)$ also indicates an increase in the dispersion of the wind speed around the mean wind speed and a higher Weibull $k$ parameter $(> 2.5)$ indicates that the variation of mean wind speed is small (Manwell et al., 2009; Sathyajith, 2006).

Power and energy production from a wind turbine

The power production P (in units of $W$ ) from a wind turbine is estimated using the wind turbine power curve $p (u)$ and the probability density function of the wind speed at the hub height of the wind turbine as follows: $P = \int_{0}^{\infty} f (u) p (u) d u$ (3)The annual energy output from a wind turbine E (in units of $Wh$ ) is estimated using the following equation: $E = η T P$ (4)where T is the number of hours in 1 year and $η$ the energy conversion efficiency of the wind turbine (Mortensen et al., 2014).

Wind speed calculation using WAsP

The WAsP software is a diagnostic model that calculates wind statistics by parameterizing the influence of orography, roughness, and obstacles. It has a built-in orographic flow model (BZ) and a simple internal boundary layer (IBL) model for surface roughness inhomogeneities (Mortensen et al., 2014). It considers the following fundamental factors when calculating wind speed:

The geostrophic balance, where the geostrophic drag law gives the geostrophic wind speed “G”: $G = \frac{U *}{K} \sqrt{{[\ln \frac{U *}{F z_{0}} - A (μ)]}^{2} + B^{2} (μ)}$ (5)where $A$ , $B$ are dimensionless constants, $U *$ the friction velocity, $K$ the dimensionless von Karman's constant $(0.4)$ , $z_{0}$ the surface roughness length, and $F$ the Coriolis parameter.

The modified logarithmic wind profile. $u (z) = \frac{U *}{K} [\ln \frac{z}{z_{0}} - ψ (\frac{z}{L})]$ (6)where $u (z)$ is the wind speed at height z, $ψ$ the stability-dependent function which is positive for unstable and negative for stable conditions, and L the Monin-Obukhov length (Mortensen et al., 2014).

A specific (but uniform) stability, roughness variations, and height variations.

Bias

The Bias is used for evaluation of data tendency. A positive (negative) bias means the WRF–WAsP coupling overestimated (underestimated) the measured/observed values: $Bias = M_{i} - O_{i}$ (7)where $M_{i}$ is the model/simulated variable and $O_{i}$ the observed variable.

The percentage Bias measures the average tendency of the model/simulated values to larger or smaller than their measured/observed values multiplied by 100 and is presented as a percentage. $% Bias = \frac{M_{i} - O_{i}}{O_{i}} \times 100$ (8)The root mean square bias (RMS Bias) is defined as the square root of the mean square (the arithmetic mean of squares) of the set. $RMS Bias = \sqrt{(\frac{{({Bias}_{1})}^{2} + {({Bias}_{2})}^{2} + {({Bias}_{3})}^{2} + \dots + {({Bias}_{N})}^{2}}{N})}$ (9)where Bias₁ is the Bias for set 1, likewise Bias₂ for set 2, and Bias_N for set N.

Methodology

Study model, domains, grid configuration, and ground-based wind measuring stations

The WRF model version 3.9.1.1 of the Advanced Research (ARW) solver was used for the mesoscale simulations of this study. The dynamical downscaling method was used for running mesoscale simulations, whereby coarse resolution output from an analysis using a General Circulation Models (GCMs) are used as initial and boundary conditions to drive a regional numerical model to simulate atmospheric parameters via nesting, considering the local conditions (Dayal et al., 2021a).

The initial and boundary conditions for the SIDS of Fiji were obtained from 6-hourly NCEP-FNL (Final) Operational Global Analysis data at $1 \circ \times 1 \circ$ grid resolution (National Centers of Environmental Prediction/National Weather Service/NOAA/U.S. Department of Commerce, 2020). The static fields for topography were obtained from USGS GMTED2010, land–water masks, land use/land cover classification and albedo, etc. were obtained and interpolated from the 21-class MODIS and MODIS FPAR, all these made available from the National Center for Atmospheric Research (NCAR) database, at a resolution of 30-arc-seconds (National Center for Atmospheric Research (NCAR) Database). Time varying sea surface temperatures (SSTs) were supplied to the model from NCEP-NOAA at $0.083 \circ \times 0.083 \circ$ grid resolution (NCEP-NOAA, SST). The well-tested Tropical Suite of the WRF physics parameterization scheme was used. The simulation period was 10 years from January 2009 to December 2018.

Figure 1 shows the WRF study domains for d01, d02, and d03. Table 1 shows the grid configuration in terms of spatial coverage, resolution, grid cells in west-east direction (e_we) and grid cells in south-north direction (e_sn). Further details can be found in Dayal et al. (2021a).

Figure 1.

WRF domain setup for d01, d02, and d03. WRF: Weather Research and Forecasting.

Table 1.

Grid setups.

Domain	d01	d02	d03
Spatial Coverage	1500 km × 1500 km	755 km × 755 km	401 km × 401 km
Resolution	15 km × 15 km	5 km × 5 km	1 km × 1 km
e_we	100	151	401
e_sn	100	151	401

Figure 2 shows the 30-arc-seconds WRF model topographical map of the study area of Fiji with marked locations of the 24 AWSs of the Fiji Meteorological Services. The wind data measurements are available on the order of 2–8 years from the 24 AWSs at 10 meters elevation. This data is used in the evaluation of the WRF–WAsP coupling methodology for Fiji.

Table 2 shows a summary of the station details in terms of their latitude, longitude, real elevation and model elevation in meters above sea level (m a.s.l), and the data period. The model terrain elevation of each AWS is the 1 km × 1 km grid cell average taken from the grid cell in which the AWS is located.

Table 2.

Details and position of AWSs.

Station name	Latitude (° N)	Longitude (° E)	Elevation (m a.s.l.)	WRF Elevation (m a.s.l.)	Data Period
Keiyasi	−17.8795	177.7552	89.8	63.7	2015–2018
Koro Island	−17.3450	179.4184	108.8	80.1	2014–2018
Korolevu	−18.2129	177.7304	25.7	41.5	2015–2018
Labasa	−16.4333	179.4000	8.5	13.1	2016–2018
Levuka	−17.6403	178.8186	41.5	43.1	2015–2018
Lomaivuna	−17.8714	178.3601	122.1	128.5	2014–2018
Momi	−17.8952	177.2668	43.8	25.9	2015–2018
Nadarivatu	−17.5676	177.9632	824.1	761.8	2014–2018
Nadi	−17.7599	177.4448	20.7	14.9	2013–2018
Nausori	−18.0464	178.5591	5.7	4.9	2013–2018
Rakiraki	−17.3404	178.2214	8.1	3.4	2013–2018
Rarawai	−17.5564	177.6814	9.3	17.0	2011–2018
RKS	−17.7260	178.5551	17.1	14.1	2014–2018
Saqani	−16.4749	179.7089	30.0	43.2	2016–2018
Seaqaqa	−16.4758	179.1578	101.8	109.5	2016–2018
Sigatoka	−18.1422	177.5039	6.7	12.0	2013–2018
Suva	−18.1475	178.4536	6.6	15.6	2015–2018
Tokotoko	−18.2186	178.1700	4.9	6.2	2013–2018
Udu	−16.1411	−179.9947	43.7	0.3	2013–2018
Viwa	−17.1494	176.9117	2.0	1.0	2013–2018
Vunisea	−19.0469	178.1654	31.9	86.3	2014–2018
Wainikoro	−16.3044	179.5586	15.1	15.3	2016–2018
Yaqara	−17.4330	177.9774	20.0	11.2	2014–2018
Yasawa	−16.6984	177.5746	40.0	5.1	2013–2018

AWSs: automatic weather stations; WRF: Weather Research and Forecasting.

WRF–WAsP coupling

The WRF–WAsP coupling methodology is divided into two parts: first, the coupling of WRF–WAsP at AWS locations using WRF model wind data from three domains (d01 = 15 km × 15 km, d02 = 5 km × 5 km, and d03 = 1 km × 1 km) into WAsP for analysis, comparison and evaluation and secondly, using the data from domain 3 (1 km × 1 km) for identified potential wind-resource sites as input boundary conditions into WAsP for micrositing. This study just focuses on the first part. Figure 3 presents a schematic view of the mesoscale–microscale (WRF–WAsP) coupling strategy.

Figure 2.

Topographical map of the study area (Fiji) with locations of the 24 AWSs.

For coupling WRF data with WAsP, decade-long (2009–2018) times series WRF simulated wind speed and wind direction data is extracted from all the three domains of $d 01 = 15 km \times 15 km$ , $d 02 = 5 km \times 5 km$ , and $d 03 = 1 km \times 1 km$ and input into WAsP alongside high-resolution terrain (elevation and topography) data, vegetation, and land cover (surface roughness) data combined in vector maps. These wind field data is from the grid cell in which the automatic weather stations are located and are input at the same elevation in the WAsP resource grid. WAsP removes the mesoscale effects of the WRF input data to geostrophic winds at each site and then applies the Wind Atlas Methodology incorporating high-resolution terrain data (Source: Global Wind Atlas Warehouse map server (http://www.viewfinderpanoramas.org) at a resolution of 3 arc-seconds = 90 m), vegetation, and land cover data (Source: Global landcover map (http://due.esrin.esa.int/page_globcover.php) at a resolution of 10 arc-seconds = 300 m) for analysis, comparison, and evaluation alongside resource mapping at a high resolution.

A domain map from the WRF model for a particular site is used initially in WAsP and the wind atlas methodology is used to remove the mesoscale effects of the WRF elevation and roughness taking the winds up to the boundary layer geostrophic winds (at around 1 km a.g.l) then a new map with high-resolution terrain and roughness is used to incorporate microscale high-resolution data for a particular site. The mesoscale effects are the tendency of the model to overestimate lower winds and underestimate higher wind speeds due to the simplification of elevation in the mesoscale model. When the winds are taken up to the geostrophic wind height all these effects are removed by the wind atlas methodology and thus high-resolution elevation and roughness data becomes useful in getting winds at a particular wind turbine hub-height.

Since a good agreement between the mesoscale wind resource mapping for Fiji at a grid resolution of 1 km × 1 km against 24 AWSs was obtained in a recent study (Dayal et al., 2021a), it is proposed to directly couple high-resolution WRF data as input into WAsP. Subsequently a two-way nesting approach has been used for the mesoscale simulations for the three domains whereby the flow of information goes from the coarser domains to the finer domains, with feedback from the inner domains. The averaged values over the grid points from the inner domain are shared to the parent domains to overwrite values at corresponding grid points. The two-way nesting approach was used so that the coarser grid results can be improved using the higher-resolution grid results of the inner domain in case of resource mapping using the data from the outer domains. In this way the results from a more resolved topography from the finest grid is used to overwrite the results of the outer grids improving its results (Skamarock et al., 2008; Wang et al., 2018). The two-way nested approach allows for interaction between the outer and inner domains. The WRF-based two-way nesting approach is performed as follows. The parent (outer) domain is first integrated one time step. Its time and space interpolated values are then specified on the nest boundaries while the nest (inner) domain is advanced with small-grid time interval (typically three steps for an outer-to-inner-domain ratio of 3:1) to reach the parent domain's time level. Then the interior values of the nest domain are averaged back to the parent domain, overwriting the parent domain solution in this overlapped region (Skamarock et al., 2008; Wang et al., 2018). In-depth details of the procedure are presented in the WRF User Manual. Since the two-way nested approach was used to improve the results of domains 1 and 2 with the high-resolution results of domain 3. Therefore, domains 1 and 2 results have also been tested for coupling with WAsP. Time series wind data from the three domains of 15 km × 15 km, 5 km × 5 km, and 1 km × 1 km are input into WAsP and are analyzed and evaluated in terms of mean wind speed, Weibull A and k parameters and wind power density for 24 AWSs. The evaluation uses the metrics of Bias, %Bias and RMS Bias.

Results

This section presents the results on the analysis and evaluation of the wind resource parameters in WAsP at AWSs for the WRF–WAsP coupling.

Wind speed prediction

Figures 4 to 6 present the Bias analysis of measured wind data used in WAsP with WRF wind data coupled in WAsP (WRF–WAsP). There is an improvement in the overall RMS bias of 1.6% from d01 to d02 and 10.4% from d02 to d03 for wind speed, 1.5% from d01 to d02 and 10.8% from d02 to d03 for Weibull $A$ parameter, 4.9% from d01 to d02 and 6.5% from d02 to d03 for Weibull k parameter. Tables 4 to 6 which presents the evaluation of WAsP-Measured (M) and WAsP–WRF (WRF) using input data from d01 = 15 km × 15 km, d02 = 5 km × 5 km, and d03 = 1 km × 1 km are provided in the supplemental data.

Figure 3.

Mesoscale microscale (WRF-WAsP) coupling strategy.

Figure 4.

Bias in wind speed (WS), A parameter and k parameter of WAsP-Measured (M) and WAsP–WRF(WRF) using input data from d01 = 15 km × 15 km. WAsP: Wind Atlas Analysis and Application Program; WRF: Weather Research and Forecasting.

Figure 5.

Bias in wind speed (WS), A parameter and k parameter of WAsP-Measured (M) and WAsP–WRF (WRF) using input data from d02 = 5 km × 5 km. WAsP: Wind Atlas Analysis and Application Program; WRF: Weather Research and Forecasting.

For few AWSs located in complex topography and/or closer to the coastline, better results (reduction in the Bias) are obtained from the coupling of the medium resolution grids of 15 km × 15 km and 5 km × 5 km in comparison with the higher grid resolution. This is also expected in some cases depending on the features of the topographical representation, as in complex topography and near coastlines the WRF high-resolution grid cell averaged wind data is over a smaller area. An average over a larger area in such cases would provide a better representation of an averaged topography.

Figure 7 presents the Bias analysis of WRF with coupled WRF–WAsP for d03 = 1 km × 1 km. The RMS bias across all 24 AWSs for the wind speed, Weibull A and k parameters and the wind power density are 0.04, 0.05, and 0.04 W/m². Since most of the WRF-modeled wind resource parameters are overestimated in comparison with measurements, a reduction in the WRF–WAsP from WRF data at AWSs will denote an improvement in the predictions by the coupling. There is an improvement in the prediction of wind speed, Weibull A parameter, and Weibull k parameter in the order of 0.5–2.0%, 0.2–2.7%, and 0.7–2.7%, respectively, from the mesoscale (WRF) model to the mesoscale–microscale (WRF–WAsP) coupling as can be seen in the plots. Each of the individual stations show better agreement in the wind resource parameters, indicating that the coupling of WRF–WAsP improves the predictions. Since the improvement in the prediction of the wind resource parameters by the coupling is significant for the high resolution of 1 km × 1 km WRF model data, it can be used into the WAsP model for micrositing wind turbines for potential wind farm sites in a future study.

Figure 6.

Bias in Wind Speed (WS), A parameter and k parameter of WAsP-Measured (M) and WAsP-WRF (WRF) using input data from d03 = 1 km × 1 km.

Figure 8 presents the Wind Power Density Bias analysis for all four scenarios, there is an improvement in the overall RMS bias of 6.8% from d01 to d02 and 17.7% from d02 to d03. For WRF with coupled WRF–WAsP for d03 = 1 km × 1 km, the RMS bias across all 24 AWSs is 1.2 W/m². There is an improvement in the prediction of wind power density in the order of 1.4–6%, from the mesoscale (WRF) model to the mesoscale–microscale (WRF–WAsP) coupling as can be seen in the plots and in Tables 3 to 6 in the Supplemental Data.

Figure 7.

Bias in Wind Speed (WS), A parameter and k parameter of WRF and WAsP-WRF using input data from d03 = 1 km × 1 km.

Figure 8.

Bias in Wind Power Density (P) for all the four scenarios.

Discussion

The higher grid resolution of 1 km × 1 km offered better coupling results between a mesoscale and a microscale model in comparison with the coarser grid resolutions of 15 km × 15 km and 5 km × 5 km. There is a significant improvement observed going from medium resolution to very high resolution in the order of 1.5–17.7% for the parameters evaluated. The coupling of the 1 km × 1 km grid resolution have better results in comparison with the studies reported in the literature (Badger et al., 2010; Bilal et al., 2016; Byrkjedal and Berge, 2008; Carvalho et al., 2013; Durán et al., 2019; Gastion et al., 2008; Mortensen et al., 2006; Silva dos Santos et al., 2016; Tammelin et al., 2013) by 2–6%, even though there are differences in the length of study period (this study uses 10 years), height of wind speed comparison (this study uses 10 m), application methodology (this study uses direct coupling at the same height as in WRF), geographical location (this study is in the tropical South Pacific), number of study validation sites (this study uses 24 AWSs), grid resolution (this study uses 15 km × 15 km, 5 km × 5 km, and 1 km × 1 km) and the mesoscale model (this study uses WRF) in some cases.

The comparison of WRF (d03 = 1 km × 1 km results) with WAsP–WRF (using d03 = 1 km × 1 km into WAsP) showed that there is improvement in the order of 0.2–6% for the wind resource parameters evaluated. The deviations in the form of bias across all stations are used to compute the improvement as it has been previously used in the reported literature (Byrkjedal and Berge, 2008; Carvalho et al., 2013; Gastion et al., 2008; Mortensen et al., 2006; Silva dos Santos et al., 2016; Tammelin et al., 2013) for similar studies. The deviations are between the observed (measured) wind resource parameters and the simulated coupled model wind resource parameters. For statistical significance, RMSE, Bias, STDE, and correlation have already been computed and evaluated in (Dayal, 2021) for measurements against WRF 1 km × 1 km results. The results of which revealed that they are within acceptable limits and the mesoscale model physics is correct (Dayal, 2021). The WRF–WAsP coupling provides opportunities for further improvement should wind measurements be available at higher elevations, where the influence of the local topography and vegetation characteristics are lower.

The proposed WRF–WAsP coupling methodology can also be effectively applied to the other SIDSs of the Southwest Pacific region and the tropics as far as these islands share similar physical, environmental, and atmospheric/meteorological characteristics.

The major contributions of the paper are as follows:

Validated mesoscale–microscale (WRF–WAsP) coupled wind resource assessment methodology for SIDS in the tropics.

Creation of a decade-long (10-years) high-resolution (1 km × 1 km) reanalysis of wind data for the whole of Fiji, covering a spatial area of 401 km × 401 km from the ground level to 1000 m elevation.

The limitations for the mesoscale–microscale (WRF–WAsP) coupling are:

The results are based on the WRF model version 3.9.1.1 for mesoscale modeling and WAsP12.1 for microscale modeling.

The coupling evaluation results are based on the 10-meter elevation WRF and ground-based wind data measurements at 24 AWS locations in Fiji.

The results may improve:

If WRF and ground-based wind data measurements from a higher elevation are used for the evaluation.

Conclusions

This study presents an analysis and comparison of the proposed WRF–WAsP coupling methodology and evaluates the improvement in the accuracy of the wind predictions for the SIDS of Fiji in the tropical South Pacific. The results revealed that the proposed WRF–WAsP coupling methodology can be used as a standard wind resource assessment methodology provided very-high resolution dynamically downscaled wind resource data is available in the order of 10 years for utility-scale wind power applications. The analysis also revealed that the 1 km × 1 km WRF model data from d03 is best for coupling WRF–WAsP for resource mapping and evaluation. The coupling methodology of WRF–WAsP improved the accuracy of the wind prediction by 0.2–6% for the wind resource parameters for Fiji.

As the first mesoscale–microscale coupling exercise for the SIDS of Fiji for the recent decade (2009–2018), our study supports the idea that the SIDS in the Pacific can utilize their wind resource to ease dependence on diesel-based power generation and can even support 100% electricity generation in the case of Fiji from renewable energy sources. It provides a methodology for mesoscale–microscale coupled wind resource assessment for the SIDS in the Southwest Pacific.

For a further study, it is recommended to evaluate the WRF–WAsP coupling methodology by extracting WRF wind data from the top of the boundary layer at 1000 m AGL, where it is assumed that there is no influence by the local terrain on the geostrophic winds and work downwards using the WAsP model.

Supplemental Material

sj-docx-1-eea-10.1177_01445987241237561 - Supplemental material for Evaluation of the mesoscale–microscale (WRF–WAsP) coupling methodology for wind resource parameters in Fiji

Supplemental material, sj-docx-1-eea-10.1177_01445987241237561 for Evaluation of the mesoscale–microscale (WRF–WAsP) coupling methodology for wind resource parameters in Fiji by Kunal K Dayal, John E Cater, Gilles Bellon, Michael J Kingan and Rajnish N Sharma in Energy Exploration & Exploitation

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The authors received no financial support for the research,authorship,and/or publication of this article.

ORCID iD

Kunal K Dayal

Supplemental material

Supplemental material for this article is available online.

References

Al-Yahyai

Charabi

Gastli

(2010) Review of the use of numerical weather prediction (NWP) models for wind energy assessment. Renewable and Sustainable Energy Reviews 14(9): 3192–3198.

Badger

Larsen

Mortensen

, et al. (2010) A universal mesoscale to microscale modelling interface. In: EWEC 2010 Proceedings online. European Wind Energy Association (EWEA).

Bilal

Solbakken

Birkelund

(2016) Wind speed and direction predictions by WRF and WindSim coupling over Nygårdsfjell. Journal of Physics: Conference Series 753: 082018.

Byrkjedal

Berge

(2008) The use of WRF for wind resource mapping in Norway. In: 9th WRF users’ Workshop.

Carvalho

Rocha

Santos

, et al. (2013) Wind resource modelling in complex terrain using different mesoscale-microscale coupling techniques. Applied Energy 108(C): 493–504.

Dayal

(2015) Offshore Wind Resource Assessment, Site Suitability and Technology Selection for Bligh Waters Fiji using WindPRO . MSc Dissertation. Uppsala University.

Dayal

(2021) Mesoscale Wind Resource Mapping of the Small Island States of the Southwest Pacific Region . Ph.D. Thesis. The University of Auckland.

Dayal

Bellon

Cater

, et al. (2021a) High-resolution mesoscale wind-resource assessment of Fiji using the weather research and forecasting (WRF) model. Energy 232: 121047.

Dayal

Cater

Kingan

, et al. (2020a) A grid sensitivity study of the Weather Research and Forecasting (WRF) model for simulating the surface winds for the small island state of Fiji. In: 22AFMC 2020, Brisbane, QLD Australia, 7–10 December 2020.

10.

Dayal

Cater

Kingan

, et al. (2020b) Evaluating of the WRF model for simulating the surface winds and the diurnal cycle of wind speed for the small island state of Fiji. Journal of Physics: Conference Series 1618: 062025.

11.

Dayal

Cater

Kingan

, et al. (2021b) Wind resource assessment and energy potential of selected locations in Fiji. Renewable Energy 172: 219–237.

12.

Dayal

Cater

Kingan

, et al. (2022) An analysis of the 10 MW Butoni wind farm in the Tropical Southwest Pacific Island of Fiji. Wind Engineering 46: 1264–1280.

13.

Durán

Meißner

Rutledge

, et al. (2019) Meso-microscale coupling for wind resource assessment using averaged atmospheric stability conditions. Meteorologische Zeitschrift 28(04): 273–291.

14.

EFL, Energy Fiji Limited (2022) Annual Report. Available at: www.efl.com.fj. Accessed on September 1, 2023.

15.

Gastion

Pascal

Frias

, et al. (2008) Wind resources map of Spain at mesoscale: Methodology and validation. In: European wind energy conference (EWEC).

16.

Government of Fiji (2021) Fiji’s Intended Nationally Determined Contribution. Available at: http://www4.unfccc.int. Accessed on April 7, 2023.

17.

GWEC, Global Wind Energy Council (2022) Global Wind Energy Report. Available at: www.gwec.net. Accessed on July 7, 2023.

18.

Kutty

Khan

MGM

Ahmed

(2019) Wind energy resource assessment for Suva, Fiji, with accurate Weibull parameters. Energy Exploration & Exploitation 37(3): 1009–1038.

19.

Manwell

McGowan

Rogers

(2009) Wind Energy Explained: Theory, Design and Application. Chichester: John Wiley & Sons.

20.

Mortensen

Hansen

Badger

, et al. (2006) Wind Atlas for Egypt: Measurements, micro- and mesoscale modelling. In: Proceedings of the 2006 European Wind Energy Conference and Exhibition, Athens, Greece, February 27 to March 2.

21.

Mortensen

Heathfield

Rathmann

, et al. (2014) Wind Atlas Analysis and Application Program: WAsP 12 Help Facility. Roskilde, Denmark: Department of Wind Energy, Technical University of Denmark.

22.

Nair

Kumar

(2021) Potential of wind energy resources around Nasawana village in Nabouwalu, Vanua Levu, Fiji. Energy Exploration & Exploitation 39(1): 409–425.

23.

National Center for Atmospheric Research (NCAR) Database. Available at: https://www2.mmm.ucar.edu/wrf/users/download/get_sources_wps_geog.html. Accessed on July 7, 2020.

24.

National Centers of Environmental Prediction/National Weather Service/NOAA/U.S. Department of Commerce (2020) NCEP FNL Operational Model Global Tropospheric Analyses, continuing from July 1999. Research Data Archive at the National Center of Atmospheric Research, Computational Information Systems Laboratory, Boulder, CO.

25.

NCEP-NOAA, SST. Accessed on July 7, 2020.

26.

REN21 (2023) Renewables 2022 Global Status Report. Paris: REN21 Secretariat.

27.

Sathyajith

(2006) Wind Energy: Fundamentals, Resource Analysis and Economics. Berlin: Springer-Verlag.

28.

Sharma

Ahmed

(2016) Wind energy resource assessment for the Fiji islands: Kadavu Island and Suva Peninsula. Renewable Energy 89: 168–180.

29.

Silva dos Santos

Santos e Silva

Faro do Amaral Lemos

, et al. (2016) Assessment of wind resources in two parts of Northeast Brazil with the use of numerical models. Meteorological Applications 23: 563–573.

30.

Skamarock

Klemp

Dudhia

, et al. (2008) A Description of the Advanced Research WRF Version 3. NCAR Technical Note, NCAR/TN-475 + STR.

31.

Tammelin

Vihma

Atlaskin

, et al. (2013) Production of the Finnish Wind Atlas. Wind Energy 16: 19–35.

32.

Wang

Bruyere

Duda

, et al. (2018) WRF-ARW User Guide V3.9.

33.

Watson

(2023) Handbook of Wind Resource Assessment. Hoboken, New Jersey: John Wiley & Sons.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.05 MB

0.00 MB