Short-Term Wind Energy Resource Prediction Using Weather Research Forecasting Model for a Location in Indonesia

Renewable energy; Wind forecasting; Wind power density; Windrose model

Indonesia has a 23% renewable energy target in its total energy mix by 2025 (as stated in the National Electricity General Plan or RUKN), reducing greenhouse gas emissions by 29-41% by 2030 and achieving Net-Zero emissions by 2060. In line with those, several studies on renewable energy development in reducing the greenhouse gas effect have been conducted, especially from the potential view. They are estimating not only national coverage, such as hydro (Pranoto et al., 2021), wind (Hesty et al., 2021), and solar (Wahyuono and Julian, 2018) but also provincial level and specific sites (Syahputra and Soesanti, 2021). Moreover, a web-based application has been developed to calculate the energy potential of a rooftop solar PV system installed in a home (Nurliyanti et al., 2021). 
      Wind power is a promising renewable energy to achieve the target because of its high efficiency and low pollution. The Ministry of Mineral and Energy Resources (MMER) of Indonesia states that Indonesia has a wind energy potential of 154.88 GW, consisting of an onshore potential of 60.65 GW and an offshore potential of 94.2 GW. According to Indonesia's wind resource assessment by Hesty et al. (2022), onshore locations on the south coasts of Java, South Sulawesi, Maluku, and NTT have high wind energy potential wind speeds of 6 to 8 m/s, power densities of 400 to 500 watt/m², and Annual Energy Production (AEP) of 4-5 GWh/year. In addition, wind energy has a large potential to be explored in the urban area, whether using Horizontal and Vertical Axis wind turbines (Krasniqi, Dimitrieska, and Lajqi 2022), and improves performance and efficiency, simplicity, and reliability of construction of wind turbines using a permanent magnet (PMGs) (Nur and Siregar, 2020). 
        However, the high reliance on seasonal variations, which causes a huge primary power generation fluctuation on a daily and annual timescale, is a significant obstacle for a 100% renewable energy source (Guenther, 2018). Atmospheric conditions and wind speed strongly influence the power generated by wind energy conversion systems (Chang, 2013a, Chang, 2013b). So unexpected fluctuations can increase system operating costs for primary backup requirements and pose a potential risk to the reliability of the power supply (Sideratos and Hatziargyriou, 2007). Network operators must overcome the challenges of intermittent wind conditions to schedule spare capacity, stability, planning, and the power system's reliability (Soman et al., 2010). Precise short-term wind speed forecasts are essential to reduce the risk of intermittent wind and allow for more penetration (Peng et al., 2016).
        Some wind power forecasting methods for approaching wind energy forecasting include statistical models, Artificial Intelligence (AI) models, and physical models. The autoregressive (AR), autoregressive moving average (ARMA), autoregressive integrated moving average (ARIMA), Bayesian approach, and gray forecasts are all statistical methods. Lopez et al. (2019) show that the seasonal ARIMA model is a fast, precise, straightforward, and adaptable load forecasting method. Artificial neural networks (ANN), fuzzy logic approaches, adaptive neuro-fuzzy inference systems (ANFIS), neuro-fuzzy networks, support vector machines (SVM), and evolutionary optimization algorithms are some of the AI methods for wind forecasting. Temperature, pressure, solar radiation, and altitude were used as inputs to the ANN by Ramasamy, Chandel, and Yadav (2015) to estimate wind speed in 11 sites in India's mountainous region. Neuro-fuzzy network method for short-term wind power forecasting was applied to the wind power forecasting of a real wind farm located in China by Xia, Zhao, and Dai (2010). The physical method is based on numerical weather prediction (NWP) using weather forecast data for large-scale area weather prediction. Tan et al. (2021) evaluated the efficacy of the weather research forecasting (WRF) model in predicting wind speed and direction up to 72 hours in advance in the western portion of Turkey. Except for low wind speeds, the model can accurately reproduce wind directions.
     Several institutions in Indonesia have issued a weather prediction system for an early warning system and education. The Meteorology, Climatology and Geophysics Agency (BMKG), the agency appointed by the Government of Indonesia to provide information and forecasts related to weather, climate, and natural disasters, that the public can access to find out about weather predictions for the next seven days. The Center for Atmospheric Science and Technology (PSTA) developed a weather prediction system for the next three days with a resolution of 5 km (Suaydhi, 2016). Meteorological Analysis Laboratory, Bandung Institute of Technology (ITB), developed a weather prediction for the next three days with a resolution of 27 km (Junnaedhi, 2017). However, there is no weather prediction system for energy purposes, especially wind energy. Therefore, the proposed short-term wind energy forecasting represents a significant scientific contribution to Indonesia's reliable large-scale wind power integration.

2.1. Data

The model relies on data from the Global Forecast System (GFS). The Global Forecast System (GFS) provides data for NOAA's (National Oceanic and Atmospheric Administration) prediction models. Global GFS data is often used as a reference for regional models or even used directly for regional predictions because of its accuracy. This input data has a resolution of 0.25 ° for the world region and has four cycles: 00, 06, 12, and 18. In this study, cycle 00 is used.

2.2.  Model Setup

The NWP model used in this study is the WRF model, a fully compressible, non-hydrostatic algorithm. To more accurately replicate airflow across difficult terrain, it uses sigma pressure in the vertical direction. The model solves the governing equations in ?ux-form, which enables the conservation of mass and scalar quantities.

The model has a single primary domain that spans the entire Indonesian territory between latitudes 7° N and 11° S and longitudes 94° E and 144° E. Using initial data from GFS, the model simulation was run for 72 hours forecast lead time, increasing its input model resolution to 5 km spatial resolution over 35 vertical pressure levels with a temporal resolution of 1 hour. The spatial resolution of 5 km is expected to be good enough for reviewing detailed weather patterns according to local conditions such as topography and coastline.

 The parameterization method configuration significantly impacts the near-surface wind field in the WRF model, particularly for complex terrain. Consideration should be given to parameterization schemes like the Surface, Land Surface (LS), and Planetary Boundary Layer (PBL) schemes that can capture the interaction between the land surface and the wind field. We used the Noah land surface model in WRF because it integrates prescribed data and dynamic modeling to simulate the surface. It also provides the user with multiple options to simulate land surface interactions (Niu et al., 2011). Land surface models and initialization datasets impacted the WRF's ability to predict accurately. The surface layer approach used in this study to compute turbulent surface fluxes is based on the Monin Obukhov similarity theory (Van et al., 2017). More details regarding the con?guration of the WRF parameter scheme are shown in Table 1.

Table 1 WRF Model Configuration and Parameterization

2.3. Model Verification

        The quantitative analysis of wind data was carried out by finding the correlation coefficient (r) and Mean Square Error (RMSE) using Equation 1 and Equation 2.

The RMSE value measures the error generated between the model data and the observations. Therefore, this RMSE value can describe accuracy; the smaller the RMSE, the better the level of accuracy. The observation data used to verify the WRF model comes from a wind measuring tower owned by Pondera/PT Hywind Energy Solution in Kadumbul Village, Pandawai District, East Sumba, with latitude coordinates 09°41'42.7" South Latitude and Longitude 120°31'55.5" East Longitude. The measuring tower is equipped with two arrangements of anemometers at various heights. Two A-B anemometers are placed at 40 and 80 m in height. At the same time, a single anemometer is placed at 60,  97, and 102 m of height. In addition, there is a wind vane installed at an altitude of 60 m and 97 m.

Figure 1(a) shows the wind measurement tower and (b) the orthomosaic map location. Pandawai District is a hilly area with the highest altitude of 255 m above sea level. In the southern part of the district is a coastal area directly adjacent to the sea. For the slope class, the Pandawai District area is dominated by the 0–8% (flat) slope class. The location where observation tower is located in a natural grassland, which is included in the less productive dry land with an elevation of 30 – 39 m above sea level.

The observation data used to verify the prediction model is 29-31 August 2021 for 10 minutes. Verification using data from August is necessary since the monsoonal type over Indonesia was identified by the flow of wind circulation that blows continuously for one particular period and in the other direction with transitional intervals in between.

Figure 1 (a) Wind measurement tower; (b) orthomosaic map; and (c) Topography of the measurement tower location

The dry season, which reaches its maximum in August, is caused by the south-easterly wind that blows from the Australian Continent to the equator from around June to August. Indonesia often has higher wind speeds during this June-July-August (JJA) month (Abdillah et al., 2022).

Figure 2 Wind measurement tower observation data on 29–31 August 2021; (a) time series; (b) mean diurnal profile; and (c) Frequency histogram

Figure 2a shows the observation data of the wind gauge tower on 29 – 31 August 2021. The average wind speed at an altitude of 40 m, 60 m, 80 m, and 97 m - 102 m are 5.9 m/s, 6.4 m/s, 6.8 m/s, and 7.1 m/s, respectively. Figure 2 b shows a diurnal profile showing the daily variation of wind speed at five altitudes; 40 m, 60 m, 80 m, 97 m, and 102 m. At an altitude of 102 m, the daily wind speed is evenly distributed throughout the day, with wind speeds between 6.29 - 8.08 m/s. Meanwhile, at an altitude of 40 m, the daily wind speed is between 4.20 - 7.98 m/s. Maximum wind speed occurs during the day at 11 AM – 2 PM. The frequency histogram of wind speed at the height of 102 m can be seen in Figure 2 c. Wind speed distribution is concentrated at low speeds, and the duration of days with high wind speeds is 7 m/s, as much as 21%.

Figures 3 - 4 show the output of the WRF model in the form of predictions of Indonesia's wind speed and direction on 29 - 31 August 2021 at 08 and 09 UTC at four altitude levels. Very few locations on land experience wind speed above 6.0 m/s..