Development of Climate Services for Renewable Energy: Statistical Post-processing of Solar Radiation Seasonal Forecast Over the Indonesian Region

Indonesia is one of the tropical regions with strong solar radiation exposure throughout the year, and this indicates the large potential for solar energy utilization in the country. Nevertheless, the utilization of solar energy in Indonesia until 2020 had only reached 10 MWp, as reported by the Ministry of Energy and Mineral Resources (ESDM), which is very small compared to the total potential of solar energy in Indonesia (approximately 112,000 GWp). One of the challenges for the development of solar energy in Indonesia is the weather and climate factors, as several weather parameters can cause intermittency in solar energy input in this region.

In the solar energy sector, a reliable forecast of potential energy input is of great importance in designing operational plans, whether it is a short-term, annual, or longer-term work plan. Global horizontal irradiance is an important quantity to determine the power generated from photovoltaic devices, and different resources are used to generate global radiation forecasts all over the world, ranging from ground-observed radiation, remote sensing observation, to numerical weather models. The European Centre for Medium-Range Weather Forecasts (ECMWF) provides solar radiation forecasts for various timescales, from hourly forecast to monthly and seasonal forecast. Whilst short-term solar radiation forecast is provided by other standard weather forecasting models, forecasts in the longer timescale are less commonly available and thus the seasonal forecast becomes a valuable information in making long-term operational plans.

A new solar radiation observation network has been installed in a number of locations across Indonesia in recent years, which allows the evaluation and modification of the seasonal forecast generated by the model. To improve the performance of the forecast, a statistical post-processing approach is implemented, by making use of measurements provided by the radiation observation network and ERA5 reanalysis dataset. To generate historical solar radiation data in all parts of Indonesia, a co-kriging interpolation of the ground-observed solar radiation is executed, using reanalysis data as an external drift in the interpolation process. The new gridded solar radiation data is then utilized to create transfer functions that represent the relationship between the statistical moments of both the numerical model output and observed radiation based on its probabilistic distributions. The transfer functions are generated in the training period, which will then be used to modify the model output in the forecast period. The implementation of the bias-correction process applied in this explorative study is aimed to provide the foundation of solar radiation prediction information that will support the operational activities of solar energy production in Indonesia.