Very Short-Term Surface Solar Irradiance Forecasting Based On FengYun-4 Geostationary Satellite

Liwei Yang, Xiaoqing Gao, Jiajia Hua, Pingping Wu, Zhenchao Li, Dongyu Jia2020 · DOI

TL;DR

This paper introduces a new algorithm for forecasting surface solar irradiance (both global horizontal and direct normal) for up to 3 hours ahead, using data from China's FengYun-4A geostationary satellite. The method leverages cloud motion vectors derived from satellite imagery and outperforms a simple persistence model, particularly for global horizontal irradiance.

What Problem Does This Paper Address?

The increasing reliance on solar energy for power generation, especially photovoltaic (PV) systems, is crucial for sustainable development. However, PV output is highly sensitive to weather, particularly cloud cover, leading to significant fluctuations that can destabilize power grids. Accurate short-term forecasting of solar irradiance is essential for effective power dispatching, grid management, and reducing operational costs. While various forecasting methods exist, very short-term (0-3 hour) forecasting, especially using readily available satellite data, remains an active area of research. Existing methods often require complex calculations or extensive ground-based data, limiting their applicability in some regions. This paper aims to address this by developing a computationally efficient forecasting method using a new generation of Chinese geostationary satellite data.

Methods

The core of the proposed algorithm involves several key steps:

  • Clear Sky Model: The European Solar Radiation Atlas (ESRA) model is used to estimate solar irradiance under clear sky conditions. This model requires inputs like aerosol optical depth (AOD) and atmospheric water vapor, which were obtained from monthly averaged data and ground-based measurements, respectively.
  • Cloud Index Calculation: The Heliosat-2 method is employed to convert satellite imagery into a "cloud index" (n⁺). This index quantifies the attenuation of solar radiation due to clouds.
  • Cloud Motion Vector (CMV) Estimation: Particle Image Velocimetry (PIV), implemented using a MATLAB toolbox, is used to track cloud movement between consecutive satellite images. This allows for the estimation of cloud motion vectors (CMVs).
  • Forecasting: The algorithm forecasts the cloud index for future time horizons (30-180 minutes) by advecting the current cloud index field using the estimated CMVs. The surface solar irradiance (GHI and DNI) is then calculated by multiplying the clear sky irradiance with the forecasted cloud index, adjusted by a clear sky index (Kc).

A significant aspect of this method is its reliance on satellite data and PIV for cloud motion, reducing the need for extensive ground-based meteorological variables or complex numerical weather prediction models.

Datasets

The study utilizes the following datasets:

  • FengYun-4A (FY-4A) Satellite Images: Visible (0.63–0.69 µm) and near-infrared (0.76–0.9 µm) channels from the Advanced Geosynchronous Radiation Imager (AGRI) onboard FY-4A were used for cloud observation and CMV calculation. Images were captured from August 2018 to September 2019 with a 2x2 km spatial resolution.
  • Ground-Based Observations: Solar irradiance data (GHI, DNI, and DHI) were collected from the Weichang national basic meteorological station in Chengde, China. This data was recorded at a 1-minute time resolution and used for validation. Data from four typical months (October 2018, January, April, and July 2019) were used to represent different seasons.

Key Insights and Ideas

  • FY-4A Satellite Data is Viable for Forecasting: The study demonstrates that data from the relatively new FY-4A satellite can be effectively used for short-term solar irradiance forecasting, opening up possibilities for utilizing this resource.
  • CMV-Based Forecasting is Promising for Very Short-Term Horizons: The use of PIV to derive cloud motion vectors and then advect cloud properties is shown to be a robust approach for very short-term (0-3 hour) forecasting, outperforming simpler methods.
  • GHI is More Predictable than DNI: The results consistently show that Global Horizontal Irradiance (GHI) is more accurately forecasted than Direct Normal Irradiance (DNI). This is attributed to DNI's higher sensitivity to rapid changes in cloud cover, atmospheric aerosols, and other microphysical properties.
  • Seasonal Variations Impact Forecast Accuracy: Forecasting accuracy varies significantly by season, with July and April proving more challenging than October and January. This is linked to more complex cloud types and higher aerosol concentrations during warmer months.
  • Computational Efficiency: The proposed algorithm is highlighted for its lower computational cost compared to methods requiring extensive meteorological variables or high-performance computing, making it more suitable for industrial applications.

Key Results

  • GHI Forecasting: The proposed FY-4A satellite-based algorithm consistently outperformed the smart persistence (SP) model for all forecasting horizons and seasons. The best performance was observed in October, with a skill score (SS) exceeding 20%. Annually, the normalized root mean square error (nRMSE) for GHI ranged from 26.78% to 36.84%, and the skill score reached a maximum of 20.44% at the 30-minute horizon, remaining positive for all horizons.
  • DNI Forecasting: The model also outperformed the SP model for DNI in July and October, with skill scores around 11-12%. However, performance was less consistent in January and April, with negative skill scores in January. The maximum skill score for DNI was 6.62% at the 180-minute horizon.
  • Overall: The algorithm is considered more accurate and reliable for GHI forecasting and slightly better for DNI forecasting compared to the SP model.

Limitations and Future Work

  • DNI Complexity: The paper acknowledges that DNI forecasting is inherently more challenging due to its high sensitivity to rapidly changing atmospheric conditions, cloud properties, and aerosol content. The current model's performance for DNI is less satisfactory than for GHI.
  • Aerosol Optical Depth (AOD) Improvement: The accuracy of the clear sky model (ESRA) is crucial. The authors plan to improve the calculation of aerosol optical thickness (AOD) using FY-4A satellite observations to enhance the clear sky model's accuracy.
  • Regional and Climatic Variations: The current study was conducted in a specific region (Chengde, China). The authors intend to validate the model in other regions with different meteorological conditions in future work.
  • Hazy Weather Forecasting: The paper suggests that forecasting surface solar irradiance under hazy weather conditions is a relevant future application, especially for the Beijing-Tianjin-Hebei region, which experiences frequent haze events.
  • Further Model Improvement: The authors note that specific technologies for DNI prediction are still developing, and their DNI forecasting is currently derived from the GHI model. Future work will focus on improving DNI prediction accuracy.

Summary updated

BibTeX
@misc{yang2020very,
  author = {Liwei Yang and Xiaoqing Gao and Jiajia Hua and Pingping Wu and Zhenchao Li and Dongyu Jia},
  title = {{Very Short-Term Surface Solar Irradiance Forecasting Based On FengYun-4 Geostationary Satellite}},
  year = {2020},
  doi = {10.3390/s20092606},
}
Figure