① Use ground measurement to calibrate satellite bias
Although the satellite data of a certain location can only guarantee that the average irradiance deviation during the day is between ±3% and 6% (depending on the climatic and topographical conditions), the recording strength of the satellite inversion model proves that the specified location is recorded The ability to compare inter-annual changes. That is to say, due to the complex terrain features and low resolution of aerosol data, there may be deviations in the model for a specific location, and this deviation will last for a long time. Therefore, if the short-term measurement activities of the satellite model can be calibrated, its long-term accuracy, that is, the ability to predict irradiance before and after measurement, should be greatly improved. In general, 6~12mo calibration reduces the length of the MBE confidence interval by half. This calibration process is often referred to as position adaptation or measurement-correlation-prediction.
By reducing the KSI error measurement, the complexity of the position adaptation method ranges from simple error correction (for example, all predicted data points use the same calibration coefficient) to more complex and usually more effective techniques (including matching measurement frequency distribution and construction Mode frequency distribution). Accordingly, according to the values of different correction coefficients, they are used in the model. In dry areas with sparse cloud cover, MBE is usually caused by aerosol parameterization problems; therefore, the method of adapting aerosols to local conditions may be very effective (for example, Figure 1).
Both the SUNY/SolarAnywhere method and the SolarGIS method are used in applications for calculating historical data. At the same time, these two methods are also playing an increasingly important role in short-term forecasting and forecasting applications. For example, the following data applications and improvements of SolarGIS will further improve the accuracy of GHI and DNI.
Use high-resolution satellite data (under-satellite point is 3km, update frequency is 15min and 30min).
●Customized Cl detection suitable for different geographical conditions based on multi-spectrum and multi-source albedo analysis; more complex measurement situations, such as snow, fog, ice, etc.; more careful handling of variable ground reflectance modes.
●Using MACC aerosol data with daily resolution and CFSR/GFS water vapor data to describe changes in the atmosphere to complete the clear sky model.
●Using the high-resolution elevation data based on the 90m digital elevation model SRTM-3 to calculate the terrain occlusion of direct radiation and scattered radiation.
Further reduce the uncertainty in the following two future development priorities:
●Improve the spatial distribution of aerosol data and water vapor data, thereby reducing deviations, in order to better represent the partial clear sky solar radiation map.
●Be more proficient in the use of multi-spectral channels and multivariate statistical analysis, and improve cloud attenuation by improving Cl quantification results to reduce deviation and RMSE.
It is possible to obtain satellite data with higher spatial resolution (1km) and time resolution (update frequency up to 5min), which will help reduce deviation and RMSE at the same time. However, using these data requires changes to all algorithms. The increased amount of data processing may reduce the computational efficiency in areas where solar energy systems are highly concentrated.