Multi-model MOS for Solar Power Forecasting

Solar MOS (SMOS), developed by Meteogroup, is a prediction system for forecasting solar radiation and/or generated power. The statistical technique employed by SMOS is called multiple linear regression analysis. Linear combinations of raw NWP model output (also known as direct model output or DMO) are fitted against an historic set of measurements of the quantity of interest. The DMO parameters that explain the variance in the observed predict and at best, yielding the lowest mean absolute error, are used as components of the regression equations. SMOS has recently been extended with a number of additional predictors.

Adding predictors from UKMO and GFS shows an improvement in terms of mean (absolute) error compared to predictors from ECMWF only. Direct model output (DMO) from ECMWF, UKMO and GFS has also been used to derive radiation related parameters like diffuse radiation, direct radiation and radiation on tilted planes. The impact of these derived DMO predictors has not been quantified but they are believed to be important as they appear in the SMOS equations many times.

Furthermore, it has been investigated if satellite observations could lead to a beneficial effect on the very short forecast range. Adding EUMETSAT cloud mask observations as additional predictors has shown to improve the forecast even further in the short-range up to ~4-5 hours into the future.