Solar radiation forecasting with wrf model in the iberian peninsula
Uncertainty for solar products assessment and benchmarking
1. Uncertainty for solar products assessment and
benchmarking
J. Polo, L. Ramírez, L.F.Zarzalejo, L. Martín, A. Navarro
CIEMAT (Energy department – Solar Platform of Almería)
4th Meeting IEA SHC Task 36 Hamburg 23-25 Oct 2007
2. Uncertainty parameters
Parameters based on deviation of data values (careful with notation)
n
( yi − g i )
MRE = ∑ × 100
Mean Relative Error (MRE) n
gi
∑ ( gi − yi ) / n i =1
i =1
Mean Bias Error (MBE) MBE = n
× 100
∑ gi / n
n
i =1
∑ ( gi − yi )2 / n
RMSE RMSE = i =1
n
× 100
∑ gi / n
i =1
Parameters based on deviation of distribution functions
KSI and OVER (Integral of KS test complete and
over critical value)
KSE KSE = ( KSI × w1 + OVER × w2) / 2
RIO = ( RMSE + KSE ) / 2
RIO
4th Meeting IEA SHC Task 36 Hamburg 23-25 Oct 2007
7. Towards standardization: open issues for
discussion
Solar Radiation Product uncertainty: users require one number
(Radiation ± U) . Candidates: RMSE, MBE, relative error…
Problems with normalization.
Model assessment: we look for more information than uncertainty.
strengths and shortcomings of models is also required.
Candidates: K-S Test in addition to uncertainty measures MBE,
RMSE, deviations at different solar elevation angles, … are useful
for this purpose.
Benchmarking of models: We should know a priori the capabilities
of different models and we want to compare their response under
the same conditions.
Candidates: RIO parameter compiles KS test and RMSE
information in one single parameter.
4th Meeting IEA SHC Task 36 Hamburg 23-25 Oct 2007
8. Future activity
Benchmarking exercise on one selected pixel for
one year of hourly global irradiation?
Elaboration of a guide for uncertainty (MESoR)?
4th Meeting IEA SHC Task 36 Hamburg 23-25 Oct 2007