Validation of a cloud detection algorithm with an All-Sky Imager
Summary
With the increase in renewables in the future power mix, fluctuations in the power production become
larger due to the intermittency of these energy sources. Accurate forecasting of solar energy
production can help to improve unit-commitment decisions and reduce ancillary costs. Various
methods can be used for solar forecasting. Most methods, e.g. satellite techniques, often overlook
small and thin clouds. The high temporal as well as spatial resolution of All Sky Imagers enables the
opportunity to take these small and thin clouds in to account, even at fast changing cloud conditions.
Recent efforts at EKO Instruments let to the development of a new cloud detecting algorithm called
TRINITY. This study aims at comparing the performance of the new TRINITY algorithm in combination
with an All Sky Imager to the existing BRBG and CDOC cloud detecting algorithms. The new algorithm
is validated using two approaches. The first method uses shortwave irradiance by determining the
clearness index and diffuse fraction as proxies for Cloud Cover Fraction. The other method calculated
the Cloud Cover Fraction by using downward longwave irradiation. Data is provided by two cases
studies, where data is collected in Utrecht (NL) and Denver (US). Results of the shortwave irradiance
method show that lowest errors where achieved by using the diffuse fraction as a proxy. Overall, the
mean absolute error of the new TRINITY algorithm was 12%, whereas the BRBG and CDOC algorithms
had errors of 17% and 14%, respectively. When differentiating for different sky conditions the TRINITY
algorithm outperforms BRBG and CDOC at clear sky conditions, whereas in overcast conditions it
outperforms the BRBG algorithm. Furthermore, the unreliable sunrise and sunset periods affect the
accuracy of the algorithms and radiation measurements. Excluding the sunrise and sunset improves
the accuracies with 11%, 15% and 2% for the BRBG, CDOC and TRINITY, respectively. Testing the effect
of the solar position on the performance of the algorithms showed that the BRBG algorithm is most
sensitive to low elevation angles, leading to higher errors. The TRINITY algorithm achieved similar
performance for all elevation angles and is more stable than the other algorithms. For elevation angles
of 35° and higher, all algorithms perform similarly. Preliminary results for using longwave downward
radiation show that the accuracies of all algorithms are comparable (53%, 54% and 56% for BRBG,
CDOC and TRINITY, respectively) with lowest errors for the BRBG algorithm. Overall, TRINITY is found
to perform best followed by the CDOC and BRBG algorithm. Accurate cloud detection by All Sky
Imagers will improve the accuracy of short-term solar forecasting.