Improving ASI-based solar radiation nowcasting by using automatic cloud type recognition methods

In the last years there is an increasing interest for developing enhanced solar nowcasting methods, fostered by the massive deployment of small-scale / residential PV systems. Reference nowcasting methods are based on the use of All-Sky-Imagers (ASI). Although ASI-based nowcasting methods has received a notable attention in the last decade, their reliability is relatively low, because the processes involved in deriving the nowcasts are prone to many uncertainties. In this work, we propose and evaluate new methods aimed at increasing this reliability. Notably, we propose the use of an automatic cloud-type recognition system in the nowcasting procedure, allowing the use of cloud-specific transmittance values for the different cloud types. The study was carried out at a location of southern Spain, using a total of 1901 samples representing all the cloud types. Each sample is composed by one-minute time resolution ASI images, GHI and DNI measurements as well as cloud base height values derived from a ceilometer. Up to 30 minutes ahead one-minute time resolution forecasts were obtained and benchmarked against reference methods that not uses cloud-specific transmittances.

In the first part of the study, a statistical analysis was conducted to determine the GHI and DNI transmittance of 11 different cloud types; to this end, a Gaussian Mixture Model (GMM) was used. For some of the cloud types, the cloud base height was also used as a parameter of the model. In a second part of the study, these cloud-specific transmittances were incorporated in the nowcasting procedure using an operational automatic cloud type recognition method. This new nowcasting method was evaluated as an operational forecasting procedure.

Results of the first study reveals, firstly, that DNI total transmittances are almost negligible for cumuliform clouds. In contrast, stratiform clouds show a wide range of transmittances. Notably, while stratus and altostratus clouds behaves as cumuliform clouds, transmittances for cirrostratus and cirrus are, respectively, 30% and 80%, approximately. For the GHI, the total transmittance values ranges from 80% (cirrus) to 30% (stratus).

Results from the second study show that the proposed method provide slightly enhanced GHI nowcasts associated with cirrus, altocumulus and cirrocumulus clouds skies. For these cloud-types skies, the reduction achieved in the rRMSE values ranges from 2% to 6%. On the other hand, the proposed method provide a clear superior performance for the DNI nowcasting, with an overall reduction of rRMSE values of around 10% for the entire dataset. The reduction in rRMSE values also show a dependence of the cloud types, ranging from 6% (cirrostratus/cirrocumulus) to 20% (cirrus clouds).