Cloud-cover information needs to be extracted from sky images and incorporated in the forecasting model. The image processing—designated as wind – ladder sector—comprises the following steps:
Step 1. The image is converted from a spherical to a rectangular grid.
Step 2. Pairs of images are used in a particle-image velocimetry (PIV) algorithm that determines the apparent velocity field for the cloud motion.
Step 3. A representative velocity vector is chosen by applying k-means clustering to the distribution of velocity vectors.
Step 4. Each pixel in the image is classified as cloud or clear-sky.
Step 5. Cloud fractions Xi (with i increasing with distance from the Sun) are computed for a set of grid elements (the wind ladder) oriented column-wise in the reverse direction of the cloud field (as defined by the representative velocity vector from step 3).
The cloud fractions computed in the last step are then suitable for use as input variables for the DNI-forecasting algorithm, as they encode information about upcoming Sun-obscuring cloud conditions.
The output of some of the procedures is presented in Figure 15.11. More details about image processing can be found in Marquez and Coimbra (Marquez & Coimbra, 2013).