At the Plataforma Solar de Almería, a novel shadow camera system is installed. Figure 1 visualizes the working principle of the shadow camera system: The system uses the inputs of six downward facing cameras placed on an 87 m high tower (CIEMAT CESA-I, Fig. 1, top left). The cameras take images of the ground every 15 s, which are then combined into an undistorted orthoimage (Fig. 1, top middle and right). The orthoimages have a spatial resolution of 5 m × 5 m and image an area of 4 km2. By comparing the current orthoimage with two reference orthoimages, taken when no shadow fell on the imaged area (“sunny reference”) and taken when the whole area was shaded (“shaded reference”), shadows are segmented. For unshaded areas in the current orthoimage, clear sky irradiance values are taken as modelled from ground measurements. The irradiances (DNI – Direct Normal Irradiance, GHI – Global Horizontal Irradiance, GTI – Global Tilted Irradiance) for the shaded areas are derived from pixel intensities of the current orthoimage relative to normalized pixel intensities of the reference orthoimages. The shadow camera system is presented and validated in Kuhn et al. (2017a). Comparing pixels of the irradiance maps to corresponding ground measurements for one-minute temporal averages, the shadow camera system shows deviations of RMSE (DNI) between 4.2 and 16.7 % and RMSE (GHI) deviations below 10 %.

Using the spatially resolved irradiance maps provided by the shadow camera system, nowcasted irradiance maps generated by all-sky imager based nowcasting systems can be validated with special focus on spatial aggregation effects. Figure 2 depicts the mean absolute error (MAE) for GHI and DNI of the WobaS-4cam nowcasting system. The WobaS-4cam system uses the inputs of four all-sky imagers and is described in (Nouri et al., 2017). The irradiance maps produced by WobaS-4cam for lead times between 0 and 15 min are compared to the reference irradiance maps of the shadow camera system for field sizes between 5 m × 5 m (one pixel) and 4 km2. Unsurprisingly, longer lead times lead to larger deviations (Kuhn et al., 2017c). However, these deviations shrink significantly if spatial aggregation effects are considered. Industrial photovoltaic (PV) plants cover areas of several km2 and the single most important parameter to predict is the average irradiance over the whole plant. Thus, these spatial aggregation effects are inherently present and play a major role for the validation of nowcasting systems (Kuhn et al., 2017d).

As shown in Sect. 2.1, shadow camera systems generate highly spatially resolved shadow and irradiance maps. By tracking cloud shadows, e.g. with the differential approach introduced in Kuhn et al. (2017e, f), future shadow positions can be estimated. The shadow camera system used for the validation presented in the previous section is located on an 87 m high tower and can thus only image a relatively small area of 4 km2. In this section, the potential application of such a system for the Andasol solar power plants (Andasol 1–3, 50 MWel each) is discussed.

If nowcasts with lead times up to 10 min are required and if cloud speeds up to 60 km h-1 (16.7 m s-1) are considered, the required imaged area around the plants must be at least 10 km in every direction. A maximum considered speed of 16.7 m s-1 is reasonable as the mean cloud speed in this region is 7.36 m s-1 and the median speed is 6.67 m s-1 (Kuhn et al., 2017e). Under otherwise unchanged conditions, greater lead times can be achieved by using more and distributed cameras.

In Fig. 3, the topographical situation for the Andasol plants is outlined: Two shadow cameras could potentially be located on a mountain approximately 20 km away from and 2 km above the plants. The maximum required range of vision is thus 30 km, which is realistic in this region (Hanrieder et al., 2015). We consider using 6 MegaPixel cameras with a viewing angle of 30∘. This results in an area element of 8 m × 17 m for the pixel the farthest away. If higher spatial resolutions are needed, cameras with higher resolutions or more cameras with a smaller field of view could be used.