Semantic segmentation of ground-based all-sky images (ASIs) can provide high-resolution cloud coverage information of distinct cloud types, applicable for meteorology-, climatology- and solar-energy-related applications. Since the shape and appearance of clouds is variable, and there is high similarity between cloud types, a clear classification is difficult. Therefore, most state-of-the-art methods focus on the distinction between cloudy and cloud-free pixels without taking into account the cloud type. On the other hand, cloud classification is typically determined separately at the image level, neglecting the cloud's position and only considering the prevailing cloud type. Deep neural networks have proven to be very effective and robust for segmentation tasks; however they require large training datasets to learn complex visual features. In this work, we present a self-supervised learning approach to exploit many more data than in purely supervised training and thus increase the model's performance. In the first step, we use about 300 000 ASIs in two different pretext tasks for pretraining. One of them pursues an image reconstruction approach. The other one is based on the DeepCluster model, an iterative procedure of clustering and classifying the neural network output. In the second step, our model is fine-tuned on a small labeled dataset of 770 ASIs, of which 616 are used for training and 154 for validation. For each of them, a ground truth mask was created that classifies each pixel into clear sky or a low-layer, mid-layer or high-layer cloud. To analyze the effectiveness of self-supervised pretraining, we compare our approach to randomly initialized and pretrained ImageNet weights using the same training and validation sets. Achieving 85.8 % pixel accuracy on average, our best self-supervised model outperforms the conventional approaches of random (78.3 %) and pretrained ImageNet initialization (82.1 %). The benefits become even more evident when regarding precision, recall and intersection over union (IoU) of the respective cloud classes, where the improvement is between 5 and 20 percentage points. Furthermore, we compare the performance of our best model with regards to binary segmentation with a clear-sky library (CSL) from the literature. Our model outperforms the CSL by over 7 percentage points, reaching a pixel accuracy of 95 %.

Accurate nowcasts of the direct normal irradiance (DNI) for the next 15 min ahead can enhance the overall efficiency of concentrating solar power (CSP) plants. Such predictions can be derived from ground based all sky imagers (ASI). The main challenge for existing ASI based nowcasting systems is to provide spatially distributed solar irradiance information for the near future, which considers clouds with varying optical properties distributed over multiple heights. In this work, a novel object oriented approach with four spatially distributed ASIs is presented. One major novelty of the system is the application of an individual 3D model of each detected cloud as a cloud object with distinct attributes (height, position, surface area, volume, transmittance, motion vector etc.). Frequent but complex multilayer cloud movements are taken into account by tracking each cloud object separately. An extended validation period at the Plataforma Solar de Almería (PSA) on 30 days showing diverse weather conditions resulted in an average relative mean absolute error (relMAE) of around 15 % for a medium lead time of 7.5 minutes and a temporal average of 15 minutes. Further reductions of the relMAE were achieved by spatial aggregation, with a relMAE of 10.7 % for a lead time of 7.5 minutes, a field size of 4 km² and a temporal average of 1 minute (during one day). Nowcasting systems described in the literature reach similar deviations but were often validated only for a few days based on a single ground measurement station, which confirms the good performance and the high applicability of the presented system. Three implementations of the system exist already demonstrating the market maturity of the system.

The share of distributed solar power generation is continuously growing. This increase, combined with the intermittent nature of the solar resource creates new challenges for all relevant stakeholders, from generation to transmission and demand. Insufficient consideration of intra‐minute and intra‐hour variabilities might lead to grid instabilities. Therefore, the relevance of nowcasts (shortest‐term forecasts) is steadily increasing. Nowcasts are suitable for fine‐grained control applications to operate solar power plants in a grid‐friendly way and to secure stable operations of electrical grids. In space and time, highly resolved nowcasts can be obtained by all sky imager (ASI) systems. ASI systems create hemispherical sky images. The associated software analyzes the sky conditions and derives solar irradiance nowcasts. Accuracy is the decisive factor for the effective use of nowcasts. Therefore, the goal of this work is to increase the nowcast accuracy by combining ASI nowcasts and persistence nowcasts, which persist with the prevailing irradiance conditions, while maintaining the spatial coverage and resolution obtained by the ASI system. This hybrid approach combines the strengths while reducing the respective weaknesses of both approaches. Results of a validation show reductions of the root mean square deviation of up to 12% due to the hybrid approach.

The incoming downward shortwave solar irradiance is harvested to an increasing extent by solar power plants. However, the variable nature of this energy source poses an operational challenge for solar power plants and electrical grids. Intra hour solar irradiance nowcasts with a high temporal and spatial resolution could be used to tackle this challenge. All sky imager (ASI) based nowcasting systems fulfill the requirements in terms of temporal and spatial resolution. However, ASI nowcasts can only be used if the required accuracies for applications in solar power plants and electrical grids are fulfilled. Scalar error metrics, such as mean absolute deviation, root mean square deviation, and skill score are commonly used to estimate the accuracy of nowcasting systems. However, these overall error metrics represented by a single number per metric are neither suitable to determine the real time accuracy of a nowcasting system in the actual weather situation, nor suitable to describe any spatially resolved nowcast accuracy. The performance of ASI-based nowcasting systems is strongly related to the prevailing weather conditions. Depending on weather conditions, large discrepancies between the overall and current system uncertainties are conceivable. Furthermore, the nowcast accuracy varies strongly within the irradiance map as higher errors may occur at transient zones close to cloud shadow edges. In this paper, we present a novel approach for the spatially resolved real-time uncertainty specification of ASI-based nowcasting systems. The current irradiance conditions are classified in one of eight distinct temporal direct normal irradiance (DNI) variability classes. For each class and lead-time, an upper and lower uncertainty value is derived from historical data, which describes a coverage probability of 68.3%. This database of uncertainty values is based on deviations of the irradiance maps, compared to three reference pyrheliometers in Tabernas, Andalucia over two years (2016 and 2017). Increased uncertainties due to transient effects are considered by detecting transient zones close to cloud shadow edges within the DNI map. The width of the transient zones is estimated by the current average cloud height, cloud speed, lead-time, and Sun position. The final spatially resolved uncertainties are validated with three reference pyrheliometers, using a data set consisting of the entire year 2018. Furthermore, we developed a procedure based on the DNI temporal variability classes to estimate the expected average uncertainties of the nowcasting system at any geographical location. The novel method can also be applied for global tilted or horizontal irradiance and is assumed to improve the applicability of the ASI nowcasts.

Solar forecasting constitutes a critical tool for operating, producing and storing generated power from solar farms. In the framework of the International Energy Agency’s Photovoltaic Power Systems Program Task 16, the solar irradiance nowcast algorithms, based on five all-sky imagers (ASIs), are used to investigate the feasibility of ASIs to foresee ramp events. ASIs 1–2 and ASIs 3–5 can capture the true ramp events by 26.0–51.0% and 49.0–92.0% of the cases, respectively. ASIs 1–2 provided the lowest (<10.0%) falsely documented ramp events while ASIs 3–5 recorded false ramp events up to 85.0%. On the other hand, ASIs 3–5 revealed the lowest falsely documented no ramp events (8.0–51.0%). ASIs 1–2 are developed to provide spatial solar irradiance forecasts and have been delimited only to a small area for the purposes of this benchmark, which penalizes these approaches. These findings show that ASI-based nowcasts could be considered as a valuable tool for predicting solar irradiance ramp events for a variety of solar energy technologies. The combination of physical and deep learning-based methods is identified as a potential approach to further improve the ramp event forecasts.