Automatic Observation of Cloudiness: Analysis of All-sky Images

The influence of clouds on Earth ́s radiation balance is a key component of the atmospheric system. Therefore, the determination of cloud cover is an important activity traditionally performed by human observations and broadcasted by meteorological services observers in Synop and Metar reports. However, human observations are subject to large errors of estimation, and their temporal resolution is very poor. We have developed an automatic system of cloud observation (SONA) that provides cloud cover percentage using neural networks in all-sky images. Our system consists on a 640x480 pixel resolution CCD camera with an infrared filter. All-sky images are stored in time periods from 1 to 5 minutes and real time processed to obtain cloud cover percentage during daytime. Clouds are also well detected during nighttime. The cloud detection algorithms are based on nonlinear multilayer neural networks for image pattern recognition. They are subjected to training processes with a backpropagation method. SONA is an inexpensive, all-weather and robust system to obtain cloud information in a wide range of sky conditions. New improvements are being developed to obtain cloud flow, cloud-base height estimation by triangulation, and an index of atmospheric dust presence. SONA products might be customized in order to satisfy requirements of diverse applications such as cloud reporting in airports, cloud and radiation nowcasting combined with satellite images for solar power plants, sky watching in astronomy observatories, public dissemination of sky conditions on beaches and resorts, etc.