Laboratory for
Climatology and Remote Sensing

Abstract:

Information about clouds is important for observing and predicting weather and climate as well as for generating and distributing solar power. Most existing approaches extract cloud information from satellite data by classifying individual pixels instead of using closely integrated spatial information, ignoring the fact that clouds are highly dynamic, spatially continuous entities. This paper proposes a novel cloud classification method based on deep learning. Relying on a Convolutional Neural Network (CNN) architecture for image segmentation, the presented Cloud Segmentation CNN (CS-CNN), classifies all pixels of a scene simultaneously rather than individually. We show that CS-CNN can successfully process multispectral satellite data to classify continuous phenomena such as highly dynamic clouds. The proposed approach produces excellent results on Meteosat Second Generation (MSG) satellite data in terms of quality, robustness, and runtime compared to other machine learning methods such as random forests. In particular, comparing CS-CNN with the CLAAS-2 cloud mask derived from MSG data shows high accuracy (0.94) and Heidke Skill Score (0.90) values. In contrast to a random forest, CS-CNN produces robust results and is insensitive to challenges created by coast lines and bright (sand) surface areas. Using GPU acceleration, CS-CNN requires only 25 ms of computation time for classification of images of Europe with 508  508 pixels.

Abstract:

Due to the significant impact of fog and low stratus (FLS) on economy, ecology and traffic systems, there is a growing demand for high-resolution information on FLS occurrence. In this study, a baseline climatology of FLS h day?1 based on data recorded from 2006–2015 by the Spinning Enhanced Visible and Infrared Imager system (SEVIRI) aboard the Meteosat Second Generation satellites is computed for Europe to provide the requested information. It is the first 10 year, spatially explicit climatology for FLS based on data with a temporal resolution of 15 min. The dataset is validated against Meteorological Aviation Routine Weather Reports (METAR) and shows good accordance with an average Heidke Skill Score of 0.45. Temporal and spatial variations in FLS frequency as well as interannual trends are analyzed. Winter shows the highest FLS occurrence, but a general decrease over the investigated period. Spring, summer and autumn show less pronounced trends and lower average FLS frequencies. Possible reasons for these distributions are discussed.

Abstract:

A method is presented for fog and low stratus detection from daytime satellite imagery based on Meteosat 8 SEVIRI (Spinning-Enhanced Visible and Infra-Red Imager) data. With its excellent spatial, spectral and temporal resolutions, this imagery is an ideal basis for operational fog monitoring. The scheme utilizes a range of pixel-based and novel object-oriented techniques to separate fog and low stratus clouds from other cloud types. Fog and low stratus are identified by a number of tests which explicitly and implicitly address fog/low stratus spectral, spatial and microphysical properties. The scheme's performance is evaluated using ground-based measurements of cloud height over Europe. The algorithm is found to detect low clouds very accurately, with probabilities of detection (POD) ranging from 0.632 to 0.834 (for different inter-comparison approaches), and false alarm ratios (FAR) between 0.059 and 0.021. The retrieval of sub-pixel and temporal effects remain issues for further investigation.

Abstract:

The distribution of fog and low stratus (FLS) is of importance in nowcasting, aviation forecasting and climate applications. This paper presents satellite-derived FLS maps as a basis for building a European FLS climatology with high spatial and temporal resolution. Averaged maps covering several winter seasons of Meteosat Second Generation (MSG) data are analysed and compared to a cloud climatology based on a 26-year record of ground-based visual cloud observations. The general patterns seen in both products are found to be in good agreement and plausible.