TESTING A COMBINED MULTISPECTRAL-MULTITEMPORAL APPROACH FOR GETTING CLOUDLESS IMAGERY FOR SENTINEL-2
- 1Dept. of Architecture and Urban Studies, Politecnico di Milano, Laboratorio di Simulazione Urbana Fausto Curti, Milan, Italy
- 2Universidad Nacional Autónoma de México, División de Estudios de Posgrado, Facultad de Arquitectura, DF, México
- 3Universitat Politècnica de Catalunya, Center of Land Policy a nd Valuations, Barcelona, Spain
Keywords: Earth Observation, Optical RS, Masking Algorithms, Cloudless Imagery, Sentinel-2
Abstract. Earth observation and land cover monitoring are among major applications for satellite data. However, the use of primary satellite information is often limited by clouds, cloud shadows, and haze, which generally contaminate optical imagery. For purposes of hazard assessment, for instance, such as flooding, drought, or seismic events, the availability of uncontaminated optical data is required. Different approaches exist for masking and replacing cloud/haze related contamination. However, most common algorithms take advantage by employing thermal data. Hence, we tested an algorithm suitable for optical imagery only. The approach combines a multispectral-multitemporal strategy to retrieve daytime cloudless and shadow-free imagery. While the approach has been explored for Landsat information, namely Landsat 5 TM and Landsat 8 OLI, here we aim at testing the suitability of the method for Sentinel-2 Multi-Spectral Instrument. A multitemporal stack, for the same image scene, is employed to retrieve a composite uncontaminated image over a temporal period of few months. Besides, in order to emphasize the effectiveness of optical imagery for monitoring post-disaster events, two temporal stages have been processed, before and after a critical seismic event occurred in Lombok Island, Indonesia, in summer 2018. The approach relies on a clouds and cloud shadows masking algorithm, based on spectral features, and a data reconstruction phase based on automatic selection of the most suitable pixels from a multitemporal stack. Results have been tested with uncontaminated image samples for the same scene. High accuracy is achieved.