Adaption of a Self-Learning Algorithm for Dynamic Classification of Water Bodies to MERIS Data

Abstract

In many global applications of remote sensing land-water masks can improve the interpretation results. Their use can be of advantage to distinguish between different types of dark areas (e.g. cloud or topographic shadows, burned areas, coniferous forests, water areas). On one hand, water bodies cannot always be classified exactly on basis of available remote sensing data. On the other hand static land-water masks of different quality are available. But the main deficiencies are caused by the fact that land-water masks represent only a temporal snapshot of the water bodies. A dynamic self-learning water masking approach was developed at first for AATSR data to combine the advantages of static mask with results of pre-classifications. This paper presents the adaption of this procedure for MERIS remote sensing data. As before with AATSR data the aim consists in integrating high-quality water masks in processing chains for deriving value-added remote sensing data products. The results for some regional examples demonstrate the quality of masks and the advantages to conventional water masking algorithms. Furthermore, it will be discussed, that it is useful for a global water mask of high quality to integrate further special masks as cloud or in particular terrain shadow masks. At least, the land-water mask plays not only an important role in complex processing chains itself is the result of a complex procedure. Beside the results have shown successful transfer of a developed processing scheme for operational deriving of actual land-water masks to data of a second sensor, the adaption to further sensors or the adaption of the processor to other object types as e.g. forest will be possible in future as part of operational monitoring systems.