Establishing the GLOBCARBON cloud detection system over land for the Along Track Scanning Radiometer (ATSR) sensor series

The GLOBCARBON initiative aims to develop a service to generate fully calibrated estimates of at-land products using data from the ATSR sensor series, MERIS and VEGETATION in combination. A fundamental requirement of such processing is the effective removal of extraneous effects such as cloud, cloud shadow, snow and atmosphere in these data. This paper reports on the implementation and extension of a new cloud and snow processing system for land for the ATSR sensor series. The snow processor is as implemented for MODIS while the cloud processor is based on APOLLO with extensions to generate probabilities of cloud presence and a further test to eliminate cloud missed by APOLLO. The system was tested over 49 images representing a wide range of cloud types and forms over three contrasting biomes – boreal forest, desert savannah and tropical evergreen forest. Excellent results for 48 out of 49 images, clearing all forms of cloud without introducing commission, have been produced. The only image where cloud remains after the processing is exhibits a situation where the threshold determination for the final shortwave-thermal test is not effective because the remanent cloud after APOLLO occupies a larger number of pixels than the clear land. However, in all 49 cases the new system represents a major improvement over the original flagging system. The approach has been adopted in GLOBCARBON for the processing of 500,000 ATSR-2 scenes and 25,000 AATSR striplines.