Assessing the value of on-board image processing in observation task planning: the cloud mask problem

View Video Presentation: https://doi.org/10.2514/6.2021-1468.vid

Clouds are an obstacle for many remote sensing applications. Earth observation satellites may have repeat cycles that allow for study of temporal variations, but for certain types of instruments, high cloud fractions can make the time gap between high quality observations unacceptably long. On-board image processing has recently been implemented on several satellites. This new capability allows for improved knowledge when planning satellite tasks, leading to improved revisit time. This article quantifies the benefit of utilizing on-board image processing in the context of determining where to observe at a given point in time for a given satellite mission with pointing capabilities. An algorithm utilizing a semi-Markov Decision Process and on-board image processing is used for observation task planning. It is shown that this algorithm provides a much greater quantity of images and a modest improvement in average scientific reward per image over a nadir-pointing strategy. The method is evaluated in a limited manner using cloud fraction as a proxy for science reward. Future work will focus on planning using more complex science reward formulations.