Earth Observation Satellite Scheduling Models: Taxonomy, Classification and Open Challenges

Abstract

Over the past three decades, a large number of optimization models have been proposed for Earth observation satellite (EOS) scheduling. The main goal of this paper is to provide an overview of these models and to relate them to the characteristics of the problems they address. To do so, we first identify the main features of EOS scheduling problems and those with the most impact on the computational complexity of the model. Based on these features, we propose a new taxonomy of satellite agility and a classification of EOS scheduling problems that allows the problem under study to be clearly identified. We then propose a classification of the optimization models into five categories, based on their time and event representations, and describe representative formulations of each category in a unified notation, together with their extensions, discussing the respective advantages and limitations. The trade-offs between the alternative representations are analyzed, showing that discrete-time formulations allow time-dependent parameters, such as transition times and image quality, to be pre-computed, which preserves the linearity of the model but makes its size depend on the adopted discretization, whereas continuous-time formulations treat these parameters as functions of the start times, which introduces nonlinearities unless simplifying assumptions are adopted. Finally, we identify the main open challenges, including the exact modeling of time-dependent transition times and the integrated scheduling of active imaging satellites.