Optimising the Synthetic Aperture Radar Repeat Imaging Strategy for the EnVision Mission

Context: Due to data return limitations, ESA’s EnVision mission to Venus targets the repeated Synthetic Aperture Radar (SAR) observation at 30 m resolution of roughly 30 % of the surface. The available downlink data volume varies throughout the mission as a function of Earth-Venus distance and visibility, so SAR observations need to be scheduled under spatial (target visibility on the Venus surface) and time (downlink capacity) constraints.The EnVision science orbit around Venus is near-polar, and science operations phase covers 6 Venus cycles, roughly 4 Earth years [1]. This yields up to 6 imaging opportunities of a typical surface target, allowing pairs and triplets of observations in different cycles under suitable conditions to be processed as stereo imagery or used for the detection of changes, respectively.

Aims: To develop the SAR-Combinatorial Analysis Synergy Tool (SAR-CAST): an automated strategy optimisation tool as part of the EnVision mission performance toolkit, to support the EnVision observation planning. Specifically, the tool shall group SAR observation budget units into Repeat Observation Combinations (ROCs such as stereo pairs or change detection triplets, optionally including a polarimetry observation), within mission, instrument and spacecraft constraints, while optimising a user-defined objective function considering both scientific value and engineering goals. These ROCs describe a combination of observations to make while the spacecraft orbits above a given band of longitudes. For a reference mission data budget, the tool generates a ROC plan that consumes this budget. The individual ROCs in the plan can be allocated to specific surface targets in a planning step downstream of SAR-CAST.

Method: A visual representation of these ROCs was developed to describe the budget allocation problem in a generic and intuitive way. Based on this representation, the optimisation problem can be translated into a mixed-integer linear program (MILP). Established tools like SCIP [2] allow finding the optimal solution if there is one. If the constraints cannot be satisfied within the given budget, we use the “Big M” to at least provide approximate, feasible solutions. For use of SAR-CAST as an interactive, explorative tool, marimo, a dataflow-based Python framework [3], is used to enable lazy recomputation of the MILP coefficients.

Results: ROC plans can be generated interactively or through a command line interface. Given a budget, the problem set-up takes around 15 minutes for a typical problem size of hundreds of constraints and thousands of variables, after which modification of most parameters requires updates to a subset of the constraints or coefficients only. This enables best-case recomputation delays as fast as 15 seconds, allowing near-real time interactive exploration of the parameter’s impact. Based on a reference scenario (not necessarily reflecting the mission which will be flown), coverage requirement performance equals or exceeds previous manual solution. Parameters related to constraints, their relaxation, and the objective function can be tweaked by the user to balance science and engineering needs.

Conclusions: With the SAR-CAST tool we have achieved automated optimisation of the SAR observation synergies at an abstract level, to support the creation of the science activity plan by the EnVision Science Working Team (SWT). In the future, the command-line interface of SAR-CAST would allow its integration into an automated science operations planning pipeline, and could exploit the lazy recomputation for sensitivity studies. Due to the large number of constraints and variables, the problem set-up takes significantly more time than the solving step (if an optimal solution can be found). Potential for improving efficiency in the variable and constraint creation has been identified. This could allow for even larger or more detailed versions of the problem to be solved in manageable timeframes. 

 

REFERENCES: 

[1] ESA. EnVision, Understanding why Earth’s closest neighbour is so different, ESA Definition Study Report, 2023, https://www.cosmos.esa.int/web/envision/links

[2] Bolusani, Suresh, et al. The SCIP Optimization Suite 9.0. 25734, Optimization Online, 26 Feb. 2024. optimization-online.org, https://optimization-online.org/?p=25734.

[3] Agrawal, Akshay, and Myles Scolnick. Marimo - an Open-Source Reactive Notebook for Python. Aug. 2023, https://github.com/marimo-team/marimo