Autonomous and Dynamic Satellite for Improved Cloud Observation
Name
dahl-marydahl-phd-aeroastro-2026-thesis.pdf
Size
17.49 MB
Format
Adobe PDF
Checksum (MD5)
da99111543945efe5233f4520dec9a25
Author(s)
Dahl, Mary
Advisor(s)
Cahoy, Kerri
De Weck, Olivier
Gizzi, Evana
Date Issued
February 2026
Publisher
Massachusetts Institute of Technology
Abstract
Clouds, aerosols, and their interactions are the largest source of uncertainty in climate models. Nadir-pointing satellites have inefficiencies: by only measuring what is directly below them, they spend resources measuring empty air and miss out on any valuable clouds that are outside of the path. Advancements in onboard autonomy and artificial intelligence can enable improved decision making and actions. A lookahead camera could obtain contextual information of where clouds are and which measurements could be made. By utilizing an onboard algorithm, the satellite could plan where and when to point its instrument, allowing it to obtain more high-value cloud data while managing power, storage, and downlink resources. This work explores the feasibility of this autonomous cloud imaging mission concept. A simulation is developed to test satellite and instrument architectures and algorithms. It leverages cloud masks which contain information on cloud cover and the location of valuable cirrus clouds. Three algorithms (greedy, nearest greedy, and distance weighted) are developed to choose satellite actions: where to slew, when to use its instrument, and when to downlink data, etc. They maximize scientific gain while limiting excessive slewing. Mission architecture options (lookahead image sizing and instrument choice) are analyzed for efficiency and impact. The timing required for taking and processing a 200 km lookahead image supports up to 90% of a LEO satellite’s uptime spent making measurements. Instruments are characterized with their relative capabilities of obtaining cloud-aerosol measurements. Combinations of instruments are assessed with a genetic algorithm to optimize instrument choice for overall scientific value while considering power and cost. Lidars are found to be the priority instrument, appearing in all optimal designs, followed by radiometers and polarimeters. Based on this, a lidar is modeled off-nadir, finding 30 degrees as the maximum angle that resolves fine features while maintaining a desired signal-to-noise ratio. This mission architecture is tested with a series of experiments in the simulation to determine the best algorithm. The algorithms are compared both by their efficacy (how scientifically valuable their measurements are) and their efficiency (how much power is expended, what they spend time doing, how much they slew) to make a recommendation for future missions. The highest performing design measures the equivalent of 2,610,000 km2 cloud data over one week, which is a 16% improvement over a traditional satellite. If focused specifically on measuring cirrus clouds, the top algorithm measures 80,000 km2 more than a traditional satellite. The results of this work demonstrate that a lookahead sensor can transform the capabilities of a satellite for measuring clouds, changing remote sensing from “collect everything” to “collect what is needed.”
MIT Department
Massachusetts Institute of Technology. Department of Aeronautics and Astronautics
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
