- The paper demonstrates a unified framework that integrates agile satellite maneuvering, onboard prediction, and disruption tolerant networking for responsive urban flood monitoring.
- The paper introduces a decentralized scheduling method using dynamic programming and DTN to optimize observation strategies under physical and communication constraints.
- Simulations with a 24-satellite constellation reveal up to a 98% improvement over static models, validating the system's efficiency and real-time capabilities.
Agile, Autonomous Satellite Constellations with Disruption Tolerant Networking for Responsive Urban Flood Monitoring
This paper presents an integrated, modular framework for scheduling and coordinating a constellation of small, fully-agile satellites equipped with inter-satellite links (ISLs) and disruption tolerant networking (DTN) capabilities, aimed at monitoring dynamically evolving geophysical phenomena, with a particular emphasis on episodic precipitation and urban floods. The main innovation lies in the unification of real-time onboard prediction, planning, spacecraft agility, and robust peer-to-peer networking for decentralized, high-cadence Earth observation. The system is validated through comprehensive simulations, yielding significant performance improvements over non-agile and ground-based planning models.
Framework Architecture
The proposed system unifies four principal components:
- Orbital Mechanics (OM): Propagates satellite positions, access to ground points (GPs), and computes the satellite's field of regard (FOR).
- Attitude Control System (ACS): Models and schedules re-orientation maneuvers with optimization (minimum-time/effort), factoring slew time and physical constraints.
- Communications: Implements DTN using the Bundle Protocol with MF-FDMA at the MAC layer, enabling robust, low-latency, peer-to-peer data exchange amid dynamic constellation topologies.
- Prediction & Planning: Incorporates physics-based predictive models responsive to observations, with value-driven dynamic programming (DP) for distributed or centralized scheduling.
A clear emphasis is placed on modularity, allowing portability and generalization to other sensing tasks or satellite hardware.
Scheduling and Value Update Algorithms
The scheduler operates as follows:
- Receives updated local and distributed state (observations and model parameters) via DTN bundles.
- Estimates the predictive value of possible future observations using either regression-based or, with trivial substitution, machine learning models (LSTM, etc.).
- Applies DP to generate a schedule maximizing total expected science value, constrained by physical and communication limits.
- Integrates feedback loops to handle ACS slew characteristics and bundle delivery latencies.
- Re-broadcasts schedule and updated model parameters after execution.
Algorithmic complexity grows linearly with planning horizon and quadratically with GPs, scaled by the number of overlapping satellite FORs. Integer programming achieves no more than 10% improvement in optimality but at orders-of-magnitude greater computational cost.
Communications and Consensus
ISL-enabled, delay/disruption tolerant networking is leveraged to avoid the high-latency bottleneck inherent in ground-based centralized scheduling, especially crucial for rapidly evolving phenomena. Rather than rigid explicit consensus, the system operates via implicit consensus, with each satellite updating its knowledge and planning dynamically in response to newly received bundles.
Simulation results indicate the delivery of all (non-low-priority) bundles within regional access gaps, with median DTN latency less than one-third of inter-access gaps, demonstrating the practical viability of decentralized, real-time consensus through limited implicit communication.
Simulation Results and Quantitative Analysis
A simulated scenario employing 24 small satellites in sun-synchronous orbits, with hydrologic modeling for five flood-prone urban regions, yields concrete numerical results:
| Scheduler Location |
Planning Horizon |
Total Observed Flood Magnitude |
% Improvement over Static |
| Onboard (Decentralized) |
10 min |
2703.7 |
+98% |
| Onboard (Decentralized) |
5 min |
2661.7 |
+98% |
| Ground (Centralized) |
99-198 min |
~2528.5 |
+90-93% |
Key findings:
- The onboard, decentralized planner captures ~7% more flood magnitude than ground-planned operations at typical ground station contact rates.
- Both agile constellation configurations deliver a ~98% improvement over static (non-agile) radars on the same satellites.
- Replanning at sub-15-minute frequency plateaus in incremental benefit, establishing a pragmatic balance between computational overhead and operational value.
Scheduling runtime is shown to be well within the capabilities of contemporary onboard processors (~2% of planning horizon even on older architectures), and faster C implementations are readily achievable.
Theoretical and Practical Implications
By integrating agile spacecraft maneuvering, onboard prediction, and DTN-based distributed scheduling, the framework demonstrates:
- Superior responsiveness to fast-evolving geophysical events.
- Robustness to communication outages and dynamic topology, crucial for smallsat missions operating with limited downlink capacity and sparse ground infrastructure.
- Quantitative methodologies for mission designers to weigh the resource/cost trade-offs between agility, communication cadence, and computational burden.
The system’s modularity and generalizability make it suitable for broader applications such as wildfire, cyclone, or snowmelt monitoring, and its use of open standards and open-source simulators enhances reproducibility and adoption.
Future Directions
Notable extensions include:
- Incorporation of higher-fidelity, potentially deep learning–based science models for improved value prediction and scheduling accuracy.
- Investigation into hybrid onboard-ground scheduling paradigms, exploiting the strengths of both approaches as dictated by mission parameters, ground infrastructure, and event timescales.
- Optimization of DTN scheduling to further reduce latency and manage network congestion.
- Application of advanced consensus and learning algorithms for further distributed intelligence at the constellation scale.
Conclusion
The presented framework establishes a significant practical advance in coordinated, intelligent sensing with agile smallsat constellations for Earth observation, demonstrating both strong empirical value gains and scalable, real-time feasibility. The work substantiates the operational benefits of distributed, onboard science-driven planning and motivates future research in onboard autonomy, resilient networking, and value-aware resource management for next-generation smallsat missions.