Risk-Priced Freshness Scheduling for Geohazard Early Warning over Coupled Sensor–Edge Observation Networks
Year: 2026
Type: preprint
Abstract: GNSS/InSAR stations, satellite fire products, UAV flood imagery and weather stations compete for the same uplink slots and edge compute. Existing Age-of-Information schedulers treat every unit of staleness as equally harmful, yet a ten-minute delay is negligible for a meteorological stream and fatal for a seismic onset. We recast freshness scheduling as risk-priced scheduling: SIV-Net predicts the geoscience value of a resource slot from link/AoI state and task descriptors, PRISM embeds freshness deadlines and a preemption penalty into a constrained MDP whose Lagrangian relaxation yields a Whittle-style index, and TANDEM couples scheduling with offload selection through a shared affinity tensor and asynchronous updates. On SMAP/MSL, Weather, EuroSAT and FloodNet, the framework reaches F1 0.9002/0.9648, MSE 0.168, accuracy 98.91% and mIoU 82.4, cutting the hazard missed-alarm rate from 2.71% to 0.96% while retaining above 89% of source-domain performance under new climate zones, sensors and seasons.
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225039432