Beyond Diagonal RIS Assisted SAGIN for Environmental Monitoring: A Deep Reinforcement Learning Approach

Space-Air-Ground Integrated Networks (SAGINs) are emerging as a critical enabler for large-scale, real-time environmental monitoring. This paper proposes a novel SAGIN framework enhanced by Beyond-Diagonal Reconfigurable Intelligent Surfaces (BD-RIS) to address the coverage, reliability, and efficiency demands of 6G-driven environmental surveillance. We present a four-layer architecture comprising satellites, high-altitude platform stations (HAPS), unmanned aerial vehicles (UAVs) equipped with BD-RIS panels, and ground users, forming an integrated sensing and communication (ISAC) system. To manage the heterogeneous resources and dynamic conditions, we develop a multi-agent deep deterministic policy gradient (MADDPG) algorithm with a hierarchical attention mechanism. This approach jointly optimizes UAV trajectories, transmission powers, BD-RIS phase configurations, and data fusion weights to maximize detection accuracy while minimizing latency and energy consumption under practical constraints. Extensive simulations validate the proposed framework, achieving a forest-fire detection performance with an F1-score of 0.982, a latency reduction of up to 60% compared to terrestrial-only networks, and a 36% improvement in energy efficiency over a 24-hour continuous monitoring mission.