Space debris fragments too small to track can still reach speeds capable of disabling a spacecraft. The number of such fragments remains uncertain, making it difficult to assess risks to satellites supporting Earth-critical services such as broadband and GPS. We combine radar detection with a Bayesian model to adaptively choose where a radar should point next, aiming to reduce uncertainty. We analyzed about 24 hours of data from each of the two European Incoherent Scatter Scientific Association (EISCAT) radars: the Tromsø ultra high frequency radar and the EISCAT Svalbard radar. We first used a Generalized Matched Filter to search for radar signals consistent with moving objects. Tests with 4,000 simulated object signals in raw voltage data led us to select cell-averaging Constant False Alarm Rate (CA-CFAR) detection. For external validation on another radar, we applied the CA-CFAR detector selected using the Tromsø recordings to the Svalbard recordings without retuning. The method found 155 events at Tromsø and 222 at Svalbard. Of these, 106 and 119, respectively, could not be confidently matched to cataloged objects. We used the Tromsø rate of unresolved events to calibrate a Bayesian population model informed by the Orbital Debris Engineering Model (ORDEM). With this model, we compared radar pointing schedules across 300 simulated worlds, giving each schedule 24 observations. Adaptive pointing with service weights reduced uncertainty at the final observation by 41.2%, 15.4%, and 5.35% relative to fixed, random, and preplanned schedules, respectively. The additional improvement over unweighted adaptive pointing was estimated at 1.55% (95% interval: -0.47% to 3.53%), a small difference that was not statistically significant. The simulations provide strong evidence that adaptive pointing can reduce uncertainty around hidden debris using the same number of observations, potentially improving the information used to assess risks to important satellite services.

Cloud seeding success depends on both the seeding material and atmospheric conditions during deployment. This study developed a more environmentally friendly cloud-seeding formulation and a machine-learning method for identifying favorable deployment windows. Six conditions were tested in a cloud chamber: distilled water, sodium chloride, cellulose nanocrystals (CNCs), CNCs passed through polytetrafluoroethylene (PTFE) tubing, CNCs with saponin, and CNCs with both modifications. A 650 nm laser and light sensor measured cloud formation onset, minimum transmitted illuminance, and persistence across five trials per condition. Compared with unmodified CNCs, the combined formulation reduced mean onset time by 61% (27.0 to 10.6 s), reduced minimum transmitted illuminance by 93.5% (3818 to 248 lux), and increased persistence by 85% (90.8 to 168.0 s). Welch’s tests produced p = 0.021, p = 0.008, and p < 0.001, respectively. Saponin produced most of the improvement, while PTFE routing provided a smaller additional benefit. In parallel, an Extreme Gradient Boosting (XGBoost) classifier was trained on approximately 41 million records from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) and Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG). Using precipitation during the following six hours to identify favorable windows, the model achieved a receiver operating characteristic area under the curve of 0.802 on a 2023 holdout set. Location and season emerged as important predictors, demonstrating the spatiotemporal dependence of cloud-seeding opportunities. Together, these results support direct testing of the combined formulation and provide a practical method for identifying favorable deployment windows.

Autonomous robotic platforms are playing a growing role across the emergency services sector, supporting missions such as search and rescue operations in disaster zones and reconnaissance. However, traditional red-green-blue (RGB) detection pipelines struggle in low-light environments, and thermal-based systems lack color and texture information. To overcome these limitations, we present an adaptive framework that fuses RGB and long-wave infrared (LWIR) video streams at multiple fusion ratios and dynamically selects the optimal detection model for each illumination condition. We trained 33 You Only Look Once (YOLO) models on over 22,000 annotated images spanning three light levels: no-light (1000 lux). To integrate both modalities, fusion was performed by blending aligned RGB and LWIR frames at eleven ratios, from full RGB (100/0) to full LWIR (0/100) in 10% increments. Evaluation showed that the best full-light model (80/20 RGB-LWIR) and dim-light model (90/10 fusion) achieved 92.8% and 92.0% mean confidence; both significantly outperformed the YOLOv5 nano (YOLOv5n) and YOLOv11 nano (YOLOv11n) baselines. Under no-light conditions, the top 40/60 fusion reached 71.0%, exceeding baselines though not statistically significant. Adaptive RGB-LWIR fusion improved detection confidence and reliability across all illumination conditions, enhancing autonomous robotic vision performance.

Space debris fragments too small to track can still reach speeds capable of disabling a spacecraft. The number of such fragments remains uncertain, making it difficult to assess risks to satellites supporting Earth-critical services such as broadband and GPS. We combine radar detection with a Bayesian model to adaptively choose where a radar should point next, aiming to reduce uncertainty. We analyzed about 24 hours of data from each of the two European Incoherent Scatter Scientific Association (EISCAT) radars: the Tromsø ultra high frequency radar and the EISCAT Svalbard radar. We first used a Generalized Matched Filter to search for radar signals consistent with moving objects. Tests with 4,000 simulated object signals in raw voltage data led us to select cell-averaging Constant False Alarm Rate (CA-CFAR) detection. For external validation on another radar, we applied the CA-CFAR detector selected using the Tromsø recordings to the Svalbard recordings without retuning. The method found 155 events at Tromsø and 222 at Svalbard. Of these, 106 and 119, respectively, could not be confidently matched to cataloged objects. We used the Tromsø rate of unresolved events to calibrate a Bayesian population model informed by the Orbital Debris Engineering Model (ORDEM). With this model, we compared radar pointing schedules across 300 simulated worlds, giving each schedule 24 observations. Adaptive pointing with service weights reduced uncertainty at the final observation by 41.2%, 15.4%, and 5.35% relative to fixed, random, and preplanned schedules, respectively. The additional improvement over unweighted adaptive pointing was estimated at 1.55% (95% interval: -0.47% to 3.53%), a small difference that was not statistically significant. The simulations provide strong evidence that adaptive pointing can reduce uncertainty around hidden debris using the same number of observations, potentially improving the information used to assess risks to important satellite services.