Ученые создали адаптивную модель BLAC-Q4DIM для точного прогнозирования ионосферных возмущений

Newswise — Traditional Global Ionospheric Maps (GIMs) and fixed-parameter models struggle to capture rapid, localized changes in electron content, particularly during geomagnetic storms or in regions with sparse GNSS reference-station coverage . Their reliance on fixed spatial scales often leads to significant errors, especially over oceans and at low latitudes. These limitations call for a fundamentally new approach to ionospheric modeling—one that can dynamically adapt to changing ionospheric conditions and evolving observation distributions. Now, researchers from Wuhan University, the China University of Geosciences, and the National Centre for Physics in Pakistan have developed a novel solution, published online in Satellite Navigation on August 17, 2026 (DOI: 10.1186/s43020-026-00213-z). The new model, named BLAC-Q4DIM, integrates adaptive clustering, Long Short-Term Memory (LSTM), and Bayesian Optimization (BO) to directly model Slant Total Electron Content (STEC) in a 4D space of latitude, longitude, elevation, and azimuth. The breakthrough lies in BLAC-Q4DIM's "closed-loop" architecture. It uses an LSTM network to predict prior estimates of clustering hyperparameters from current spatiotemporal features and historical parameter evolution, and then refines these estimates through BO using a composite objective function. This allows the model to automatically adjust its parameters to prevailing conditions. Unlike conventional models that rely on empirically fixed hyperparameters, BLAC-Q4DIM treats these critical settings as learnable variables that evolve with the observation geometry, ionospheric structure, and space-weather conditions. Tested over two 30-day periods, including the extreme G5 geomagnetic storm of May 2024, the system showed dramatic improvements. The best-performing variants, BLAC-HDBSCAN and BLAC-DPMEANS, achieved Root Mean Square (RMS) errors of just 0.96–1.01 Total Electron Content Units (TECU) during the quiet period and 1.34–1.40 TECU during the disturbed period—representing an 80–85% RMS reduction relative to standard GIMs and more than 58% relative to the best fixed-parameter Q4DIM baseline. The model also demonstrated remarkable robustness under low-latitude, sparse-station, and geomagnetically disturbed conditions, maintaining RMS errors below 2 TECU at low latitudes and below 1.44 TECU under the sparsest reference-station configuration. Its ability to preserve clustering quality and modeling accuracy across vastly different geomagnetic regimes underscores the power of its adaptive, data-driven design. "The core innovation is that we've transformed empirically fixed clustering hyperparameters into learnable variables that dynamically evolve with changes in observation geometry, ionospheric structure, and space-weather conditions," the authors said. "By integrating 4D LOS-STEC modeling, multi-paradigm clustering, LSTM-based temporal learning, and Bayesian posterior optimization into a unified framework, BLAC-Q4DIM can 'understand' and adapt to the environment. It is no longer a static map but a self-aware system that adjusts its own rules as conditions change—whether under sparse or dense observation distributions, or under conditions ranging from a quiet day to a severe geomagnetic storm. The closed-loop mechanism, involving LSTM prior prediction, BO posterior correction, and periodic retraining, is what gives the model its long-term stability and resilience." This adaptive capability translates directly into real-world utility. By substantially reducing ionospheric delay modeling errors, BLAC-Q4DIM can significantly enhance the integrity and accuracy of GNSS-dependent services, from autonomous vehicle navigation and precision agriculture to drone delivery and maritime operations. The framework also holds promise for space-weather monitoring, helping scientists track and respond to geomagnetic disturbances with higher precision. Crucially, its computational efficiency further supports near-real-time applications, with the best-performing variant requiring as little as 3.5 s per window for LSTM prior prediction and BO posterior correction. This sets a new benchmark for next-generation resilient positioning systems, demonstrating BLAC-Q4DIM’s ability to combine high modeling accuracy with practical computational efficiency for near-real-time ionospheric modeling in operational settings. ### References DOI Original Source URL https://doi.org/10.1186/s43020-026-00213-z Funding information This work was supported by the National Natural Science Foundation of China under Grant No. 42574036. About Satellite Navigation Satellite Navigation (ISSN: 2662-1363; ISSN: 2662-9291) is the official journal of the Aerospace Information Research Institute. The journal aims to report innovative ideas, new results, and progress in the theories, techniques, and applications of satellite navigation. The journal welcomes original articles, reviews and commentaries.