Adaptive Partial State Updates for MagNav Based on Map Uncertainty

Magnetic Anomaly Navigation (MagNav) provides an absolute positioning solution for GNSS-denied environments. However, Extended Kalman Filter (EKF) performance and stability is fundamentally constrained by variable map quality. While some global regions feature high-quality aeromagnetic surveys, many historical datasets are sparsely sampled, resulting in severe localized errors. Filters often attempt to handle these inaccuracies by appending a First-Order Gauss-Markov (FOGM) state to track map errors. While helpful, these methods fail to capture highly non-linear or rapidly changing map errors and introduce observability issues. Because a standard scalar measurement update simultaneously affects multiple coupled filter states such as position, map biases, and aircraft Tolles-Lawson coefficients, allocating the measurement residual correctly becomes mathematically ambiguous during unmodeled map errors. Furthermore, state augmentation often assumes a uniform error propagation model which fails to account for spatial volatility across a single flight path. Previous research applied a binary threshold to either accept or completely reject magnetic updates based on map uncertainty. While this improves filter stability over a naïve EKF, a strictly binary approach ignores the relative reliability of different map regions and the varying linearity of the individual filter states. Additionally, the previous work relied on extensive prior knowledge of the survey data and map-processing to attempt to quantify map uncertainty. This research applies a coherence-based method to MagNav to detect problematic map areas by comparing real magnetometer flight data with the underlying map. Unlike previous approaches that attempt to quantify map covariance at every individual point, this coherence method does not require extensive prior knowledge of the original survey data and can be applied to any arbitrary map. We leverage this information to inform a partial state update framework, dynamically scaling how updates are distributed across the state vector. This allows the filter to adaptively weight updates, aggressively updating the states when coherence is high and minimizing updates when the coherence is low.