A Bayesian framework for explosive yield estimation using statistical signal characterization and attenuation modeling

We investigate explosive yield estimation from infrasound signals generated by controlled ground explosions at the Hukkakero ammunition disposal site in northern Finland. Since 1988, highly repeatable blasts with yields of approximately 20 tons TNT equivalent have been conducted annually, providing a valuable reference dataset.

Explosive yield is estimated using a Bayesian framework that explicitly accounts for uncertainties in source characteristics and transmission loss statistics. Spectral characteristics of the signals are extracted using the multichannel maximum-likelihood (MCML) method, providing robust inputs for yield estimation. Propagation effects are represented through an updated statistical transmission loss law derived from extensive full-wave simulations under realistic atmospheric conditions. Rather than relying on deterministic scaling relations, transmission loss is incorporated as a probability distribution within the Bayesian formulation as a function of frequency and effective sound speed ratio.

Applying this approach to historical infrasound observations from the IMS array IS37 (northern Norway, ~320 km from Hukkakero) yields probabilistic explosive energy estimates with physically meaningful uncertainty bounds. The results demonstrate improved robustness and reduced bias compared with traditional methods, particularly for regional-distance observations where atmospheric effects strongly influence signal amplitudes.