Cluster-based waveform surrogate model for three-dimensional propagation of the sonic boom : an application to Carancas

The amplitude of infrasonic arrivals at ground level depends on both atmospheric propagation conditions and the three-dimensional geometry of the source. Temperature and wind speed gradients, as well as the source's directivity, can create shadow zones where the geometric acoustic approximation incorrectly predicts no arrival. For instance, during the 2007 Carancas meteorite entry, the I08BO infrasound station was located at the boundary between the geometric arrival zone and the shadow zone. In such configuration, conventional ray-tracing models were unable to simulate the recorded signals while a full wave code, such as Flhoward3D, can simulate arrivals in both zones. However, a small variation in the trajectory azimuth or elevation, or in the sound speed profile, can sharply change the dynamics of the arrivals at the station. To invert the trajectory, we rely on a surrogate model capable of reproducing the discontinuities in the numerical signal predictions.

To address this challenge, our surrogate construction approach proceeds in three steps. First, the parametric domain is partitioned using clustering techniques applied to numerical signals. Each cluster is then associated with a physical behavior, such as a shadow or light zone. Second, a principal component analysis (PCA) is performed for each cluster. Third, the relationship between the PCA coordinates and the input parameters is approximated using least-squares regression.

We compare the method's performance to that of a global surrogate model. Next, the method is applied to invert the Carancas trajectory angles using only arrivals at a single infrasonic station. This work paves the way for inferring wind or gravity wave profiles when infrasound propagation is highly sensitive to small atmospheric variations.