Assessing uncertainties in infrasound network performance modelling: application to the Euro-Mediterranean and Southeast Asian region

Abstract

We propose a modelling technique to confidently estimate and optimize the performance of any infrasound network to remotely monitor sources of interest such as volcanic eruptions, while considering realistic atmospheric specifications along the propagation path, source frequency and noise levels at the station. To provide a more realistic picture of the network performance, we define a confidence level accounting for propagation and atmospheric uncertainties. Therefore, we consider 'numerical' uncertainties linked to the approximations made in the used propagation model, errors of the developed mathematical model and atmospheric uncertainties derived from measurement campaigns. In parallel, we perform a sensitivity analysis to determine how each input parameter contributes to the developed mathematical model output as well as to the attenuation model output. Such study is helpful for model simplification and uncertainty reduction by identifying, and thus paying more attention to the most influential model inputs. Below 1 Hz, the effect of 'numerical' errors on network performance modelling dominates. The same situation is observed during strong and stable downwind stratospheric winds along propagation paths. Conversely, when propagation occurs upwind, atmospheric uncertainties become predominant as the frequency increases. This method is then applied to assess the performance of the International Monitoring System (IMS) infrasound network in the Euro-Mediterranean and the Southeast Asian regions. We highlight a frequency, seasonal and spatial dependence of uncertainties in the modelling. Below 1 Hz, large errors are predicted in the shadow zone but the overall error is less than 20 dB. Above 1 Hz, errors with same order of magnitude are also observed, when strong stratospheric jets prevail. But during weak stratospheric duct, uncertainties associated to the modelled attenuation may exceed 30 dB. Such studies lead to significant improvement in assessing detection capability of infrasound network, which is of great interest for monitoring artificial or natural explosive sources like volcanic eruption. In particular this work will contribute into designing and prioritizing maintenance of any given infrasound network, in order to provide even better and more accurate predictions.