Development of an adaptive multi-method algorithm for automatic picking of first arrival times: application to near surface seismic data
Accurate picking of first-arrival times is important in many seismic studies, particularly in seismic tomography and reservoirs or aquifers monitoring. Many techniques have been developed, mainly for seismological purposes, in order to pick first arrivals automatically or semi-automatically. However, these techniques do not reach the accuracy required in shallow seismics due to the complexity of near-surface structures and low signal-to-noise ratio. We propose here a new adaptive algorithm to automatically pick first arrival in near-surface seismic data by combining three picking methods: multi-nested windows, higher order statistics, and Akaike information criterion. They benefit from combining different properties of the signal in order to highlight first arrivals and finally to provide an efficient and robust automatic picking. This strategy mimics the human first-break picking, where a global trend is first defined at the beginning of the picking procedure. The exact first breaks are then sought in the vicinity of each point suggested by this trend. Three successive phases are combined in a multistage algorithm, each of them characterizing a specific signal property. Within each phase, the potential picks and their error range are automatically assessed and sequentially used as prior constraints in the following phase picking. Since having realistic estimates of the error in picked traveltimes is crucial for seismic tomography, our adaptive algorithm automatically provides picked arrival times with their associated uncertainties. We demonstrate the accuracy and robustness of the implemented algorithm using synthetic, pseudo-synthetic and real datasets that pose challenges to classical automatic pickers. A comparison of both manual and adaptive picking procedures demonstrates that our new scheme provides more reliable results even under different noisy conditions. All parameters of our multi-method algorithm are self-adaptive, thanks to the sequential integration of each sub-algorithm results in the workflow. Hence, it is nearly a parameter-free algorithm, which is straightforward to implement and demands low computational resources.