Dense, short‐term deployments of seismograph networks are frequently used to study upper‐mantle structure. However, recordings of variably emergent teleseismic waveforms are often of lower signal‐to‐noise ratio (SNR) than those recorded at permanent observatory sites. Therefore, waveform coherency across a network is frequently utilized to calculate relative arrival times between recorded traces, but these measurements cannot easily be combined or reported directly to global absolute arrival‐time databases. These datasets are thus a valuable but untapped resource with which to fill spatial gaps in global absolute‐wavespeed tomographic models.

We developed an absolute arrival‐time recovery method (AARM) to retrieve absolute time picks from relative‐arrival‐time datasets, working synchronously with filtered and unfiltered data. We also include a relative estimate of uncertainty for potential use in data weighting during subsequent tomographic inversion. Filtered waveforms are first aligned via multichannel cross correlation. These time shifts are applied to unfiltered waveforms to generate a phase‐weighted stack. Cross correlation with the primary stack or the SNR of each trace is used to weight a second‐higher SNR stack. The first arrival on the final stack is picked manually to recover absolute arrival times for the aligned waveforms.

We test AARM on a recently published dataset from southeast Canada ( ∼10,000∼10,000 picks). When compared with the available equivalent earthquake–station pairs on the International Seismological Centre (ISC) database, ∼83%∼83% of AARM picks agree to within ±0.5  s±0.5  s . Tests using synthetic P‐wave data indicate that AARM produces absolute arrival‐time picks to accuracies of better than 0.25 s, akin to uncertainties in ISC bulletins.