Abstract: Accurate and reliable picking of the first arrival time is one of the critical steps in micro-seismic monitoring. Aiming at the problem of low accuracy of first arrival picking for micro-seisms under low signal-to-noise ratio, the traditional short term averaging/long term averaging algorithm is improved by introducing weight factor according to the change of signal amplitude to improve the accuracy of initial pickup. In order to further reduce the pickup error, variational mode decomposition （VMD） is optimized based on cross-correlation coefficient and permutation entropy criterion, and decomposition layers are determined adaptively. Then, the signals of 2−3 s before and after the initial pickup are decomposed by VMD, and the Kurtosis-Akaike information criterion （AIC） values of the decomposed intrinsic mode functions （IMF） are calculated to get the arrival time of each IMF, and the secondary arrival time is obtained by comprehensively weighting the picking results and energy ratios of each IMF. Simulation results show that the improved STA/LTA can reduce the initial picking error by more than 0.01 s at low SNR; compared with empirical mode decomposition （EMD） and wavelet packet decomposition, the adaptive VMD decomposition can reduce the picking error again, and the finalaverage picking error is less than 0.023 s. The first arrival time picking results of real micro-seismic signals show that the proposed algorithm can identify the first break of P-wave quickly and effectively, and the error is smaller than that of manual picking, which shows that the algorithm is effective and the picking accuracy is high.