To detect the natural gas leakage of the pipeline underwater, a distributed optical fiber sensor system is introduced based on hybrid configuration of Sagnac/Mach-Zehnder interferometers. Based on time domain statistic features, the probability features of amplitude values distributed near the average can be extracted with time domain analysis of optical fiber interferometric signals. A principal components analysis model based on statistic features is established to monitor gas pipeline leakage condition. The squared prediction errors (SPE) statistics is used to discriminate pipeline condition, and compared to the traditional method of distributed optical fiber pipeline leakage detection which just depends on whether null frequency existing or not. The missing alarm rate (MAR) is calculated. The results show that the proposed time-domain statistical feature vector can describe the characteristics of pipeline condition, the SPE statistics in leakage condition can discriminate pipeline leakage accurately, and MAR decreases to about 1/10 of that with the traditional method based on null frequency.