Abstract: Microseisms play an important role in long-term risk monitoring for geothermal extraction, gas storage construction and operation, and CCUS. Conventional microseismic techniques were developed for short-term monitoring, e.g. shale oil hydro-fracturing, and suffer from such limitations as high acquisition costs and low automation levels in long-term monitoring applications. To address the application requirements of long-term monitoring, an automated processing flow based on machine learning is established and tested using sparse high-quality microseismic data acquired from hot dry rock fracturing with shallow-hole receivers. Comparative analysis with a traditional processing flow demonstrates the reliability of machine-learning methods for long-term microseismic monitoring. The results show that neural network-based denoising and phase identification algorithms significantly improve the S/N ratio and P- and S-phase identification. The automatic quality control algorithm based on the Gaussian Mixture Model effectively eliminates false picks and further enhances phase identification. Microseism locations obtained from this automated workflow are comparable to those derived from the traditional method with manual quality control. These results confirm the feasibility of this processing flow for long-term monitoring through automated processing of sparse shallow-hole microseismic data.