IEEE TRANSACTIONS ON SIGNAL AND INFORMATION PROCESSING OVER NETWORKS, XXX, XXX, 2020 1 Distributed Sensor Networks based Shallow Subsurface Imaging and Infrastructure Monitoring Fangyu Li, Maria Valero, Yifang Cheng, Tong Bai, and WenZhan Song Abstract—Distributed sensor networks can be used as passive seismic sensors to image and monitor subsurface and under- ground activities. Passive seismic surface-wave imaging adopts background ambient sounds from a far-field energy source. Because high frequency components decay a lot between the neighboring stations, conventional sparse sensor networks cannot image small-scale and shallow objects. In this paper, we propose to use local seismic spatial autocorrelation coefficients, obtained by the combinations of independent dense sensor network measurements and pre-processed readings of its neighbor(s), to perform real-time collaborative imaging of the shallow subsurface objects. First, we derive the high-frequency spectral coefficient based shallow subsurface imaging method. Then, we apply the proposed approach to image a shallowly buried pipeline and obtain promising results. Furthermore, using a time-lapse manner, we demonstrate that the water leakage from the buried pipeline can also be detected using distributed computations between sensors. Comparisons and analysis of field deployments are made to validate the effectiveness and performance of the proposed method. Index Terms—shallow subsurface imaging, high-frequency components, seismic interferometry, infrastructure, I. I NTRODUCTION S HALLOW subsurface imaging is of great importance for understanding underground infrastructures, especially in civil engineering [1]–[3]. The seismic ambient noise imaging methods leverage background noises, such as traffic, railways as well as natural sources like long-distance earthquakes, as sources [4]–[6]. Utilizing the source energy, depth information is extracted from the dispersive frequency-phase velocity curves associated with the surface-wave propagation [7]. Compared with the intrusive investigation using borehole survey, passive surface wave methods have gained increasing popularity recently due to their non-intrusive nature [8]. Based on the Rayleigh wave analysis, the spatial autocorrela- tion (SPAC) method was proposed by Aki [9], which has been generalized in [10] and restated in [11]. The SPAC method extracts the scalar seismic velocity irrespective of the number of sources and the azimuths of the surface waves [12]. Also, SPAC methods have a higher resolution than f − k methods [13]. Cho et al. [10] proposed to use power-special densities to obtain a higher resolution than SPAC. The research is partially supported by NSF-1663709 and Southern Company. F. Li, M. Valero and W. Song are with the Center for Cyber-Physical System, University of Georgia, Athens, GA 30602, USA, e-mail: (fangyu.li@uga.edu, maria.valero@uga.edu, wsong@uga.edu). Y. Cheng is with the Department of Earth Sciences, University of Southern California, Los Angeles, CA, USA, e-mail: (chengyif@usc.edu). T. Bai is with Department of Geoscience, University of Wisconsin–Madison, Madison, WI 53706, USA (e-mail:tbai4@wisc.edu). To image shallow subsurface, high frequency information should be considered. As stated in [14], high frequency surface waves have short wavelength, only travel in a shallow depth, and are very sensitive to shallow infrastructures. However, because of the sparse deployment of traditional seismic arrays, high-frequency components decay a lot between the geophone stations, and cannot be used for seismic imaging. Taking advantage of the merging sensor network technique with a dense sensor array [2], [4], [7], [15]–[17], we can obtain seismic data with more adequate high frequency components for shallow subsurface imaging with an improved resolution. In [18], a deployment with the high spatial resolution was adopted for urban structure analysis thanks to the dense seismic array. Moreover, another ambient noise based method- interferometry has also been used to obtain 1D velocity model in Japan [19] and the depth of bedrock in Singapore [8]. Fig. 1: Shallow infrastructure imaging based on distributed sensor networks. The surface sensor network performs the ambient noise imaging to map velocity variances caused by underground infrastructures, e.g. pipeline. However, current shallow subsurface imaging approaches typically involve manual collection of raw seismic data from sensors to a central server, as well as the post-processing and analysis. Thus, they do not have the capability of obtaining information in a real-time and automatic manner. Sensor network technology is a distributed approach with the ubiquitous in-situ computing and real-time processing ability [2]. Thus, distributed sensor network (DSN) based shallow subsurface imaging can be used to capture the temporal evolution processes, such as water leakage. We can implement the imaging system in a time-lapse manner, which has been widely used in monitoring velocity variations caused by underground mining [20], volcano eruptions [21], oil/gas