# Context

## Introduction

The presence of induced seismicity at geothermal power generation sites has been recognized for decades (Ward, 1972; Allis, 1982). Commonly these events are caused by temperature and pressure change within a reservoir as a result of fluid injection, although seismicity has also been attributed to reservoir volume changes and changes in fluid chemistry (Allis, 1982; Sherburn et al., 2015). Most of this seismicity is of magnitude <3.0, termed microseismicity, and normally presents limited hazard to local population and infrastructure, although the degree to which humans are affected varies considerably from location to location (Cladouhos et al., 2010). Importantly, microseismicity can provide very useful information about the movement of fluid and pressure within the reservoir as well as size and distribution of the reservoir fracture system and therefore has implications for geothermal resource management. However, the small magnitude of the events, high levels of anthropogenic noise and, potentially, highly attenuating geology often make detecting such events difficult. One way to address these difficulties is to use a matched filter detection technique.

Matched filter earthquake detection uses waveform cross-correlation between continuous seismic data and known earthquake recordings to identify additional events in a seismic catalogue. Correlation-based detection offers improved performance over traditional, amplitude-based techniques due to its ability to detect signals in noisy data and when multiple, repetitive events are closely spaced in time (Gibbons et al., 2006). This significantly increases the number of events detected without increasing the rate of false detections. These advantages make matched filter detection ideal for monitoring microseismicity in areas of geothermal power generation, which are characterized by numerous noise sources and the possibility for dense clusters of small-magnitude, induced seismic events. The purpose of this study is to investigate the performance of matched filtering on a four-year dataset for Ngatamariki and Rotokawa geothermal fields on the north island of New Zealand, focusing on increasing the number of detections triggered using standard methods and assessing what any additional detections might contribute to our knowledge of the processes at play within the reservoirs. The specific objectives are as follows:

## Objectives

• Objective 1: Perform matched filter earthquake detection on the full 2012-2015 Mercury seismic dataset

• Compare rates of matched filter detection with Mercury power plant operations, especially injection

• Locate and double-difference relocate detected events

• Characterize location and extent of microseismicity in the context of well locations and rates of injection with time. How do the locations relate to what we know about fluid migration within the reservoirs?

• Objective 2: Determine source parameters, including magnitude and focal mechanisms, of matched filter detections

• Relate source parameters to reservoir processes (i.e. increased fluid pressure, thermal contraction, subsidence)

• Objective 3: Perform subspace earthquake detection on the full 2012-2015 Mercury seismic dataset

• Compare subspace detection results with results of matched filter detection and assess the performance of both

# Geologic and geophysical setting

Both the Ngatamariki and Rotokawa geothermal fields are located in the southern Taupo Volcanic Zone (TVZ) on the North Island of New Zealand, approximately 15 and 17 km north of the town of Taupo, respectively (Figure \ref{Figure1}). Rotokawa is a high-temperature (>300°C at 1–2.5 km below sea level) reservoir with a roughly circular footprint measuring approximately 6 km across (Sherburn et al., 2015). Ngatamariki, also a high-temperature system at >280°C and similar in depth to Rotokawa, measures roughly 7 km$$^{2}$$ (Chambefort et al., 2016). Sequences of andesites, rhyolites and volcaniclastic sediments overlie greywacke basement at roughly 3 km below sea level at both fields, although lateral heterogeneity of these units is considerable as is the depth to basement (McNamara et al., 2016; Chambefort et al., 2014). The geological structure at Rotokawa is dominated by three faults cutting through the field, which have been modeled based on offsets in well cuttings of the basement greywacke and Rotokawa Andesite (Wallis et al., 2013; McNamara et al., 2016). These faults, from West to East, are the Production Field Fault (PFF), Central Field Fault (CFF) and Injection Field Fault (IFF) (Figure \ref{Figure1}). At Ngatamariki, two important features dominate the local geological structure. The first is the Aratiatia Fault Zone in the southern end of the field (Figure \ref{Figure1}) and the other is a shallow (<2 km) intrusive body, the presence of which was confirmed by drill cuttings from wells NM04, NM08 and NM09 (Chambefort et al., 2014). Given the low-permeability nature of much of the host rock, reservoir permeability at both fields is thought to be dominated by fractures and faults within the reservoirs along the prevailing NE-SW structural trend (McNamara et al., 2016).

\label{Figure1} Map of the Ngatamariki and Rotokawa geothermal fields, North Island, New Zealand. Red triangles are seismograph sites and green dots represent template events used in the matched filtering process. Black lines represent faults from the GNS Active Faults Database (Langridge et al., 2016), unless labeled, in which case they represent the inferred faults mentioned above as published in previous works (Wallis et al., 2013; Sherburn et al., 2015).The goldenrod boundaries are the Boseley