MACHINE LEARNING INVESTIGATION OF INJECTION-SEISMICITY IN ROTOKAWA GEOTHERMAL FIELD

Aug 5, 2026

Understanding the injection–seismicity relationship in geothermal reservoirs helps characterize reservoir connectedness. Fault and reservoir complexity in real fields limit the applicability of simple analytical models. This study applies time-series feature engineering with machine learning to link fluid injection signals to microearthquakes in New Zealand’s Rotokawa geothermal field, using four years of injection data (2012–2016), windowing, and tsfresh-derived features.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Microseismicity in geothermal operations

Theoretical and modeling studies of induced seismicity

Challenges in forecasting injection–seismicity links

Role of machine learning in geoscience

Gap: ML correlations with injection-well microseismicity

MACHINE LEARNING INVESTIGATION OF INJECTION-SEISMICITY INROTOKAWA GEOTHERMAL FIELD Pengliang Yu1, David Dempsey2, Aimee Calibugan3, Rosalind Archer1 1Department of Engineering Science, University of Auckland, Auckland, New Zealand 2 Department of Civil and Natural Resources Engineering, University of Canterbury, Christchurch, New Zealand 3 Mercury NZ, Rotorua, New Zealand [pengliang.yu@auckland.ac.nz](pengliang.yu@auckland.ac.nz) Keywords: Microseismicity, injection, time series feature engineering, seismicity rate, significant features, p-value ABSTRACT Understanding the injection-seismicity relationship in geothermal reservoirs can provide insight into reservoir connectedness. One challenge is that, in real fields, fault and reservoir complexity make it difficult to apply simple analytical models to understand the data. Here, we use a machine learning technique called time-series feature engineering to study relationships between aspects of fluid injection and microearthquakes in Rotokawa geothermal field, New Zealand. We took four years of injection data between 2012 and 2016 and sliced it into smaller sub-windows. For each window, the average seismicity in a look-back period was computed, and then binary label of 1 was assigned if it exceeded a threshold. Automatic time series feature extraction from the raw and transformed injection data in each window was performed using Python package tsfresh. Significant features of the data were identified on the basis of distribution discrepancy between the two labels. The results show that the injection rate at some wells is a predictor of longterm (fortnightly) earthquake rates. At other wells, there is a poor correlation between injection rate and seismicity. We have been unable to find any link between rapid changes in injection rate and seismicity spikes, as suggested by some theoretical models. 1. INTRODUCTION Microseismicity is a common phenomenon resulting from hydrothermal fluid circulation in geothermal operations (Giardini, 2009; Hopp et al., 2020; Majer et al., 2012) . These small earthquakes are generally only detectable by sensitive networks, although occasionally felt events can occur. The locations, rates and magnitude frequency statistics of the microearthquakes carry information about pressure changes (Dempsey & Suckale, 2016) in a geothermal system and, by extension, its relative connectedness. However, this information is often underutilized, as it is difficult to interpret. Numerous studies have investigated the mechanisms of induced seismicity and explored links with fluid injection. It is widely understood that injected fluids increase pore pressure and decrease the effective stress on critically stressed faults (Healy et al., 1968; King Hubbert & Rubey, 1959) and this can promote fault slip that is detectable as earthquakes. If well-located, the seismicity can reveal faults that may operate as fluid conduits or baffles in the system. Segall and Lu (2015) used analytical and numerical models to explain the reasons for possible post-injection seismicity, and highlighted that a poroelastic surge may be the main reason for seismic events shortly after well shut-in. This was supported by the findings of Deng et al. (2020) who modelled wastewater disposal at multiple wells near the town of Cushing, Oklahoma and showed that poroelastic stress changes can affect the regimes on a preexisting fault where shear slip is promoted or inhibited. In contrast, Turuntaev (2018) used a rate-state model to estimate earthquake rates at the Basel geothermal project in Switzerland and showed only a small increase after shut-in. Dempsey and Riffault (2019) developed analytical and numerical models of seismicity rate changes when injection rates are reduced and found a decline, occasional quiescence, and eventual recovery of the seismicity rate to a new equilibrium after a rate reduction. These studies suggest the relation between injection and seismicity is

FAQs

What is the main goal of the machine learning investigation in this study?

The study aims to understand how fluid injection relates to microearthquakes in the Rotokawa geothermal field by learning predictive relationships from time-series injection data.

How was the injection data processed before feature extraction?

Four years of injection data from 2012 to 2016 were divided into smaller time windows, and seismicity rate metrics were computed for each window using a look-back period to create binary labels.

What key findings were reported about injection rate and earthquake rates?

Injection rate at some wells predicts long-term (fortnightly) earthquake rates, while other wells show poor correlation; additionally, rapid injection-rate changes were not linked to seismicity spikes as some theoretical models suggest.

Aug 5, 2026

Upload by Theodore