Tailings dams — engineered impoundments holding the waste slurry from mining operations — represent one of the most consequential geotechnical hazards in resource-rich nations. The 2019 collapse of Dam I at Brumadinho, Brazil, killed 270 people, released nearly ten million cubic metres of tailings into a major river system, and underscored a troubling gap: the dam had passed multiple safety inspections in the months preceding failure, yet satellite radar (InSAR) data later showed it had been deforming for at least eighteen months beforehand.
This project develops and validates a framework that transforms raw satellite displacement time-series into operationally actionable early-warning alarms, with no requirement for prior failure events to calibrate against. Rather than searching post-hoc for patterns that fit a known outcome, the framework is grounded in the physics of how complex systems approach catastrophic transitions — the same thermodynamic principles that describe critical slowing-down in ecological, climatic, and materials systems. Three complementary unsupervised detectors, each targeting a physically distinct precursor regime, are combined into a single ensemble that provides complete coverage across all six monitored locations at Brumadinho, with worst-case lead times of eight days or more. A reproducible codebase, a 32-test validation suite, and a pre-registered cross-dam transfer protocol provide the scientific infrastructure for extending the framework to new dam sites.
Aim
The project aims to establish a physically motivated, parameter-transparent, and operationally deployable precursor detection framework for InSAR-monitored tailings dams. Specifically, it aims to:
Demonstrate that multi-coupled thermodynamics (MCTE) observables — cross-timescale Pearson correlation and Laplacian spectral distance — encode physically meaningful precursor signals in satellite displacement data that classical statistical indicators miss or detect late.
Show that three unsupervised detectors operating on a shared twelve-dimensional feature substrate are sufficient to cover the full range of precursor regimes observed at a single dam site, without hyperparameter tuning on the target event.
Provide a rigorous sensitivity and false-alarm characterisation that goes beyond lead-time reporting, enabling practitioners to select an operating point matched to their risk tolerance and monitoring context.
Establish the methodological case that single-event InSAR precursor inference is intrinsically an unsupervised problem, and document why supervised regression approaches fail structurally in this setting.
Objectives
Building on the validated Brumadinho framework, the next phase of the project will pursue the following concrete objectives:
Execute the pre-registered cold-transfer test on a second tailings dam collapse event with no parameter re-tuning, using a frozen protocol with explicit falsification conditions.
Extend the cold-start data-acquisition pipeline — which reconstructs per-cluster displacement time-series directly from Sentinel-1 satellite products via an automated processing chain — to a prospective monitoring mode, enabling near-real-time feature extraction and alarm generation at active facilities.
Develop the Hotelling-persistent operating point (false-alarm rate under 3% on a realistic stable-dam null) into a deployable high-stakes alarm module, and validate its specificity on at least two additional dam datasets.
Quantify the spatial-consensus fusion rule’s performance across multiple sites, establishing whether within-dam spatial coherence of deformation is a reliable specificity lever in settings where only two or three pixel clusters are available.
Engage with regulatory bodies and tailings facility operators in Australia and Brazil to map the framework’s detection thresholds against existing trigger action response plans, ensuring the alarm taxonomy is operationally interpretable.
Significance
Tailings storage failures are among the most destructive industrial accidents globally, and their frequency has not decreased despite improvements in inspection practice. The Brumadinho collapse — and the near-simultaneous Feijão disaster — demonstrated that conventional inspection regimes can miss deformation-driven failure modes that are already visible in freely available satellite data. The scientific and humanitarian significance of reliable automated early warning is therefore high.
This project is significant for three reasons. First, it addresses a genuine methodological gap: no existing framework combines physically motivated multivariate features with principled unsupervised detectors and a rigorous false-alarm characterisation for this application. Second, it produces a reproducible, open-source tool that lowers the barrier for operators of small and mid-sized facilities who lack in-house InSAR expertise. Third, it provides a template — multivariate physical features plus complementary threshold-principled detectors — that transfers directly to other geotechnical monitoring problems: open-pit slope stability, underground mine subsidence, and critical infrastructure settlement.
For Curtin University, which operates in one of the world’s most active mining jurisdictions, the ability to contribute a peer-reviewed, pre-registered, and openly reproducible dam safety framework has direct relevance to industry engagement, graduate training, and research translation.
Ideal Candidate
We are looking for a self-motivated PhD candidate with strong quantitative and computational skills, including proficiency in Python and familiarity with time-series analysis, signal processing, or statistical machine learning. A background in one or more of the following is highly desirable: geophysics, remote sensing, geotechnical engineering, applied mathematics, or computational physics. Experience with satellite InSAR data processing or unsupervised anomaly detection methods is an advantage but not required. The candidate should demonstrate excellent problem-solving ability, scientific rigour, and capacity for independent research. Additionally, the applicants should meet the eligibility criteria for entry into a PhD program at Curtin University.
This project is open to International and Domestic applicants.
Internship
This project includes an industry internship with WSP, a global engineering and professional services firm, co-funding research into tailings dam stability monitoring. The intern will work alongside WSP engineers to apply and extend the satellite-based precursor detection framework to real active facilities, integrating complementary ground-based data streams including GPS displacement measurements and seismic monitoring networks. This placement provides the candidate with direct exposure to industry dam safety practice, regulatory engagement, and the translation of research outputs into operational monitoring workflows — bridging academic research and real-world geotechnical risk management.
Scholarship
If you are identified as the preferred candidate for this project, you may be considered for an RTP scholarship.
Enquires and How to Apply
For enquires about this opportunity contact Professor Victor Calo at Victor.Calo@curtin.edu.au
To formally apply submit an Expression of Interest to Professor Victor Calo during the Central Scholarship round (July 1st – July 31st 2026)