Advanced analysis of satellite data reveals ground deformation precursors to the Brumadinho Tailings Dam collapse

The collapse of Dam I at the Córrego do Feijão iron ore mine complex in Brumadinho, Brazil, on January 25, 2019, was a catastrophic event. The dam, inactive since 2015 and built using the upstream method, contained approximately 11.7 million m³ of tailings. Despite passing stability tests months prior, the dam experienced a sudden and complete failure, resulting in over 200 deaths and widespread environmental damage. While the dam was monitored using various geotechnical techniques (survey monuments, inclinometers, piezometers, and ground-based radar), none indicated significant precursors to failure. This lack of warning highlights a critical need for improved monitoring strategies. The expert panel report attributed the collapse to a flow (static) liquefaction mechanism, a combination of internal creep and suction loss induced by heavy rainfall. Understanding the pre-collapse deformation is crucial for determining the cause and improving future monitoring frameworks for similar dams worldwide. This study aims to address this gap by analyzing ground surface displacements using satellite interferometric synthetic aperture radar (InSAR) data in the 17 months leading up to the collapse. The use of InSAR, particularly the advanced ISBAS technique, offers the potential for extensive coverage and millimetre-level accuracy, overcoming limitations of traditional InSAR techniques in vegetated areas. The freely available Sentinel-1 C-band SAR imagery provides a high-resolution temporal view of the dam's deformation.

Various InSAR techniques are used for land surface motion measurement, including monitoring tailings impoundments. Previous studies using InSAR on the Brumadinho Dam I revealed displacement rates up to 25 cm/year at the rear of the tailings beach and more subtle rates near the center. However, these studies primarily showed deformation consistent with consolidation settlement. The challenge in detecting precursory accelerating displacements was likely limited by the lack of measurements and accuracy degradation over vegetated surfaces, a common problem with conventional InSAR techniques due to poor coherence. This study uses the ISBAS technique to overcome this limitation and to achieve high spatial coverage and millimetre-level accuracy.

This study utilized freely available Sentinel-1 C-band SAR imagery from two overlapping descending orbit tracks (tracks 53 and 155) acquired between August 2017 and January 2019. The data were processed independently using the Intermittent Small Baseline Subset (ISBAS) technique, a coherent scatterer technique known for its ability to provide high-quality data even in vegetated areas. ISBAS is a multi-temporal stacking technique which helps to reduce the impact of atmospheric effects on displacement measurements. Average line-of-sight (LOS) rates of motion were derived for each track. Additionally, time-series analysis was conducted to investigate the nonlinear deformation history. The study compared the ISBAS results with those from a conventional Small Baseline Subset (SBAS) approach. A stereo analysis, using the overlapping tracks, was attempted to resolve the LOS motion into east-west and up-down components. Daily rainfall data from the nearest meteorological station were incorporated to assess the relationship between rainfall and deformation. Finally, inverse velocity analysis was performed on the time-series data to predict the timing of the dam collapse. This involved fitting a straight line to inverse velocity values to extrapolate the predicted collapse time. A sensitivity analysis assessed how early and reliably a correct collapse date could be predicted using successive InSAR observations.

The ISBAS analysis revealed widespread ground surface displacements across the tailings beach in the 17 months before the collapse. Both tracks showed consistent displacement patterns, with velocities ranging from -7 mm/year to 9 mm/year. The conventional SBAS approach, in contrast, provided significantly less coverage. Time-series analysis revealed nonlinear deformation patterns, with some areas exhibiting a sudden acceleration of deformation from December 2018, coinciding with increased rainfall. This accelerated deformation was not consistent with consolidation settlement. Locations at the back and front of the tailings beach and on the dam wall showed accelerated displacements in the period leading up to the collapse. The analysis of deformation using the inverse velocity method at various locations on the dam allowed for prediction of the collapse time, with the actual collapse occurring within the predicted time intervals. A sensitivity analysis indicated that a reliable prediction could have been made as early as 44-75 days before the actual collapse, depending on the specific location. A similar InSAR analysis of the Sul Superior dam at the Gongo Soco mine showed deformation consistent with consolidation settlement, indicating no imminent stability issues. This contrasts sharply with the anomalous deformation observed at Dam I prior to its collapse.

The findings demonstrate the potential of advanced InSAR techniques, such as ISBAS, for detecting precursory deformation in tailings dams. The observed anomalous deformation, not attributable to consolidation, strongly suggests that the Brumadinho dam collapse was potentially foreseeable. The ability to predict the collapse time interval, even with a relatively short period of observation, highlights the value of this monitoring approach. The contrast between the deformation patterns at Brumadinho Dam I and the Sul Superior dam emphasizes the importance of comprehensive monitoring considering the entire dam structure. The study highlights the limitations of traditional geotechnical monitoring methods that may lack spatial coverage and temporal resolution. The results support the use of satellite-based InSAR monitoring as a valuable tool for early warning systems in tailings dam management.

This study demonstrates the effectiveness of the ISBAS InSAR technique in detecting precursory deformation indicative of impending failure in tailings dams. The analysis of the Brumadinho dam collapse revealed anomalous deformation patterns that could have led to an earlier warning. The successful prediction of the collapse time interval highlights the potential of this methodology for improved dam safety. Future research should focus on refining prediction models, investigating the influence of various factors (such as rainfall and material properties) on deformation patterns, and developing integrated monitoring systems that combine InSAR with other techniques.

The study's analysis is limited by the availability of Sentinel-1 data, which has a specific revisit time and limitations in the ability to perform precise stereo analysis due to the small angular separation between the tracks. The accuracy of the predicted collapse time is also affected by the temporal resolution of the satellite data. While rainfall data is used, the study does not directly investigate the precise hydrological mechanisms influencing the deformation. The study focuses on two specific dams; further research is needed to confirm the generalizability of these findings to other dams and geological settings.