LifeSim

Brumadinho

Feijão Dam B-1 Tailings Storage Facility at Minas Córrego do Feijão mine in Brumadinho, Minas Gerais, Brazil.

🡇 Download Model

Overview

On January 25, 2019, the Feijão Dam B-1 Tailings Storage Facility (TSF) at Minas Córrego do Feijão mine in Brumadinho, State of Minas Gerais, Brazil failed catastrophically. A devastating downstream mudflow inflicted extreme to complete damage to many structures, vehicles (Figure), and transportation infrastructure, and 270 fatalities were caused. The Feijão Dam B-1 event provides a unique opportunity to demonstrate how LifeSim can be used to analyze a historic TSF failure event, to provide insights into how life loss can be estimated for these types of events, and to show how this approach could be used to inform consequences estimation and risk assessments for existing, or proposed, tailings storage facilities.

Figure: Car dragged by mud during Brumadinho dam failure. [?]

Data

The complete Brumadinho validation study analysis is documented in the Morrill-Winter and Johnstone (2025) abstract [?].

The following provides a short description of the data sources used for validation.

Examples of key sources include the Robertson et al. (2019) expert panel report [?], a book by the journalist Arbex (2022) [?], post-disaster investigations and support by Brazil’s Federal and State of Minas Gerais legislatures, ministries and agencies (Assembleia Legislativa do Minas Gerais, 2019) [?]; (Government of Brazil, 2019a) [?], (2019b) [?], reports from media outlets (e.g., Globo [2019]) [?], journal papers by Lumbroso et al. (2021) [?] who developed a post-failure model of the event using the HRW Life Safety Model and by Passos and de Freitas (2021) [?] who describe the immediate post-failure response and longer-term work by the Bombeiros of Minas Gerais to locate and recover the remains of the victims, and work by the Instituto Medico Legal in Belo Horizonte to organize, identify and return these remains (da Rocha, Silva, & da Silva, 2022; Quierati, 2019) [?] [?] and to determine causes of death (Araujo et al., 2022) [?].

Hydraulics

Four breach outflow and runout models of the event were identified by this study: a pre-failure model developed by Walm Engineering to inform Emergency Action Plan (EAP) development (Vale & Walm Engenharia, 2018a, 2018b) [?] [?], post-failure models developed by Lumbroso et al. (2021) [?] and by Gibson et al. (2022) [?], and more recent work by Adria et al. (2023) [?] to investigate the properties and arrival times of the runout downstream to the Rio Paraopeba. This study used the Hydrologic Engineering Center's River Analysis System (HEC-RAS) model developed by Gibson et al. (2022) [?] as the 100% (i.e., “historic”) event.

Structure Inventory

Web-based sources such as Google, Esri, and OpenStreetMap [?] were used to provide before and after base mapping and imagery of the full extent of the impact zone and surrounding region.

Structures and outdoor areas located in and around the tailings runout zone were compiled starting with a baseline dataset from Lumbroso et al. (2021) [?]. More structures, outdoor areas, and details were then added using pre-failure Google and Esri satellite imagery, and OpenStreetMap basemap sources. Multiple structure datasets were created to model variations in population sizes and distribution, number of stories, stability criteria and structure location.

Road Network

Development of the road and trail network started with a downloaded copy of the OpenStreetMap dataset [?]. Additional GIS editing effort was required to ensure that the network model had correct nodal connections at street intersections.

Destinations

A related GIS points layer was created to define the set of target evacuation destinations outside of the impact zone. This was based on Vale’s Plano de Ação de Emergência de Barragens de Mineração (PAEBM) (Vale & Walm Engenharia, 2018a, 2018b) [?] [?].

Emergency Planning Zones

A GIS polygon layer was created to define the Emergency Planning Zones (EPZs) that are used to vary the community warning and evacuation times. This allowed the LifeSim modelers to set up different ranges of warning times. Seven EPZ zones were created, each with validated Warning Issuance Delay, First Alert, and Protective Action Initiation curves applicable per zone.

Alternative

A total of 59 alternatives were created and grouped into twelve simulations that focused on key goals of the study for the sensitivity runs.

For the event validation alternative, to simulate the lack of an official warning, the Imminent Hazard ID Time was set to 1 hour after the breach. Zero Hazard Communication Delay is assumed.

Modeling Conclusion

The Feijão Dam B-1 Tailings Dam Failure case study successfully demonstrated how LifeSim can be used to develop a useful model of this historic event. The majority of the event validity life loss estimates encompass the observed consequences.

It was possible to characterize the population-at-risk (PAR) for a range of location types (e.g., residential, industrial, agricultural), for variations by time of day, ranges of TSF breach outflow scenarios, and to explore how variations in EAPs and community preparedness can influence life loss outcomes.

While the study confirmed the strengths of the LifeSim simulator and methodology, and no fundamental gaps were found, modifications and adjustments to existing methods could increase the value of the simulator for modelling TSF failure and runout events. This includes:

• Develop a better understanding of how tailings flows affect buildings, vehicles, and people via mechanisms such as adhesion, fouling, and deposition. Examples include more building damage, loss of transport (vehicles cannot move, vehicles stall), impacts on pedestrians, and compromised first-responder rescues (deposited tailings become barriers to emergency vehicles).

• Investigate how object (buildings, vehicles) stability calibrations can be adjusted to account for the physical properties of tailings flows, and whether there is a need to adjust the mortality rate curves.

• Add the “Fire Drill” Protective Action Initiation response pattern to represent well-trained industrial PAR.

• Since the Feijão Dam B-1 failure is only one of four types of Tailings Dam Breach Analysis (TDBA) event identified by the Canadian Dam Association’s (CDA) 2021 Bulletin, the LifeSim methodology (and possibly the simulator) may need to be extended to also support modeling TSF failure events that have both a supernatant pond and liquefiable tailings.

• Study other TSF incidents and failures with different characteristics such as: larger communities, larger PAR, the potential for earlier detection and warning, medium- to lower-intensity tailings flows, longer downstream runout distances, lower life loss rates, and successful evacuations.

Model Version

The original LifeSim model was developed in 2022 using LifeSim version 2.0. The public model was updated in 2025 to LifeSim version 2.1.5 and produced zero observed changes from the original documented results.