A series of facilitated multi-stakeholder workshops will be held quarterly at both case study sites. These interactions will be developed to foster multi-stakeholder governance (Utting 2001) model championed by the United Nations <> for managing risks in tailings storage. These workshops will be structured to address specific challenges, including a framework for communication between the different stakeholder groups, a formal commitment agreement for each stakeholder group, disaster management plans, information disclosure strategies, and formulation of the structure and functions of executive committees. These multi-stakeholder interactions are intended to encourage all dimensions of sustainable development as well as address the issues of disaster preparedness and decision-making processes through stakeholder collaboration. This will initiate a people-centred preventative approach to risk management. These multi-stakeholder interactions will be conducted for each of the two case studies, focusing particularly on the local context of each site. The workshops will be open to everyone. Specifically invited groups will include mine workers, mine owners, mine operators, local politicians, the local and national press, scientists and engineers working in domains relevant to each challenge area, and local residents in and near areas that could be affected by dam failures, tailings leachates or dust. The outcomes of these workshops will be generalised for future implementation across all tailings storage sites in Brazil. Based on discussions with Brazilian regulators, we anticipate these will become regulatory requirements. The planned research program will develop the body of evidence to justify this.

The early workshops will be focused on developing the framework for communication between the different stakeholder groups. This will make clear to participants how they can best request information from other stakeholders and to whom can they address their questions, concerns, or protests. The information disclosure strategies will also evolve during these early workshops supported by the framework for communication. This will initiate a partnership between the different stakeholders, listening to each other's views and sharing information that each wants and needs to make decisions. In particular, close attention will be paid to the capabilities and desires of the at-risk and most marginalised populations whose participation in the workshops will be subsidised. The commitment agreements will be the focus of the later workshops. They will seek to establish a voluntary code of practice for each stakeholder group and detail the ongoing commitments each party is prepared to make to the others.

Other workshops will focus on the enhancement of each dam’s disaster management plan, focusing particularly on an all-of-society approach to disaster response (<>). Finally, the structure and functions of an executive committee to manage and ensure the continued use and enhancement of the multi-stakeholder governance model will be developed by participants themselves. This framework is intended to ensure the full and meaningful participation of all stakeholder groups.

There are a number of socio-cultural challenges specific to Brazil that are factors contributing to the higher rates of industrial accidents and disasters. Migueles et al. (2007) have explored the deep anthropological roots of Brazil’s difficulties with operational discipline. These include short-termism exhibited by many managers because of a persisting commodity mentality, a hangover of Brazil’s history as a previous colony of exploitation. The hierarchical culture and long distances to power, endemic risk and uncertainty avoidance (Hofstede 2001), and low propensity to trust (Zanini 2016) are all factors contributing to poor risk management and awareness. Co-investigator Migueles will lead this work package seeking to improve our knowledge on how to implement adequate governance systems, using metrics to assess each of these factors throughout the project, she will seek to radically enhance the evidence base for such multi-stakeholder governance models in Brazil, and more broadly in developing nations. <>

Stochastic structural reliability analysis of the physical dam under its environmental conditions and given its known history (if available) can be used to estimate the chances of dam failure whose immediate consequences would severely impact human life, property, agricultural and environmental resources. Brazilian regulators are currently commissioning geophysics reports and geotechnical investigations including soil profiles, longitudinal piezometry and water levels in borings, “as-is” dam geometry and profile characterisation, and dam-break assessments of all Brazilian dams (including one of the case-study dams) in the state of Minas Gerais that surpass minimal risk criteria. These data and investigations are important but currently insufficient to the main goal of assessing the safety of tailings storage in dammed ponds.

We will conduct reliability analyses on the two dams’ physical vulnerability using pre-existing reliability software and site-specific data already collected by Brazilian collaborators, enhanced by the data and analyses untaken in the case studies. Advanced simulation approaches including finite-element modelling and boundary-element modelling, informed by site-specific field data, will be used in the case studies to identify failure modes and possible thresholds or limit states. Static and dynamic fluid–structure interaction loading and analyses, including susceptibility to liquefaction, will be used to predict slope stability thus determining the actual safety factors of the dams. These analyses will support the development of and provide validation for a more tractable, reduced-order (simplified yet reliable) model that can interpret local data already being collected by Brazilian authorities, but has previously had limited usefulness. Because this reduced-order model is general, it will be applicable to all upstream construction dams in Brazil. The simplified assessment approach will be implemented on the regulator’s existing data collection portal. This approach will partially automate the analysis, dramatically enhancing the regular data collected by the regulator. By making the collected data interpretable, it will provide guidance and contextualisation for relative risk ranking among dams across Brazil. The rankings will inform the regulator’s provision of resources. he on-line monitoring of structures will deliver augmented near-real-time situational awareness, empowering the regulator to respond to emergent conditions. This facility will also support the development of an early-warning system for dam failures that is independent of the dam operators. Under this work package a generic stochastic model of upstream construction dams and their failure risks will be developed that can be parameterised by the dam operators and evaluated by the regulators and which, in combination with the product of the next work package, will produce outputs understandable by local residents.

The routings and movement of tailings flows in the downstream after dam failure, with the distributions of personnel, assets and property in the inundation area will be explored with extensive field study on the two case studies to assess the immediate vulnerability of at-risk communities. Data collection exercises including site surveys, aerial LIDAR scans, demographic data and satellite image data will all be explored for their utility to develop a comprehensive vulnerability model for the two case studies. This will be developed along with a best-practice framework for data collection. The experience gained during the case studies will lead to guidance for cost-effective approaches to constructing a general model and collecting relevant data.

The risk assessment, built upon a GIS-based model with spatially explicit consideration of the local geography, will lead to accessible visualisations to support preventative decision making. These visualisations will be three-fold in their intended purpose. (1) The regulator ANM requires dam-break assessments for each dam, however, they lack the capacity to independently validate these submissions. The first intended use of the tools developed will be to provide ANM with a facility to independently validate dam-break assessments received from mining operators. The tools, like the dam reliability assessment tools, will be added to the regulator’s data collection portal, partially automating this process, empowering them to engage with operators and sharply enhancing their ability to hold them accountable. (2) The second intended use is to address the asymmetry of information between at-risk populations and the mine operators. The risk map visualisations will support at-risk communities to engage in the risk reduction and management process. Access to colour-coded maps of the likely inundation areas will provide qualitative and quantitative evidence to requests for preventive action such as resettlement of structures and communities. (3) Finally, the visualisations will provide a critical element of the mine worker training within this project. Brazil lacks a culture of whistleblowing and unsafe practices reporting. Disaster reduction requires an all-of-society approach, and site workers are an essential component in risk management. In order to instill a culture of equitable engagement of all stakeholders in risk management and disaster prevention, it is essential that mine workers appreciate their role. If presented with accessible visualisations of the possible consequences of a disaster, we believe this could become a driving motivator to initiate the required change.

We will quantify potential secondary impacts from mining operations which can be long-lasting and insidious. These include impacts on human health and agro-ecological systems affected by dam failures—but also runoff and leaks from dams—from transported environmental contaminants, including heavy metals, acid drainage, and organic toxicants which can create often severe and wide-ranging environmental problems downstream. Their impacts are in addition to any immediate fatalities or physical devastation caused by a dam failure, and they include both chronic and acute effects on aquatic life, herbivores (grazers) on affected uplands, and humans exposed incidentally or through food chains or water supply. In some case, locally increased rates of malaria and yellow fever have occurred. There is ample evidence of the potential magnitudes of these effects from tailings runoff and dam failures in Brazil and elsewhere (Cesar et al. 2013, Kříbek et al. 2014, Castilhos et al. 2015). Likewise, ecological risks on threatened and endangered species are of global significance because of the uniqueness of Brazilian ecosystems.

Once in the terrestrial system, toxic metals and organic toxicants can affect soil fauna and, consequently, can compromise important ecosystem services, including the reduction of soil erosion rates, soil fertility and the unfitness of vegetables for human consumption (Suthar and Singh 2008). Given the fact that some soil organisms are able to uptake high amounts of metals, such elements can be transferred through soil food web, causing important disturbances to soil ecosystems (Straalen et al. 2005, Suthar and Singh 2008). Last, but not least, the deposition of gold mining in soil can generate technogenic soils (anthroposols), affecting pedogenesis and natural soil biological activity (Naeth et al. 2012, Huot et al. 2015). The ecotoxicological implications behind the formation of technogenic soils derived from land disposal of hazardous materials are still poorly investigated in the tropics.

Bioassays with ecologically representative organisms have been used to predict metal noxious effects on soil biota, since they can complement metal determination in ecological risks assessments (Niemeyer et al. 2017). Earthworms (Eisenia andrei, Eisenia fetida) and collembolans (Folsomia candida) have been widely used in bioassays for evaluating the toxicity of mining wastes and metal-polluted soils (Niemeyer et al. 2010, Bianchi 2013, Cesar et al. 2013). In addition, Hinton and Veiga (2002) suggest that earthworms are good bioindicators to evaluate mercury contamination in gold mining areas. The earthworms consume high amounts of soils, are sensitive to toxicants in the soil, are abundant in tropical and temperate soils and play a relevant role in the terrestrial food web (Cesar et al. 2014, Sivakumar 2015, Usmani and Kumar 2015). Thus, the integration of ecotoxicological and geochemical data related to the deposition of mining wastes in soil and aquatic systems are often required to understand the ecological impacts related to the dam collapse. This kind of information is essential to (i) priorize the monitoring of hazardous dams and make prognosis for hypothetic scenario of a dam breaking; (ii) define the ecological receptors at most risk; (iii) identify key-contaminants that play a crucial role in the toxicity and (iv) help establishing action in emergency plans and to select the best strategies for remediation.

Quantifying the spatially explicit risks from such impacts is essential for giving agency to vulnerable populations and in regulatory considerations for ranking sites in terms of costs of remediation. Ecotoxicological assessment is necessary to reveal the potential impacts. These impacts depend on the chemical composition of the tailings and soils, which vary widely across sites. Quantitatively characterising risks to the environment depend on access to reliable ecotoxicological data as well as software tools to integrate this information. This work package will collect bioassay data which uses the presence of biological species to inform us of the toxicological properties on the materials and matrix. Substantial ecotox data are currently being collected (Cesar et al. 2015, Bidone et al. 2018), suitable for risk analysis using RAMAS software (Spencer and Ferson 2001). These results will inform spatially explicit risk assessments of possible vulnerabilities in the surrounding region. Stochastic, high-level fate-and-transport modelling based on fluvial mechanics, geomorphology of mudflows, and dust dispersion will be used to create multiple risk maps that will feedback to the risk analysis and to improve public awareness.

Risk does not tend to follow the traditional segregation of scientific disciplines. Disasters are never isolated events; they are a nexus of external drivers, human factors, and engineering failure. They only matter if they effect an exposed vulnerable populations. Safe tailings storage requires a multi-disciplinary, trans-sectoral approach to risk management. The risk management work package incorporates four research components (i) identification of the external drivers leading to accidents, (ii) safety thresholds relating to water penetration, the risk of liquefaction, and dam properties, (iii) local disaster response coordination and evacuation procedures, and (iv) strategies to deal with uneven scientific and engineering ignorance.

The identification of external drivers will explore the operational and environmental factors that increase the risk of major disasters in tailings storage. The Fundao and Brumadinho disasters will be used, along with the literature to identify the key physical and organisational indicators suggestive of increased risk using standard methods. The environmental drivers will be explored via modelling of the tailings pond catchment area, chemical and physical properties to better understand the correlations of interaction with the dam structure, and assessment of risk in extreme environmental scenarios. These will be explored from a historical, statistical perspective because complexity limits mechanistic understanding.

Water penetration and liquefaction of the dam structure are the primary accident drivers for dam structural failure <>. Co-investigator Braga is pioneering the use of electroresistivity to model the extent of water penetration into tailings dams, producing full-field 3D maps of the soil water content. The project will assess how the reliability of a traditional approach with limited piezometry samples and water level meters varies with sample size. A mechanistic study of the failure modes will be conducted including cyclic and static liquefaction processes in order to develop site-specific safety thresholds for water penetration. Site-specific thresholds for sensor failure or missing data, and thresholds for dam safety features such as the beach length. These thresholds can be used to enhance the proposed early warning system with validated risk-informed thresholds. The decision thresholds could be used to trigger evacuation or other disaster reduction strategies minimising the possible impacts of failures or preventing them entirely by triggering a suspension in mining operations that cause micro-seismic activity which is the trigger for liquefaction dam failure.

The regulator ANP has helped to develop the Plano Nacional de Contingência (national contingency plan), a comprehensive disaster management coordination plan. This plan for offshore installations includes a mobilisation plan of all necessary bodies in Brazil, and an organisational structure during the immediate recovery phase. Under the proposed project, ANP will undertake a knowledge transfer to ANM, leading to the development of a disaster coordination plan for tailings storage. In addition to the primary deliverable of this activity, further coherence across the two sectors will be sought to harmonise regulatory strategies and influence unified policy documents. Because both regulators fall under the Ministry of Mines and Energy, existing organisational structure supports the exchange of staff and knowledge. This structure will be of great benefit to this project and promises to enhance the impact of the research. The University of Liverpool and ANP have entered into an agreement to co-support a PhD student from ANP to conduct research aligned to this project. Their research will seek the enhance and generalise the relative risk ranking methodology developed in this project for tailings dams to offshore installations. They will enhance the current indices approach used by ANP to incorporate online system monitoring and data acquisition.

Finally, with more than 2000 dams across Brazil that surpass the minimum risk criteria of ANM, the challenge facing the regulator is vast. This issue is compounded by uneven ignorance across these sites, the wide discrepancy in the ease of access to the sites, inconsistencies in governance between states, and a number of different operators each working to different design codes and international codes of practice. There is a significant danger that some dams may be overlooked, where there is greater ignorance of scientific, engineering and demographic information, or at sites that are in remote regions and thus more challenging to access. It is therefore essential that all the tools developed for the regulator's data collection portal should be insensitive to ignorance, or lack of data. That is not to say they should make unjustified assumptions to handle this epistemic uncertainty. All the mathematical tools within the project will be developed around the principles of imprecise probability, allowing us to make conservative quantitative estimates without relying on tenuous or unjustified assumptions. These methods lend themselves to taking a value-of-information approach to analysis. The provision of the limited resources available to the regulators will allow them to optimise their investigation efforts where further information would be most useful.