(1) Determination of reservoir boundaries

Using the Atmospheric Correction Module, it can be accurately compensated for atmospheric effects. In this study, the atmospheric correction using the FLAASH model [34,35,68] was performed and Landsat 8 images were fused using the Gram—Schmidt Pan Sharpening method [69–72]. After calculating the NDWI using the images after atmospheric correction and image fusion, zero was used as the segmentation threshold to extract the water body, and the manual editing was used to complete the extracted water body boundaries in the ArcGIS software. Figure 9 shows the boundaries of two reservoirs, Retiro Baixo and Três Marias.

**Figure 9.** (**a**) Retiro Baixo and Três Marias reservoirs. The larger reservoir on the left is Três Marias, and the smaller one in the lower right corner is Retiro Baixo (**b**).

(2) Sediment index used to estimate the level of sediment concentration

The sediment index was obtained using the SI calculating method. Figure 10 shows the pre- and post-disaster sediment concentrations. Contrasting the area of high sediment concentration where the river enters into the Retiro Baixo reservoir (Figure 10a–c), it had a larger area of 3.66 km2 compared with 1.89 km2 on 20 December 2018 and 2.49 km2 on 27 April 2019 with a 2-month interval spanning this disaster. It can be observed that the sediment plume had a great impact on the reservoir of the Retiro Baixo Plant, over 300 km from the failure location, while less impact on the reservoir of the Retiro Baixo Plant. This result is consistent with Vale's evaluation [73]. It appeared that small river gradients and obstruction of the reservoir barriers played an important role in slowing down the tailings waste moving into the São Francisco River. It is not hard to find out that this disaster event had little impact on the Três Marias reservoir (Figure 10d–f).

**Figure 10.** Level of sediment concentration of Retiro Baixo and Três Marias. (**a**) Level of sediment concentration of Retiro Baixo on 20 December 2018; (**b**) level of sediment concentration of Retiro Baixo on 22 February 2019; (**c**) level of sediment concentration of Retiro Baixo on 27 April 2019; (**d**) level of sediment concentration of Três Marias on 20 December 2018; (**e**) level of sediment concentration of Três Marias on 22 February 2019; (**f**) level of sediment concentration of Três Marias on 27 April 2019.

#### *4.4. Tailings-Dam-Failure Trend Analysis Based on Spatiotemporal Pattern Mining*

Using the emerging spatiotemporal hot spot analysis method, it can be found that the Brumadinho dam disaster in Brazil belongs to the Consecutive Hot Spot area (Figure 11). This disaster happened 3 years and 2 months after the Mariana dam disaster (5 November 2015), which was considered the worst environmental disaster in Brazil [74,75]. The Brumadinho and Mariana dam disasters both occurred in Minas Gerais, Brazil, and the two dams were both owned by Vale, a Brazilian multinational corporation engaged in metals and mining. Furthermore, based on the world mine tailings failure records [14], tailings dam failures have been recorded several times in this area. As a result, this area belongs to the Consecutive Hot Spot area, and the risk of tailings dam failure in this area would be relatively high if the necessary pond's management and the engineering safety measures were not carried out.

**Figure 11.** Spatiotemporal hot spot analysis of tailings dam failures. The location of Brumadinho dam disaster is marked with red circles.

According to Figure 11, New Hot Spots appear in Africa and South America which belong to developing regions, where the mining industry has been an important economic pillar in recent years [76,77]. Consecutive Hot Spots mainly lie in Eastern South America and Western Pacific islands where there are a lot of tailings left by mining, but due to poor management, tailings failure is easy to occur. The Sporadic Hot Spot is in Southwest South America and Oscillating Hot Spots mainly lie in Asia and America. China has many Oscillating Hot Spots and there have been some particularly serious tailings failure disasters, such as the 8 September 2008 dam break accident in Shanxi [78]. Different hot spots have different characteristics of disaster occurrences. These characteristics can be influenced by the mining history, mining features (e.g., man-made or natural), etc. [79–81]. The recommendation is that different hot spots should be treated differently. The areas of New Hot Spots and Consecutive Hot Spots are the ones that deserve the most attention. Local governments should adopt appropriate risk management strategies to monitor and change the trend. The risk assessment and monitoring of tailings reservoirs should be adopted and implemented. In this regard, some risk assessment and monitoring methods of mountain disasters can be used as references [82–85].

#### **5. Discussion**

#### *5.1. Cause Analysis of This Disaster*

5.1.1. Lack of Stability Management during the Maintenance Stage

Some experts believed that Brazil's weak regulatory structures and regulatory gaps allowed the dam's failure [86]. This dam was built in 1976 using the "upstream" method, in which coarse rubble, compacted soil, and dried tailings were used to build the dam (Figure 12). This is similar to the Fundão dam which failed in November 2015, killing 19 people and causing an environmental catastrophe, compared with a more expensive and strong method using solid rocks to contain the waste. The water leak was first observed near its base in July 2018, and then repairs were carried out [87]. José de Gouveia, the worker of Vale, said that the dam exhibited a small leak soon after the rainy season, and leaking water was observed in several places at the bottom [87]. The possible pore pressure build-up would have resulted in a decrease in effective stress and initiated a failure.

**Figure 12.** Cross section of Brumadinho dam "Barragem I" from the west to the east [88,89].

#### 5.1.2. Extreme Weather Effects

In this section, weather effects are examined. Daily precipitation data in Brumadinho were obtained from the World Weather Online website [90]. It appeared that precipitation increased the water content of the tailings pond before the dam failure event, increased the pore pressure, and thus induced failure initiation, which could be an important triggering factor for the tailings dam collapse (Figure 13). Although there was less rainfall in January than in February and March 2019, the rainfall in January 2019 sometimes reached the peak of monthly rainfall in some years from the perspective of multi-year rainfalls (Figure 14).

**Figure 13.** Daily precipitation in Brumadinho during the period from January to March 2019 [90].

**Figure 14.** Monthly precipitation in Brumadinho during the period from 2009 to 2019 [90].

El Niño and La Niña events are complex weather patterns resulting from variations in ocean temperatures in the Equatorial Pacific [91]. Their circulation is a global scale climate oscillation. Generally, the impacts of most El Niño events include above-average rainfalls in southeastern South America, eastern equatorial Africa, and the southern USA [92]. Extreme weathers in Brumadinho are linked to El Niño conditions from September to January typically [93]. A major El Niño event has been recorded since October 2018, and there were unusual rainfalls before the disaster (Figure 14) [94]. The increase of soil water content created conditions for the dam instability.

#### *5.2. Lessons Learned and Perceptions about Safety Management of Tailings Ponds*

The catastrophic tailings dam failure disaster that occurred in Brumadinho, Brazil is worthy of reviewing the safety management of tailings ponds in order to reduce the occurrence of such incidents. Tailings pond management has the characteristics of a heavy task and wide involvement, and requires the cooperation of different departments and joint law enforcement. The participation of multiple departments also illustrates that tailings pond management is a comprehensive and complex task involving disaster, development, industry, finance and taxation, resources, ecology, water conservancy, meteorology, and other fields [95–97]. At the same time, the successful completion of this task requires interdisciplinary integration and close cooperation of the related fields. Based on the above considerations, the following recommendations are made as a reference:


pond is stopped, the follow-up treatment of tailings, the ecological restoration, and land use planning of the mining area after the tailings treatment [108,109]. The above plans finally form a "whole chain" plan, and the system engineering theory would be introduced into it. On the basis of interdisciplinary integration, the overall optimal operation of the tailings pond system could guide the entire life cycle of the tailings pond. During the management of tailings ponds, the problem of tailing pond failures is the most important part. Such failures are mainly related to geotechnical engineering. Engineering measures should be taken to evaluate and control the water content in the reservoir and pore water pressure. To evaluate the slope stability a range of geotechnical datasets are necessary, including tailing granulometry, hydraulic conductivity, effective porosity, water level in the pond, and pore pressure under meteorological stresses. Based on these datasets, hydraulic simulations, followed by slope stability simulations will lead to the establishment of design requirements.


#### **6. Conclusions**

Tailings reservoir materials are easy to cause harm to the environment, and damfailure disasters often occur in production mining areas where there are population and production equipment. Therefore, the damages caused by such disasters are often more serious than others, and more attention should be paid. In this study, we carried out a disaster investigation of the Brumadinho tailings failure event. A detailed analysis of the Brumadinho tailing dam failure disaster was carried out using medium to high-resolution satellite images covering the entire affected areas of the event, including the place where the disaster occurred to the transport of tailings in the river and its impact on downstream reservoirs. Especially, the research of the diffusion of sediment in the reservoir was done in order to assess the impact of tailing waste, and discuss whether the waste reached the dams of two hydroelectric plants: Retiro Baixo and Três Marias or even the São Francisco River. Different from those caused by common landslides and debris flows, the disasters caused by the tailings dam failure are more serious and could affect larger areas due to the tailing waste pollution, and they should be paid more attention. On the other hand, the impacts of climatic factors on this event were also discussed in order to make people pay attention to the relationship between extreme weather events and nature disasters. Most importantly, the temporal and spatial characteristics of tailings dam failure were analyzed and summarized by building a global tailings dam failure database. The following conclusions are drawn:


**Author Contributions:** Conceptualization, Y.C.; methodology, D.C. and Y.C.; formal analysis, D.C.; supervision, Y.C.; validation, Y.C. and Z.L.; writing—original draft, D.C. and Y.C.; writing—review and editing, Z.L. and J.I.; funding acquisition, Z.L. and Y.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Natural Science Foundation of China (reference numbers 41941019 and 42077238), and the Chinese Academy of Sciences through the International partnership program (reference number 131551KYSB20160002). Part of this work was also supported by the Shaanxi Province Science and Technology Innovation team (reference number 2021TD-51), the Fundamental Research Funds for the Central Universities, CHD (reference numbers 300102260301 and 300102261108), and by the European Space Agency through the ESA-MOST DRAGON-5 project (reference number 59339). The authors are grateful for the Key Research Program of Frontier Sciences, CAS (grant number QYZDY-SSW-DQC006), and the Chinese Academy of Sciences President's International Fellowship Initiative (grant number 2021VCB0003).

**Data Availability Statement:** Our research data are from relevant open data websites, which can be obtained according to the links listed in our references.

**Acknowledgments:** The authors would like to thank all colleagues who participated in this study.

**Conflicts of Interest:** The authors declare no conflict of interest.
