3.2.2. Waterbody Extraction

In order to carry out the research of suspended sediment information in the reservoirs, it is necessary to obtain an accurate extent of the reservoirs. Due to the water spectral characteristics of the near-infrared band absorbing strongly, but reflecting highly in the green band (Figure 5), the Normalized Difference Water Index (NDWI) was proposed by Mcfeeters [43] as follows:

$$\text{NIDWI} = \text{(Green } - \text{NIR)} / \text{(Green } + \text{NIR)} \tag{1}$$

where Green and NIR are reflectance factors in green and near-infrared bands, corresponding to Bands 3 and 5 of Landsat 8 imagery. After calculating the NDWI, we used 0 as the segmentation threshold to extract the water body, and the water body boundaries were manually extracted in the ArcGIS software.

**Figure 5.** Remote sensing reflectance (Rrs) of clear water (blue), water with chlorophyll content (green), and water with sediments (orange). The green, red, and NIR bands of Landsat 8 images are drawn in the above figure. Note that the figure is modified from Sherry's research [44].

#### 3.2.3. Suspended Sediment Detection

Remote sensing techniques have been widely used to measure qualitative parameters of water bodies [45], including turbidity [46], chlorophyll-a [47], Colored Dissolved Organic

matters (CDOM) [48], Secchi disk depth [49], and water temperature [50]. Here, the concentration of suspended sediment is chosen to examine the degree of waste pollution of Paraopeba River after the Brumadinho dam disaster. The suspended sediment is one of the most important water quality parameters, which directly affects the optical properties of water, such as transparency, turbidity, watercolor, and aquatic ecological conditions [51]. In particular, the level of water turbidity is dependent on the concentration of suspended sediment in the water body. With the increase of suspended particles, it is more difficult for light to travel through the water, and as a result the turbidity of the water increases accordingly. To date, many remote sensing quantitative models have been developed to monitor suspended sediments in water bodies, and several researchers used both the single and double band algorithms to calculate the concentration of suspended sediment of the water body [52–54]. The reflectivity of suspended sediment water is higher in the green and red bands (Figure 5). According to the above band reflectance characteristics, Wang et al. [55] proposed the concept of sediment index as follows:

$$\text{SI} = \text{(Green} + \text{Red)} / \text{(Green} / \text{Red)} \tag{2}$$

where SI is the sediment parameter, Green and Red are the reflectance in green and red bands, corresponding to 30 m resolution bands 3 and 4 of the Landsat 8 OLI image. Compared with the field-measured data, the correlation coefficient between the measured data and SI value is 0.89 [55], which shows that this method can directly and quantitatively reflect the relative concentration distribution of suspended sediments. The following indicators (Table 2) were used as the criteria to divide different suspended sediment water bodies (M represents the average, D represents the standard deviation, and MIN represents the minimum value):

**Table 2.** Level of sediment concentration in water bodies.


#### 3.2.4. Spatiotemporal Pattern Mining

Spatiotemporal pattern mining is often used to analyze data distribution and patterns in space and time. The emerging spatiotemporal hot spot analysis regards data cubes as input and identifies statistically significant hot and cold point trends over time. Using this method, the spatiotemporal hot spots of tailings dam failure database were analyzed. In this study, five main hot spots including New Hot Spot, Consecutive Hot Spot, Sporadic Hot Spot, Oscillating Hot Spot, and No Pattern were detected. Their definitions are listed in Table 3.
