15:30 〜 17:00
[HDS05-P12] An Exploration of Atmospheric Reanalysis Usage to Assess Internal Erosion Susceptibility
キーワード：残渣ダム、浸透流
Mining Tailings Storage Facilities (TSFs) store enormous amounts of waste products known as tailings: finely ground rock particles, chemicals, minerals, or liquid. Tailings need to be kept safely and securely, but TSFs have failed worldwide and caused catastrophic disasters such as environmental pollution, ecosystem damage, and loss of life.
Nevertheless, recent studies suggest the frequency and severity of tailings dam failures are increasing. TSF failures are caused by overtopping, slope instability, earthquakes, seepage, structural reasons, defective foundation, erosion, or subsidence (listed in descending order of frequency). About one-fourth causes remain unknown. Water in the air, soil, and ground directly or indirectly contributes to TSF failures. Consequently, hydrological monitoring is also vital for failure prevention. The most common cause is overtopping, and heavy rainfall induces the risk. We also consider long-term precipitation influence. Slope instability, the second most common cause, is often led by partial saturation. Seepage – water flowing from one place to another via porous material – is the fourth common cause. A study shows that slope instability and seepage caused 14 % and 8 % of TSF failures from 1915 to 2016, respectively, and the summation exceeds the ratio of overtopping failures, which is 21%.
While in-situ measurement is indispensable for reliable soil saturation or seepage monitoring, modern soil-related datasets at a global scale can advance the monitoring technology. This study explores the potential of atmospheric reanalysis usage to locate where subsurface deterioration can happen. The strengths of atmospheric reanalyses, which are generated from millions of observations and numerical models, are their coverage and spatial and temporal resolution.
Initially, a hydrological condition analysis was conducted using atmospheric reanalysis near recent TSF failures. A few cases showed large seasonal variabilities of soil moisture without preceding heavy rainfall events. Then, it is hypothesized that the TSF foundation decays easier if the weathered layer has a greater volume of expansive soil. The expansive soil generates many cracks by undergoing a dry-wet cycle, and its strength attenuates. Accordingly, soil with the following characteristics is likely to be susceptible to internal erosion: larger seasonal variabilities of soil moisture, more frequent dry-wet cycles, and a thicker weathered layer.
This research uses ERA5, one of the atmospheric reanalyses. Since the spatial resolution is a 0.1-degree grid (approximately 11 km), it is possible to examine the vicinities of failed TSF sites. We can classify subsurface states estimated from "total precipitation" and "volumetric soil water layer 4" (100-289 cm) using machine learning techniques. Since there is a poor correlation between the two variables, and time-lagged analysis did not improve the correlation coefficients, they are practically independent.
After applying an unsupervised machine learning technique over randomly sampled data, six clusters are determined. The fifth cluster, which is 5.8% of the total, shows more moisture at 2m depth, small coefficients of variation of soil moisture, and rainy climate conditions. While anomalously wet conditions do not appear in this cluster, their yearly dry-wet cycles are distinct. The hydrological parameters near seepage failure sites (14 cases) are examined as well. The results are similar to the fifth cluster's pattern and indicate that the hypothesis is valid.
The lack of proper investigation before TSF construction is known as one of the reasons for foundation problems. Extending this study, we can develop an indirect examining technique with modern geophysical data. That can mitigate the initial site assessment issue.
Nevertheless, recent studies suggest the frequency and severity of tailings dam failures are increasing. TSF failures are caused by overtopping, slope instability, earthquakes, seepage, structural reasons, defective foundation, erosion, or subsidence (listed in descending order of frequency). About one-fourth causes remain unknown. Water in the air, soil, and ground directly or indirectly contributes to TSF failures. Consequently, hydrological monitoring is also vital for failure prevention. The most common cause is overtopping, and heavy rainfall induces the risk. We also consider long-term precipitation influence. Slope instability, the second most common cause, is often led by partial saturation. Seepage – water flowing from one place to another via porous material – is the fourth common cause. A study shows that slope instability and seepage caused 14 % and 8 % of TSF failures from 1915 to 2016, respectively, and the summation exceeds the ratio of overtopping failures, which is 21%.
While in-situ measurement is indispensable for reliable soil saturation or seepage monitoring, modern soil-related datasets at a global scale can advance the monitoring technology. This study explores the potential of atmospheric reanalysis usage to locate where subsurface deterioration can happen. The strengths of atmospheric reanalyses, which are generated from millions of observations and numerical models, are their coverage and spatial and temporal resolution.
Initially, a hydrological condition analysis was conducted using atmospheric reanalysis near recent TSF failures. A few cases showed large seasonal variabilities of soil moisture without preceding heavy rainfall events. Then, it is hypothesized that the TSF foundation decays easier if the weathered layer has a greater volume of expansive soil. The expansive soil generates many cracks by undergoing a dry-wet cycle, and its strength attenuates. Accordingly, soil with the following characteristics is likely to be susceptible to internal erosion: larger seasonal variabilities of soil moisture, more frequent dry-wet cycles, and a thicker weathered layer.
This research uses ERA5, one of the atmospheric reanalyses. Since the spatial resolution is a 0.1-degree grid (approximately 11 km), it is possible to examine the vicinities of failed TSF sites. We can classify subsurface states estimated from "total precipitation" and "volumetric soil water layer 4" (100-289 cm) using machine learning techniques. Since there is a poor correlation between the two variables, and time-lagged analysis did not improve the correlation coefficients, they are practically independent.
After applying an unsupervised machine learning technique over randomly sampled data, six clusters are determined. The fifth cluster, which is 5.8% of the total, shows more moisture at 2m depth, small coefficients of variation of soil moisture, and rainy climate conditions. While anomalously wet conditions do not appear in this cluster, their yearly dry-wet cycles are distinct. The hydrological parameters near seepage failure sites (14 cases) are examined as well. The results are similar to the fifth cluster's pattern and indicate that the hypothesis is valid.
The lack of proper investigation before TSF construction is known as one of the reasons for foundation problems. Extending this study, we can develop an indirect examining technique with modern geophysical data. That can mitigate the initial site assessment issue.