Metallic ore mining causes heavy metal pollution worldwide. However, the fate of heavy metals in agrosystems with long-term contamination has been poorly studied. Dongchuan District (Yunnan, southwest China), located at the middle reaches of the Xiaojiang River, is a well-known 2000-year-old copper mining site. In this work, a survey on soil heavy metal contents was conducted using a handheld X-ray fluorescence instrument to understand the general contamination of heavy metals in the Xiaojiang River Basin. Furthermore, river water, soil, and rice samples at six sites along the fluvial/alluvial fans of the river were collected and analyzed to implement an environmental assessment and an evaluation of irrigated agrosystem. V, Zn, and Cu soil levels (1724, 1047, and 696 mg·kg−1, respectively) far exceeded background levels. The geo-accumulation indexes (Igeo) showed that cultivated soils near the mining sites were polluted by Cd and Cu, followed by Zn, V, Pb, Cr, Ni, and U. The pollution index (Pi) indicated that rice in the area was heavily polluted with Pb, Cr, Cd, Ni, Zn, and Cu. The difference in orders of metal concentrations between the soil and rice heavy metal contamination was related to the proportion of bioavailable heavy metals in the soil. The crop consumption risk assessment showed that the hazard quotient exceeded the safe threshold, indicating a potential carcinogenic risk to consumers. The Nemerow integrated pollution index and health index indicated that the middle of the river (near the mining area) was the heaviest polluted site.
Due to the complex mechanisms of rockburst, there is no current effective method to reliably predict these events. A statistical learning method, support vector machine (SVM), is employed in this paper for kimberlite burst prediction. Four indicators \(\sigma_{\theta } ,\sigma_{c} ,\sigma_{t} ,W_{\text{ET}}\) are chosen as input indices for the SVM, which is trained using 108 groups of rockburst cases from around the world. Data uniformization is used to avoid negative impact of differing dimensions across the original data. Parameter optimization is embedded in the training process of the SVM to achieve optimized predictive ability. After training and optimization, the SVM reaches an accuracy of 95% in rock burst prediction for validation samples. The constructed SVM is then employed in kimberlite burst liability evaluation. The model indicated a moderate burst risk, which matches observed instances of rockburst at a diamond mine in north Canada. The SVM method ignores the focus on rockburst mechanisms, instead relying on representative indicators to develop a predictive model through self-learning. The prediction results show an excellent accuracy, which means this method has a potential application in rockburst prediction. 相似文献
