In fragile karst environments that have seen past and current human exploitation of agricultural and forest resources, the quantification of underground riverine sediment has been widely used to evaluate subterranean stream basin erosion. These measures are highly influenced by both precipitation and anthropogenic factors; therefore, soil erosion control measures must be urgently designed and applied. In this study, 17 years of sediment discharge across the Nandong underground river system in southwest China was monitored. To achieve this goal, the Mann–Kendal mutation test and proxy indicators were used to estimate the general influence of human activities and precipitation on sediment discharge. The results showed that: (1) Both anthropologic disturbance and rainfall have impacted the sediment discharge, although the influence of the anthropologic factor on sediment discharge was greater (61.53%), and (2) rainfall showed a hysteresis effect on sediment discharge. We obtained three different stages based on the mutation points and variation characteristics of the studied sediment discharge resulting from different driving forces, from 1998 to 2014. Prior to 2004, in the whole basin, the decrease of sediment yield was the result of the Natural Forest Protection Project. During the period from 2004 to 2008, due to continuous droughts, flood disasters, and intensive cultivation practices on the steeper hillslopes, the total sediment discharge of the whole basin increased. After 2009, the sediment discharge decreased due to the development of soil conservation projects and mushrooming reservoirs. These findings are expected to provide insights into watershed management and ecological restoration in fragile karst ecosystems, specifically, in southwestern Chinese river systems. More research must be conducted to monitor, with in situ measurements and observations, possible extreme events that can determine the exact erosion control measures that need to be designed and applied.

Serious soil and water loss affects the economy and the living quality of the population in faulted basins. Since 2002, China has carried out large-scale ecological restoration projects in karst areas. However, the karst faulted basins have experienced complex climatic changes, which makes it difficult to assess the ecological restoration effects quantitatively. Using the improved Budyko model, basin factor (n) and sediment content factor (C) were divided into climate influence and ecological restoration influence by second-order approximation, and the change ratio of climate influence and ecological restoration on sediment yield and loss was quantitatively calculated in the representative basin of the Nandong Underground River System (NURS). This was carried out in order to effectively distinguish the effects of climate change and ecological restoration on runoff and sediment change in the NURS. Furthermore, this study sought to understand the mechanism of runoff and sediment change and to evaluate the ecological restoration in the faulted basin to provide a scientific basis for the next stage of soil and water conservation policy formulation. The results showed that: (1) Using Pettitt to test the abrupt change of water and sediment in the NURS from 1987 to 2018, we found that 2002 was the abrupt change year of water and sediment in the basin. Before and after the mutation, the runoff and sediment yield decreased significantly, with the change rates of −15.5% and −51.8%, respectively. The decrease in precipitation and the increase in E₀ were not significant, which were −4.5% and 1.4% respectively. (2) By calculating the correlation coefficient and the double logarithm function, it was found that the maximum temperature is the main climatic factor affecting the underlying surface factor n, and precipitation is the main climatic factor affecting the sediment concentration c. (3) Climate change reduced runoff by 77.6%. Ecological restoration has reduced sediment by 51.3 percent. The sensitivity of runoff to maximum temperature was 3.61. The sensitivity of grain yield to precipitation and NDVI was 5.37 and 3.26, respectively. The results show that climate is the main factor of runoff reduction, and ecological restoration is the main factor of sediment reduction. Ecological restoration has greatly reduced sediment production, and ecological engineering has made remarkable progress. However, the reduction of runoff caused by climate change should be paid more attention. The results of these studies will help to better formulate land use management policies for soil and water conservation.

Since 2008, soil and water treatment and ecological restoration have been applied in the karst areas of Southwest China, but the effect of the treatments in karst fault basins is not clear. As a typical watershed of a faulted basin, studying the influence of landscape pattern changes on runoff and sediment yields in the Nandong underground river system (NURS) helps to establish the relationship between watershed and runoff and sediment changes. It provides a theoretical basis and effective method for water and soil management assessment, and soil and water treatment in karst fault basins. The vegetation topographic factor (VTF) was constructed using the normalized vegetation index (NDVI), digital elevation model (DEM) and water-system map from 2000 to 2018. On the basis of VTF classification, the vegetation topographic landscape index (VTLI) was calculated using the FRAGSTATS software, and the effects of VTLI changes on NURS runoff and sediment yield were analyzed. The study found the following: (1) PD, IJI, LSI and SHDI were positively correlated with runoff and sediment yield (p < 0.01), and the correlation coefficients were 0.693, 0.668, 0.551 and 0.582 and 0.62, 0.635, 0.627 and 0.63, respectively. AI and CONNECT were significantly negatively correlated with runoff and sediment yield (p < 0.01), and the correlation coefficients were −0.551 and −0.596, −0.627 and −0.446, respectively. The LPI and DIVISION index were not significantly correlated with discharge, but were significantly correlated with sediment yield (p < 0.05), and the correlation coefficients were −0.179 and 0.271, respectively. (2) The interpretation of VTLI for runoff increased from 0.639 to 0.778, and the interpretation of sediment transport decreased from 0.809 to 0.613. (3) In urban areas, VTF decreased and was persistent. VTF increased in the basin mountain edge area. In mountainous areas, VTF was unchanged, but has an antipersistence trend. The NURS ecological restoration project had achieved obvious results, and the change in the watershed increased runoff production and reduced sediment production. The capacity of soil and water conservation in the high slope area of the mountain edge of the basin recovered and showed a trend of sustainable development. Due to the urban expansion brought about by economic development, the capacity of soil and water conservation around the city has declined, and it showed a sustainable development trend. Policymakers should strengthen the ecological environment of urban areas and coordinate development within mountainous areas.

Previous unsupervised feature selection methods typically model global data distributions and local geometric structures separately or via parallel constraints. Such decoupled modeling often leads to structural inconsistencies due to the lack of a unified representation of sample relationships. Moreover, existing local structure-based methods generally rely on predefined and fixed adjacency graphs, which often contain noise and redundancy, further degrading the learning performance. To address these issues, this paper proposes a novel Unified Structural Matrix Framework (USMF) for unsupervised feature selection. Specifically, a dynamically learnable, non-negative, symmetric structure matrix is introduced to consistently capture both global and local structural relationships within a single variable. This matrix preserves the global data structure via a trace-maximizing term, similar to weighted principal component analysis, while maintaining local geometric relationships via weighted distance constraints in the embedding space. Furthermore, a Frobenius norm (F-norm) consistency constraint is introduced to preserve neighborhood information and stabilize structure learning. Concurrently, an l2;1-norm regularization is imposed on the projection matrix to achieve effective feature selection. An alternating optimization algorithm is designed to solve the proposed model. Extensive experiments on multiple public datasets demonstrate that USMF outperforms existing state-of-the-art unsupervised feature selection methods in terms of Clustering Accuracy (ACC) and Normalized Mutual Information (NMI).