高级检索

基于多种机器学习算法的东北地区潜在湿地分布模拟

Simulation of potential wetlands distribution in northeast China using multiple machine learning algorithms

  • 摘要: 潜在湿地的分布范围对于湿地的科学规划、有效管理和适当恢复至关重要。在潜在湿地分布模拟制图中,如何选择高精度的机器学习模型以实现高置信度的湿地资源评价仍需开展深入研究。本研究以中国东北地区为研究区,构建了综合考虑水文、土壤、植被和地形因子的潜在湿地模拟指标体系,应用地理大数据和多种机器学习算法(随机森林、支持向量机、深度神经网络和极端梯度提升),模拟东北地区潜在湿地的分布并进行空间格局分析。研究结果表明,4种算法的模型性能均较好,接收者操作特征曲线下的面积(AUC)均大于0.69,其中基于随机森林算法模拟的潜在湿地分布具有最高的精度,总体精度为85.57%,Kappa系数为0.71。东北地区潜在湿地面积为128 790 km2,主要分布在降水量为400~600 mm、土壤类型为半水成土且覆盖有草甸和湿生植物的区域。研究结果可为东北地区乃至中国湿地资源的评价提供重要基础数据。

     

    Abstract: The distribution of potential wetlands is crucial for the rational planning, effective management, and efficient conservation of wetland ecosystems. Wetlands are among the most productive and valuable ecosystems on Earth, providing essential ecological services such as water purification, climate regulation, flood control, and biodiversity support. However, wetlands worldwide are under significant threat due to climate change, urbanization, agricultural expansion, and other human activities, leading to their rapid degradation and loss. Therefore, understanding and simulating the distribution of potential wetlands is not only a scientific challenge but also a critical step toward sustainable wetland management and restoration. Despite the growing interest in using machine learning algorithms for ecological modeling, the optimization of these algorithms for simulating potential wetland distribution remains underdeveloped. This gap in knowledge motivated our study, which was conducted in the northeastern region of China. Northeast China is home to some of the most extensive and ecologically significant wetland resources in the country, including the Sanjiang Plain and the Songnen Plain. These wetlands are vital habitats for migratory birds and endangered species, and they play a crucial role in regional hydrological cycles and carbon sequestration. However, wetland loss and degradation in this region have been severe in recent decades, driven by agricultural expansion and urbanization. Thus, there is an urgent need to develop robust models to predict and understand the distribution of potential wetlands in this ecologically sensitive area. To address this challenge, our study leveraged geographic big data and a variety of machine learning algorithms, including random forest (RF), support vector machine (SVM), deep neural network (DNN), and extreme gradient boosting (XGBoost). These algorithms were chosen for their ability to handle high-dimensional data, capture complex nonlinear relationships, and provide reliable predictions. We integrated a comprehensive set of environmental factors, such as hydrology, soil properties, vegetation types, and topographic features, to construct a potential wetland distribution simulation system. This approach allowed us to simulate the distribution of potential wetlands in northeast China and identify the environmental conditions most conducive to wetland formation. The results of our study demonstrated that all four machine learning algorithms performed satisfactorily in simulating potential wetland distribution, with AUC values exceeding 0.69, indicating strong predictive capabilities. Among these algorithms, the random forest model achieved the highest accuracy, with an overall accuracy of 85.57% and a Kappa coefficient of 0.71. These metrics suggest that the random forest algorithm is particularly well-suited for modeling complex ecological systems like wetlands, where multiple environmental factors interact in nonlinear ways. Our findings revealed that the potential wetland area in northeast China is approximately 128 790 km². These potential wetlands are most likely to form in regions characterized by annual precipitation ranging from 400 to 600 mm, semi-hydric soils, and vegetation dominated by swamps and meadows. These environmental conditions align well with the known ecological requirements for wetland formation, such as adequate water availability, suitable soil moisture retention, and vegetation adapted to wet conditions. The results of this study provide a vital data foundation for the evaluation and conservation of wetlands in northeast China and even across the country. By identifying areas with high potential for wetland formation, our model offers valuable insights for prioritizing wetland restoration efforts and designing effective management strategies. Furthermore, the methodology developed in this study can be adapted and applied to other regions, contributing to global wetland conservation efforts. In conclusion, this research highlights the importance of integrating geographic big data with advanced machine learning techniques to address complex ecological challenges and support sustainable wetland management.

     

/

返回文章
返回