基于随机森林方法研究鄱阳湖典型洲滩植被群落分布与表层土壤因子耦合关系

准确识别湿地植被与土壤的相互作用是湿地恢复和保护的重要前提。水文情势是影响鄱阳湖地区植被分布的关键因子,而植被分布格局则会对湿地土壤养分累积与赋存形态产生影响。本文利用优化的随机森林(Random Forests)算法,基于多环境变量预测了鄱阳湖典型低滩植物(虉草,Phalaris arundinacea Linn)和高滩植物(南荻,Triarrhena lutarioriparia L. Liu)的分布,进而分析这两种植被表层土壤养分累积差异。结果表明:随机森林模型对虉草和南荻预测的精度分别达到了89.6%和89.3%。模型给出了土壤因子的重要性排序。按土壤因子与虉草分布的密切相关程度,重要性依次为全钾、氨氮、有机质、含水率、全氮、有效磷、全磷、pH和硝态氮;按土壤因子与南荻分布的密切相关程度,重要性依次为全钾、pH、有机质、全氮、全磷、硝态氮、氨氮、含水率和有效磷。从植被分布与土壤因子的偏依赖图中可得出,南荻分布区较虉草分布区土壤酸性更强;虉草分布与土壤全氮、氨氮含量呈负相关关系,南荻分布则与土壤全氮含量呈正相关关系,而与氨氮关系不显著;虉草分布与土壤全磷含量正相关,而南荻则与全磷负相关;虉草和南荻与土壤有效磷相互作用关系较弱;此外虉草分布区钾含量低,二者负相关,而南荻分布区钾含量高,二者正相关。随机森林方法适用于模拟复杂的非线性关系,给出了单个土壤因子与植被之间的偏依赖关系,易于给出生态学意义上的解释,在研究湿地植被与环境因素的相互作用关系中有极大的推广价值。

Study on Relationships Between Vegetation Community Distribution and Topsoil Factors Based on Random Forests in Shoaly Wetlands of Poyang Lake

Abstract: Accurate identification of the interaction between vegetation and soil is an important premise of wetland restoration and protection. Hydrologic situation is the key factor affecting the distribution of wetland vegetation; while the vegetation distribution pattern can affect the accumulation and occurrence form of soil nutrient. Based on soil environmental factors, optimized Random Forests was used to predict the distribution of Phalaris arundinacea Linn and Triarrhena lutarioriparia L. Liu, which grow in low elevation and high elevation of shoaly wetlands respectively. And then we analyzed soil nutrient accumulation of the two plants. Results show that the accuracy of 89.6% and 89.3% for Phalaris arundinacea Linn and Triarrhena lutarioriparia L. Liu respectively was achieved by Random Forests. According to the model, The importance of soil factors which closely related to Phalaris arundinacea Linn followed a decreasing order of total potassium, ammonia nitrogen, organic matter, soil water content, total nitrogen, available phosphorus, total phosphorus, pH, nitrate nitrogen; The importance of soil factors which closely related to Triarrhena lutarioriparia L. Liu followed a decreasing order of total potassium, pH, organic matter, total nitrogen, total phosphorus, nitrate nitrogen, ammonia nitrogen, soil water content, available phosphorus. According to the partial dependent plot, pH value in Phalaris arundinacea Linn was significantly higher than Triarrhena lutarioriparia L. Liu. Distribution of Phalaris arundinacea Linn was negatively correlated to total nitrogen and ammonia nitrogen; distribution of Triarrhena lutarioriparia L. Liu was positively related to total nitrogen while week relationship between Triarrhena lutarioriparia L. Liu and ammonia nitrogen was found. Total phosphorus was positively related to Phalaris arundinacea Linn while negatively related to Triarrhena lutarioriparia L. Liu. Weak relationships were found between available phosphorus and the two kinds of plant. Total potassium was negatively correlated with Phalaris arundinacea Linn while positively correlated with Triarrhena lutarioriparia L. Liu. Soil water content was positively related to Phalaris arundinacea Linn while negatively related to Triarrhena lutarioriparia L. Liu. Random Forests is suitable for simulating complex nonlinear relations, and can show the partial dependence relationship between individual soil factors and vegetation, so that we can explain the results in the ecological sense. Random Forests is of great value in the study of the interaction between wetland vegetation and environmental factors.