基于随机森林的河流总磷预测模型及影响因素分析

Prediction model and influencing factors of total phosphorus concentration in river based on random forest method

通过采集不同时间段淮河干流19个典型采样点的水样,获得水体中总磷浓度数据,针对河流总磷浓度时空分布差异大,受影响因素多和非线性的特点,基于随机森林算法,选择(气候特性、水动力、土壤类型和流域特性等)特征变量,构建模型对河流中总磷浓度进行预测,然后通过均方差增量参数对影响河流总磷浓度时空分布因素的重要性程度进行评估。研究结果表明,基于随机森林算法的模型可较好地模拟淮河水体中总磷浓度,模拟的一致性相关系数可达到0.83;对影响河流中总磷分布的因素进行评估发现,气候因素(降雨、温度)及水动力因素(流量)是最重要的因素;地表黏土含量对于水体中总磷的贡献要高于粉沙及沙粒的贡献;面源污染是淮河干流中总磷的主要来源,其中旱作农田的重要性系数高于灌溉农田。

By collecting water samples from 19 sampling points in the mainstream of the Huaihe River in different periods,the concentration of the total phosphorus in water was obtained.It can be noticed that the patial and temporal distribution was of a great difference and the totaol phosphorus appeared with a nonlinearity characteristics affected by lots of factors.The random forest method was employed to predict the total phosphorus concentration in river by choosing the variables of climate,hydropower,soil type and basin properties.Then,the influencing factors that affecting the spatial and temporal distribution of the total phosphorus concentration were analyzed by the square deviation of incremental parameter.Results showed that the concordance correlation coefficient of the simulation of the total phosphorus concentration in Huaihe River using random forest method could reach 0.83.Rainfall,temperature and river flow were the most important factors by analyzing the factors affecting the total phosphorus concentration distribution in rivers.The contribution of the surface clay to total phosphorus concentration in water was higher than that of silt and sand.Non-point source pollution was the main source of the total phosphorus of the Huaihe River,and the importante coefficient of the dry farmland was higher than that of the irrigated farmland.