基于GIS和BP神经网络的农区地下水硝态氮含量分布特征分析

Spatial analysis of groundwater NO_3^--N concentration in agriculture-dominated regions based on GIS-based BPNN

针对于非点源污染机理模型在实际运用中的限制,将人工神经网络引入地下水非点源污染格局的模拟和预报中,建立了基于G IS的BP神经网络模型用以模拟分析农区浅层地下水NO3--N含量及其空间分布特征。结果表明,以农田氮盈余、地下水埋深、30~60cm土层砂粒含量和土壤有机质4个因素为输入因子,以地下水NO3--N为输出因子,通过网络训练以及观测点缓冲区半径的设定与调整,BP神经网络模型有效地模拟了山东省桓台县地下水NO3--N含量及其空间分布特征,并且有较高的精度。该研究可为华北平原农区地下水质管理提供分析工具与决策依据,是对非点源污染机理模型的有益补充。

Aiming at the practical difficulty of processed-based non-point model in groundwater pollution management, an artificial neural network was introduced for modeling and prediction of non-point pollution. A GIS- based Back Propagation Neural Network （BPNN） was developed for modeling groundwater NO3^--N concentration. Field nitrogen surplus, groundwater depth, soil sandy content at 30-60 in depth and soil organic content were included as input vectors of the BPNN. By designation of buffer zone around sampling well, the BPNN simulated NO3^--N concentration well and effectively captured the general trend of the spatial patterns of the NO3^--N concentration. The study provides a practical tool for analysis and management of groundwater nitrate pollution in North China Plain and serves as a supplement of processed-based non-point pollution.