基于CA的溢油扩散时空过程模拟及敏感性分析

Spatio-Temporal Simulation and Sensitivity Analysis of Oil Spill CA Model

为了解决传统溢油元胞自动机(cellular automata,CA)模型难以获得的模型参数问题,引入逻辑回归方法到溢油CA模型中,并构建基于逻辑回归的溢油CA模型。该模型仅需要设置起始影像和影响因子(包括海风、洋流、距离、温度与盐度)等,便可以获取模型参数并模拟出溢油扩散范围的动态变化过程。将模型应用到深海溢油实验中以验证本模型在海上溢油模拟情况,结果表明模拟结果的总精度为96.6%,Kappa系数达0.899,能够准确地模拟出溢油扩散范围。通过参数敏感性分析可发现,温度与盐度因子对模型的影响最弱,其次是距离因子。对溢油扩散影响最为显著的是海风和洋流因子。若忽略海风与洋流的影响,模拟结果与观测结果将产生较大的偏差,并无法呈现出溢油扩散中的漂移等特征。

In order to overcome the difficulty of obtaining parameters incellular automata（CA）model for oil spill,this paper introduces logistic regression method into CA model and presents a new oil spill model based on logistic-regression CA model.This model can easily obtain model parameters and simulate the dynamic changes of oil spill by using only a few inputs,such as the initial image,impact factor.The model was applied to DeepSpill experiment to verify its effect in simulation of oil spill.Experiments show that the simulation results are consistent with the real situation.The total accuracy and Kappa coefficient of simulation results is 96.6%and 0.899 respectively.After the sensitivity analysis of parameters,we found that the impact of temperature and salinity on simulation results is the weakest,followed by proximity variable.The most important variables in the model are the winds and currents.If we don＇t take winds and currents into consideration,the simulation results will deviate from the verification image for lacking the important characteristics—drift behavior of oil spill.