Simulation of oil spill using logistic-regression CA model

Abstract

Cellular automata (CA) are considered to be effective models to simulate the behavior of oil spills for overcoming the difficulty of obtaining parameters in numerical models of oil spills. Besides, CA models are convenient to combine geographic information system (GIS) to display the simulation results. This paper presents a new oil spill simulation based on logistic-regression CA model, which easily obtain the weights of the impact factors. The model also can simulate the dynamic changes of oil spill using only a few inputs, such as the initial image, impact factors, and their weights. It was applied to simulate the oil spill in DeepSpill project using five factors, the distance factor, wind, current, temperature, and salinity. Experiments showed that the simulation results are consistent with the verification image with the total accuracy and Kappa coefficient of simulation results as high as 96.8% and 0.834 respectively. We also study the influence of sampling ratio on simulation results. The accuracy improves with the increasing ratio. However, the performances improve only slightly when the ratio reaches 20%. We also analyze the sensitivity of temperature, salinity, winds, currents, and distance. Experiments showed that the simulation results will only expanse around the original area without considering the current and wind. The simulation results will have big model error without considering distance factor. However, less model error occurs in the simulation results without using temperature and salinity.