Research of influence of sample size on normal information diffusion based on the Monte Carlo method: risk assessment for natural disasters

Abstract

Information diffusion techniques and Monte Carlo methods have been widely used in solving all kinds of problems of small samples with incomplete data in the field of natural disaster risk assessment such as environmental resource rating, flood monitoring and temperature changing. Data are not the only thing that matters for natural disaster risk assessment, but with enough data, we can accurately predict the time, place, scale and loss of future disasters. It is important to simulate the complete data scene when there is a minimum sample size. In this paper, we collect temperature data from 3050 meteorological stations in China and use the Monte Carlo simulation method to investigate the effect of sample size on estimating the normal information diffusion. The results show that (1) for the same sample, the information diffusion method is significantly better than the traditional histogram method. (2) Using the hard histogram estimation method, the recommended sample size is 85 or more, which is slightly larger than the traditional threshold value (i.e., 30), while using the information diffusion estimation method, the recommended sample size decreases to 45 or more. (3) Simulation experiments show that, with insufficient samples, both estimation methods, i.e., the information diffusion and the traditional histogram methods become invalid because of its poor correlation, low robustness, high RMSE and variance values. These results indicate that the Monte Carlo simulation method and information diffusion technique have certain practical reference value in the research of natural disaster risk assessment in the case of a small sample.