| 中国森林火灾发生规律及预测模型研究 |
| |
| 引用本文： | 吴恒,朱丽艳,刘智军,孔雷,郭小阳,张锋.中国森林火灾发生规律及预测模型研究[J].世界林业研究,2018,31(5):64-70. |
| |
| 作者姓名： | 吴恒 朱丽艳 刘智军 孔雷 郭小阳 张锋 |
| |
| 作者单位： | 1.国家林业局昆明勘察设计院, 昆明 650216 |
| |
| 基金项目： | 国家林业局昆明勘察设计院项目“林地‘一张图’蓄积量更新研究”（2014071501）。 |
| |
| 摘 要： | 量化分析森林火灾发生规律能为预测和防治森林火灾提供科学依据。文中采用四参数Weibull分布描述了我国森林火灾发生次数和火场面积分布规律，运用Spearman相关系数分析承灾主体因子、灾害管理因子、孕灾环境因子与森林火灾发生次数、面积间关系，基于全国森林火灾数据分别建立灰色系统理论模型、BP人工神经网络模型和时间序列ARIMA模型，并采用Markov随机过程改进已建立模型。结果表明，我国森林火灾发生次数分布呈左偏正态分布，火场面积呈倒J型分布，火灾次数和火场面积分布模型拟合决定系数分别为0.63和0.66；承灾主体、孕灾环境和灾害管理对森林火灾次数和火场面积影响程度依次减小，人工林面积、累年年平均气温、年降雨量平均差值、年最低气温平均日数与森林火灾发生具有明显相关性，影响森林火灾的因子与森林火灾发生次数、火场面积间存在指数型关系；不同模型对森林火灾发生次数和火场面积拟合优度次序为BP模型、GM（1，1）-Markov模型、BP-Markov模型、GM（1，1）模型、ARIMA模型、ARIMA-Markov模型，采用Markov过程能显著改进GM（1，1）预测模型对火灾随机性的预测效果，可以更好地反映森林火灾发生规律。
|
| 关 键 词： | 森林火灾 灰色模型 Markov链 人工神经网络 ARIMA模型 气象因子 中国 |
| 收稿时间： | 2018/3/7 0:00:00 |
| 修稿时间： | 2018/7/15 0:00:00 |
A Study of Regularity and Prediction Model for Forest Fire in China |
| |
| Wu Heng,Zhu Liyan,Liu Zhijun,Kong Lei,Guo Xiaoyang and Zhang Feng.A Study of Regularity and Prediction Model for Forest Fire in China[J].World Forestry Research,2018,31(5):64-70. |
| |
| Authors: | Wu Heng Zhu Liyan Liu Zhijun Kong Lei Guo Xiaoyang and Zhang Feng |
| |
| Institution: | 1.Kuming Survey and Design Institute, State Forestry Adminitration, Kunming 650216, China2.Yan'an Forestry Bureau, Yan'an 716000, Shaanxi, China3.Huxian Forestry Bureau, Xi'an 710300, China |
| |
| Abstract: | Quantitative analysis of forest fire regularity provides a scientific basis for forest fire prediction and effective prevention. 4-Parameters Weibull model was used for describing forest fire frequency, burned area and its distribution. Spearman indicator was applied to analyze the correlation between forest fire frequency & burned area and the various factors, including hazard bearing body, environmental factor, and disaster management. Grey model, BP artificial neural network model and ARIMA model were established based on the forest fire data nationwide, and Markov chain was used to modify these models. The results showed that forest fire occurrences are presented as left skewed normal distribution and the burned area as inverted J shape, while their coefficients of fitting are 0.63 and 0.66 respectively; the impacts of hazard bearing body, environmental factor, and disaster management on forest fire frequency and burned area go down gradually, while plantation area, mean annual temperature, amplitude of annual rainfall difference, days with annual minimum temperature significantly relate to forest fire frequency and burned area, and exponential relationship exists between forest fire frequency & burned area and the influencing factors. The fitting for forest fire frequency and burned area by various models showed different goodness, which could be ordered by BP model > GM (1, 1) -Markov model > BP-Markov model > GM (1, 1) model > ARIMA model > ARIMA-Markov model, and Markov process can significantly modify the predictability and randomness of GM(1,1), which could contribute to better prediction of fire regularity. |
| |
| Keywords: | forest fire grey model Markov chain artificial neural network ARIMA model meteorological factor China |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《世界林业研究》浏览原始摘要信息 |
| 点击此处可从《世界林业研究》下载免费的PDF全文 |
|
