欢迎访问《应用生态学报》官方网站,今天是 分享到:

应用生态学报 ›› 2022, Vol. 33 ›› Issue (6): 1547-1554.doi: 10.13287/j.1001-9332.202206.026

• 研究论文 • 上一篇    下一篇

混合效应模型在林火发生预测中的适用性

张珍1,2, 杨凇1,2, 朱贺1,2, 王光玉3, 郭福涛1,2, 孙帅超1,2*   

  1. 1福建农林大学林学院, 福州 350002;
    23S技术与资源优化利用福建省高校重点实验室, 福州 350002;
    3不列颠哥伦比亚大学林学院亚洲森林研究中心, 加拿大温哥华 BC V6T 1Z4
  • 收稿日期:2021-08-23 接受日期:2022-03-24 发布日期:2022-12-15
  • 通讯作者: *E-mail: sun_sc@yeah.net
  • 作者简介:张 珍, 女, 1996年生, 硕士研究生。主要从事林火生态与管理研究。E-mail: zhen17803477566@163.com
  • 基金资助:
    国家重点研发计划战略性国际科技创新合作重点专项(2018YFE0207800)资助。

Applicability of mixed effect model in the prediction of forest fire

ZHANG Zhen1,2, YANG Song1,2, ZHU He1,2, WANG Guang-yu3, GUO Fu-tao1,2, SUN Shuai-chao1,2*   

  1. 1College of Forestry, Fujian Agricultural and Forestry University, Fuzhou 350002, China;
    23S Technology and Resource Optimization Utilization Key Laboratory of Fujian Universities, Fuzhou 350002, China;
    3Asia Forest Research Centre, Faculty of Forestry, University of British Columbia, Vancouver, BC V6T 1Z4, Canada
  • Received:2021-08-23 Accepted:2022-03-24 Published:2022-12-15

摘要: 林火是森林生态系统的重要影响因子,建立科学准确的林火预测预报模型对林火管理工作至关重要。本研究以不同气象因子为主要预测变量,基于Logistic回归和广义线性混合效应模型建立福建省林火发生预测模型,通过对比Logistic基础模型和广义线性混合效应模型的拟合度和预测精度,研究混合效应模型在林火预报中的适用性。结果表明: Logistic基础模型的受试者工作特征曲线下面积(AUC)值为0.664,验证准确率为60.4%。添加随机效应后,模型的拟合和检验精度均获得了提升。其中,考虑行政区划和海拔差异效应的两水平混合效应模型的表现最优,其AUC值和验证准确率分别比基础模型提升0.057和6.0%。用此混合效应模型对福建省各地区的林火发生概率进行预测的结果表明,福建省西北部和南部为林火中高发区域,西南部和东部为林火低发区域,与实际观测的火点分布一致。混合效应模型在数据拟合和林火发生预测方面均优于Logistic基础模型,可作为林火预测和管理的重要工具。

关键词: 林火预测, 随机效应, Logistic回归, 混合效应

Abstract: Fire is an important influencing factor in forest ecosystems. Establishing an accurate forest fire forecasting model is important for forest fire management. We used different meteorological factors as predictors to construct a forest fire prediction model in Fujian Province, based on Logistic regression and generalized linear mixed effect model. We compared the fitness and prediction accuracy of the two models, judged the applicability of the mixed effect model in forest fire forecasting. The results showed that the AUC and accuracy values of the Logistic base model were 0.664 and 60.4%, respectively. Models considering random effects gave better fitting and validating statistics. Among them, the two-level mixed model containing both area and altitude difference effects performed best, with increases of 0.057 and 6.0% for the AUC and accuracy values, respectively. By applying the model to predict the probability of forest fires in Fujian Province, we found that the middle-incidence and high-incidence areas of forest fires distributed in northwest and south Fujian, whereas the low-incidence areas of forest fires distributed in southwest and east Fujian, which was consistent with the observed data. The data fitting and forest fire prediction of the mixed effects model was better than those of the Logistic basic model. Therefore, it could be used as an important tool for forest fire prediction and management.

Key words: forest fire forecasting, random effect, Logistic regression, mixed effect