基于气象因素的广义加性模型在福建省林火预测中的应用

doi:10.13466/j.cnki.lczyyj.2024.04.011

摘要/Abstract

摘要：

林火的准确预测对其预防与管理具有重要意义。基于福建省2010—2020年林火和气象数据,分别采用Logistic回归模型和基于6种平滑样条基[高斯过程平滑样条基(GP)、三次回归样条基(CR)、薄板回归样条基(TP)、Duchon样条基(DS)、B-样条基(BS)、P-样条基(PS)]拟合的广义加性模型进行林火预测,并对各模型的预测效果进行评价。结果显示:1)Logistic回归模型在训练集上的准确率为74.80%,在测试集上的准确率为75.97%。广义加性模型的预测精度整体优于Logistic回归模型,其中由TP样条基拟合的广义加性模型表现最佳,其训练集和测试集的准确率分别比Logistic回归模型分别提高了3.86%和2.52%。2)基于最优广义加性模型预测结果,对福建省的森林火险等级进行划分。结果表明,中高火险区主要集中在福建省西北和东南地区,西部和东部地区为低火险区。广义加性模型能够更好地捕捉复杂的非线性关系,适用于复杂环境下的林火预测。

关键词: 林火预测, Logistic回归模型, 广义加性模型, 薄板回归样条

Abstract:

Predicting forest fire occurrences is crucial for fire prevention and management.This study used historical forest fire and meteorological data from Fujian Province(from 2010 to 2020)to apply the Logistic Regression Model and Generalized Additive Model(GAM)with six types of smooth spline bases[Gaussian Process Smoothing Spline Basis(GP),Cubic Regression Spline Basis(CR),Thin Plate Regression Spline Basis(TP),Duchon Spline Basis(DS),B-Spline Basis(BS),and P-Spline Basis(PS)]to predict forest fire occurrences.By comparing the performance of these models,their effectiveness in forest fire prediction was evaluated.The results indicate the following.1)The logistic regression model achieved an accuracy of 74.80% on the training set and 75.97% on the test set,demonstrating its baseline performance.Overall,the predictive accuracy of the GAM was generally superior to that of the Logistic Regression Model,with the TP spline basis-based GAM performing the best.Its accuracy on the training and test sets was improved by 3.86% and 2.52%,respectively,compared to the logistic regression model.2)Based on the optimal GAM,the forest fire risk levels in Fujian Province are delineated.The results revealed that areas with moderate to high fire risk are primarily concentrated in the northwest and southeast regions,while the Western and Eastern regions exhibit low fire risk.GAM arebetter at capturing complex nonlinear relationships,making them suitable for predicting forest fire occurrences in complex ecological environments.

Key words: forest fire prediction, logistic regression model, generalized additive model, thin plate regression spline

中图分类号:

S762.2

陈国富, 李春辉, 陈振雄. 基于气象因素的广义加性模型在福建省林火预测中的应用[J]. 林草资源研究, 2024,(4): 94-102.

CHEN Guofu, LI Chunhui, CHEN Zhenxiong. Application of Generalized Additive Model Based on Meteorological Factors in Forest Fire Prediction in Fujian Province[J]. Forest and Grassland Resources Research, 2024,(4): 94-102.

图/表 6

表1

表2

表3

表4

表5

表6

参考文献 35

[1]	BAKER R. Forest history:International studies on socioeconomic andforest ecosystem change[J]. Forest Ecology and Management, 2002, 159(3):293-293.
[2]	FLANNIGAN M, STOCK B, TURETSKY M, et al. Impacts of climate change on fire activity and fire management in the circumboreal forest[J]. Global Change Biology, 2009, 15(3):549-560.
[3]	RASBASH D J. Effects of fire on items which may have helped cause thefire[J]. Fire Safety Journal, 1984, 7(4):293-294.
[4]	LIANG Hao, ZHANG Meng, WANG Hailan. A neural network model for wildfire scale prediction using meteorological factors[J]. IEEE Access, 2019, 7:176746-176755.
[5]	苏漳文, 曾爱聪, 蔡奇均, 等. 基于Gompit回归模型的大兴安岭林火预测模型及驱动因子研究[J]. 林业工程学报, 2019, 4(4):135-142.
[6]	NAMI M H, JAAFARI A, FALLAH M, et al. Spatial prediction of wildfire probability in the Hyrcanian ecoregion using evidential belief function model and GIS[J]. International Journal of Environmental Science and Technology, 2017, 15(2):373-384.
[7]	ZHANG Yang, LIM S, SHARPLES J J. Modelling spatial patterns of wildfire occurrence in South-eastern Australia[J]. Geomatics,Natural Hazards and Risk, 2016, 7(6):1800-1815.
[8]	GUO Futao, WANG Guangyu, SU Zhangwen, et al. What drives forest fire in Fujian,China:Evidence from logistic regression and random forests[J]. International Journal of Wildland Fire, 2016, 25(5):505-519.
[9]	曾爱聪, 蔡奇均, 苏漳文, 等. 基于MODIS卫星火点的浙江省林火季节变化及驱动因子[J]. 应用生态学报, 2020, 31(2):399-406. doi: 10.13287/j.1001-9332.202002.015
[10]	CHANG Yu, ZHU Zhiliang, BU Rencang, et al. Predicting fire occurrence patterns with logistic regression in Heilongjiang Province,China[J]. Landscape Ecology, 2013, 28(10):1989-2004.
[11]	朱政, 赵璠, 王秋华, 等. 林火发生预报模型研究进展[J]. 世界林业研究, 2022, 35(3):26-31.
[12]	COLLINS L, MCCARTHY G, MELLOR A, et al. Training data requirements for fire severity mapping using Landsat imagery and random forest[J]. Remote Sensing of Environment, 2020, 245:111839.
[13]	MOHAJANE M, COSTACHE R, KARIMI F, et al. Application of remote sensing and machine learning algorithms for forest fire mapping in a Mediterranean area[J]. Ecological Indicators, 2021, 129:107869.
[14]	孙立研, 刘美玲, 周礼祥, 等. 基于气象因子深度学习的森林火灾预测方法[J]. 林业工程学报, 2019, 4(3):132-136.
[15]	纪守领, 李进锋, 杜天宇, 等. 机器学习模型可解释性方法、应用与安全研究综述[J]. 计算机研究与发展, 2019, 56(10):2071-2096.
[16]	赖界亨, 卢洵, 王克英, 等. 基于广义加性模型的调温负荷测算方法[J]. 广东电力, 2023, 36(6):40-49.
[17]	简盈, 张云雷, 丁兆成, 等. 基于Tweedie-GAM模型的鱚属鱼类鱼卵丰度与栖息环境的关系研究[J]. 中国海洋大学学报(自然科学版), 2022, 52(9):43-53.
[18]	段鹏, 陈文波, 杨欢, 等. 生境破碎化过程对流域生境质量的影响[J]. 生态学报, 2024, 44(14):6053-6066.
[19]	梁慧玲, 林玉蕊, 杨光, 等. 基于气象因子的随机森林算法在塔河地区林火预测中的应用[J]. 林业科学, 2016, 52(1):89-98.
[20]	张珍, 杨凇, 朱贺, 等. 混合效应模型在林火发生预测中的适用性[J]. 应用生态学报, 2022, 33(6):1547-1554. doi: 10.13287/j.1001-9332.202206.026
[21]	LI Yudong, FENG Zhongke, CHEN Shilin, et al. Application of the artificial neural network and support vector machines in forest fire prediction in the Guangxi autonomous region,China[J]. Discrete Dynamics in Nature and Society, 2020. 5612650:1-14.
[22]	PRADHAN B, SULIMAN B H D M, AWANG B A M. Forest fire susceptibility and risk mapping using remote sensing and geographical information systems(GIS)[J]. Disaster Prevention and Management, 2007, 16(3):344-352.
[23]	OLIVEIRA S, OEHLER F, SAN-MIGUEL-AYANZ J, et al. Modeling spatial patterns of fire occurrence in Mediterranean Europe using multiple regression and random forest[J]. Forest Ecology and Management, 2012, 275:117-129.
[24]	PEDERSEN E J, MILLER D L, SIMPSON G L, et al. Hierarchical generalized additive models in ecology:An introduction with mgcv[J]. PeerJ, 2019, 7:e6876.
[25]	何培, 辛士冬, 姜立春. 基于广义加性模型的樟子松树干削度方程研建[J]. 北京林业大学学报, 2020, 42(12):1-8.
[26]	BENINI E, PONZA R. Nonparametric fitting of aerodynamic data using Smoothing Thin-Plate Splines[J]. AIAA Journal, 2015, 48(7):1403-1419.
[27]	WOOD S N, AUGUSTIN N H. GAMs with integrated model selection using penalized regression splines and applications to environmental modelling[J]. Ecological Modelling, 2002, 157(2):157-177.
[28]	VILAR L, WOOLFORD D G, MARTELL D L, et al. A model for predicting human-caused wildfire occurrence in the region of Madrid,Spain[J]. International Journal of Wildland Fire, 2010, 19(3):325-337.
[29]	ZUMBRUNNEN T, PEZZATTI G B, MENENEDZ P, et al. Weather and human impacts on forest fires:100 years of fire history in two climatic regions of Switzerland[J]. Forest Ecology and Management, 2011, 261(12):2188-2199.
[30]	GUO Futao, SU Zhangwen, WANG Guangyu, et al. Understanding fire drivers and relative impacts in different Chinese forest ecosystems[J]. Science of the Total Environment, 2017,605-606:411-425.
[31]	CHUVIECO E, COCERO D, RIANO D, et al. Combining ndvi and surface temperature for the estimation of live fuel moisture content in forest fire danger rating[J]. Remote Sensing of Environment, 2004, 92(3):322-331.
[32]	韩鹏, 郭桂祯, 李鑫磊, 等. 基于地理探测器的福建省台风灾情影响因素分析[J]. 遥感技术与应用, 2023, 38(2):487-495. doi: 10.11873/j.issn.1004-0323.2023.2.0487
[33]	国家林业局. 全国森林防火规划(2016—2025年)[EB/OL].(2016-12-29)[2021-07-06]. https://www.ndrc.gov.cn/fggz/fzzlgh/gjjzxgh/201705/W020191104624246431007.pdf.
[34]	SHANG Dongfang, ZHANG Fan, YUAN Diping, et al. Deep learning-based forest fire risk research on monitoring and early warning algorithms[J]. Fire, 2024, 7(4):151.
[35]	BARMPOUTIS P, PAPAIOANNOU P, DIMITROPOULOS K, et al. A review on early forest fire detection systems using optical remote sensing[J]. Sensors, 2020, 20(22):6442.

数据类型	统计量	风速/(m/s)	最高气温/ ℃	日照时数/h	相对湿度/%	降水量/mm
训练数据	平均值	1.73	23.70	5.69	74.71	3.32
	最小值	0.10	-2.50	0.00	16.00	0.00
	最大值	21.70	39.70	12.70	100.00	273.50
	标准差	1.47	7.44	3.80	12.91	10.45
	变异系数/%	84.93	31.41	66.79	17.27	314.88
测试数据	平均值	1.91	24.33	4.96	76.45	3.18
	最小值	0.00	2.20	0.00	24.00	0.00
	最大值	13.20	40.30	12.40	100.00	61.90
	标准差	1.62	7.53	3.97	12.63	8.35
	变异系数/%	84.55	30.93	80.19	16.52	262.75

气象因素	估计值	标准误差	P值
截距	-1.327 6	0.066 8	<0.001
风速	-0.492 3	0.032 7	<0.001
最高气温	-0.677 6	0.028 7	<0.001
日照时数	1.027 1	0.037 5	<0.001
相对湿度	-0.227 5	0.031 8	<0.001
降水量	-2.500 6	0.223 1	<0.001

模型	参数项(截距α)			非参数项
模型	自由度	卡方检验	P	平滑项	自由度	卡方检验	P
高斯过程平滑样条(GP)	-2.416 0	1.435	0.092	风速(f₁)	7.037	203.57	<0.001
				最高气温(f₂)	7.192	760.19	<0.001
				日照时数(f₃)	7.259	635.95	<0.001
				相对湿度(f₄)	1.002	95.04	<0.001
				降水量(f₅)	5.376	142.66	<0.001
三次回归样条(CR)	-1.258 9	0.225 5	<0.001	风速(f₁)	7.219	215.01	<0.001
				最高气温(f₂)	7.250	766.94	<0.001
				日照时数(f₃)	7.096	650.93	<0.001
				相对湿度(f₄)	1.002	98.47	<0.001
				降水量(f₅)	5.468	132.87	<0.001
薄板回归样条(TP)	-1.205 1	0.105 5	<0.001	风速(f₁)	6.780	213.8	<0.001
				最高气温(f₂)	7.223	787.1	<0.001
				日照时数(f₃)	7.161	682.6	<0.001
				相对湿度(f₄)	1.002	106.2	<0.001
				降水量(f₅)	3.616	109.6	<0.001
Duchon样条(DS)	-1.329 0	0.383 5	<0.001	风速(f₁)	7.437	219.5	<0.001
				最高气温(f₂)	7.688	774.7	<0.001
				日照时数(f₃)	7.506	667.9	<0.001
				相对湿度(f₄)	1.000	101.4	<0.001
				降水量(f₅)	4.596	116.4	<0.001
B-样条(BS)	-1.370 0	0.180 9	<0.001	风速(f₁)	5.094	194.8	<0.001
				最高气温(f₂)	6.027	778.1	<0.001
				日照时数(f₃)	6.610	666.8	<0.001
				相对湿度(f₄)	1.006	101.4	<0.001
				降水量(f₅)	2.828	114.1	<0.001
P-样条(PS)	-1.416 3	0.251 1	<0.001	风速(f₁)	5.092	195.1	<0.001
				最高气温(f₂)	5.916	778.5	<0.001
				日照时数(f₃)	6.465	662.6	<0.001
				相对湿度(f₄)	1.004	101.7	<0.001
				降水量(f₅)	2.868	114.0	<0.001

模型	AIC	BIC	阈值	AUC	准确率/%
模型	AIC	BIC	阈值	AUC	训练集	测试集
Logistic回归模型	10 890.80	10 934.64	0.446	0.795	74.80	75.97
GP	9 713.04	9 940.67	0.395	0.841	78.64	76.17
CR	9 724.55	9 953.43	0.442	0.841	78.43	76.66
TP	9 755.65	9 937.80	0.390	0.840	78.66	78.49
DS	9 732.18	9 964.19	0.417	0.840	78.56	77.69
BS	9 759.36	9 939.01	0.418	0.840	78.54	77.92
PS	9 760.40	9 938.43	0.422	0.840	78.62	77.80

模型检验指标	Logistic回归模型	GAM(TP)
平均绝对误差	0.333 9	0.302 0
均方根误差	0.414 7	0.395 9