比较逻辑斯蒂与地理加权逻辑斯蒂回归模型在福建林火发生的适用性

林火预测预报是科学有效进行林火管理的前提,是林业管理部门和科研工作者的广泛关注的领域。逻辑斯蒂回归(Logistic Regression,LR)是目前国内外广泛应用于森林火灾预测的模型方法,然而近年来有学者发现该方法没有充分考虑林火影响因子的空间相关性和异质性,从而导致模型拟合结果偏差。地理加权逻辑斯蒂回归(Geographically weighted logistic regression,GWR)模型考虑到了模型变量之间的空间相关性,有效提高的模型的拟合能力。为探讨GWLR模型在福建林火预测上的适用性,本研究应用LR和GWLR两种方法分别建立福建省森林火灾与气象因子的预测模型,通过模型拟合能力对比,判断在GWLR的适用性。研究以2000—2005年福建地区森林火灾卫星火点数据和每日气象因子为基础,将全样本分为60%的建模数据和40%的校验数据,并重复5次,建立5个样本组。选择在5个样本组中3个及以上表现显著的变量进入最终模型。研究结果表明GWLR在模型拟合度、模型残差、空间自相关性以及预测准确率等方面均优于LR模型,说明充分考虑模型变量的空间异质性有助于提高模型的预测精度,同时也验证了GWLR在福建地区林火预测上的适应性。此外,模型参数结果显示,"日最高地表气温"、"日最低地表气温"、"日平均风速"、"24小时降水量"、"日最高本站气压"、"日照时数"、"日最高气温"和"日最小相对湿度"8个因子对福建省林火发生有显著影响,研究结论为福建地区林火预测预报提供了新的方法。

Comparing the application of logistic and geographically weighted logistic regression models for Fujian forest fire forecasting

Forest fire forecasting is a key component of effective and science-based forest management and has been comprehensively addressed in the scientific literature. The logistic regression (LR) technique has been used in forest fire prediction models. However, some scholars have recently reported that the technique does not adequately consider the spatial correlation and heterogeneity of fire impact factors, which results in poorly fitting models. In contrast, geographically weighted logistic regression (GWR) models consider the spatial correlation of model variables, which improves the model''s goodness of fit. In order to explore the applicability of the GWLR model in Fujian forest fire forecasting, the present study used both the LR and GWLR methods to establish forecast model for forest fires and meteorological factors in Fujian Province, and the model fitting ability of two models were compared. Based on the forest fire and meteorological data for Fujian from 2000 to 2005, the original dataset was randomly divided into training (60%) and validation (40%) samples, with five replications and five sample groups, and predictors that were significant (a=0.05) for at least three of the five sample groups were included in the final models. The goodness of fit, residual error, spatial autocorrelation, and prediction accuracy of the GWLR model were all better than those of the LR model, and the GWLR comprehensively explained the spatial heterogeneity of model variables and helped to improve the prediction accuracy of the model. The study also verified the suitability of the GWLR model on the forest fire forecasting in Fujian area. In addition, the results also indicated that the occurrence of Fujian forest fires is significantly affected by eight parameters, including minimum and maximum surface temperature, daily average wind speed, daily precipitation, highest station pressure, hours of sunshine, daily maximum temperature, and daily minimum relative humidity. Therefore, the GWLR model may provide a new technique for the prediction of forest fires in Fujian Province.