滇中云南松林地表可燃物含水率预测模型

Prediction model for surface fuel moisture in Pinus yunnanensis forest in central Yunnan

云南松针叶富含油脂,防火期含水率低,是滇中地区林火主要地表可燃物。在2020年防火期持续采集滇中地区云南松林地表可燃物含水率数据,使用相关性分析、公因子方差、膨胀系数和多重预测回归模型,探究地形、气象、林分等因子与含水率的关系,利用离差标准化法调整模型系数及完成模型精度评价。结果表明:云南松林地表可燃物含水率影响因子排序为温度>湿度>风速>坡向>郁闭度>坡度>海拔>风向,坡向、林分郁闭度指标方差膨胀系数VIF>10,存在共线性;因此选择温度、湿度、风速、坡度、海拔构建含水率预测回归模型E_(1),k、s、g代表云南松枯枝、松针和小灌枯枝-枯草,Y_(k1)、Y_(s1)、Y_(g1)平均拟合度为74.35%,平均误差率为32.06%,误差率偏高。以强相关(r>0.70)因子温度、湿度、风速为自变量重构含水率预测增强回归模型E_(2),其Y_(k2)、Y_(s2)、Y_(g2)平均拟合度为83.99%,平均误差率为17.09%,其拟合度、误差率均优于E1。在E2基础上选择具有现实意义的弱相关性因子坡向、坡度、海拔、郁闭度为调整因子,并运用离差标准化法转化为系数,构建含水率预测系数校正回归模型E_(3),其Y_(k3)、Y_(s3)、Y_(g3)平均拟合度89.72%,平均误差率为8.48%。E_(3)精度优于E_(1)、E_(2),其Y_(k3)、Y_(s3)、Y_(g3)拟合优度分别提升9.69%、2.11%,4.84%、10.77%,8.41%、4.33%,误差率降低15.65%、6.89%,11.24%、13.69%,18.01%、5.24%。增加校正系数可提高模型预测精度,同时模型因子易获取,便于林火管理者野外快捷、精准、实时预测滇中云南松林地表可燃物含水率,为林火防控提供技术支持。

Needle litter of Pinus yunnanensis,the main surface fuel of wildfire in the central Yunnan Province,China,is highly flammable for its high oil content and low moisture.We monitored the moisture contents of surface fuels of P.yunnanensis forests in central Yunnan during the fire prevention period in 2020.Correlation analysis,common factor variance,variance inflation factor(VIF),and multiple prediction regression model were used to explore the relationships between topographical,meteorological,and stand factors and the moisture contents of surface fuels.The model coefficients were adjusted by deviation standardization method,and the model accuracies were evaluated.The results showed that the factors affecting surface fuel moisture of P.yunnanensis forests in descending importance were temperature,humidity,wind speed,slope direction,canopy density,slope,elevation,and wind direction.The VIFs of slope direction and canopy density were more than 10,showing a high degree of collinearity.Therefore,regression model E_(1)was constructed using temperature,humidity,wind speed,slope,and elevation.The average goodness of fit for moisture of Y_(k1),Y_(s1),and Y_(g1)was 74.35%,and the average error rate was 32.06%;the symbols k,s,and g were branch litter of pine,needle litter of pine,and shrub twig litter and grass litter,respectively.An enhanced regression model E_(2)was reconstructed with temperature,humidity,and wind speed as independent variables,which showed significant correlations with the target(r>0.70).The average goodness of fit of Y_(k2),Y_(s2),and Y_(g2)was 83.99%,and the average error rate was 17.09%,which outperformed those of E_(1).Slope direction,slope,elevation,and canopy density,which showed insignificant contributions to surface fuel moisture,were selected as adjustment elements and converted as correction coefficients by the deviation standardization method,to reconstruct the regression model E_(3).The average goodness of fit was 89.72%,and average error rate was 8.48%for Y_(k3),Y_(s3),Y_(g3).The accuracy of E_(3)outperformed E1 and E_(2),with the mean goodness of fit of Y_(k3),Y_(s3),Y_(g3)being improved by 9.69%vs 2.11%,4.84%vs 10.77%,and 8.41%vs 4.33%,and the error rates being reduced by 15.65%vs 6.89%,11.24%vs 13.69%,and 18.01%vs 5.24%,respectively.The added correction coefficients can improve the prediction accuracy of the model,and the model factors are easy to obtain.These findings are useful for forest fire prevention and management,and provide technical support for rapid,accurate,and real-time prediction of surface fuel moisture.