【目的】林分冠层可燃物水分参数在森林植物生理状态与火灾风险评估中发挥着重要的作用,本研究主要探讨了利用野外高光谱探测数据估算植被冠层可燃物水分特征参数的方法,为快速无损地获取植被冠层水分特征信息以评估森林可燃物火险提供新的途径。【方法】针对马尾松(Pinus massoniana)、火炬松(Pinus taeda)、香樟(Camphora officinarum)和石楠(Photinia serratifolia)共4种林分的冠层样本,通过野外采样测定和室内实验获取了样本叶片的光谱反射数据、叶面积、鲜质量和干质量,以及林分样本的叶面积指数(LAI)等数据,并分别对LAI和水分特征变量[可燃物含水率(FMC)、等效水厚度(EWT)、冠层水平可燃物含水率(FMC_C)和冠层水平等效水厚度(EWT_C)]及其相关性进行了分析。采用偏最小二乘回归法(PLSR)和人工神经网络(ANN)法对水分特征参数FMC_C和EWT_C进行预测估算。【结果】4种林分的FMC_C和EWT_C汇总样本的估算和验证过程表明,PLSR方法的决定系数均大于ANN法,而均方根误差(RMSE)则均小于ANN法。在4种不同类型林分冠层可燃物的水分特征变量EWT_C与FMC_C值估算中,采用PLSR方法分析的R²均值为0.73,RMSE_cv均值为0.22,马尾松表现最优(EWT_C的R²=0.80,RMSE_CV为0.02;FMC_C的R²=0.74,RMSE_CV为0.19);而采用ANN法的 R_{CV}^2均值为0.76,RMSE_CV均值为0.20,采用ANN-2作为输入的马尾松FMC_C估算 (R_{CV}^2=0.91,RMSE_CV为0.12)和火炬松的EWT_C值估算 (R_{CV}^2=0.90,RMSE_CV为0.01)表现最优。【结论】EWT_C与FMC_C均可从反射光谱中获得较为准确的估算,FMC_C作为一种基于质量的变量,通过PLSR和ANN法均可获得高于基于面积的变量EWT_C的估算精度。在决定光谱发射率方面,植物质量可能比植物面积发挥更大的作用,即可燃物的质量因素对光谱反射率水分特征反演的影响更大,叶片面积影响则相对较小。总体来看,野外探测的林分冠层高光谱数据用于估算可燃物水分特征参数的效果较好,可辅助森林可燃物火险预测预报和风险等级评估。

【Objective】Canopy fuel moisture parameters play a crucial role in assessing the physiological status of forest plants and fire risk. This study primarily explores methods for estimating vegetation canopy fuel moisture characteristic parameters using field hyperspectral data, providing a new approach for rapidly and non-destructively acquiring vegetation canopy moisture characteristic information to assess forest fuel fire risk.【Method】For canopy samples from four forest types, including Pinus massoniana, P. taeda, Camphora officinarum and Photinia serratifolia, spectral reflectance data, leaf area, fresh weight, dry weight, and leaf area index (LAI) were obtained through field sampling and laboratory experiments. Correlations between LAI and moisture characteristic variables (FMC, EWT, FMC_C and EWT_C) were analyzed. Partial least squares regression (PLSR) and artificial neural networks (ANN) were used to predict and estimate the moisture characteristic parameters FMC_C and EWT_C.【Result】The correlation between LAI and FMC_C (R²=0.86) was significantly higher than that between LAI and EWT_C(R²=0.04), while the correlation between LAI and EWT (R²=0.52) was slightly higher than that between LAI and FMC (R²=0.50). The estimation and validation processes for the aggregated samples of FMC_C and EWT_C across the four forest stand types showed that the coefficient of determination for the PLSR method was higher than that for the ANN, while the root mean square error was lower. In the estimation of EWT_C and FMC_C values for canopy fuel moisture characteristic variables in the four different forest types, the mean R² value was 0.73 and the mean RMSEcv was 0.22 when using the PLSR method. P. massoniana showed the best performance (FMC_C: R²=0.80, RMSE_CV=0.02; EWT_C: R²=0.74, RMSE_CV=0.19). When using the ANN method, the mean R² value was 0.76 and the mean RMSEcv was 0.20. The best performance was observed for the estimation of FMC_C in P. massoniana (R_{CV}^2=0.91, RMSE_CV=0.12) and EWT_C in P. taeda (R_{CV}^2=0.90, RMSE_CV=0.01) using ANN-2 as input.【Conclusion】Both EWT_C and FMC_C can be accurately predicted from reflectance spectra. As a mass-based variable, FMC_C can achieve higher estimation accuracy than the area-based variable EWT_C through both PLSR and ANN methods. Plant mass may play a more significant role than plant area in determining spectral emissivity, indicating that fuel mass factors have a greater impact on the inversion of spectral reflectance moisture characteristics, while leaf area has a relatively smaller influence. Overall, the hyperspectral data of forest stand canopy obtained through field measurements are effective in estimating fuel moisture characteristic parameters, and can provide valuable support for forest fuel fire risk prediction, forecasting, and level assessment.