落叶松受雅氏落叶松尺蠖危害程度光谱检测

Spectral Detection of Damaged Level of Larch Affected by Jas's Larch Inchworm

近年来蒙古国雅氏落叶松尺蠖灾害不断加剧,逐渐逼近大兴安岭地区,将威胁我国北方森林生态系统安全。以现代遥感监测方法替代传统检测方法,及早掌握该虫害发生发展规律对防控有重要意义。为快速、大范围遥感监测雅氏落叶松尺蠖灾害,利用光谱分析技术研究了该害虫危害下落叶松受害程度检测模型。通过实测健康和轻度、中度、重度受害落叶松光谱,计算与比较不同受害程度落叶松原始光谱和去除包络线光谱的敏感度,揭示光谱敏感波段及去除包络线光谱敏感性。然后对去除包络线光谱进行一阶导数变换获得光谱特征参数并分析其随受害程度的变化特征,构建基于CART(分类与回归树)算法的落叶松受害程度光谱检测模型。研究表明:去除包络线光谱敏感性比原始光谱更显著,尤其在480~520nm(蓝边)、640~720nm(红谷、红边)、1 416~1 500nm(短波红外谷)等波段内光谱敏感度介于0.1~2.0,而且出现了敏感峰现象。随受害程度增加,去除包络线光谱敏感性增强趋势比原始光谱更明显;在蓝边波段上去除包络线光谱敏感峰位置向短波方向移动,即502nm→490nm,而在红谷及红边、短波红外谷等波段上光谱敏感峰位置向长波方向移动,即664nm→672nm和1 436nm→1 448nm;红谷位置和短波红外谷位置归一化反射率以及红谷和短波红外谷面积呈上升趋势。在蓝边与红边波段内去除包络线光谱一阶导数对受害程度有明显响应,出现了波峰现象。随害虫危害程度加剧红边位置蓝移(718nm→700nm),红边斜率及面积和蓝边斜率及面积呈下降趋势。基于此,利用红边斜率、红谷位置和短波红外谷位置归一化反射率、红谷和短波红外谷面积、蓝边斜率及面积等去除包络线光谱特征参数构建的CART模型对落叶松受害程度有很好的检测能力。与多元线性回归模型相比,CART模型检测精度更高,其Kappa系数达0.875。研究结果对雅氏落叶松尺蠖灾害的防治有参考价值。

In recent years,the Mongolian Jas’s Larch Inchworm disaster have been increasing and gradually approaching the Daxinganling area,which will threaten the safety of the forest ecosystem in the north of China.It is necessary to replacethe Mongolian traditional detection method with the modern remote sensing monitoring method,and it is important to control the rule of occurrence and development of the insect at the first opportunity.A detection model was developed forrapid and wide range remote sensing monitoring of Jas’s Larch Inchworm damagebased on spectral analysis technology.Spectra in situ measurements were conducted for healthy larch and mild,moderate,severe damaged larch to compute andcompare the raw spectra and continuum removal spectra.On this basis,spectral sensitive bands and sensitivities of continuum removal spectra were uncovered.Then some spectral characteristic parameters were obtained by the first derivative transformation of continuum removal spectrum,and its change features vs damaged level of larch were analyzed to construct the detection model based on CART(classification and regression tree)algorithm.The results showed that thesensibility of continuum removal spectra was more significant than that of raw spectra,especially in 480~520 nm(blue edge),640~720 nm(red edge)and 1 416~1 500 nm(short wave infrared valley).In these bands,the value for the sensitivity of continuum removal spectra was between 0.1 and 0.2.At the same time,it appeared sensitive peaks in the sensitive bands.With the increase of damaged level,the sensibility of continuum removal spectra had an enhanced trend which was more evident compared with raw spectra.In addition,the sensitivepeak position of the continuum removal spectra in the blue band was shifted tothe short wave direction,that is,502 nm→490 nm,and the spectral sensitivepeak position in the red valley and red edge,short wave infrared valley movedto the long wave direction,that is,664 nm→672 nm and 1 436 nm→1 448 nm.Furthermore,the normalized reflectance of red valley position and short wave infrared valley position as well as areas of red valley and short wave infrared valley were found a trend of rising.First derivative of continuum removal spectra had obvious in response for damaged level in blue edge and red edge.Moreover,the band peaks were arisen.The red edge position blue shift arose(718 nm→700 nm)as the damaged level increased,while the slopes and areas for blue edge and red edge decreased.Hence,the CART model that was established based on several continuum removal spectral parameters such as red edge slope,normalized reflectance of red valley position and short wave infrared valley position,areas of red valley and short wave infrared valley and both slope and area of blue edge had superior detection ability in the damaged level of larch.Comparedto multi-linear regression model,the CART model performed better with the Kappa value of 0.875.These results will play important roles on remote sensingmonitoring of the damage of Jas’s Larch Inchworm.