【作者】 王琪；

【导师】 张晓丽；

【作者基本信息】 北京林业大学 ， 农业硕士（专业学位）， 2022， 硕士

【摘要】 松材线虫病(Bursaphelenchus xylophilus)是一种世界性的重大检疫性森林病害,松树在感染病害后具有发展速度快,危害大,防治难的特点,造成严重的经济损失。疫情调查是松材线虫病防治工作的基础,开展及时且高效的松材线虫病害调查监测工作是遏制疫情蔓延的前提。人工疫情调查方法基于样点的调查,时效性较差,导致防控措施滞后。无人机遥感近年发展很快,但对于区域病害监测,由于成本高难以实现全覆盖连续监测,使得难以掌握区域病害程度,预测病害蔓延趋势。卫星遥感技术具有低成本、监测周期较短的特点,可用于大范围、快速监测松材线虫病。本文以安徽省六安市霍山县为研究区,以松材线虫病为研究对象,研究松材线虫病灾害程度的遥感监测方法,利用样地调查数据,使用2012至2021年的Landsat时间序列影像,研建松林枯死率监测模型,实现松林灾害程度的定量监测,并基于时间序列气象数据及人文社会经济数据,分析病害程度影响因子。论文的主要研究内容及结果如下:(1)基于Landsat影像时序多特征的松林空间分布信息提取。分别应用研究区2013年和2021年的年内多时相Landsat影像,提取年内时间序列归一化植被指数(Normalized Difference Vegetation Index,NDVI)、归一化差异指数(Normalized Difference Index,NDI)、增强植被指数(Enhanced Vegetation Index,EVI)、最佳时相的光谱反射率特征以及主成分分析第一分量的纹理特征,利用随机森林算法结合十折交叉验证进行变量筛选和地物分类,提取研究区2013年和2021年的松林分布范围,作为松材线虫病灾害程度监测的潜在风险区,两年松林的分类精度分别为89.29%和94.94%,满足后续建模要求。(2)构建病害监测指数和松林枯死率监测模型。结合松材线虫生物学特性和对松林的危害特征,利用多时相Landsat数据构建病害监测指数,并将其应用于枯死率的估测。以地面调查的松林枯死率为因变量,分别采用植被指数特征组合、病害监测指数特征组合以及两者结合的特征组合为自变量,利用多元线性回归、偏最小二乘回归、随机森林等建立林木枯死率监测模型,从建模集和预测集的决定系数(R~2)、均方根误差(RMSE)、平均绝对误差(MAE)三个方面对比不同模型的精度。特征变量筛选结果表明,植被指数和构建的病害监测指数结合的建模效果最好,其次是病害监测指数特征组合,最后是植被指数特征组合,说明建立的病害监测指数(MDI)对枯死率的估测精度有一定的提高作用;两者结合的特征组合反映的信息更加全面;不同建模方法的对比分析可知,植被指数和病害监测指数结合建立的偏最小二乘回归模型拟合效果最好,模型的决定系数R~2=0.442,均方根误差RMSE=6.8%,平均决定误差MAE=5.7%,适用于区域尺度松材线虫病害的快速、有效监测。(3)基于时间序列Landsat影像的历史灾害程度反演。应用(2)建立的监测模型,基于Landsat影像对2012年-2021年(剔除2016年无可用影像年份)的松林枯死率进行定量反演,并根据枯死率利用阈值法将病害程度分级(健康、受害和严重受害),得到历史灾害程度等级分布图。结果表明,研究区病害分布在空间上的特征为疫情集中在东北部和中南部,总体呈现点块状分布;在时间上呈现重度灾害占比逐渐减少的变化特征。(4)病害发生的影响因子分析。分析2012年、2015年、2020年这三年受害松林与地形因子(高程和坡度)、气象因子(年平均气温、四月和五月平均降水量)、人为因子(人口密度、与居民点距离、与道路距离)共七个因子之间的关系,利用松材线虫病发生规律的先验知识,进一步证明了历史灾害程度反演的有效性。因子分析结果表明,松林在年平均温度越高、四月和五月平均降雨量越少、海拔较低、地形坡度较为平缓、人口密度较大以及距离农村居民点越近的情况下更易感染松材线虫病,为松材线虫病重点防控区域的选择提供依据。研究表明,提出的病害监测指数一定程度上可以提高松林枯死率的估算精度,使用该指数对历年灾害程度进行反演是可行的。本文利用多光谱遥感技术对松材线虫病进行区域尺度的监测,可连续、快速监测区域灾害程度,为无人机遥感监测提供基础,对松材线虫病监测和防治措施的制定具有重要意义。更多还原

【Abstract】 Pinewood nematode(Bursaphelenchus xylophilus)is a significant quarantine forest disease worldwide.Pine trees infected with the condition have several characteristics such as rapid development,considerable damage,and difficulty to control,causing severe economic losses.The epidemic investigation is the basis of pine nematode control,and timely and efficient investigation and monitoring of pine nematode disease is the key to containing the spread of the disease.Manual epidemic survey methods require sample points of investigation,which are less time-sensitive and lead to lagging prevention and control measures.Remote sensing by drones has developed rapidly in recent years.Still,for regional disease monitoring,it is challenging to achieve full-coverage continuous monitoring due to high costs,making it difficult to grasp the extent of regional disease and predict the trend of disease spread.Satellite remote sensing technology has the characteristics of low cost and short monitoring period,which can be used for large-scale and rapid monitoring of pinewood nematode disease.This paper takes Huoshan County,Liuan City,Anhui Province as the study area,takes pine wilt nematode disease as the research object,and studies the remote sensing monitoring method of pine wilt nematode disease disaster level.The study also analyzed the factors influencing the extent of disease based on time series meteorological data and human and socio-economic data.The main research contents and results of the paper are as follows.(1)Spatial distribution information extraction of pine forests based on Landsat images with multiple features in time series.The annual time series Normalized Difference Vegetation Index(NDVI),Normalized Difference Index(NDI),Enhanced Vegetation Index(EVI),spectral reflectance features of the best time series,and principal components were extracted from Landsat images of the study area in2013 and 2021,respectively.The random forest algorithm combined with ten-fold cross-validation for variable selection and feature classification was used to extract the distribution of pine forests in the study area in 2013 and 2021 as a potential risk area for pinewood nematode hazard monitoring,and the two-year time series of Normalized Difference Vegetation Index(NDVI),Normalized Difference Index(NDI),Enhanced Vegetation Index(EVI),spectral reflectance characteristics of the best time phase,and texture features of the first component of principal component analysis.The classification accuracy of the pine forests in 2013 and 2021 was 89.29%and 94.94%,respectively,which met the requirements of subsequent modeling.(2)Construction of disease monitoring index and pine forest mortality rate monitoring model.The disease monitoring index was constructed using multi-temporal Landsat data and applied to estimate the dieback rate by combining the biological characteristics of pinewood nematode and its damage characteristics to pine forests.The ground surveyed pine forest mortality rate was used as the dependent variable,and the combination of vegetation index characteristics,disease monitoring index characteristics,and the combination of both were used as independent variables to establish the forest dieback rate monitoring model using multiple linear regression,partial least squares regression,and random forest,respectively.The accuracy of the models was compared in terms of coefficient of determination(R~2),root mean square error(RMSE),and mean absolute error(MAE)of the modeled and predicted sets.The results of feature variable screening showed that the combination of vegetation index and the constructed disease monitoring index had the best modeling effect,followed by the combination of disease monitoring index features.Finally,the combination of vegetation index features indicates that the established disease monitoring index(MDI)had a specific effect on the accuracy of the estimation of dieback rate;the combination of the two combined features reflected more comprehensive information;the comparative analysis of different modeling methods showed that The partial least squares regression model combining vegetation index and disease monitoring index had the best fit,with R2=0.442,RMSE=6.8%,and MAE=5.7%,which is suitable for rapid and effective monitoring of pine nematode diseases at the regional scale.(3)Historical hazard level inversion based on time series Landsat images.The monitoring model established in(2)was applied to quantitatively invert the mortality rate of pine forests from 2012 to2021(excluding the year 2016 when no images were available)based on Landsat images,and the disease level was graded(healthy,damaged and severely damaged)based on the mortality rate using the threshold method to obtain the historical disaster level distribution map.The result showed that the characteristics of the disease distribution in space are concentrated in the northeast and central southern part of the study area,with an overall point-block distribution;The change in time is characterized by a gradual decrease in the proportion of severe disasters.(4)Analysis of factors influencing disease occurrence.The relationships between the affected pine forests and topographic factors(elevation and slope),meteorological factors(average annual temperature,average precipitation in April and May),and anthropogenic factors(population density,distance from settlements,and distance from roads)in the three years of 2012,2015,and 2020 were analyzed.Moreover,the validity of the inversion of historical hazard levels was demonstrated using a priori knowledge of the occurrence pattern of pine wilt nematode disease.The results of the factor analysis showed that pine forests were more susceptible to pine wilt nematode infection at higher average annual temperatures,lower average rainfall in April and May,lower elevation,gentler topographic slope,higher population density,and closer to rural settlements,providing a basis for the selection of crucial pine wilt nematode prevention and control areas.The research result showed that using a disease monitoring index can improve the estimation accuracy of pine forest dieback rate to a certain extent,and it is feasible to use this index for inversion of historical disaster extent.This paper uses multispectral remote sensing technology to monitor pine wilt nematode disease on a regional scale,which can continuously and rapidly monitor the regional disaster extent,provide a basis for UAV remote sensing monitoring,and is of great significance for pine wilt nematode disease monitoring and the formulation of control measures.更多还原