原产于美洲的墨西哥至哥斯达黎加的紫茎泽兰，大约20世纪40年代由中缅边境传入云南南部，现已广泛分布于云南、四川、贵州、广西、西藏、重庆、湖北和台湾等省、市和自治区。由于紫茎泽兰具有很强的繁殖能力，现已在中国西南地区泛滥成灾，对生物多样性和当地的生态环境造成了严重威胁。对此国内外利用遥感技术进行了紫茎泽兰监测研究，尽管在某些特定的样点研究上有成功的案例，但是在我国西南地区进行紫茎泽兰遥感监测存在很多困难。我国西南地区植被覆盖度比较大，植被多样性存在方式复杂，林下紫茎泽兰较多，受高层树冠的遮挡，给遥感监测研究带来了很大难度。另外，由于缺乏对紫茎泽兰波谱特性的细致分析，目前没有成熟的遥感监测指示指标，从而不能进行长时间大范围紫茎泽兰时空分布的有效监测。本文的研究结果包括以下几个方面：（1）通过对典型样地紫茎泽兰长期光谱测量，提取了不同生长阶段紫茎泽兰波谱特征值。结果表明紫茎泽兰植物的不同生长阶段，其可见光谱段内的光谱曲线会随着各种色素的变化而变化，使得其在不同生长阶段呈现不同的光谱特征，特别是花期（3-4月）和发新枝叶期（8-9月）的反射率差别最明显，通过模拟发现，即使林下的紫茎泽兰，在开花盛期，也有可能在林上冠层反映出来，成为紫茎泽兰区别周围环境的指示性特征。（2）利用遥感植被指数的长时间序列的数据，提取典型样点，对比分析了紫茎泽兰和农作物、林地、草地等其他植被NDVI年变化的差异，以多年同一时间平均NDVI构建植被变化特征时间谱VTP （Vegetation Temporal Spectrum）。结果表明，农作物的VTP具有明显的受物候期和耕作方式影响的特点，其NDVI值随着农作物的生长、发育不断增大，到收割时NDVI急剧下降。对于双季农作物，其时间谱呈双峰模态；对单季农作物，其时间谱呈单峰模态。对于落叶林，其时间谱从叶子萌发到凋零期呈高NDVI值的平台模态。紫茎泽兰的时间谱和农作物、落叶林的差异比较大，但与常绿针叶林的时间谱特征相似，都具有年度变化较缓的特点。时间谱是区分紫茎泽兰和其他植被类型的有效特征。（3）针对已知紫茎泽兰分布区域，利用我国环境减灾卫星数据，通过几何纠正和配准研究，构造了研究区时间谱的数据，并在此基础上，利用监督分类，选择典型地类训练区进行最大似然分类，在30米像元尺度上很好地区分了紫茎泽兰和农地、林地，水体等其他地物类型，验证了利用时间谱识别紫茎泽兰方法的有效性。(4) 由于在西南地区，雨季分布时间比较长，很难获取完整的遥感植被指数序列，利用紫茎泽兰在开花期和非开花期植被ＮＤＶＩ的变化特点，通过构建CDI(Croften Weed Detection Index），建立紫茎泽兰的识别标志，进行紫茎泽兰的提取，为西南地区紫茎泽兰分布信息的提取提供了理论依据，尽管识别的精度不如监督分类的方法高，但是可以进行快速的信息提取。(5) 利用MODIS植被指数数据，通过多年平均的NDVI方法提高数据的信噪比，降低云雪和阴影的影响，分析了缅甸-云南-四川一线的1km大小的剖面，对该剖面进行时间谱的监督分类，获得了紫茎泽兰在该路径上的空间分布，利用监督分类和CDI方法给出了近30年来紫茎泽兰的时空分布强度。结果表明：AVHRR数据的质量比较差，结果噪音比较大，MODIS 数据的结果比较客观地反映了紫茎泽兰近几年来的分布。紫茎泽兰主要沿着两条线路从西南向东北进行传播，CDI强度总的趋势是逐渐减弱，但是有些强分布点随机分布在传播的路径上。

Originally growing in South America (from Mexico to Costa Rica), Crofton weed was introduced into the south of Yunnan Province in 1940s via the border between China and Burma. Now it has been widely spread out in Yunnan, SiChuan, GuiZhou, GuangXi, Tibet, ChongQing, HuBei and Tai Wan. Because of its strong reproductive capacity, it has been causing serious evil consequences in the south east of China which is risking severely the biological diversity and the living environment in those regions. To help handle this situation, there have been researches on utilizing the remote sensing technology to monitor the spread of the Crofton weed and there have been a few successful cases in some specific sampling regions. It is fairly difficult though to monitor the Crofton weed in the south east of China because of the big vegetation coverage and the complexity of the vegetation diversity in the region. The Crofton weed is hidden under the forest and is sheltered from the canopy of the trees. All these impact negatively the monitor of the Crofton weed by using the remote sensing technology. Aside from this, there are no mature indicators for the remote sensing monitoring due to the lack of detailed analysis about the spectral features of the Crofton weed which makes it almost impossible to effectively monitor the Crofton weed in a larger scale for a longer time. The results out of this research include a few items as follows:(1) Based on the spectral measurement of the weed for a longer time in a few typical sampling regions, its spectral feature values at various growth phases are retrieved. The result indicates its spectral curve at visible light band varies according to the change of the various pigments which makes it show itself with different spectral features at different growing phases. Especially during the weed’s blooming period (from March to April) and its new branches and leave generating period (from August to September), the reflectivity difference between the weed and other vegetations becomes very significant. It is found via the simulation that when the weed is in its bloom, the difference between the Crofton weed and the others can still be possibly reflected on the canopy of the trees even the weed is hidden under the trees. This can provide the directive features to distinguish the weed from its surroundings.(2) The comparison and analysis have been done on the NDVI of the Crofton weed, crops, woodland and grassland in the typical sampling regions by using the temporal series data of the remote sensing vegetation index. The VTP (Vegetation Temporal Spectrum) is defined to reflect the vegetation feature changes by using the average value of the NDVIs at the same time over multiple years. The results indicate that the VTP of the crop is impacted strongly by the phenophase and the tillage methods. The crop’s NDVI is increased along with crop’s growth and it gets decreased dramatically when it is harvested. For the bi-seasonal crop, its VTP presents with two peaks; while for the mono-seasonal crop, its VTP presents with only one peak; for the deciduous forest, its VTP presents like a platform with a high NDVI value from the leaves’ germinating to leaves’ withering. Being quite different from the crop and the deciduous forest, the Crofton weed’s VTP is similar to the coniferous forest which is slightly changing during the time period of one year. The VTP is one of the features capable of distinguishing the Crofton weed from other vegetations. (3) Based on the known regions with the Crofton weed, the data of the temporal spectrum in the research regions has been obtained by conducting geometric correction and coregistration against the images from the HJ-1-A/B. Furthermore, by using the maximum likelihood classification on the images after the selection of the typical target training areas of interest, the Crofton weed is well distinguished from the crops, woodland, water and other surface objects which proved the validity of the VTP method in recognizing the Crofton weed. (4) It is difficult to acquire the complete remote sensing vegetation index series because that the rainy season is too long in the south east of China. However, based on the changes of its NDVI during its blooming and non-blooming periods, the CDI (Crofton weed Detection Indicator) is defined as the indicator to extract the Crofton weed which provides the theoretical basis for extracting the distribution information of the Crofton weed in the south-east region of China. Although the precision of this method is not as high as the supervised classification, it is quicker in extracting the information. (5) Based on MODIS vegetation index data, the average NDVI of several years (from 2001 to 2007) is obtained to improve the signal-noise ratio and to reduce the impact of the clouds, snow and shadows. Selecting Burma-Yunnan-SiChuan linearly as the profile at the scale of 1km, the spatial distribution along this profile is obtained by conducting the supervised classification on the temporal spectra which indicates that the distribution of Crofton weed is not consecutive; there is a long time period distribution in Dali of Yunnan and Xichang of SiChuan with a big density while there is less distribution both time wise and space wise in other areas. The difference is given quantitatively by using CDI which reflects the average results over several years.