基于物候参数和面向对象法的濒海生态脆弱区植被遥感提取

用遥感数据快速准确地提取植被信息对生态环境监测和发展气候模型具有基础性和关键性的意义。由于中国黄河三角洲地区植被类型插花分布,传统遥感提取方法精度较低。该文选取近代黄河三角洲典型生态脆弱区为研究区,基于MODIS和LANDSAT8数据,通过提取研究区的物候参数和不同分辨率遥感影像的融合处理,根据植被类型斑块大小确定分割尺度,根据典型植被类型的物候特征、光谱和空间等特征值构建分类规则,利用分区策略自上而下进行面向对象分类。结果表明,该方法总体精度为80.75%,Kappa系数0.79,高于传统物候和面向对象分类方法。广生态幅的棉田与其他植被的光谱混淆是传统面向对象分类方法植被分类精度低的主要原因,利用物候参数进行植被分区能规避棉田和自然植被的光谱混淆,有利于对植被类型的区分。分类结果与当地植被分布情况相符,可以用于研究区植被类型的精细提取。

Extraction of vegetation information in coastal ecological vulnerable areas from remote sensing data based on phenology parameters and object-oriented method

Abstract: Obtaining good vegetation classifications based on remote sensing data is important for ecosystem forecasting and improvement of global climate modeling. However, the classification result using the traditional methods is not accurate in the modern Yellow River Delta due to interspersed distribution of the vegetation types. The work reported here concerns the use of multi-sensor and multi-temporal remote sensing data in order to alleviate this problem by the object-oriented method assisted with the phenology parameters. Landsat 8 OLI and MODIS data were chosen because of the advantages such as being free of charge and stable to offer dataset. Meanwhile, these 2 types of data can bring a proper combination because they show the characteristics of vegetation respectively in space and time. Taking the typical ecologically vulnerable area of the Yellow River Delta as the study area, this study used the 250-meter 16-day MODIS vegetation indices products (MOD13Q1) to build time series curves of NDVI (normalized difference vegetation index) for different vegetation types, which were later smoothed by logistic time function method to fit NDVI data. Then, the different bands of Landsat 8 data were fused using the Gram-Schmidt (GS) method to obtain the 15-meter resolution image. A set of phenology indicators, including start of season, end of season, season length, peak NDVI, accumulative area of NDVI during growth period, and integral result between peak value and baseline value were extracted by the maximum curvature method. The extracted phenology images of the vegetation (250 m resolution) were resampled to 15 m resolution and merged with Landsat 8 image. Further, we employed a multi-resolution segmentation method according to the patch size of different vegetation types. Then, the classifier rules utilizing the phenological features and spectral characteristics of typical vegetation types were developed to map the vegetation in the study area, and we applied a partitioning strategy to carry out object-oriented classification. Finally, the classification results were compared with that from traditional methods. It indicated that the overall accuracy is 80.75% and Kappa coefficient is 0.79, higher than traditional phenology and object-oriented classification methods. In addition, we found that the low accuracy of the traditional object-oriented classification method is mainly caused by the cotton fields that had wide ecological fitness, which leads to the confusion of the cotton fields with other vegetation types. And the disadvantage for the confusion of the cotton fields and the natural vegetation is avoided by the phenology-assisted method, which is beneficial to distinguish the vegetation types. Therefore, the combination of vegetation phenological parameters and object-oriented method can solve the problem of spectral confusion effectively, and is suited for extraction of the vegetation types in small-scale areas like the coastal vulnerable areas. Moreover, statistical results on vegetation area indicated that the classification results accurately reflected the real situation of local vegetation distribution. In the study area, the vegetation coverage rate is high, and the proportion of natural vegetation and artificial vegetation is near to 1. In the natural vegetation type, the Suaeda, Reed-Tamarix and Suaeda community are the main vegetation types, accounting for 77.98%. The cotton is the dominant vegetation in the plant vegetation types, accounting for 71.16%, and less for non-salt vegetation types. Therefore, this method of the study provides support for vegetation survey in coastal vulnerable areas.