湿地专题图的更新无论是对湿地研究还是湿地管理和保护都具有重要价值。但是由于湿地本身具有显著的时空动态性和空间异质性特征，使得大空间范围湿地专题图的更新面临着周期长、时效性差的挑战。为应对这一挑战，实现大空间范围湿地地图的快速更新，本文提出一种通过提取历史湿地专题图中的信息，对新遥感影像进行自动化的湿地分类制图方法——“迭代解译再组织” ⅡR (Iterative Interpretation and Reorganization)。针对湿地空间异质性强、稳定样本获取困难等特点，ⅡR方法通过分别获取不同湿地类型的空间信息和类别属性两个步骤完成湿地的自动更新。为验证该方法在大空间范围湿地图更新中的应用效果，随机选取了位于高海拔地区、高纬度地区、低纬度地区和滨海地区等不同自然地理环境的4个湿地保护区(若尔盖湿地保护区、莫莫格湿地保护区、鄱阳湖湿地保护区和黄河三角洲湿地保护区)进行验证。结果表明ⅡR方法的湿地制图的总体精度在70%—90%，总体上优于传统监督分类方法。ⅡR方法对于大的时空尺度背景下湿地专题图更新面临的高时空动态特征具有较好的解决能力。

Timely updating of wetland map is vital to wetland research and management. However, the highly hydro-dynamic characteristics and great spatial heterogeneity of wetlands pose challenges in updating large-scale wetland thematic maps in a timely manner. General mapping methods, such as supervised or object-oriented classification, are time consuming and can be easily affected by cognitive differences. To address this issue, we propose an automatic updating method of wetland map, namely, iterative interview and reorganization.
We aim to design a method that can transfer the knowledge from existing wetland thematic maps into the classification of a new remote sensing image. At the same time, the method should be robust for different geographical conditions.
Rather than adapting samples between different domains, Iterative Interview and reorganization (ⅡR) tries to obtain the precise spatial distribution of ground objects first and then defines the properties of the spatial distributions by matching their spatial features. The method can tackle complex situations caused by changes of ground objects.
This automatic method achieves an overall accuracy ranging from 70% to 90%, similar to the results of general supervised classification. In some cases, ⅡR has better performance in the identification of detailed information than that the support vector machine or maximum likelihood classification, such as for boundaries of ground objects and slender targets. To examine the performance of this method, we choose four wetland reserves with various geographical environments across China, including the Momoge Nature Reserve in high latitude, the Zoige Reserve in high altitude, Poyang Lake in a hot-humid area, and Yellow River Delta along a coastal area. Both overall and individual accuracies of various wetland classes in the four study areas are higher than those of the general supervised classification. Furthermore, ⅡR can automatically detect new classes such as paddy field.
Without extra samples, ⅡR achieves better classification in four study areas of different landscapes. This method is not only adaptable for eliminating unfavorable factors, such as terrain or clouds, but also more flexible and robust when dealing with different wetlands and phenological changes, demonstrating that the ⅡR method can be applied in large-scale thematic map updating. ⅡR can have a consistent interpretation of the same wetland class because all procedures are carried out without expert knowledge. In conclusion, ⅡR can meet the needs of automatic updating of large-scale wetland thematic maps.