城市湿地是城市的重要生态系统，具有多样的生态与社会服务功能，与人类密切相关。湿地城市是一个城市在湿地生态保护方面规格高、分量重的一项荣誉，是由《湿地公约》国际组织授予“湿地城市”称号的城市，是城市湿地保护的先锋与榜样。然而湿地城市的有效期仅有6年，期满后需再次评估才可确定是否继续授予该称号。因此实时监测城市湿地的分布特征与变化状况是湿地城市非常必要的需求。湿地类型复杂多样提取难度大，目前基于遥感提取的城市湿地类型不够精细、城市湿地分类方法体系有待完善，传统逐年分类的动态监测方法耗时长、工作量大。由此针对湿地城市开展城市湿地的遥感提取方法研究非常必要。本研究在国际上与城市湿地保护密切相关，具有很好的前瞻性，在国内既可支持中国推进湿地城市认证，同时也有助于总结城市湿地保护的中国经验，在城市湿地的遥感提取研究中具有重要的理论科学意义与实际应用价值。

本研究利用密集时序的Sentinel数据构建基于“光谱-几何-频率”特征的城市湿地逐级细分方法提取2020年中国首批6个湿地城市的城市湿地类型。同时构建基于连续变化检测的城市湿地逐年分类方法提取2015~2022年6个城市的城市湿地。基于精细湿地类型与湿地要素指标分别分析2015~2022年城市湿地的时空变化特征，并揭示6个城市的城市湿地分布特征模式、城市湿地类型与要素变化模式、城市湿地保护与恢复模式。创新性的实现了针对6个城市8年时间包含17种湿地类型的城市湿地动态监测与变化模式总结。本研究的主要特色内容与创新结果如下：

（1）构建基于“光谱-几何-频率”特征的城市湿地逐级细分方法，提取包含17种湿地类型的城市湿地。该方法分别利用K折随机森林方法、湿地细分规则及年内积水频率逐级提取6个城市2020年城市湿地。研究结果发现，①城市湿地整体精度在91%以上，Kappa在0.87以上，大部分城市湿地类型的生产者精度与用户精度可以达到80%以上。②与其他湿地数据产品对比，本研究的城市湿地提取范围较为准确且类别更多更准确，非湿地与其他数据产品的一致性面积较高。本研究实现了分类精度可达91%，类型可达17种的城市湿地遥感提取。

（2）构建基于连续变化检测的城市湿地逐年分类方法，提取得到2015~2022年的城市湿地。该方法采用连续变化检测与分类算法的思想开展变化检测并针对变化与不变区域采用不同方式进行包含17种湿地类型的城市湿地逐年分类。研究结果发现，①变化检测的整体精度均在80%以上，大部分城市不变区域的生产者精度高于变化区域；②2015~2022年城市湿地与非湿地分类结果与10m数据产品的空间一致性较好，一致性面积占比达到70%。对比传统逐年湿地分类，本研究在有效减少工作量的前提下实现了2015~2022年城市湿地的逐年提取。

（3）基于精细湿地类型分析2015~2022年中国首批湿地城市的城市湿地时空变化特征。研究结果发现，①6个城市的湿地面积占比均在5%以上，内陆草本沼泽与永久性河流是内陆城市的主要湿地类型，两者在湿地中的占比达30%以上，浅海水域、滨海滩涂、滨海木本沼泽与滨海草本沼泽是滨海城市的主要湿地类型，占比超过50%；②自然湿地是分布最多的类型，在湿地中占比达60%以上，2015~2022年期间大部分城市的自然湿地占比增加，人工湿地占比减少。

（4）基于水体、植被与土壤的湿地要素分别选取水体指标（NDWI）、植被指标（NDVI）与土壤湿度指标（SMMI），分析2015~2022年中国首批湿地城市的城市湿地时空变化特征。研究结果发现，①针对水体特征，常德与常熟的增加趋势明显，海口的减少趋势明显；②针对植被特征，常德的减少趋势明显，东营与哈尔滨的增加趋势明显；③针对土壤湿度特征，银川、海口与哈尔滨的下降趋势明显；④指标变化稳定区域的面积占比最高，可以达到80%左右。

（5）总结6个城市的城市湿地分布特征模式、城市湿地类型与要素变化模式、城市湿地保护与恢复模式，揭示城市湿地的分布特征与变化规律。研究结果发现，①城市湿地分布特征模式包含滨海湿地城市、内陆平原湿地城市与内陆山区湿地城市，其湿地保护率均大于50%，滨海城市的湿地率大于10%，内陆城市大于5%；②总结城市湿地类型与要素变化模式发现稳定湿地占比最多可达60%，2018年后稳定湿地面积占比增加，退化湿地面积减少，恢复湿地的程度更高；③城市湿地的保护与恢复呈现“重要湿地保护区”+“法律与应用结合的保护政策”+“湿地恢复工程”的模式。

本文聚焦于中国首批湿地城市，构建包含17种湿地类型的城市湿地逐级提取方法与逐年分类方法实现城市湿地基于遥感的动态监测，为中国首批湿地城市提供2015~2022年的城市湿地数据集。本研究对基于遥感的城市精细湿地类型动态监测体系的构建具有理论指导意义，对湿地城市的湿地动态监测、保护、管理与合理应用有实际应用价值。

Urban wetland is an important ecosystem of cities, which has a variety of ecological and social service functions, and is closely related to human beings. Wetland City is a high standard and important honor for a city in terms of wetland ecological protection. It is a city awarded the title of "Wetland City" by the International Organization of the "Convention on Wetlands of International Importance Especially as Waterfowl Habitat ". Wetland City is a pioneer and example of urban wetland protection. Wetland City's term of office is only six years, and it needs to be re-evaluated after expiration to determine whether to continue to grant the title. Therefore, it is very necessary to monitor the distribution and change of urban wetlands in time, which is also the demand of wetland cities. It is difficult to extract wetlands because wetlands are complex and diverse. At present, the types of urban wetlands extracted by remote sensing are not sufficiently detailed, and the classification method system of urban wetlands needs to be improved. The traditional dynamic monitoring method of year-by-year classification is time-consuming and burdensome. Therefore, it is necessary to carry out research on remote sensing extraction methods of urban wetlands for wetland cities. The research is closely related to urban wetland protection in the international arena and is forward-looking. It supports China's promotion of wetland city certification and helps summarize China's experience in urban wetland protection. It has good practical application value. The remote sensing of urban wetlands extraction has important theoretical scientific significance and practical application value.

This paper uses dense Sentinel time series data to construct a systematic subdivision method for urban wetlands based on "spectral-geometric-frequency" characteristics, extracts the urban wetland classification results of the first six wetland cities in China in 2020. It also constructs an annual urban wetland classification method based on continuous change detection, extracts the classification results of urban wetlands in six cities from 2015 to 2022. Spatial-temporal characteristics of urban wetlands from 2015 to 2022 were analyzed, respectively, based on detailed wetland types and wetland element indicators. The urban wetland distribution pattern, urban wetland type and element change pattern, and urban wetland protection and restoration pattern of six wetland cities were revealed. This paper innovatively implemented dynamic monitoring and change patterns of urban wetlands, including 17 wetland types in six cities in 8 years. The main features and innovative results of this study are as follows:

(1) Construct an urban wetland subdivision method based on "spectral-geometric-frequency" characteristics, and extract urban wetlands containing 17 types of wetlands. This method used K-fold random forest method, wetland subdivision rules and annual water accumulation frequency to extract wetlands gradually The results of the study found that: ①the overall accuracy of urban wetland classification results in 2020 was above 91%, and Kappa was above 0.87. The accuracy of producers and users of most types of urban wetlands can reach more than 80%; ②Compared with other wetland data products, it is found that the urban wetland extraction range in this study is more accurate and has more categories, the consistency area of the non-wetland classification results and other datasets is high. The research has achieved remote sensing extraction of urban wetlands with classification accuracy of up to 91% and wetland types of up to 17 types.

(2) Construct an annual classification method for urban wetlands based on continuous change detection, and extract the classification results of urban wetlands from 2015 to 2022. This method adopts the idea of continuous change detection and classification algorithm to carry out change detection. Based on the change information, the urban wetland classification results are obtained year by year, and then based on the changed and unchanged areas, different methods are used to extract the annual wetland containing 17 types. The study found that: ① The overall accuracy of change detection in 6 cities is above 80%, and the accuracy of producers in the unchanged area of most cities is higher than that in the changed area; ②Comparing with the land cover data products, the spatial consistency is good, the consistency area of most cities accounts for 70%, and the water consistency area accounts for the highest proportion 90%. Under the premise of effectively reducing the workload, the study realized the annual extraction of urban wetlands from 2015 to 2022.

(3) Based on detailed wetland types, the spatio-temporal variation characteristics of urban wetlands from 2015 to 2022 are analyzed. The results of the study found that: ①The proportion of wetland area in the six cities was more than 5%. Inland herbaceous swamps and permanent rivers were the main types of wetlands in inland cities, accounting for more than 30% of wetlands. Shallow waters , coastal tidal flats, coastal woody and herbaceous swamps are the main types of wetlands in coastal cities, accounting for more than 50%; ②Natural wetlands are the most distributed type, accounting for more than 60%. The proportion of natural wetlands in most cities will increase between 2015 and 2022, and the proportion of artificial wetlands will decrease.

(4) Based on the wetland elements of water, vegetation and soil, the water index (NDWI), vegetation index (NDVI) and soil moisture index (SMMI) were selected respectively to analyze the temporal and spatial variation characteristics of urban wetland from 2015 to 2022. The results of the study found that : ① For the water characteristics, Changde and Changshu had a significant increase trend, while that of Haikou had a significant decrease trend; ②For the vegetation characteristics, Changde had a significant decrease trend, while that of Dongying and Harbin had a significant increase trend; ③For the soil moisture characteristics, Yinchuan, Haikou and Harbin had an obvious downward trend; ④Proportion of the area with stable index changes is the highest among the six cities , can basically reach about 80%.

(5) Summarize the distribution, change and protection patterns of urban wetland distribution. The results of the study found that: ① Urban wetland distribution pattern includes coastal wetland cities, inland plain wetland cities and inland mountainous wetland cities. ②The proportion of stable wetlands can reach up to 60%, and has increased after 2018. The proportion of stable wetlands in Haikou is as high as 95.95% after 2018. The area of degraded wetlands has decreased, and the degree of wetland restoration is higher, and wetland cities have strengthened the protection and restoration of wetlands when wetland cities have passed the certification. ③ The protection and restoration of urban wetlands presents a model of "important wetland protection areas" + "protection policies combining law and application" + "wetland restoration projects".

This paper focuses on the first batch of wetland cities in China, constructs a systematic extraction method and an annual classification method to realize dynamic monitoring of urban wetlands, and provides urban wetland data for the six cities from 2015 to 2022. This study has theoretical significance for a detailed dynamic monitoring system for urban wetland types based on remote sensing, and has practical application value for dynamic monitoring, protection, management and rational application of wetlands in wetland cities.