Objective  The sand flat polders in the Yangtze River Delta are the products of the continuous inning of tidal flats along and silting of sandbanks amidst the Yangtze River. They are mainly distributed along the lower reaches of the Yangtze River. With thousands of years of natural evolution and human settlement activities, they have formed a unique landscape. In the Yangtze River Delta, the formation of shoreline spans nearly 7,000 years, the natural conditions and social environment are distinctive in different periods, and the land generation mechanism under different backgrounds finally leads to the diversified features of the water landscape network. Since sand flat polders are typically distributed linearly across several counties and cities along the rivers concerned, they vary greatly in different areas in terms of landscape morphology, and relevant historical documents and drawings are fragmented and scattered that entail a systematic combing and integration, which is so complex that has prevented domestic and foreign scholars from paying attention to sand flat polders. It is necessary to systematically divide the morphology type and identify the distribution range of the sand flat polders in the Yangtze River Delta. In view of the large differences in landscape morphology of the sand flat polders in different areas and the difficulties in data collection, this research suggests that the method of typology be used to systematically classify and identify the landscape types and distribution of the sand flat polders. Moreover, the ML-based AI technology, with strong remote sensing image identification capability, can also be used to replace human brain in processing regional distribution identification of the sand flat polders.

Method  Based on the analysis of historical documents related to the generation mechanism and distribution characteristics of the sand flat polders in the Yangtze River Delta, and in combination with the 1917 survey map of the Military Commission of the Republic of China, the 1937 survey map of the Japanese Land Survey Department and other historical mapping maps, as well as the “keyhole” remote sensing image data of the United States Geological Survey (USGS), this research judges the types of the sand flat polders in the Yangtze River Delta, and analyzes the morphological characteristics of each type of the sand flat polders through a case study of typical polder areas, with the data obtained being used as the original data to mark all types of the sand flat polders in the Yangtze River Delta. Moreover, the research adopts the CNN-based SegNet semantic model to segment the “keyhole” remote sensing image with the water network texture as the main identification feature, based on which the distribution range of these types of sand flat polders is predicted. The predicted results are modified by confidence cluster analysis and manual visual interpretation. From the perspective of landscape architecture, the research analyzes the regional distribution characteristics and development dynamics of the water network of the sand flat polders in the Yangtze River Delta, and summarizes the wisdom of landscape construction of these sand flat polders.

Results  This research finds that the sand flat polders in the Yangtze River Delta can be divided into three types: 1) the inner sand flat polders represented by Guazhou area; 2) the outer sand flat polders represented by Yangzhong area; 3) the new reclamation polders at the river estuary represented by Haimen area. The results show that the change of physical geographical environment is the dominant factor influencing the landscape structure, distribution and water network development of the sand flat polders. For example, the inner sand flat polders are distributed in the early inland zone, the outer sand flat polders are distributed in the middle and late riverside zone, and the new reclamation polders at the river estuary are distributed in the late estuary zone. In the early years, the Yangtze River carried a balanced amount of mud and sediment, and the population increased steadily, making the water network structure of the polders close to the density and spatial dispersion of the settlements. In the middle and late periods, taking Yangzhou-Zhenjiang as the boundary, the polder network and settlement density in the east were much higher than those in the west, mainly resulting from the acceleration of riverbed silting rate in the east area. Besides, due to the population surge in the late period, the density of the polder water network in the estuary of the Yangtze River gradually increased and became more regular from west to east.

Conclusion  The research shows that the typology method can be used to extract and classify landscape features of the sand flat polders, and the ML-based AI technology can help identify the distribution range of different types of polders. It is found that the landscape structures of the three types of the sand flat polders in the Yangtze River Delta show diachronic evolution. The formation, evolution and development of the sand flat polders are closely related to such factors as development degree of beach land, silting rate, population growth rate and tidal influence range. The formation mechanism driven by multiple factors well reflects the environmental adaptability of the sand flat polders. More specifically, the Chinese ancient working people used and transformed nature while constantly adapting to nature, presenting a relatively mature state of dynamic balanced development, which can reflect the landscape construction of the environmental adaptability characteristic of the sand flat polders in the Yangtze River Delta. This research reveals the process of mutual selection and mutual adaptation between the construction of traditional human settlements and natural systems along the Yangtze River, in order to provide a basis for the protection of historical and cultural heritage, the construction of ecological corridors and the promotion of comprehensive and coordinated development in the Yangtze River Delta region. The research can also provide a reference for similar researches on human settlement environment.