*2.4. Materials*

In this study, Landsat images and the Digital Elevation Model (DEM) were used for analyzing the ecological landscape. Google Earth provided higher resolutions to recognize the local environmental features and help extract the historical environmental features in the Corona images. The remote sensing data are listed in Table 2. Because the city outskirts where the archaeological sites are located were less affected by urban sprawl before 1995 and have changed a lot under urbanization in recent years, as referred to in some other studies [75,76], we chose the phases of 1995 and 2017 for comparison. It is important to note that these images cannot reflect all the ecological information in the past. However, the multi-source remote sensing images in early time and variance analysis of two periods can help us learn the ecological characteristics in history combined with some documentaries.


**Table 2.** Remote sensing images used in this study.

The data sets of Landsat and DEM were provided by the Geospatial Data Cloud site, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn).

### **3. Study Background**

From 7 to 5 ka B.C., the plain of Ningbo was rarely influenced by marine corrosion except for three rising sea level events caused by the combined effects of regional tectonic subsidence and climate changes [77]. The relative variation of sea level did not cause cultural decline for a long period. The local culture flourished again with the retreat of the coastline, according to the historical changes map of Hangzhou Bay [78]. This area had been a continent of lakes for a long time after that, inferred by the strata information and the changes of paleovegetation composition [79]. Ningbo, with an age-old agricultural civilization, has been proven to have originated as early as 5.8 ka B.C. [80]. The area with a network of rivers and a humid subtropical monsoon climate provided a suitable environment for the development of civilization. This is supported by the evidence of archaeological excavations and the site distribution in early times [81].

At present, Ningbo a fast-developing city that is experiencing the universal problems of urban sprawl and population growth, which generate challenges for the conservation of archaeological sites, especially the early relics of important historical value. As shown in Figure 3a, all the archaeological sites are located in the transition area between the mountainous area and plain on the outskirts of Ningbo in Zhejiang Province. According to the FAO 1990 system of soil classification, all the sites are mainly on the boundaries of Humic Acrisols (ACu) and Cumulic Anthrosols (ATc), as in Figure 3b. The soil is the synthesis of physical, chemical, and biological characteristics. It is closely related to the ecological service system and human activities. Especially the presence of anthrosoils can be used to detect historical human habitation. It was formed due to long-term human activity, such as irrigation or anthropogenic organic matter [82,83]. Humus is the colloidal part of soil organic material with the nutrients and water molecules retained between cosmids [84]. The two types have a significant effect on promoting the formation of an ecosystem and settlement by ancients. Besides that, elevation and proximity to a river are also the important factors for ancient living (Figure 3c,d). Most of sites distribute in the gentle slope of the piedmont area or the river terrace with a certain altitude to avoid flood risk. At the same time, the proximity to rivers ensured adequate water supply for agricultural irrigation and daily life. These integrated demands for living environment causes that the archaeological sites almost locate in the transition of ecological environments. There are 158 archaeological sites from the Han–Jin Dynasties (206 B.C.–420 A.D.) located in this study area. Only three ancient cities named Gouzhang, Yinxian, and Maoxian (currently undiscovered) have been documented as the earliest cities in the basins of Yuyao River, Yinjiang River, and Yongjiang River, respectively [85]. The other sites are ancient tombs. This research focused on the site selection pattern of the three ancient cities compared with other tomb sites at the same time.

**Figure 3.** The spatial distribution of archaeological sites in (**a**) a remote sensing map and (**b**) a soil classification map based on the FAO 1990 system (provided by Cold and Arid Regions Sciences Data Center at Lanzhou (http://westdc.westgis.ac.cn)). The spatial characteristics were analyzed based on (**c**) the elevation range of archaeological sites and (**d**) the proximity to the river.

The ancient city of Gouzhang was discovered in 2004–2012. Its surface is covered with farmland and settlement (see Figure 4). After archaeological excavation, the profile was found to be an irregular rectangle with a 1200 m perimeter and 100,000 m<sup>2</sup> area, whose culture accumulation was from the Warring States Time (475–221 B.C.) to Jin Dynasty (265–420 A.D.) [86,87]. The location of Gouzhang is circled by the Yuyao River in the south with mountains in the east and west. Another ancient city of Yinxian was discovered in 2015~2018. It was inhabited for nearly more than 800 years from the Qin Dynasty (221–206 B.C.) to early Sui Dynasty (589 A.D.) [88]. Yinxian is situated on the mountain of less than a 50 m height in the transition zone between the mountainous area and residential area, with encircling rivers. The site's perimeter is about 760 m, and area is more than 38,000 m2. Its internal residential zone is mainly located in the mountain (see Figure 5). The Yinxian site has convenient transportation in terms of both land and water ways because it is next to the Ningbo Plain in the north and connects to the Xiangshan Port in the south. In terms of the last undiscovered city with the longest documented history, ancient Maoxian should be situated between the east of Maoshan Mountain and west of Ashoka Mountain [85,89], according to the records, which are mostly from the Song–Ming Dynasties (960–1644 A.D.). An ecological structure analysis was applied to discuss the possibility of exploring its site location under the impact of urban sprawl. Because of the ubiquitous ecological transition zone, the methodological application can also be tested in other areas.

**Figure 4.** The scenery of the Gouzhang site (photographed in December 2018) and the ancient port discovered by archaeologists [87].

**Figure 5.** The western wall ruin of the Yinxian site and its classic unearthed relics [88] in the Han and Jin Dynasties.

### **4. Results and Discussion**

### *4.1. Ecological Pattern Analysis*

The ecological environment in the study area was classified in Landsat images. As the results show in Figure 6, the urban sprawl ratio changed from 14.42% in 1995 to 30.73% in 2017 by calculating the settlement area. In contrast, the ratio of the forest area dropped a little, but stabilized at a higher equilibrium of more or less 50%. Generally speaking, the urban sprawl is marked by the tendency of multiplying human activities, but the geomantic environment made up of forest, mountains, and river remains basically invariant. Considering the locations of archaeological sites at the edge of mountains and the stable mountainous environment in early times, we assumed that the major environmental constitutions did not substantially change in ancient times compared with early remote sensing images. This has been proven by archaeological excavations [84,85].

The pattern of patches in an ecological service is helpful for analyzing the location preference and ecological changes of ancient sites. As shown in Table 3, most ancient sites were located in forest and farmland in 1995. After the extension of the settlement area, some locations of sites transferred to patches of settlement. However, the forest is still the major ecological type preferred by ancient sites. With long-term urban development, patches have gone through fragmentation, which is revealed by

the greater number of patches and smaller area of each patch. The shapes of forest patches became more complicated due to the increasing mean perimeter but declining mean area at the same time.

**Figure 6.** The ecological classification based on Landsat images in (**a**) 1995 and (**b**) 2017.


**Table 3.** Classification statistics of the patches where the relics are located.

As demonstrated by the analysis of shape metrics in Figure 7, high values of fractal dimension index (FRAC) that reflect high shape complexity of patches occur at several large patches of forest in both 1995 and 2017. The CONTIG index demonstrates the high spatial contiguity for most of the forest patches, representing strong ties in the ecosystem of 1995, but they decreased to a certain degree in 2017 due to the occupation by the settlement patches. The results of edge effect analysis depend on the ecological pattern.

**Figure 7.** Ecological metrics analysis of 1995 is shown in (**a**), thereinto, the patch number 17 is the location of Yinxian and 19 represents the location of Gouzhang. Ecological metrics analysis of 2017 is shown in (**b**), in which patch number 28 is the location of Gouzhang and 101 is the location of Yinxian.

### *4.2. Edge E*ff*ect of the Ecological Transition Zone*

In the study area, the archaeological sites are located along the border of the mountainous area in the ecological transition zone. Therefore, we tried to quantify the edge effect in the ecotope according to their spatial positions. Six scales were tested to discuss the edge effect changes with the varying edge width from 100 to 1000 m, which represented the buffer size for each site. Among the 158 sites, only the two ancient cities of Gouzhang and Yinxian are presented as an example in Figure 8.

**Figure 8.** Ecological patches distribution of Gouzhang and Yinxian sites. Edge effects with six edge widths (100, 200, 400, 600, 800, and 1000 m) were calculated.

The edge intensity of different ecological types in which the archaeological sites are located are shown in Table 4. Most of the edge intensity increases as the scale also increases, because of more edge complexity and constitutions in the higher scope of the edge width. Considering the different ecological types that existed in 1995 before extensive urbanization, the forest-domain area had few patches and simple shapes, and thus had a weak edge effect compared to other types of farmland, bare land, and settlement area. In general, the ecological types of forest and water make the edge effect more positive to be distinguished. Things are different in 2017, when all ecological types have a basically equivalent edge intensity for the ancient sites on any scale but a higher value than in 1995. This is the result of universal fragmentation along the edge of the forest and river. Therefore, the ecological structure (the constitution and fragmentation of different ecological types) is the reason for the edge effect discrepancy of ancient sites. It could have an effect on the site selection pattern.

**Table 4.** Comparison of the edge intensity for different ecological types the ancient sites are located in.


From the matrix scatter graphs (see Figure 9), all the sites have no obvious relation with the low scale of 100 m, but the edge intensity shows approximate linear growth between adjacent widths with the augmentation of scales. The linearity becomes distinct, gradually referring to the R<sup>2</sup> values of 0.242, 0.591, 0.768, 0.889, and 0.944 from the scale of 100 to 1000 m in 1995 and 0.179, 0.564, 0.696, 0.881, and 0.920 in 2017 with the same order. In practice, considering the redundant information which results from the linear relation of the neighboring scales, one optimum scale can be selected from the edge effect analysis at a medium or large scale. In the next step, we want to discuss how to choose the best analysis scale to recognize a site location using edge effect and further distinguish the ancient cities from other sites.

**Figure 9.** Edge effect relationship between neighboring scales in (**a**) 1995 and (**b**) 2017.

In order to validate the method of edge effect when distinguishing between a relics area and a no relics area, 158 relics and random points with an equal number were compared at six scales (see Figures 10 and 11). In general, the separability aggrandizes based on the growth of average values and the enlarged difference between relics and random points as the scales increase. Moreover, the separability was more prominent at large scales in 2017. In Figure 11, the ancient cities of Gouzhang and Yinxian show the different degrees of separability among all the relics. The edge widths of 200 and 400 m are the scales that could be used to separate Gouzhang from most of the sites through the edge effect of 1995. After the ecological structure changes in 2017, the separability of Yinxian is better obtained through edge effect analysis at medium and large scales from 400 to 1000 m. The separability of the edge effect in the transition zone is helpful for differentiating the ecological patterns of ancient sites.

**Figure 10.** Comparison of the edge intensity between the locations of relics and random points in (**a**) 1995 and (**b**) 2017.

**Figure 11.** The separability of the ancient cities of Gouzhang and Yinxian based on the average edge intensity in (**a**) 1995 and (**b**) 2017.

By the analysis above, we could conclude that the different variation tendency of the edge effect was based on the discrepancies of the ecological structure for the locations of ancient cities. This manifested in different site selection patterns by ancient peoples. In this study, two patterns were recognized. One is the ecologically balanced pattern that the preferred living environment was constituted of mountains, rivers, and other ecological types. Edge effect shows a unique trend of changes in low–medium scales in the processing of urbanization. Because of rich ecological types and a strong edge effect in the early ecosystem, ancient peoples could access abundant resources and materials. At the same time, the ecological environment remained stable and reduced the influence of patch fragmentation to some extent. Therefore, the growth of edge effect is slow in certain scales under the background urban sprawl. The ancient city of Gouzhang is the representative of this pattern. The other site selection pattern is shown from the preference of the mountainous (also the forest-domain pattern in this study) ecological environment. The variations of edge effect are basically consistent with the fragmentation of the transition zone in front of the mountain. The main ecological constitution is simple and not stable with low resistance to environmental changes in general. Therefore, the edge effect changed a lot during the settlement expansion in this area. However, this could help us highlight the edge effect of the ecological transition zone and separate the important but unremarkable environment in the early ecosystem. The site selection of ancient Yinxian is this pattern. The edge effect in the ecological transition zone was found to have some relevance in terms of ancients' environmental cognition. Then, we aimed to discuss the two site selection patterns further with the ecological network method in a geomantic environment of closed spaces.

### *4.3. Ecological Networks and Centrality Importance*

The geomantic analysis of site locations is concentrated on closed spaces, which are the minimum circles constituted of mountains and rivers. The internal closed area labeled as Zone 1 is the minimum circle consisting of the circling mountains, while the external closed space Zone 2 consists of mountains stretching far away. Both the spaces are traversed through by the flowing river. As an example, for Gouzhang City (see Figure 12a), we can identify the core geomantic environment of Zone 1 surrounded by adjacent mountains of Dawanshan and Xiaowanshan and some river branches, and the external closed space of Zone 2 is comprised of the mountains of Mashishan and Changmingshan in the distance. This site location may be selected by referencing both the landform and water position in the geomantic environments, as shown in Figures 1 and 2. About 35 km away in the southeast is the ancient city of Yinxian. The closed spaces can also be constructed by the minimum and maximum closed circles (see Figure 12b). The Yongxin River runs through the spaces and approaches traffic corridor to the Xiangshan Port in the south.

**Figure 12.** Geomantic analysis of the ancient cities of (**a**) Gouzhang and (**b**) Yinxian as shown in Google Earth images from 31 December 1984.

As shown by the geomantic analysis above, both sites of Gouzhang and Yinxian generally follow geomantic principles and also acclimatize to local circumstances. Of course, this landscape of mountains and rivers usually restricts the desirable location to the ecological transition zone, thus causing the edge effect, as we analyzed. Furthermore, additional ecological meanings behind the geomancy in terms of site selection were discussed by network analysis in wider ranges of Zone 2.

For the network constructed by ecological patches and links between neighbors in the geomantic environment of closed space, Gouzhang has a higher density of vertices and edges than Yinxian, as demonstrated in Table 5. Its circuitry reaches 0.91, indicating a higher number of loops in the network, and the connectivity is 0.94, which shows that most of the vertices are well-connected in the network. It can be concluded that the ecological network is more stable, with a real benefit for the long-time exchange of material and energy, than that of Yinxian. In addition, the high ratios of edge to vertex show that the network structure is more complicated and complete. The network complexity of Yinxian rises up with the ecological fragmentation. It can be concluded that the complex and stable ecological network will bring small changes in the closed environmental space.


**Table 5.** Network analysis of Gouzhang (GZ) and Yinxian (YX) regions.

In order to quantify the importance of each patch in the ecosystem, centrality importance was introduced in our study (see Figure 13). In the network of the closed space in the Gouzhang region in 1995, the vertex of V5 representing the patch where the ancient city is located has the highest centrality of 1.10. It reflects that the selected location plays an important role in the ecological network. The adjacent patches of V54 (Mashishan Mountain) and V80 (Yuyao River) also have high centrality or closeness. It increases the importance of this site location. Although the patch number rises significantly in 2017, the centrality of the site location (V33) is still high, and the neighboring vertex of river patch V124 and forest patch V102 retains the high centrality. The river and mountain (covered with forest in the study area) always play important roles in the network, as in the geomantic environment. For the Yinxian site, its location of forest patch V26 (the Chengshan Mountain) had very high centrality in 1995. Until 2017, the centrality importance of Chenshan Mountain reduced to 0.40 at the medium level. However, the neighboring patches of V66 and V31, with high centrality, could make up the loss of its importance. We can conclude that the mountain and river patches play important roles in the geomantic environment and ecological network, and the locations of high centrality had important ecological meanings for site selection in ancient times. The site locations have the advantages of accessing the life necessities in ancient times because of their high centrality importance in the ecosystem.

**Figure 13.** Ecological network analysis of Gouzhang in (**a**) 1995 and (**b**) 2017, as well as Yinxian in (**c**) 1995 and (**d**) 2017. The red dots represent the patches where the two ancient cities are located.

Based on the above analysis of the edge effect and ecological network, some interrelations between the ecological structure and environmental preference were explained. In ancient times, the site location may have been selected for its complexity and diversity of ecological types (for example, the balanced portions of mountain, river, farmland, and others), which would have presented a stable ecological and geomantic environment for long-term occupation. Even though there has been an appreciable impact of urban sprawl in the whole area, large variation of the ecological structure would have not taken place. People could have accessed enough goods in the ecosystem to maintain their life, supported by its strong edge e ffect and high centrality. The waterborne and land transport also permitted them to settle there. As for another classic site selection pattern, few ecological types and a simple structure usually lead to a relatively weak stabilization in the local environment. At the same time, the strong edge e ffect represents the high possibility of energy flowing to anthropological needs in the ecosystem. Therefore, the regional environment is susceptible to change, especially in the background of resource shortage and population swell in the process of urbanization, sometimes accompanied by the growth of the settlement area and edge intensity. It also proves that a large edge intensity is the indicator of a high possibility of environmental satisfaction for site selection, whether in ancient or modern times. In addition, the preferred mountain-domain area can provide not only an auspicious living space in geomancy but also a defensive fort and energy hub. The study cases of Gouzhang and Yinxian correspond to the above two patterns, respectively. To some extent, the ecological significance behind ancient site selection pattern can help us to learn more about human–environment relationship in ancient times.

### *4.4. Exploration of Undiscovered Archaeological Site*

The environment of Ningbo is classic for its mountainous area and dense network of rivers. It provides suitable locations for ancients' settlement, which has been proved by the analysis of ancient site selection patterns in this research. The ecological method to recognize the environmental preference of this region is also applicable to other similar regions in the middle and lower reaches of the Yangtze River. Although the cultural and historical background varies along the Yangtze River, there are some universal environmental views shared among the local ancient people about how to utilize natural conditions and select site locations. Therefore, the site selection pattern may have a certain degree of similarity. For methodological validation of edge e ffect and ecological network, and a deeper revelation of cultural landscape in the Ningbo Plain, we analyzed the ecological structure and geomantic environment in the area of Maoshan Mountain in the basin of the Xiaojiajiang River. This comprehensive analysis was also used to find the new clues of the lost city of Maoxian.

There are 29 ancient sites in the transition zone around Maoshan Mountain, of which 16 were located in forest patches, three in settlement patches, and 10 in other patches in 1995. Until 2017, 14 were situated in forest patches, 13 in settlement patches, and only two in other patches. Therefore, the forest always provides the major environment for site locations. As the settlement region increases during urbanization, it plays a more important role in the ecological environment where the relics are located. The ecotone between the forest and settlement has significant meanings for ancient living. Although Maoxian has not ye<sup>t</sup> been discovered, we can speculate its possible location around Maoshan Mountain in the basin of the Xiaojiajiang River, according to archaeological discoveries [68,69]. To narrow the scope further, the suspected zones of MX1~MX6 were divided by geomantic analysis and regarded as advantageous environments for ancients selecting a residential location through the systematic evaluation of the constitution of the mountain, water, and orientation in these closed spaces. The ecological transition zone was built to explore the ancient Maoxian, as demonstrated in Figure 14.

We divided this transition zone into 214 grids with a 400 m width. The edge e ffect analysis was conducted to reflect the edge intensity of each central point in a grid with the scales of 400, 600, 800, and 1000 meters. From the scatter diagram presented in Figure 15, the edge intensity of neighboring scales had a significant linear relationship in both 1995 and 2017, which coincides with the previous analysis results displayed in Figure 9. When considering the enlarged disparities of the edge e ffect on a large scale, a 1000 m scale was chosen for further analysis.

**Figure 14.** Geomantic analysis of the possible location of the Maoxian site. The river system was extracted from a Corona image of 1964 and partly refers to the Google image of 1984. The closed spaces of MX1–MX6 are mainly constituted of mountain and water with a suitable orientation.

**Figure 15.** Edge intensity calculation with different scales in the transition zone around Maoshan Mountain: (**a**) the evenly distributed sampling points for edge intensity computing, and (**b**–**d**) the edge intensity relationship between different neighboring edge widths.

As shown in Figure 16a, Maoshan Mountain is composed of homogeneous forest patches and thus has a low edge effect. All the archaeological sites are located in this median zone of edge intensity, which imply that they had a similar ecological structure in early times. In Figure 16b, the medium zone appears differentiated for the ecological structure changes under the influence of urbanization development. The ancient sites are generally distributed in high-value regions in the ecosystem of 2017. These high anomaly regions signify the changes of ecological structure from patch fragmentation. In terms of the proximity of the environmental pattern and nearness of the spatial position, Maoxian might have similar high separability of edge effect as the ancient city of Yinxian. Therefore, the regions of MX2–MX5 with high edge intensity in front of mountains should be focused on when exploring the ancient site of Maoxian.

**Figure 16.** Regional edge effect distribution obtained by the Kriging interpolation method for (**a**) 1995 and (**b**) 2017.

In the ecological network, the central patches play the role of connecting medium and interchange hubs in the network. In this transition zone along Maoshan Mountain, the ancient sites are distributed in the range of centrality transition from high to medium, as shown in Figure 17. What is different between the two periods is that the forest patches comprise the major part in 1995, with high values, but settlement patches have a central role in the ecosystem for urbanization in 2017. This change reveals that centrality transfers from the natural domain to human domain in the ecological network. Like Yinxian, the undiscovered city of Maoxian may have displayed a high centrality in the natural ecosystem, and with patch fragmentation under settlement sprawl, its central status may have decreased in the network for the unstable ecological network. The regions of MX2 and MX3 match this condition on the basis of the edge effect analysis result. The suspected area can thus be narrowed down to a smaller space.

By further interpretation of Google Earth images, some suspected archaeological features were extracted explicitly in the suspected zones of MX2 and MX3, as shown in Figure 18a, including the possible city walls and the reservoir nearby, but these were not found in other places around Maoshan Mountain. The size of this suspected area is 80 × 60 m. The reservoir might have been part of the city used for daily life and irrigation. Spatial analysis with the ArcGIS tool was used as an auxiliary method to confirm the importance of this location. The tomb groups in the target area around Maoshan Mountain in the basin of Xiaojiajiang River are concentrated in the standard deviation ellipse (SDE) and encircle the mean center, which was computed by the weight of relics count in the spatial analysis tool of ArcGIS and represents the residential center or high hierarchy position that may have some links with the ancient city of Maoxian. This result shows that the suspected site next to the mean center may have a central status among the contemporaneous sites in Figure 18b. Therefore, it seems that the site selection of ancient Maoxian may also emphasize the ecological structure, as analyzed by the edge effect and network structure. More important for us, this suspected site has been affirmed to have been built in a very early time, following a preliminary field investigation (sees Figure 18c–e), though we will not know more details before archaeological excavation. However, this methodological application demonstrates that the edge effect and ecological network are worth applying when trying to find the location of an ancient site.

**Figure 17.** The centrality distribution in the ecological network of (**a**) 1995 and (**b**) 2017.

(**b**)

(**a**)

**Figure 18.** Suspected archaeological target from remote sensing interpretation and a field investigation, (**a**) the interpretation of archaeological features in a Google image, (**b**) the centrality analysis of tomb groups in the same period of ancient Maoxian in ArcGIS, (**<sup>c</sup>**,**d**) ancient wall of the suspected site, and (**e**) the reservoir next to the suspected site now covered with dense grass.
