Anthropogenic influencing mechanisms on earthen sites preservation under the spatial-temporal variation of land utilization

Abstract

Rapid urbanization and land use change increasingly threaten earthen site conservation. As a primary medium of human activities, land use reflects diverse human behaviors. Clarifying the correlation between earthen site preservation and land use is essential to uncover the mechanisms of influence of anthropogenic activities. This study selected 44 castles of the Ming Great Wall in Shaanxi Province, China, as research objects. Through applying the gravity model, multiple linear regression, and the Levenberg-Marquardt algorithm, a theoretical model was established to reveal the relationship between the linear absence ratio of sites and land use. The results have shown that construction land negatively affects site preservation in an exponential decay pattern, while agricultural and ecological lands support conservation, following exponential and power-law decay patterns, respectively. By 2035, approximately 60% of the castles will face challenges due to urban growth. Two protection zones were comprehensively determined. This study provides valuable insights for heritage conservation and urban planning.

Introduction

Earthen sites are invaluable cultural heritages with significant historical, artistic, cultural, social, and scientific value^1,2. Owing to their unique materials and rammed earth construction techniques, earthen sites inherently possess a fragile structural nature³. As a result, most earthen sites have experienced ongoing degradation due to external factors like wind and rain, leading to deterioration such as scaling, cracking, sapping, and so on^4,5,6. Such deterioration has been accelerating the degradation process^7,8. However, rapid urbanization in recent years has made the impact of anthropogenic activities on the earthen sites increasingly prominent, such as residential construction, farmland development, and agricultural irrigation⁹. Compared to natural factors, human activities tend to have more direct and severe effects on the degradation of these valuable sites^10,11,12,13. Therefore, it is urgent and necessary to research the impact of human activities on the preservation of earthen sites.

According to the relationship between human activities and archeological sites, many scholars have constructed quantitative indicators to characterize the impact of anthropogenic activities from different perspectives. The Geomorphologic Impact linked to Tourist cave conditioning (GTU) was utilized to identify geomorphic surfaces associated with tourist use, which facilitates the assessment of the impact of human activities on the site's surfaces¹⁴. Urbanization Intensity Index (UII), including building area, nighttime lighting, and population density data, has been applied to evaluate the risk of anthropogenic damage faced by 79 World Heritage sites¹⁵. In addition, the Heritage Impact Assessment tool (HIA) has been used to identify the effects of human activities on cultural heritage sites by qualitatively comparing the visual integrity across four world heritage sites¹⁶. Regarding linear earthen sites, the positive correlation between population density and the conservation status of earthen sites has been revealed by proposing the Linear Absent Section Ratio (L-ASR) indicator, which quantitatively characterizes the damage caused by anthropogenic activities¹⁷. These studies highlighted the strong correlation between human activities and the conservation status of cultural heritage sites. However, a considerable gap remains in quantitative evaluation and predictive studies, especially concerning earthen sites. Thus, developing a dynamic, numerical model that can assess and predict cultural heritage site preservation status is essential. Such a model will deepen our understanding of the long-term impacts of anthropogenic activities and provide a more refined framework for heritage conservation.

Machine learning techniques have been proven to have the potential to support analysis in this process, and it has been widely applied in the conservation of cultural heritage¹⁸. When it comes to specialized studies utilizing machine learning for architectural heritage protection, the majority of efforts are concentrated in two main directions: one involves the identification, analysis, and classification of building structural components, such as columns and vaults^19,20, and the other is surface damage detection, which is used to assess the health condition of cultural heritage^21,22. In the earthen site conservation field, 3D scanning technology and the Random Forest algorithm, a form of machine learning algorithm, were applied to investigate the geometric and spatial regularity of rammed pits on a typical rammed layer interface of the Ming Great Wall²³. Similarly, Random Forest analysis in conjunction with point cloud data was adopted in another study to classify deterioration regions on wall surfaces²⁴. Machine learning algorithms, including Support Vector Machines (SVM)²⁵, BP neural networks²⁶, Fuzzy C-means (FCM)²⁷, and K-means clustering²⁸, have been integrated into the vulnerability assessment system for earthen sites. To address the limitation of these methods’ excessive dependence on experts, some scholars have proposed the entropy weight-TOPSIS method, which is entirely based on objective weight determination²⁹. The entropy weight method calculates the weights of various indicators using actual disease data. When combined with the TOPSIS method, it can more objectively evaluate the vulnerability of earthen sites. However, the emphasis of evaluations in these research works lies mainly in the static vulnerability assessment of the specific earthen sites under present conditions. Consequently, establishing a measurable, dynamic, and anticipatory evaluation model can play a crucial role in enhancing and complementing the existing assessment system for earthen sites.

Land utilization is a critical indicator that reflects the type and intensity of human activities^30,31,32. As the physical foundation of human activities, the land is utilized in various ways that unfailingly change and impact the surrounding environment, including earthen sites. According to the China Multi-Period Land Use Land Cover Change Remote Sensing Monitoring Dataset (CNLUCC), the land was categorized into three main types: construction land (namely the land used for constructing various buildings and infrastructure, including built-up land), agricultural land (namely the land used for agricultural production, including crop land), and ecological land (namely the land with important ecological functions, including wood land and grass land and water body)^33,34. After the field investigation, it was found that the earthen sites are primarily located in residential areas, surrounded by these three types (Fig. 1). Additionally, remote sensing image slices from different periods (1980-2020) collected from the Landsat 8 satellite of the United States, further illustrate the increasing risks faced by the sites during the urbanization process, with a significant expansion of construction land. For instance, by taking the Qingshui Castle as an example through comparing multi-temporal land use data within a 1 km radius around the site from 1980 to 2020, a significant increase in the construction land surrounding this site due to urban development was observed (Fig. 2), which indicated an increment in the intensity of human activities within the preservation environment of earthen sites. Therefore, it is crucial to explore the dynamic relationship between land use changes and earthen site preservation and uncover the anthropogenic influencing mechanisms of the preservation of earthen sites under spatial and temporal variations in land use, which forms the foundation for scientifically conducting conservation work and developing conservation plans and policies.

Fig. 1: Photos of some earthen sites surrounded by three typical land use types.

a Construction land; b Agricultural land; c Ecological land.

Fig. 2: Land use evolution within a 1 km radius around the earthen sites from 1980 to 2020.

(taking Qingshui Castle as an example).

Located in the north of Shaanxi Province, China, Yulin represents cultural resource-based cities and is home to many earthen sites³⁵. In recent years, Yulin has experienced rapid economic growth, with an urbanization rate exceeding 60%³⁶. This accelerated urbanization has resulted in the drastic alterations of spatial layout and structure of land use in this city³⁷. As one of the world cultural heritage sites, most of the Ming Great Wall in Shaanxi Province were located within Yulin city, and its preservation status has been inevitably severely impacted by human activities³⁸. In this study, to reveal the mechanisms by which anthropogenic activities contribute to the destruction of earthen sites, the 44 castles (called “sixty-six castles” in history) along the Ming Great Wall in Shaanxi Province were selected as the research objects³⁹. By quantitatively analyzing the dynamic correlation between three land utilization types and earthen sites preservation status, a comprehensive predictive model for the L-ASR of earthen sites was constructed, and the potential damage of these sites by the time of 2035 as well as the priority of conservation efforts can be further discussed. The results provide reliable guidance for the current protection work of earthen sites. In addition, the achievements of this study can not only effectively fill the gap in dynamic prediction within previous earthen sites assessment systems but also offer valuable references for the conservation of other types of archeological sites.

Methods

Study area

As the central distribution zone of the Ming Great Wall in Shaanxi Province, Yulin City represents a culturally significant region in China, currently undergoing rapid development and economic transformation⁴⁰ (Fig. 3a–c). Since the implementation of the Western Development Strategy in the 21st century, Yulin has experienced substantial economic growth, with its GDP ranking first among the non-provincial capital cities in western and central China in 2023 (Fig. 3d), and its urbanization has been accelerating (Fig. 3e). Nevertheless, development challenges such as unstable land use structures, a sensitive and fragile ecological environment, and the decline of cultural and tourism resources have become increasingly prominent. The rapid urban expansion has inevitably affected the sustainable preservation of earthen sites in Yulin⁴¹. As an essential component of the defensive system of the Ming Great Wall, 44 castles responsible for management, military deployment, monitoring functions, etc. Historical records in the Yulin Fu Zhi indicate that the castles were initially built with dimensions of approximately 32 chi in height, 12 chi in base width, and 6 chi in top width (1 chi ≈ 0.32 meters)⁴². The angle of the walls’ cross-section is approximately 84°. All these castles were built using rammed earth techniques, with measured individual rammed layers ranging from 10 cm to 20 cm thick. Over time, these sites have suffered from constant anthropogenic activities. In this study, 44 castles were selected as research objects (Table 1), which are mainly located in Yulin City. Focusing on the quantitative relationship between the change of land use around 44 castles and the L-ASR of these castles, this research aims to reveal the influencing mechanisms of anthropogenic activities on the destruction of earthen sites under the spatial and temporal variation of land utilization.

Fig. 3: Study area.

a Geographic location of China; b Geographic location of Shaanxi Province; c Geographic location of 44 castles along the Ming Great Wall; d GDP of Yulin city; e Urbanization rate of Yulin city.

Table 1 Basic information of the research earthen sites⁴²

Land use data

The multi-period land use data (1980-2020) employed in this paper were collected from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences (http://www.resdc.cn/) with a spatial resolution of 30 m (Fig. 4). The land use data within a 1 km radius of these earthen sites were extracted by the “Extract by mask” tool in GIS (Fig. 5).

Fig. 4

Land-use data sets from 1980 to 2020.

Fig. 5

Land use data processes.

L-ASR data

The Ming Great Wall is a typical representative of linear earthen sites in China. The authors applied the L-ASR index proposed by our group to characterize the preservation status of such 44 castles since their construction, and the specific formula was as follows¹⁷:

$$L-AS{R}_{{\rm{i}}}=1-(R{L}_{i}/CL)$$

(1)

where ${L-{ASR}}_{i}$ is the linear absent ratio of earthen sites, used to quantitatively characterize the length of absent walls in the entire site, i is the year. In this research, the authors only focused on two main periods, 2009 (${L-{ASR}}_{2009}$) and 2024 (${L-{ASR}}_{2024}$). RL_i is the length of the wall remains of the site. RL₂₀₀₉ and RL₂₀₂₄ represent the length of the wall remaining at these sites in 2009 and 2024, respectively, and CL is the original length of the wall constructed originally. The Shaanxi Ming Great Wall Resource Survey Report contains textual and graphical documentation of the earthen site preservation status by 2009 and their original length data. This report was produced by the Shaanxi Provincial Bureau of Cultural Relics and the Shaanxi Provincial Bureau of Surveying and Mapping between 2007 and 2009⁴³. By scaling the plan view drawn in this report in AutoCAD, RL₂₀₀₉ and CL can be directly measured. Moreover, with the aid of UAV, the authors scanned 17 representative sites of these 44 castles in the field in July 2024, collecting the data on the current remains of the site’s walls, namely RL₂₀₂₄⁴⁴. The flowchart of the data acquisition and process was shown in Fig. 6.

Fig. 6: Processes of calculating L-ASR₂₀₀₉ and L-ASR₂₀₂₄.

(taking Qingshui Castle as an example)

Gravity model

According to the first principle of geographic economics, near things are more related than distant things. In previous study¹⁷, our research team has empirically tested the distance-decay assumption by fitting the population density of surrounding villages (as a proxy for human activity) against the L-ASR index (quantifying site degradation), and the significant inverse correlation obtained before have confirmed that heritage site degradation decreases with increasing distance from population centers. As a classic spatial interaction model, the Gravity Model has been widely applied in land use planning, urban development, and environmental impact studies, particularly in scenarios where the intensity of interaction or influence is known to decay with distance^45,46. Moreover, compared to other models, such as the Maximum Entropy Model (MaxEnt)⁴⁷, calibrating cellular automata (CA)⁴⁸, and artificial neural network (ANN)⁴⁹, the Gravity Model offers a significant advantage in its relatively low data requirements while maintaining robustness⁵⁰. Therefore, the Gravity Model was introduced to quantify the impact of human activities on earthen sites by assessing the intensity of the interaction between these sites and three types of land. The simple form of the gravity model can be expressed as follows:

$${W}_{ij}=K{P}_{{\rm{i}}}{P}_{{\rm{j}}}f({d}_{ij})$$

(2)

where W_ij is the interaction intensity of i and j, and W_con, W_agr, W_eco are the interaction intensity between the site and construction, agricultural, and ecological land, respectively; K is a constant coefficient; P_i and P_j are the qualities of the interacting entities, often expressed in terms of area, population or economic indicators. For simplicity in analysis, the castles were treated as abstract points. Hence, P_i is 1, P_j denotes the area of different types of sites, and P_con, P_agr and P_eco are the areas of the construction land, agricultural land, and ecological land, individually, which can be obtained from land-use data sets using GIS. f(d_ij) is the distance-decay function, and their three common forms are as follows⁵¹.

$$f({d}_{ij})={{d}_{ij}}^{-\alpha }(\alpha > 0)$$

(3)

$$f({d}_{ij})=\exp (-\beta {d}_{ij})(\beta > 0)$$

(4)

$$f({d}_{ij})=\exp (-\gamma {{d}_{ij}}^{2})(\gamma > 0)$$

(5)

Equations (3)-(5) are power-law, exponential and Gaussian decay functions respectively, where the parameters α, β and γ can be obtained as the optimal values by using the LM algorithm⁵²; Larger values of α, β, or γ result in steeper decay curves, meaning the interaction intensity decreases more rapidly as the distance increases; d_ij is the average distance from a single class land-use spot to the castle point. For the convenience of the subsequent fitting calculation, ${P}_{i}{P}_{j}\,f({d}_{ij})$ in Eq. (2) was defined as $\widehat{{W}_{ij}}$, and $\widehat{{W}_{con}}\,$, $\widehat{{W}_{agr}}$, $\widehat{{W}_{eco}}$ to denote the values of the construction land, agricultural land, and ecological land, correspondingly.

Multiple linear regression

In this paper, the multiple linear regression model was utilized to construct the evaluation and prediction model characterizing the comprehensive impacts of the three types of land use on the earthen site to explore the influencing mechanisms of anthropogenic activities on the destruction of earthen sites^53,54:

$$Y={C}_{0}+\mathop{\sum }\limits_{i=1}^{P}{C}_{i}{\hat{W}}_{ij}+\varepsilon$$

(6)

where Y is the combined impact indicator of these three types of land use on the earthen site, C₀ is the constant term, C_i is the regression coefficient, i∈ (1~p), p takes 3, and $\widehat{{W}_{ij}}$ is the values of $\widehat{{W}_{con}}$, $\widehat{{W}_{agr}}$, $\widehat{{W}_{eco}}$, respectively. $\varepsilon$ is the error term, E ($\varepsilon$) = 0, and R² is used to evaluate the fit of the model, with a value of 0-1, and the larger the R² is, the better the model fit is.

CV-Markov algorithm

In this study, the CV-Markov algorithm was employed for dynamic land use forecasting, absorbing the advantages of Traditional empirical-statistical models and dynamic models^55,56. Implemented within the framework of Geographic Information System (GIS) technology and the PLUS platform, the CV-Markov algorithm demonstrated remarkable predictive accuracy and substantial application potential⁵⁷. The steps are as follows: Initially, land use maps from 2005 and 2020 were extracted from the land use database. Subsequently, transition matrices were constructed through Markov chain analysis, enabling the quantification of probabilities for land use conversions between different categories. Next, twelve carefully selected assessment indicators, encompassing natural and socioeconomic factors, were employed to generate land use transition potential maps.

Results

Land use development characteristics

In this research, GIS was used to statistically analyze the area of construction land, agricultural land, and ecological land within 1 km radius of the 44 castles from 1980 to 2023 (Fig. 7). Over the past 40 years, the proportion of the area of construction land has significantly increased, whereas the proportion of agricultural land has decreased dramatically. The proportion of ecological land has remained relatively stable. The entire period can be roughly divided into four stages (I–IV) based on land use changes. Among them, Stage I (before 2005) and Stage III (2008–2018) are the steady development stage, while Stage II (2005–2008) and Stage IV (2018–2023) are the mutation stage. The emergence of these two mutation stages (Stages II and IV) coincides closely with the introduction of relevant national policies regarding the conservation of cultural heritage sites. For example, in 2006, the state successively introduced policies such as Administrative Measures for the Protection of World Cultural Heritage (November) and Regulations on the Protection of the Great Wall (December), which rigorously restricted construction activities around the site, directly triggering the first mutation period. Similarly, the release of the Overall Work Plan for the 13th Five-Year Great Wall Protection Project in 2016, and Shaanxi Province’s clarification of the Great Wall Protection and Construction Control Zone as a provincial heritage protection unit in 2017 contributed to the second mutation period.

Fig. 7

Evolution of the three land use categories from 1980 to 2023.

Correlation between each land use type and earthen site conservation

The land is the intuitive reflection of human intention, and we focused on the influence of a single type of land use on their surrounding earthen site, to explore the correlation between ${L-{ASR}}_{2009}$ and W_con, W_agr, and W_eco. Given the unavailability of land use data in 2009 and the short-term stability of land use, W_con, W_agr, and W_eco were calculated based on land use data in 2010 and then were fitted with ${L-{ASR}}_{2009}$ separately (Table 2). However, in the calculation of Eq. (2), there is no conclusion about the optimal attenuation mode for each type of land, so based on Eqs. (3), (4) and (5), the values of α, β, and γ were obtained by the advanced LM algorithm in these three standard modes (Table 3) to choose the optimal attenuation modes.

Table 2 Data on the architectural information of 44 castles in 2009 and L − ASR₂₀₀₉

Table 3 Calculation results and fitting effects of three attenuation function parameters

The fitting results (Table 3 and Fig. 8) indicate that all three types of land use affect the preservation of earthen sites, and their intensity of influence gradually decreased with increasing distance.

Fig. 8

The fitting of theoretical value of L-ASR₂₀₀₉ based on single land use type and actual L-ASR₂₀₀₉.

Based on the premise ($\alpha$, $\beta$, γ > 0), it was revealed that ecological land performs well under the power-law decay model (α_eco = 0.358993, R² > 0.4), and it is beneficial to the preservation of the site; Otherwise, after a careful consideration of the range of values of α, β, and γ, the accuracy of results, and the fitting efficiency, it was shown that agricultural land also has a positive effect on the preservation of the earthen sites, and the decay of its influence intensity is in line with the exponential model (β_agr = 0.000252, R² > 0.5). In contrast, the construction land negatively impacts the status of earthen sites, and the aforementioned results are verified by fitting the exponential decay model (γ_con = 0.004507, R² > 0.7).

Field investigations further validated that the relationships between the three land types and the earthen sites align with real-world conditions. Construction land negatively impacts on the sites primarily due to intense human activities, such as soil extraction, factory and house construction, and the development of road networks, all of which cause severe and direct physical damage. In contrast, agricultural land, through its crops, acts as a natural barrier, protecting the sites from harm by birds, insects, animals, and human activities. At the same time, the flat areas surrounding the castle sites provide an ideal environment for farming. This means the sites and human agricultural activities are mutually reinforcing. For ecological land, human activities are primarily focused on research, conservation, and educational purposes, and the intensity of human activities is lower compared to the other two types of land use. Additionally, vegetation in ecological land, including shrubs and trees, offers similar protective benefits, with its lower levels of human activity further reducing the impact on the sites.

Then, we obtained the following formula for the interaction intensity between a single type of land use and the earthen site:

Construction land:

$$\left\{\begin{array}{l}{W}_{con}=0.00458{\hat{W}}_{con}+0.211295\\ {\hat{W}}_{con}={P}_{i}{P}_{j}f({d}_{ij})\\ f({d}_{ij})=\exp (-0.004507{d}_{ij})\end{array}\right.$$

(7)

Agricultural land:

$$\left\{\begin{array}{l}{W}_{agr}=-0.00033{\hat{W}}_{agr}+0.747283\\ {\hat{W}}_{agr}={P}_{i}{P}_{j}f({d}_{ij})\\ f({d}_{ij})=\exp (-0.000252{d}_{ij})\end{array}\right.$$

(8)

Ecological land:

$$\left\{\begin{array}{l}{W}_{eco}=-0.00164{\hat{W}}_{eco}+0.721716\\ {\hat{W}}_{eco}={P}_{i}{P}_{j}f({d}_{ij})\\ f({d}_{ij})={{d}_{ij}}^{-0.358993}\end{array}\right.$$

(9)

Construction of the L-ASR predictive model

Human activities play a complex function on the preservation status of earthen sites, and the combined impacting mechanisms of the multiple land use categories on the earthen site conservation was explored. In the Linear Regression model: $\overline{L-{ASR}}=\widetilde{{C}_{0}}+\widetilde{{\text{C}}_{1}}* {W}_{con}+\widetilde{{\text{C}}_{2}}* {W}_{agr}+\widetilde{{\text{C}}_{3}}* {W}_{eco}+\varepsilon$, where $\overline{L-{ASR}}$ is the theoretical value based on these three land use types. The constant K in original W_con, W_agr, W_eco can be combined with the coefficients $\widetilde{{\text{C}}_{1}}$、$\widetilde{{\text{C}}_{2}}$、$\widetilde{{\text{C}}_{3}}$, so the actual variables involved are$\widehat{{W}_{con}}$、$\widehat{{W}_{agr}}$、$\widehat{{W}_{eco}}$ whose values can be obtained, $\varepsilon$ is the error term. Finally, the following equation can be constructed as follows:

$$\overline{L-AS{R}_{i}}={{\rm{C}}}_{0}+{{\rm{C}}}_{1}* {\hat{W}}_{con}+{{\rm{C}}}_{2}* {\hat{W}}_{agr}+{{\rm{C}}}_{3}* {\hat{W}}_{eco}+\varepsilon$$

(10)

Finally, C₀, 0,5695, C₁= 0.0026, C₂ = –0.0001, C₃= −0.0007, R² = 0.8, were obtained (Fig. 9 (a)).

Fig. 9: Fitting of theoretical and actual values of L-ASR₂₀₀₉ and L-ASR _2024.

a The L-ASR₂₀₀₉; b The L-ASR₂₀₂₄.

Validation of the L-ASR predictive model

To verify the reliability of the model, based on the 1 km land use data around 44 castles extracted from the land use data of Yulin City in 2023 (released in August 2024), the authors statistically sorted out the area and average distance of each three types of land use, and then calculated the theoretical value of L-ASR₂₀₂₄ according to the Eq. (10) (Table 4). After that, the authors fitted the theoretical values with the actual values of L-ASR ₂₀₂₄, which were obtained during the team’s field investigation by UAV (Fig. 9b). According to Fig. 9, the model showed a higher accuracy with the value of R² = 0.61 and Pearson’s r = 0.78.

Table 4 Actual values of L-ASR₂₀₂₄ and the data involved in the calculation

Discussion

In the predictable model, the area of land use patches of the same type ${P}_{j}$ and their average radius ${d}_{{ij}}$ from the earthen site are the two major factors that affect the interaction intensity between the land usage and earthen sites. The authors assigned the mean area $\bar{{P}_{j}}$ of these three types of land use within 1 km around the site in 2010 to ${P}_{j}$. Based on Eq. (10), we explored the patterns of the combined effects of the three land use types on the earthen sites as a function of distance ${d}_{{ij}}$, to propose a reasonable management radius for the earthen sites.

By examining the graph of $\overline{L-{ASR}}$ versus ${d}_{{ij}}$ (Fig. 10a), it was clear that the value of $\overline{L-{ASR}}$ rose sharply with the distance ${d}_{{ij}}$, peaking at point A before declining at point B, which marked a slope mutation. At point A, where ${d}_{{ij}}$ was 30 m and $\overline{L-{ASR}}$ was 0.98; At point B, where ${d}_{{ij}}$ was 531 m and the slope mutated and approached 0, indicating that anthropogenic impacts on earthen sites stabilized and became minimal over 531 m. Notably, the 2019 Great Wall Protection Master Plan has defined two protection zones for earthen sites, which were highlighted in orange in Fig. 10b⁵⁸: the protection scope (R ≥ 50 m) and the construction control zone (R ≥ 550 m, extending 500 m beyond the protection scope). Thus, the theoretical minimum widths for the two management zones were calculated as 30 meters and 531 meters, respectively, while the official minimum protection boundaries are set at 50 meters and 550 meters. The comparison shows a difference of about 20 meters, primarily within the scope of protection. Meanwhile, the width of the construction control zone showed an accuracy of 99.8%, with the theoretical value at 501 meters and the official value at 500 meters. These findings have validated the reliability of the evaluation model and also confirmed the scientific rationale behind the current official protection boundaries. However, it is also observed that the 30-meter threshold represents a critical, non-negotiable protection baseline. Since the theoretical boundary falls within the officially designated 50-meter area, integrating both approaches would contribute to a more refined and precise protection management. Therefore, based on the analysis, the following recommendations for the protection of earthen sites are proposed: (1) The minimum protection area should extend 30 to 50 meters outward from the castle walls. Before defining this zone, a site classification and value assessment is essential to ensure that conservation needs are met while minimizing costs. Buffer zones would help balance urban development and heritage preservation effectively. For sites with higher conservation value, the protection area can be expanded, while in cities with greater land development demand, the protection boundaries may be reduced. (2) The minimum range of the construction control zone should extend at least 500 meters beyond the boundary of the scope of protection. In practical applications, the integrity and coherence of the surrounding ecological environment and landscape should also be considered to enhance overall conservation outcomes.

Fig. 10: Combined impacts of the three land use types on the earthen site at different distances.

a Change curves; b Management thresholds.

Additionally, the model offers potential for castle predictive conservation. The PLUS model, a meta-cellular automata model integrated with the CA-Markov algorithm, is commonly used for patch-scale land use/cover (LULC) changes simulation and land use prediction^59,60,61. Based on the CA-Markov method of PLUS model, this research then proceeds to predict the land use distribution of Yulin city in 2035 by using the land use data in 2005 and 2020, with a 15-years step interval (Fig. 11). The final prediction Kappa coefficient obtained is 0.81, with the overall prediction accuracy of 87%. The area of the same land use type within the 1 km radius of each earthen site and the average distance indicator were then calculated, which were applied to the predictive model constructed in this paper (Eq. (10)) in order to anticipate the potential damage status of the 44 castles in 2035. Finally, according to the existing classification criteria in the field of earthen sites (Du et al., 2021), we classified the future L-ASR in 2035 of these 44 castles into the following five levels (Table 5): Excellent (E) (0–10%), Good (G) (10–25%), Fair (F) (25–50%), Poor (P) (50–75%), and Very Poor (VP) (75–100%), which can help prioritize the future conservation of earthen sites.

Fig. 11

Evolution of construction land, agricultural land and ecological land from 2005 to 2035.

Table 5 Predictation of L-ASR₂₀₃₅ and classification of 44 castles

From Fig. 12, the L-ASR of earthen sites will increase significantly from 0.33 in 2009 to 0.61 in 2035. Furthermore, according to the prediction results in 2035, 0% of the castles will be classified as E, 2.27% as G, 45.45% as F, 15.91% as P, 29.55% as VP, and 6.82% as unknown owing to the data limitation (Table 5). Overall, 45.45% of the earthen sites are likely to be in poor and very poor preservation status. Thus, they should be prioritized for planning and conservation, including Anbian Castle, Pingyi Castle, Boluo Castle, Yuhe Castle, Zhenchuan Castle, Huaiyuan Castle, Shenmu Castle, Qingshui Castle, Zhenqiang Castle, and Gaojia Castle. On the other hand, some of these castles have been reinforced because of their location in the city center, such as the Dingbian Castle, Yulin Wei, and Suide Wei, which have now become essential tourism resources and cultural emblems of the city. Even though the speed of decreasing is slower from 2024 to 2035, the predicted mapping has suggested that the large-scale construction activities will occur within 1 km radius of these castles over the next 10 years, hence resulting in them facing a severe situation. Therefore, the relevant authorities should place emphasis on such endangered sites and then take timely and effective protection measures.

Fig. 12

L-ASR in 2009, 2024, and 2035.

However, there are still some limits in this research: (1) earthen sites are not affected only by anthropogenic activities (construction, farming, grazing, etc.) but also by numerous other factors, like animals, and natural disasters (e.g., earthquakes, wind, and rain erosion). Therefore, there is a need for a more comprehensive evaluation and forecasting framework that incorporates a broader range of influencing factors, such as policy, economic elements, and environmental dynamics, to improve prediction accuracy in future studies. (2) since the latest land use data available is from 2023, there is a one-year difference between the data used for prediction and the actual L-ASR₂₀₂₄, which may affect the accuracy of the model’s predictions; Furthermore, land use change is driven by a complex interplay of policy shifts, economic transformations, and other socio-political factors, making it highly dynamic and difficult to predict. Although the PLUS model is generally considered accurate and reliable, its limitations in simulating the mutual attraction and evolution between open spaces and urban land under different policies ultimately impact the prediction precision. (3) land use was employed as a key index for anthropogenic activity and was categorized into three broad classes: construction land, agricultural land, and ecological land. Nevertheless, this classification may oversimplify the complex interactions between land use and heritage site preservation. Within each major category, diverse subtypes of land use exist, and their impacts on heritage sites can vary significantly. According to Current Land Use Classification (GB/T 21010-2017), for instance, irrigated land and dryland involve different cultivation practices and crop types within agricultural land, which may greatly influence soil moisture dynamics and microbial community composition in distinct ways, threatening the conservation of earthen sites⁶². Future research should consider refining land use classifications and incorporating more nuanced evaluation indicators to more accurately assess and forecast the preservation conditions of heritage sites. (4) owing to the constraints of research conditions, the research team only investigated 17 earthen sites due to the limited higher factors, which may cause the samples to be biased, thereby affecting the model’s prediction accuracy. In the future, to improve the model, we aim to comprehensively enlarge more external factors affecting the preservation of the earthen sites, and enhance prediction accuracy regarding the damage of the earthen sites, devoting to the efficient preservation planning of the earthen sites of the Ming Great Wall.

In this paper, 44 earthen sites of the Ming Great Wall in Shaanxi Province, China, were selected as the research objects. The gravity model, multiple linear regression, and Levenberg-Marquardt (LM) algorithm were utilized to build a quantitative predictive model for the Linear Absent Section Ratio (L-ASR) of earthen sites, based on the dynamic correlation between three land types (construction land, agricultural land, ecological land) and earthen sites preservation status. This study provided data support and valuable references for earthen sites conservation planning and scientific management. The main findings are as follows:

First, from 1980 to 2023, the proportion of the area of construction land within a 1-km radius of the earthen sites in Yulin has increased significantly, while that of agricultural land decreased, and the ecological land remained relatively stable. The entire land use transformation was divided into four stages (I–IV), with the mutation periods (II and IV) highly corresponding to the timing of relevant national policy implementations;

Second, while the construction land was detrimental to site conservation status with an exponential decay pattern, agricultural land and ecological land contribute positively to site preservation, following an exponential decay pattern and a power-law decay pattern, respectively;

Third, based on the relationship between these three land use types and the preservation status of earthen sites, a comprehensive predictive model for L-ASR of earthen sites has been constructed. By applying this model, two levels of protection zones were defined for earthen sites: a strictly controlled zone (R ≥ 30 m) and a protective buffer zone (R ≥ 531 m), which corresponded to the two-tier protection zones specified by the government.

Last but not the least, the damage prediction of earthen sites in 2035 indicates that 0% of the castles will be rated as Grade E, 2.27% as Grade G, 45.45% as Grade F, followed by 15.91% and 29.55% of them being rated as Grade P and Grade VP, respectively, emphasizing a serious challenge for the future preservation of earthen archeological sites. This study helps to delineate the priority protection levels for earthen sites, providing data support and a scientific basis for their future conservation planning and management.