Introduction

Agricultural heritage represents a distinctive land-use pattern and agrarian landscape shaped by prolonged human-nature interactions, embodying the co-evolution and adaptive dynamics of rural settlements, agricultural practices, and their surrounding environments1,2,3,4,5. Acknowledged by the GIAHS and UNESCO World Heritage List, these landscapes deliver essential ecosystem services, support local food security, and preserve unique cultural identities2,6,7. They are not static relics but dynamic social-ecological systems that constantly adapt to changes and can maintain or quickly restore their spatial functions after disturbances, showing typical resilience8,9,10. However, rapid environmental changes and human activities present significant challenges to their resilience and sustainability11,12.

A primary consequence of these pressures is the degradation of the Agricultural Production Landscapes (APLS)—the core productive areas that sustain the system’s food security and cultural practices13,14. Such degradation disrupts essential ecological processes and traditional agricultural practices, while diminishing landscape connectivity and functional redundancy, thereby heightening system vulnerability15,16. Consequently, identifying core APLS, preserving connectivity of these areas, and ensuring sufficient landscape functional redundancy are critical for strengthening the capacity of the system to withstand disturbances, adapt to changes, and sustain the distinctive structure and function of the system17,18. Research that comprehensively elucidates the complexities of managing agricultural heritage systems amid dynamic evolution and enhances landscape resilience is urgently needed, posing a significant scientific challenge in heritage conservation.

Within the production-living-ecological space (PLES) framework, production space (PS) generates the goods and services essential for societal development through agricultural, industrial, and commercial activities19,20,21,22. Living space (LS) accommodates residential and daily activity needs to enhance the quality of the human environment23. Ecological space (ES) maintains ecological balance through natural and semi-natural areas, providing fundamental support for biodiversity conservation24. The PLES framework coordinates land functions to manage the conflict between the expansion of production and LS and ecological protection25,26. It also establishes a resource recycling mechanism to maintain regional stability, forming a resilient closed loop27,28. The Hani Rice Terraces, located in Yunnan Province, China, and designated as a World Heritage site, exemplify a dynamic rice-based agricultural heritage system. Characterized by a “four-element integrated” landscape pattern (forest-village-terrace-water system), the balanced configuration of PLES reflects the spatial arrangement of these elements, fostering ecological sustainability and socio-economic development within the heritage area29,30. In this system,the Agricultural Production Landscapes (APLS) represent the core component of the Production Space (PS), encompassing the iconic rice terraces and other arable lands that form the foundation of agricultural output and cultural identity31,32. However, the Hani Terraces are subject to growing environmental and social pressures, thereby increasing system vulnerability11,33. This is especially evident in APLS, as manifested by the gradual reduction in scope and the degradation of functions34,35.

Research indicates that comprehensive studies on complex interactions and equilibrium within PLES in agricultural heritage systems remain limited36. To establish resilient protection mechanisms, spatial attribute states must be mapped, and precise spatial control patterns developed to inform decision-making for conservation objectives. Although “three zones and three lines” protection measures are implemented within the buffer zone, delineating development, farmland, and ecological protection boundaries often overlooks multi-scale spatial relationships, rendering them inadequate for addressing the dynamic evolution of APLS37,38,39,40. When production, living, and ecological spaces are considered as an integrated complex system, understanding the interactions among factors influencing landscape patterns and designing proactive interventions to enhance structural and functional resilience remain challenging41,42,43. Currently, effective spatial network optimization methods to integrate the multidimensional functions of landscapes are lacking44. Furthermore, approaches to improve landscape connectivity and redundancy, based on the integrated impacts of PLES, are insufficient to promote resilience and support the conservation and management of heritage sites.

Prior research indicates that resilient systems can absorb disturbances and reorganize during changes22,45. System spatial resilience is underpinned by five key components: diversity, redundancy, network connectivity, modularity, and adaptability46,47. These components serve as critical tools for evaluating landscape vulnerability and mitigating the impacts of environmental changes and human activities47,48,49,50. Based on this concept to solve the above-mentioned problems, this study focuses on dynamically integrating the multifunctional spatial structure under the interrelationship of PLES and proposes strategies to enhance resilience in terms of network connectivity, redundancy, and spatial pattern. It also addresses the following interrelated sub-questions based on this scientific issue: 1. How to effectively identify and quantify the spatial balance states of production, living, and ecological spaces within agricultural heritage systems? 2. How to delineate the core areas of APLS based on PLES balance to establish a foundation for the protection network? 3. How to construct an optimized agricultural protection network to improve connectivity and redundancy, thereby enhancing landscape resilience? 4. Compared with traditional single-function-based protection methods, in which aspects and to what extent does the PLES-guided network demonstrate superior spatial resilience performance?

It is hypothesized that the balanced distribution of the three spatial types is shaped by the intricate interplay of natural and socio-economic factors, exhibiting spatial threshold relationships that can be leveraged to identify areas critical for sustaining APLS51. Using multi-source data, including POI, and the MaxEnt model, the distribution patterns of the three spatial types within the buffer zone of the Hani Terraces heritage site were modeled. Key influencing factors and their spatial thresholds under a balanced state of complex interactions were identified. Based on the spatial thresholds of PLES equilibrium, core APLS source areas essential for maintaining agricultural system functionality were delineated. Finally, the minimum cost distance method was employed to construct an agricultural protection network linking these core areas, aiming to enhance overall landscape connectivity and redundancy through an optimized spatial configuration.

This study advances knowledge in several domains. Methodologically, it combines PLES balance analysis with network optimization techniques to develop an innovative approach for cross-scale collaborative protection of complex agricultural heritage sites, rooted in spatial resilience principles. Practically, it provides managers of dynamic rice agricultural heritage sites, such as the Hani Terraces, with actionable spatial management frameworks and decision-support tools to bolster landscape spatial resilience. The study presents outcomes of PLES analysis and the construction of a protection network. By addressing the intricate interactions within PLES, it establishes a replicable methodological framework. This framework seeks to enhance understanding of effective management and protection strategies for rice agricultural heritage sites facing disturbances, such as environmental changes and human activities.

Methods

Study area overview

The Yuanyang Hani Terraces Cultural Landscape Heritage, located in southern Yunnan Province, China, at the southern extremity of the Ailao Mountains, experiences a subtropical monsoon climate. Elevations range from 700 to 2900 m, with a relative altitude difference of 2000 m, resulting in a pronounced vertical climate gradient52. This environment has fostered the development of a landscape integrating forests, villages, terraced fields, and water systems, reflecting harmonious interactions among production, livelihood, and ecological systems34,46. In 2013, the terrace-dominated region, primarily in Yuanyang County, was inscribed on the UNESCO World Heritage List, encompassing a core area of 16,603.22 ha and a buffer zone of 29,501.01 ha, characterized by high mountains, deep valleys, and river systems53. For over a millennium, the Hani and other ethnic groups have adapted to the constraints of mountainous terrain with limited flat land by transforming slopes into rice terraces, a practice recognized for its global exemplary value and substantial benefits35,54,55.

The Hani Terraced Fields Cultural Landscape Heritage serves as a valuable case study for examining the interplay between mountainous rice heritage characteristics and the coupling of livelihood, production, and ecosystem services35,56,57,58 (Fig. 1). In recent years, motivated by World Heritage designation, local authorities have implemented measures to restore degraded and abandoned rice terraces within the heritage area. By 2023, the restored terrace area reached 328.87 ha2,34,53. Nevertheless, the heritage site faces significant challenges, including intensified climate change, frequent natural disasters, and declining agricultural labor, which threaten the sustainability of the current terrace extent46,59,60. Previous research, utilizing settlement connectivity, has quantified farmland abandonment risk, revealing that over 55% of the land in the region is classified as high-risk or unstable14. Analysis conducted using high-resolution Landsat 8 imagery from 2014, 2019, and 2024, processed with the maximum likelihood classification method in ENVI software for landscape feature identification, indicates a trend of decline and fragmentation in the APLS system within the buffer zone over the past decade (Fig. 2).

Fig. 1: Map of the geographic location of Yuanyang Hani Terraces.
figure 1

The image indicated the study area by showing the map of China and the county-level administrative divisions of Yunnan Province.

Full size image
Fig. 2: APLS change maps.
figure 2

a Status of APLS in 2014, b Status of APLS in 2019, c Status of APLS in 2024.

Full size image

These circumstances underscore the limitations of existing management frameworks30,61,62. Current heritage management strategies, including the delineation of permanent basic farmland protection boundaries, have not fully adapted to evolving spatial dynamics63. Amid ongoing transformations, there is a pressing need to develop a more dynamic and integrated protection approach to strengthen heritage spatial resilience14.

Method overview

This study initially classified multi-source point data from the agricultural heritage site according to PLES criteria64. Subsequently, the MaxEnt model was employed to assess the potential impact range of each spatial unit, with model outcomes validated spatially using land-use classification data65. The assessment delineated the potential impact range of PLES and characterized the spatial distribution of balanced and unbalanced PLES states. Furthermore, key environmental factors contributing to the imbalance among the three spaces were examined under the maximum entropy index, identifying stable and unstable drivers of heritage space dynamics. Finally, a comprehensive resistance surface was developed through raster overlay based on multifactor PLES coupling. By applying the minimum cost-path principle, an optimized framework was proposed to enhance the spatial resilience of the APLS protection network. This framework connects the core functional areas through the agricultural conservation network and assesses its effectiveness through key indicators of spatial resilience: connectivity, redundancy, and spatial structure distribution (Fig. 3).

Fig. 3: Spatial network optimization framework.
figure 3

The image visually presents the research process through indicative arrows.

Full size image

Data collection

The production, living, and ecological functions of the Hani Terraces are often interwoven, with indistinct boundaries, rendering traditional areal land-use data inadequate for capturing the transitional characteristics of their spatial patterns. Consequently, this study employed point data to better represent these transitions. For example, in the collected POI data, factories, highways, and enterprises support production functions during the urbanization of the terraces; government offices, recreational facilities, and villages provide essential infrastructure for local residents’ daily activities; and parks, green spaces, and scenic areas primarily contribute to ecological functions. However, POI data alone are insufficient to fully encompass mountainous agricultural production and natural ecological components. To improve the comprehensiveness of the PLES impact analysis and more accurately depict the spatial structure influenced by multiple factors, this study incorporated the centroids of land types with distinct functional attributes as supplementary point data sources for agricultural production and ecological elements66.

This study employed a multi-source data collection approach. A total of 3291 POI data points were acquired from Google, Baidu, and Amap using data crawling tools, complemented by central points of land parcels with varying functional attributes. The dataset exhibits a substantial sample size, high representativeness, and uniform spatial distribution. Classified within the PLES framework, the complete dataset was utilized as training samples for subsequent modeling. This methodology facilitates the exploration of diverse spatial attributes in human-nature interactions, elucidates the intricate internal structure of PLES, and delineates specific land-use patterns within the study area (Table 1).

Table 1 Classification of POI data
Full size table

To elucidate the balanced-spatial distribution characteristics of PLES, raster data exhibiting strong correlations with heritage spatial resilience and representing critical environmental and anthropogenic influences were selected as environmental variables. These variables were categorized into natural and socio-economic factors, both of which are closely associated with PLES distribution67,68.

All variables were normalized and resampled to a consistent 10-m spatial resolution prior to input into the MaxEnt model. Socio-economic factors, notably population density and nighttime light index, exert significant influence on spatial distribution by indicating population aggregation and economic activity intensity, thus serving as primary drivers of changes in residential patterns and ecosystem services67. Rural settlement points, associated with prolonged human activity, provide critical support for spatial resilience and exhibit high stability69,70.

Natural environmental factors encompass topography (slope and elevation), hydrology (proximity to water systems), and soil conditions, which are critical for ecological and agricultural productivity51,71,72. The NDVI reliably indicates ES quality, whereas temperature and precipitation influence its spatial distribution72,73,74. Furthermore, supplementary data derived from land-use types and landscape pattern analysis facilitate the characterization of land-use spatial configurations (Table 2).

Table 2 Data of environmental factors
Full size table

Simulation of PLES potential impact range

The MaxEnt model was developed to forecast potential species habitats. It generates a probability distribution model using species occurrence data and environmental variables to predict species distributions in unsampled regions73,74. This approach, which extrapolates broader patterns from localized data, aligns methodologically with the objectives of this study. Recently, the MaxEnt model has been utilized in simulations of agricultural suitability and analyses of spatial patterns in human activities and cultural distributions75,76,77.

This study conceptualizes PLES as distinct “spatial states” or “functional types,” comparable to species distribution ranges. The distribution of these spatial types is shaped by various environmental and socio-economic factors exhibiting spatial heterogeneity. The MaxEnt model integrates multiple environmental variables to examine their nonlinear associations with specific spatial types. Following calibration, the model produces a probability distribution map that delineates the potential influence zones of different PLES types within the study region. This enables the extrapolation of broader spatial patterns from localized point data78.

Kernel density analysis was initially performed on classified point data within a GIS platform to establish a baseline for comparison with MaxEnt model outputs. Subsequently, the MaxEnt model integrated PLES-classified point data with environmental variables to simulate the potential spatial extent of PLES impacts. For each model iteration, 25% of the sample data were randomly assigned for testing, with the remaining 75% allocated for training. This procedure was repeated ten times, and the mean output was calculated. Principal factors affecting PLES distribution patterns were determined, and model performance was evaluated using the mean AUC value derived from the ten iterations.

To validate the model predictions against empirical data, a conventional land-use classification approach was applied. The 2021 WorldCover land-cover dataset from the European Space Agency, with a 10 m resolution, was utilized. Land-use types were classified into production, living, and ecological functions based on predefined PLES criteria79. These outcomes were compared with MaxEnt predictions to assess their validity.

APLS protection network optimization

To enhance the regional-scale balance of PLES and identify primary sources of APLS under the combined influence of PLES, raster layers of PLES potential impact ranges, generated by the MaxEnt model, were integrated into the ArcGIS platform. Initially, impact values for PS, LS, and ecological space were computed for each raster cell. The standard deviation of these impact values was calculated to assess the variability in their balanced impact intensities, providing a measure of spatial impact heterogeneity for PLES components. The standard raster deviation (StdDev) was calculated using the following equation:

$$StdDev=\sqrt{\frac{{[M(P)-\mu ]}^{2}+{[M(L)-\mu ]}^{2}+{[M(E)-\mu ]}^{2}}{3}}$$
$$\mu =\frac{S\,(P)+S\,(L)+S\,(E)}{3};\,{\rm{S}}(P)=\frac{M(P)}{AUC(P)};\,{\rm{S}}(L)=\frac{M(L)}{AUC(L)};\,{\rm{S}}(E)=\frac{M(E)}{AUC(E)}$$

In this equation, M(P), M(L), and M(E) represent the spatial thresholds for PLES derived from the MaxEnt model, μ denotes the arithmetic mean of the PLES factor products, and AUC indicates the model accuracy. The parameter S represents the ratio of the potential distribution range threshold to AUC across different PLES categories. This ratio evaluates the relative efficacy of the potential distribution range threshold compared to model accuracy, facilitating standardized comparisons across indicators and supporting a more robust evaluation of the model.

The standard deviation raster was normalized to a 0–1 scale, with higher values indicating greater spatial balance. The PLES balance index layer was subsequently generated. To delineate core areas optimal for sustaining agricultural production within highly balanced PLES regions, a spatial overlay analysis was performed between regions with high PLES balance indices and the agricultural PS distribution layer. Overlapping areas were extracted, and patches exceeding 10 ha were selected to minimize fragmentation. These patches were designated as core source areas for agricultural development.

Spatial dispersion of core APLS source areas, coupled with barriers such as construction land, water systems, and roads, restricts their connectivity. Solely delineating core areas falls short of supporting refined management. Accordingly, this study aims to develop a conservation network to integrate these core source areas, thereby bolstering spatial connectivity and facilitating evidence-based decision-making for refined management.

The development of the APLS protection network relies on cost-path analysis. Initially, a resistance surface was created to represent the impedance of the landscape to APLS connectivity. The PLES balance index raster, produced by the MaxEnt model, was linearly inverted based on extreme values to generate resistance rasters under varying spatial influences. Regions with higher balance indices exhibited lower resistance values post-inversion, indicating greater suitability for corridor connectivity. Within this resistance raster, resistance coefficients derived from the MaxEnt model function as entropy weights, encapsulating the influence of each environmental factor (i.e., Percent Contribution, PC) as determined by the MaxEnt model. The magnitude of these coefficients indirectly reflects the spatial suitability assessed by the MaxEnt model based on key environmental factors.

Using four independent resistance raster datasets and previously identified core areas as source points, corridor networks were constructed with the Linkage Mapper tool. The maximum cost detection radius for corridor searches was set to 20 km. For each resistance layer, minimum cost-path (LCP) calculations were performed independently to delineate the lowest-cost potential agricultural protection network linking the source areas.

APLS protection network advantage verification

To assess the performance of the agricultural protection network in enhancing spatial resilience, a quantitative evaluation was conducted across three dimensions: connectivity, redundancy, and spatial layout. Cross-validation with two indicators per dimension ensured the reliability of the results.

For connectivity, Current Flow Centrality (CFC) and the standard deviation of the Cost-Weighted to Euclidean Distance Ratio (CEDR) were employed. Higher CFC values indicate greater network efficiency, while a larger CEDR standard deviation reflects increased landscape resistance affecting corridors, resulting in reduced network stability.

For redundancy, inactive redundant paths were excluded, and the number and proportion of effective networks relative to total mileage were calculated. A greater number and higher proportion of effective networks signify enhanced redundancy and spatial resilience to disturbances. Kernel density analysis was performed on the generated corridors, with the ratio of mean density to standard deviation indicating the concentration and dispersion of corridor density, thereby reflecting the spatial distribution of redundancy.

Regarding spatial layout, the mileage of corridors passing through terraced fields and their variation over a decade was quantified, alongside the proportion of these corridors relative to the total network mileage. A higher proportion indicates a stronger network response to terraced field dynamics and more targeted protection. Average Nearest Neighbor (ANN) analysis in GIS was also applied, where a higher ANN value signifies a more uniform node distribution across the study area.

Finally, results were normalized or reverse-normalized based on their influence on network spatial resilience, and radar charts were generated for comparative analysis. These findings provide scientific guidance for the future development and protection strategies of APLS in heritage sites.

Results

Potential distribution characteristics of PLES

Analysis of the distribution and interactions of PLES is essential for network optimization and spatial resilience enhancement. Core density analysis of POI indicates high concentration of PS, partial clustering of LS within the core heritage area with relatively uniform distribution elsewhere, and predominant concentration of ES at the center of the heritage area, exhibiting overall balance (Fig. 4). Although this approach effectively identifies concentrated PS, partially clustered LS, and relatively balanced ES, it is limited in capturing complex spatial structures, particularly spatial transitions and interactions.

Fig. 4: Map of the PLES influence range under kernel density.
figure 4

a The potential impact range of PS, b The potential impact range of LS, c The potential impact range of ES.

Full size image

To investigate this issue, the MaxEnt model was utilized to simulate the potential impact ranges of PLES using multi-source data. The model revealed pronounced spatial heterogeneity in PLES distribution within the Hani Terraces buffer zone (Fig. 2). The potential impact range of PS is predominantly concentrated in the central and eastern regions of the terrace core area, exhibiting a compact and contiguous pattern. In contrast, the LS impact range is primarily located in the central buffer zone and the western core area, aligned along valleys and roads. The ES impact range is extensive, encompassing the central core area and the western buffer zone. Regarding interactions, a degree of coupling exists between PS and ES, with terraces near villages typically exhibiting better management. However, LS expansion, driven by population growth and lifestyle changes, has led to the reduction or fragmentation of PS. In certain areas, PS and ES are interlinked; for example, terraced fields near forests or water sources support agricultural production while contributing to ecological regulation (Fig. 5).

Fig. 5: Map of the influence range of PLES under the maximum entropy index.
figure 5

a The potential impact range of PS, b The potential impact range of LS, c The potential impact range of ES.

Full size image

Verification of the MaxEnt model’s identification results was conducted through land-use classification. Results indicate that PS predominantly occurs in the central and eastern terrace regions, LS is distributed along valleys forming village clusters, and ES is extensively distributed across mountainous areas. These findings align closely with the MaxEnt model’s outputs, confirming its effectiveness in delineating the three spatial types. In contrast to traditional land-use classification, which relies on rigid boundaries, the MaxEnt model employs a threshold-based approach. This method more accurately captures the transitional characteristics and interactions among the three spatial types, providing a robust foundation for optimizing and managing spatial networks (Fig. 6).

Fig. 6: Land-use-based PLES classification.
figure 6

a The land-use map, b The PLES map of land-use classification.

Full size image

Key factors influencing PLES distribution characteristics

Across ten iterations of the MaxEnt model, the average AUC values for production, living, and ecological elements were 0.834, 0.880, and 0.774, respectively. All AUC values exceeded 0.6, indicating robust model performance. The model quantified factor contributions using Percent Contribution (PC) and Permutation Importance (PI). Results revealed that low vegetation cover, high population density, and intensive economic development positively correlated with PS growth. Road accessibility, NDVI, and settlement distribution significantly influenced LS. Natural factors, including NDVI, DEM, and water systems, exerted a strong effect on ES distribution (Table 3).

Table 3 The influence and importance of different environmental elements under the PS-LS-ES categories
Full size table

Network construction and optimization results

Standard deviation was calculated using the PLES potential impact range layer generated by MaxEnt to delineate PLES balance threshold areas (Fig. 7a). Results indicate that high-threshold PLES balance areas are unevenly distributed, primarily concentrated in the central and eastern core zone, and forming linear corridors along mountains and rivers in peripheral regions. These areas exhibit a high degree of coordination among production, living, and ecological functions, with strong functional coupling.

Fig. 7: Map illustration of the core agricultural production area.
figure 7

a Spatial threshold map of PLES balance state, b Regional relationship between terracing and high threshold, c APLS core source location.

Full size image

Overlay analysis of the identified high-threshold PLES balance areas with existing APLS (Fig. 7b) revealed a significant coupling between APLS and the PLES balance pattern. Notably, APLS, particularly terrace landscapes, are predominantly located at the boundaries of high-threshold balance areas. Conversely, in low-threshold areas with marked PLES imbalance, APLS distribution is more dispersed. This finding underscores the substantial influence of PLES balance on agricultural production layouts and facilitates precise identification of critical APLS patches within balanced areas essential for system functionality. After filtering out smaller patches, 89 core APLS source areas, covering a total of 3976 ha, were identified (Fig. 7c). Accurate delineation of these core APLS source areas provides a critical spatial framework for establishing a conservation network linking these key nodes.

Using the 89 identified core APLS source areas, the least-cost-path method was employed to develop an agricultural protection network. To incorporate the multidimensional functions of the landscape, individual networks for PS, LS, and ES were constructed using single-cost surfaces, followed by network optimization with a composite PLES cost surface (Fig. 8).

Fig. 8: The enhancement network of agricultural production landscapes’ spatial resilience under PLES balance.
figure 8

The image shows the network layout of corridors under different spatial impacts and the various factors considered under the PLES-balanced-spatial impact.

Full size image

Analysis of single-factor-driven networks revealed 223 corridors (485.54 km) influenced by PS, 222 corridors (336.66 km) by LS, and 199 corridors (393.51 km) by ES. Comparative evaluation highlighted the limitations of these networks. The PS network exhibited uniformity but lacked granularity; the LS network provided detailed patterns but showed reduced connectivity in the northwest and southeast; and the ES network offered extensive coverage but insufficient representation in the southwest and northeast. These shortcomings underscore the inadequacy of single-dimensional approaches for optimizing spatial networks, which fail to address the comprehensive protection requirements of complex agricultural heritage landscapes.

Conversely, optimization using a composite PLES cost surface increased the number of corridors to 227, with a total length of 395.70 km. By integrating multidimensional factors into a unified cost surface, this network achieved a balanced and refined corridor configuration. Notably, it enhanced connectivity in the northeastern heritage area, streamlined redundant connections in the northwest and southeast, and improved linkage within the central ecological core. This optimized network effectively links core APLS source areas while maintaining PLES equilibrium in corridor selection. Agricultural activities along these corridors, including reclamation, conservation, and development, foster balanced PLES development and mitigate conflicts among spatial functions. This methodology provides a framework for optimizing agricultural protection networks within PLES balance thresholds, promoting a four-element integrated landscape pattern and enhancing connectivity and spatial resilience (Fig. 8).

Optimized network’s spatial resilience advantages

To evaluate the effectiveness of the optimized network in improving system spatial resilience, a quantitative analysis of its connectivity, redundancy, and spatial configuration was conducted and compared with networks derived from single-type spatial elements (Fig. 9). Results indicate that the optimized network, constructed under the PLES framework, exhibits substantial improvements in critical spatial resilience metrics. It significantly enhances landscape spatial resilience through improved connectivity and redundancy.

Fig. 9: Network advantage numerical comparison chart.
figure 9

The image individually illustrates the network advantages under each spatial impact in the form of a radar chart, with a comparative display within the same chart.

Full size image

In assessing connectivity, a network efficiency map was constructed to evaluate the rationality of the spatial configuration (Fig. 10). The network demonstrated elevated CFC values, indicating an equitable distribution of connection strength and node significance. The CEDR exhibited a low standard deviation, suggesting enhanced stability in network connectivity. Although the composite network displayed marginally lower values for the proportion of unstable areas traversed and CFC compared to ecological and residential networks, it exhibited superior overall connectivity. Regarding redundancy, the composite network featured a high proportion of effective networks in terms of number and mileage, with a more uniform corridor kernel density distribution, reflecting greater redundancy. This implies a robust capacity to sustain functionality despite partial node or corridor disruptions. Spatially, the composite network traversed a higher number of agricultural transition zones, indicating greater adaptability to agricultural spatial dynamics. Additionally, the Nearest Neighbor Index revealed a more balanced-spatial distribution pattern of the network.

Fig. 10: Network connectivity map.
figure 10

a PLES-balanced network connectivity, b PS network connectivity, c LS balanced network connectivity, d PLES-balanced network connectivity.

Full size image

In contrast, networks comprising single spatial elements exhibit specific strengths but lack comprehensive spatial resilience. The PS network demonstrates robust connectivity yet exhibits deficiencies in redundancy and spatial layout balance. The Life Space Network features high node density but is limited by inadequate connectivity and redundancy balance. The ES network, with the highest density, offers strong connectivity and stability but is constrained by suboptimal spatial layout balance and network functionality efficiency. Comparative analysis reveals that the APLS protection network, optimized based on PLES balance thresholds and integrating multidimensional functions, enhances connectivity between critical agricultural source areas and improves both connectivity and redundancy within the Hani Terrace agricultural heritage landscape under network conditions, thereby bolstering system spatial resilience.

Discussion

This study investigates the Hani Terraces, a representative rice-based agricultural heritage system, elucidating the intricate interactions among PLES and their influence on APLS spatial resilience. The results underscore the critical role of PLES equilibrium in bolstering the spatial resilience of agricultural heritage systems and provide novel insights and approaches for developing effective agricultural conservation frameworks.

Within the framework of dynamic conservation and development of agricultural heritage, precise delineation and coordination of PLES are essential for sustainable land-use practices79. This study employs a data-driven MaxEnt model coupled with threshold analysis to determine the potential impact range of PLES in the complex rice-farming heritage site of the Hani Terraces. Notable spatial disparities in PLES distribution were observed in the buffer zone of the Hani Terraces, aligning with prior research on spatial pattern dynamics in agricultural heritage sites80,81.

The spatial concentration of PS underscores the regional reliance on traditional agricultural practices, whereas the clustering of LS along valleys and roads indicates a strong connection between human settlements, transportation infrastructure, and water resources. ES exert a wide-ranging influence, emphasizing their critical role in sustaining ecological equilibrium. As a Globally Important Agricultural Heritage Site, the Hani Terraces derive their core value from the harmonious integration of production, living, and ecological spaces. Nevertheless, rapid urbanization and declining rural labor availability have led to challenges such as terrace abandonment and ecological deterioration, which threaten landscape distinctiveness, cultural heritage, and local food and ecological security. Consequently, precise identification of the balanced and conflicting dynamics within PLES is essential for formulating effective protection and management strategies.

Spatial statistical analysis (standard deviation calculation) of PLES impact ranges derived from the MaxEnt model quantified the balance and conflict states of PLES, with balanced regions predominantly located in the central and eastern core areas. Overlay analysis of these balanced regions with actual APLS revealed that APLS, primarily terraces, are predominantly situated in high-threshold PLES zones. This finding underscores the significant influence of PLES balance on the distribution and sustainability of agricultural heritage landscapes. However, APLS in these high-threshold zones exhibit a fragmented spatial distribution.

This study quantified the contributions of environmental variables to the distribution of PLES, identifying spatial-functional drivers critical to the spatial resilience of the Hani Terraces landscape. In contrast to research emphasizing solely climatic or natural factors, socio-economic and environmental variables exert comparable influence, occasionally predominating in aspects of PS and LS. Elucidating these dominant factors and their interactions is essential. For example, road enhancements may support PS and LS but impair ES by fragmenting ecological corridors. Similarly, higher population density may bolster labor availability for LS and PS, yet degrade the environment if excessive. This insight into multifactor interactions underpins the adoption of a composite cost surface, rather than a single-space-based approach, for designing the agricultural protection network. Integrating these interrelated and often conflicting factors is vital for formulating spatial strategies that effectively balance PLES and enhance landscape spatial resilience. Nevertheless, achieving equilibrium among these factors and comprehending their interactions remains critical for developing robust resilience-enhancing strategies.

This study proposes a strategy to enhance landscape spatial resilience through the development of an optimized agricultural protection network based on PLES balance. The approach relies on a comprehensive understanding of PLES distribution, balance, and critical influencing factors. Overemphasis on single-space functions in previous practices has increased the vulnerability of agricultural heritage. For instance, excessive development of PS can degrade soil and water resources, while inadequate attention to LS optimization may diminish community engagement and vitality. Similarly, prioritizing ES protection without integrating its interactions with PS and LS can result in economic stagnation and landscape fragmentation. Therefore, this study advocates for the integrated management of APLS from a PLES balance perspective. The spatial optimization network developed here exemplifies this approach. Employing the minimum-cost-path method, the network links core agricultural source areas within PLES-balanced regions while accounting for multi-space resistance factors of PS, LS, and ES. This ensures that the network corridors are both efficient for spatial connectivity and optimized to promote PLES balance while minimizing conflicts. Quantitative evaluation demonstrates that the agricultural protection network significantly improves landscape spatial resilience by enhancing connectivity and redundancy. Compared to networks based on single-space factors, it exhibits higher CFC, lower standard deviation of CEDR, more effective pathways, and a more uniform density distribution. These attributes validate the efficacy of the optimized network in bolstering landscape spatial resilience.

The findings are summarized as follows: 1. The MaxEnt model, combined with PLES balance threshold analysis, surpasses conventional approaches by effectively capturing the spatial heterogeneity and functional interactions within complex agricultural heritage landscapes, thereby facilitating precise identification of critical agricultural source areas.2. The optimized network, constructed based on these source areas, exhibits a superior structure compared to networks derived from single-function spatial analyses.3. Quantitative evaluation reveals that the optimized network significantly enhances landscape connectivity and redundancy, thereby bolstering system spatial resilience.

In exploring ways to enhance the resilience of agricultural heritage, this study focuses on the “production-living-ecological space” balance. Despite the need to consider the multidimensional interactions of natural, social, and economic factors, this approach highlights the key role of spatial resilience in supporting the dynamic adaptation of agricultural heritage. The focus on PLES stems from its imbalance being a core challenge for the Hani Terraces, as seen in issues like farmland fragmentation, settlement expansion, and ecological degradation. Spatial network connectivity and redundancy are crucial for buffering environmental disturbances and maintaining system resilience. However, resilience is a multi-scale and multi-element system property. Focusing solely on the spatial dimension may overlook hidden mechanisms like social capital, cultural heritage, and economic resilience. Future research will build a multidimensional resilience indicator system covering “space, society, institutions, and culture” to explore pathways for enhancing the overall system resilience of the Hani Terraces. This study also has certain limitations that need to be addressed in future research. First, the spatiotemporal resolution and consistency of multi-source data are achieved through post-processing, which may affect the accuracy of model identification results. Second, the profound impact of cultural factors (such as traditional knowledge and community governance patterns) on heritage resilience was not adequately quantified in this study. Future research could combine qualitative methods like participatory mapping and interviews with quantitative models to more comprehensively incorporate the cultural dimension. While the MaxEnt model is relatively robust, exploring its integration with other methods such as multi-criteria decision-making analysis and intelligent optimization algorithms may provide a more powerful decision-support framework for addressing complex trade-offs and dynamic feedback.

Despite these limitations, this study presents a novel paradigm for proactively managing and enhancing the spatial resilience of agricultural heritage systems. Grounded in a “balanced-spatial-functions” framework, the approach emphasizes the identification of critical agricultural nodes and the optimization of spatial connectivity. The research addresses the fundamental scientific question of effective management and protection of rice agricultural heritage in dynamic environments. A method integrating multifactor threshold analysis with spatial network optimization is proposed to strengthen spatial resilience. This approach provides heritage managers of the Hani Terraces and analogous systems with a practical decision-support tool, enhancing connectivity and redundancy. The method offers significant theoretical innovation and substantial potential for practical application.