Spatial constraints or spatial dynamics? The spatial spillover effect of networks of flood regulation service flows on land-use degree

Flood disaster is one of the most frequent and serious natural disasters in the world (Najibi & Devineni 2018). In recent years, China has experienced frequent floods and caused serious economic losses. Flood regulation service is an important part of flood risk management which represents the ability to store and detain floods and reduce runoff in a certain area. Ecosystem services have significant spatial transfer characteristics (Chen et al. 2014). When flooding becomes a medium or carrier, flood regulation services are formed and a spatial connection of ecological service delivery is established between different regions. The supply of flood regulation services is related to the soil type and land use structure of a spatial unit (Stürck et al. 2014). Some studies have shown that with the continuous expansion of construction lands, the supply of flood regulation services will gradually weaken, thereby exacerbating the imbalance between the supply and demand of flood regulation services (Wu et al. 2019). Although some researchers have explored the response of ecosystem flood regulation services to land-use change, there are few studies on the spatial impact of changes in flood regulation services on land-use degree (LUD). Therefore, the identification, measurement, and spatialization of flood regulation services are necessary to reflect the direct response and feedback of the degree of land use to the increase or decrease of service supply.

Affected by topographical factors, low-lying coastal lowlands are areas with frequent natural disasters (Ayyam et al. 2019). The built-up areas in China's coastal areas have expanded rapidly in recent years. Although the construction of flood control projects has increased, the flood damage caused by typhoon intrusion and severe rainstorms still caused a lot of losses (Duy et al. 2018). It has greatly restricted the development of social and economic security, and the coastal areas need to improve the disaster-bearing capacity of land use to cope with floods. According to relevant studies, 25% of China's coastline is highly vulnerable, and more than 5 million people are exposed to the risk of flooding (Sajjad et al. 2018). Coastal systems and low-lying areas will increasingly suffer from negative impacts such as land inundation, coastal flooding, and coastal erosion as a result of accelerated sea-level rise due to intensified climate change in the future (Chen et al. 2017; Nerem et al. 2018). At the same time, the rapid growth and development of cities in coastal areas have led to the continuous migration and concentration of population and assets to coastal areas, which has also increased the risk of storm floods. In particular, the catastrophe of storms and floods will cause serious security risks and social stability problems in coastal lowlands (Cazenave et al. 2014; Rahmstorf 2017). Therefore, the role of land use management in disaster reduction needs to be emphasized. The realization of the cooperative, forward-looking and adaptive development of high-risk areas of the flood is conducive to the improvement of the safety and resilience of coastal cities (Lu et al. 2020).

At present, various fields have begun to pay attention to the issue of land-use change, and LUD is a practical indicator that reveals the breadth and depth of land use. Studying the dynamic change of LUD in a certain region can not only understand the degree of human disturbance to the land but also measure the social and economic development of the region (Wu et al. 2014; Li et al. 2017, 2018). Land-use change is influenced by many factors, such as population, climate, land policy, rapid urbanization, and household income (Hao et al. 2015; Wang et al. 2018; Li et al. 2021a; Zhai et al. 2021a). On the other hand, the response relationship between land use and ecosystem services has also attracted research attention. Ecosystem services are the various benefits that humans obtain from the ecosystem (Brauman et al. 2007), including tangible material products and intangible services (Wallace 2007; Braat & Groot 2012; Danley & Widmark 2016). The products and services of the ecosystem are concrete manifestations of the structure and function of land use, and the types and structures of land use are the expressions of the pattern of ecosystem products and services (Xue & Ma 2018). It can be seen that land use includes the content of terrestrial ecosystem services, and ecosystem services provide support for part of the way humans use land. Land-use change plays a decisive role in maintaining ecosystem services, and profoundly affects the structure and function of ecosystems, which in turn causes changes in the types, areas, and spatial distribution patterns of ecosystems, thereby changing the ability to provide ecosystem products and services (Wang et al. 2017, 2021; Xu et al. 2019).

Previous studies focused on the relationship between land use and ecosystem services such as food production, climate regulation, water conservation, biodiversity protection, entertainment, and culture, but few studies on the combination of flood regulation services and land use. As flood regulation service plays an indispensable role in reducing flood disaster risk and maintaining regional ecological security, its impact mechanism on space cannot be ignored either. The University of Vermont proposed the ‘Service Path Attribution Networks’ (SPANs) module based on the Artificial Intelligence for Ecosystem Services project (Bagstad et al. 2013). The SPANs model integrates various simulation models widely used in ecology and geography. Its goal is to employ probabilistic Bayesian networks to analyze the movement of ecosystem service flow from supply areas to demand areas, ultimately generating a spatial map (Bagstad et al. 2011). Some researchers have mapped the supply area, beneficiary area, and connection area of flood regulation service, which provides a reference for the network analysis of ecosystem service flows (Syrbe & Walz 2012). These previous studies have provided many useful insights, but there are still some research deficiencies. First, these studies have analyzed the relationship between the supply area and the profit area, but how to draw a clear service flow path and quantify the service flow is worthy of further consideration. Second, some relevant studies have been too macroscopic in their handling of spatial scales, making it difficult to capture subtle spatial changes and pattern variations. Third, the impact mechanisms and spatial effects of flood regulation services on land use are not clear. Taking the Hainan Island in China as an example, the flow path map for flood regulation services was generated using ArcGIS 10.2. This involved utilizing the Digital Elevation Model and river hydrological network. The spatial transfer process of ecosystem services from supply areas to demand areas was clarified by examining the spatial balance between supply and demand. By quantifying the spatial service flow, the matrix of spatial interaction intensity at the county scale can be incorporated into the spatial weight matrix. Subsequently, the Spatial Durbin Model (SDM) is employed to investigate the impact mechanism of land use under the constraints of flood regulation services. The research results contribute to providing practical guidelines for improving flood regulation networks, enhancing land use management, and regulating the spatial spillover effect.

Most ecosystem services can be used by humans after they are produced, delivered, and transformed. Driven by natural and man-made effects, ecosystem services transfer in time and space in the process of realizing benefits, thus forming ecosystem service flows (Mononen et al. 2016; Dolan et al. 2021). The basic spatial units of ecosystem service flows include the service-providing area, connection area, and service-using area. The ecosystem that can provide products or ecosystem services is called the service-providing area, which is the basic unit for the research on the formation and change mechanism of ecosystem services. Correspondingly, the area where the service is consumed is called the service-using area (Luck et al. 2003, 2009). The concept of a service-connecting area is an intermediate area that exists under the condition of spatial heterogeneity between service-providing area and service-using area, which will affect the process variables of ecosystem service flows (Syrbe & Walz 2012). Some ecosystem service flows can transfer services directly to the service-using area through the service-providing area, while some regions or landscape systems in the delivery path are called ‘sinks’, which can absorb the delivered products or services and reduce the delivery efficiency of ecosystem service flows (Chen et al. 2019). In nature, some ecosystem services need to be transferred from the service-providing area to the service-using area with the help of carriers such as non-biological factors or biological factors (Kontogianni et al. 2010). For example, a directional service flow is formed with the flood as the carrier, and flood regulation services are transferred from the service-providing area to the service-using area in a certain direction.

Based on the temporal and spatial connection characteristics of the ecosystem service flows between the service-providing area and the service-using area (Zhai et al., 2021b; Zhang et al. 2021), the ecosystem service flows that carry flood regulation is intertwined along their respective directions to form a complex network. As the nodes of the network of ecosystem service flows, the supply source and human demand play the role of supply, transformation, processing, and use. The network of ecosystem service flows closely connects the ecosystem and human society.

The network of ecosystem service flows is based on the spatial interaction force of the ‘source-sink-matrix’ model, which has spatial attributes and exhibits spatial spillover effects. This spatial spillover effect can be regarded as a pure spatial externality (Zhai et al.,2021b). Ecosystem service flows have strong spatial scale characteristics. Ecosystem services provide welfare for humans through spatial flows of different scales (Bagstad et al. 2013; Cimon-Morin et al. 2013). In this dynamic process, the ecosystem service flows will play a key role in ecological penetration, which will have a direct impact on urban and rural land use, changes in landscape patterns, and economic consumption (Baró et al. 2016; Green et al. 2016). The communication of ecosystem service flows between ecological ontology and its adjacent affected areas can improve the marginal benefits of ecological space and its influence on land use. Due to the existence of regional externalities, the network of ecosystem service flows exhibits spatial spillover effects in the spatial process. On the one hand, due to the network nature of ecosystem service flows, they link the economic activities of various regions into a whole and make the land use of one region affect the land use of adjacent areas through the diffusion effect, which is a direct effect. On the other hand, flood regulation services will change the attractiveness of the region, improve the location advantage of the service-using area, and speed up the flow of production factors. The accumulation or diffusion of production factors in different regions may affect the growth or decline of land use in other regions.

The research data are derived from five components: land cover data, elevation data, soil data, meteorological data, and social and economic data. (1) Land cover data is obtained using Landsat TM/OLI remote sensing images from the Geospatial Data Cloud (Chinese Academy of Sciences 2021), serving as an open-source, high-resolution remote sensing image. After downloading data for the study area in 2015 and 2018, ENVI 5.1 software was employed for the second-phase remote sensing image preprocessing, including radiometric calibration, atmospheric correction, mosaicking, and cropping. Object-oriented classification is then applied. The classification results are iteratively refined, achieving an interpretation accuracy exceeding 80% through a combination of Google Maps and field research. (2) Elevation data with a precision of 12.5 m is obtained from ASF Data Search (Alaska Satellite Facility 2021). This digital elevation map features high detail accuracy, enabling the visualization of more intricate contour lines. (3) Soil data is sourced from the Resource and Environment Science and Data Center (Chinese Academy of Sciences 2020), covering various soil types and their primary attribute characteristics in China. (4) Meteorological data is acquired from the National Climate Center (China Meteorological Administration 2020). This dataset possesses advantages such as a high number of monitoring stations, temporal precision, and comprehensive coverage of meteorological elements. (5) The social and economic data extracted from the Hainan Statistical Yearbook (Hainan Provincial Bureau of Statistics 2022), an official publication by the statistical department of Hainan Province. This publication systematically records the requisite social, economic, resource, and agricultural data for this study.

Interception rates and soil moisture were determined using empirical values from some local studies, and this study sets ⁠, and equal 20, 10 and 10% (Chen et al. 1998, 2020; Liu et al. 2012, 2013). and are constants calculated based on the correlation analysis between precipitation and depth of soil infiltration. When the soil type is sandy loam, and are equal to 0.3951 and 4.4691, and when the soil type is clay loam, and are equal to 0.7647 and 0.74 (Jiang 2006; Ye et al. 2010). It should be noted that in addition to soil type, the infiltration depth of precipitation is also related to many factors, such as precipitation duration, watershed slope, and groundwater level, which all determine the amount of rainfall infiltration. Therefore, the relationship between precipitation and depth of soil infiltration is not necessarily simple linear. However, the emphasis lies on introducing a straightforward method for estimating the total amount of soil water in the region. Therefore, the relationship between precipitation and infiltration depth calculated by experimental data and empirical data is only used for estimation. The natural breaks method was employed to categorize the computed results of the supply. The natural breaks method is a data grouping method that divides data into several groups by identifying breakpoints in the dataset. This division ensures that the within-group variation is minimized while maximizing the variation between groups (Fraile et al. 2016).

The path of flood regulation service flow is determined mainly by the flooded ditch. There are four main points in the analysis of the watershed features of the Hainan Island based on DEM data as follows: (1) DEM data preprocessing. Depressions and flat land will affect the formation of river networks, leading to the phenomenon of pseudo river channels and parallel flow directions, so that a complete watershed network cannot be generated, which is inconsistent with the actual continuous river network. Consequently, in analyzing the hydrological characteristics of the watershed, it is necessary to address depressions in the DEM data to establish a defined channel for water flow. The ‘Fill’ tool in ArcGIS 10.2 can automatically identify the boundaries of depressions and level them based on the elevation information from the DEM data. The calculation steps can be referred to in the study by Planchon & Darboux (2002). (2) Calculation of flood flow direction. The determination of flow direction can be divided into single flow direction and multi-flow direction, and the flow direction obtained by different algorithms is also different. Considering the accuracy of river network extraction within the basin and its alignment with the actual terrain, the eight directions (D8) flow model is adopted. This model demonstrates robust capabilities in handling depressions and flat land, coupled with a straightforward calculation method. (3) Extraction of the flood flow network. River extraction and classification are defined by Stream Link, Stream Order, and Stream to Feature, which is provided by the hydrology toolset of ArcGIS. The network classification of flood flow uses the Strahler method. All connections without tributaries are classified as Grade 1. When the same level of river convergence, the classification of the river network will increase (Li et al. 2021b). (4) Watershed extraction. Watershed extraction is based on the river, flow direction, and water outlet to determine its spatial scope. From the perspective of hydrology and geography, its area must have a corresponding relationship with the river. The watershed corresponding to each flood flow path can be extracted by using Snap Pour Point and Watershed of ArcGIS.

The temporal and spatial analysis of LUD can characterize the change process of land-use intensity in each region. Concerning previous research results and considering the actual state of land use in the study area, this study divides the current state of land use in the Hainan Island into six types: construction land, croplands, forest, grasslands, water bodies, unclassified land (Zhang et al. 2019). The definitions of the detailed LUD are summarized in Table 3.

The driving forces leading to land-use changes come from many aspects, such as population, wealth, economy, and urbanization. The focus is on the influence of social and economic development on the LUD. Combining the geographical characteristics of the Hainan Island and the availability of statistical data, potential socioeconomic driving factors selected include population, GDP, urbanization rate, total retail consumption sales, investment in fixed assets, the proportion of the secondary industry, the proportion of the tertiary industry, disposable personal income of urban residents, disposable personal income of rural residents, gross industrial output value and gross agricultural output value. The values of these variables can be found in Tables S2 and S3, Supplementary material. The descriptive statistics of each variable are shown in Table 4.

The spatial weight matrix is an important means to describe the spatial relationship between regions. W is the spatial weight matrix of the county-level administrative unit in the Hainan Island. is the element in row i and column j of W. and represents the spatial weight matrix under the value of the ecosystem service flows for 2015 and 2018, respectively. In contrast, the spatial weight matrix under geographical proximity is represented as ⁠. or is the standardized value of the flood regulation service flowing from county j to county i. The assignment principle of is equal to 1 when there is a common boundary between county j and county i, otherwise, it is equal to 0. and are directed adjacency matrixes and is an undirected symmetric matrix. Taking data availability into account, the spatial matrix is constructed based on service flow paths at level 3 and above in the network.

Table 5 shows socioeconomic indicators related to land use in 2015 and 2018. First, GDP, TRCS, IFA, and GIOV are highly correlated with the LUD at the significance level of 1%, and the correlation coefficients are all above 0.50. Second, POP, PTI, DPIUR, and DPIRR are highly correlated with the LUD at the significance level of 5%. Finally, UR, PSI, and GAOV can be excluded, which have no significant correlation with the LUD.

First, through the spatial weight matrix of ecosystem service flows (⁠ and ⁠), POP, GDP, and IFA have an impact on the LUD. In 2015, when other variables remain unchanged, for every 1% increase in GDP, the LUD will increase by 3.54%. (−0.1060 in 2015) is the regression coefficient of the space lag item of LUD, which is significant at the 0.05 level, indicating that the LUD of the related area has a significant negative influence on the LUD of the local area in that year. W_POP (1.1968 in 2015) is the regression coefficient of the spatial lag of POP, which is significant at 0.01 level, indicating that POP has a spatial spillover effect between regions with ecosystem service connections, and POP has a positive effect on the LUD. In 2018, for every 1% increase in GDP and IFA, LUD will increase by 5.60 and 2.70%, respectively, when other variables remain unchanged. (−0.0617 in 2018) is significant at 0.01 level, and it shows that the LUD has a negative effect under the network of ecosystem service flows. In other words, when the LUD of the county where the flood regulation service outflow increases, the LUD of the county where the flood regulation service inflow tends to decrease. W_GDP (0.7833 in 2018) is the regression coefficient of the spatial lag of GDP and is significant at 0.1 level, indicating that GDP has a positive effect on LUD.

Second, in the network of geographical proximity, both POP and GDP have an impact on LUD. Except for the constant term, there were no significant independent variables, indicating that POP, GDP, and IFA had no significant effect on LUD in 2015. In 2018, for every 1% increase in GDP, LUD will increase by 4.76% when other variables remain unchanged. W_POP (−1.9529 in 2018) is significant at 0.1 level, and it shows that POP has a negative spatial spillover effect between geographically adjacent regions. is not significant, which indicates that the LUD of county units had no significant influence on the LUD of neighboring counties in 2015 and 2018. Moreover, neither 2015 nor 2018 passed the likelihood ratio test, indicating that under the weight matrix of geographical proximity, the SDM cannot be rejected as an OLS regression.

Finally, the results of the SDM in 2015 and 2018 were compared. The spatial lag term of the dependent variable has an enhanced impact on the LUD of the Hainan Island through the network of ecosystem service flows. The direct contribution of GDP and IFA to LUD changed from 2015 to 2018, in which the contribution of GDP was still significant and strengthened, and IFA changed from insignificant to significant. In the spatial lag of independent variables, the positive spillover effect of POP on the LUD of the Hainan Island is reduced, while the GDP is the opposite. However, the spatial lag of the dependent variable has no significant effect on the LUD through the spatial weight matrix of geographical proximity. At the same time, GDP promotes the LUD significantly, while the negative spillover effect of POP on the LUD of the Hainan Island increased.

The supply capacity of flood regulation service of the Hainan Island is attenuated from the central area to the surrounding areas, while the demand level is much higher in the surroundings than in the central area. The flow path of flood regulation services extracted by hydrological analysis tools reveals the spatial transfer relationship between service-providing areas and service-using areas among counties. This information is helpful to form the adjustment strategy of the supply and demand relationship of flood regulation service and provides effective guidance for the management and protection of natural resources in rapidly urbanized areas (Shen & Wang 2021). This study suggests the following measures to improve the flow network of flood control service in Hainan Island: enhance the supply capacity of the service-providing area, reduce the risk of flood disaster in the service-using area and control the quality of flood regulation.

The results show that forest provides more flood regulation services than grassland and reservoirs, but the area of forest is decreasing. A series of flood disasters tells us that forest plays an irreplaceable special role in preventing natural disasters (Sheng & Zou 2017). The interception and soil fixation of forests can greatly reduce surface runoff, avoid soil erosion, and effectively alleviate soil erosion (Cooper et al. 2021). Moreover, the forest also has obvious effects on reducing hazards such as debris flow and collapse. With the increase in forest coverage, the flood disaster caused by heavy rain in the region can be alleviated, thus reducing disaster losses. The practice has proved that afforestation is the fundamental measure to control floods (Bhattacharjee & Behera 2018). Cultivation of forest resources should be given high priority to increase existing forest resources. During forestry construction, adherence to the laws of nature and scientific planning is imperative. It is essential to strike a balance between the economic forest and ecological forest structures.

Based on the comprehensive assessment of ‘Hazard-Exposure-Vulnerability’, it is concluded that the risk of flood disaster in the eastern, northern, and western parts of the Hainan Island is higher than central area, and these areas include a large number of contiguous croplands, towns of a certain size and scattered rural settlements. Common methods to reduce flood risk include engineering measures and non-engineering measures (Kundzewicz et al. 2018; Xia & Chen 2021). Engineering measures are to improve the resistance to flood disasters through the construction of hardware facilities, such as building flood control levees, enhancing the flood resistance of buildings. Non-engineering measures indirectly affect human behaviors and actions through appropriate mechanisms, such as financial and legal policies. This study suggests that the control of flood disaster risk should be embedded in the whole process of flood disaster, including all stages of preparation, response, and recovery. Preparation includes building and consolidating flood control facilities, monitoring, and forecasting flood disasters, and improving the public's self-rescue and mutual rescue capabilities with public participation. The response includes emergency rescue, evacuation, and resettlement. The recovery process after the incident is not only to help the victims rebuild their homes quickly but more importantly, to start a new round of risk analysis and risk assessment procedures.

Due to the limited regulation capacity of the ecosystem itself or affected by cumulative pollution and external pressure, it is easy to damage the structure and function of the ecosystem. To ensure the healthy development of the network of flood regulation services in the Hainan Island, the maintenance and management of the flow path must be carried out according to the characteristics of the regional water system and related planning. It mainly includes the maintenance and management of eco-embankment, ecological buffer zones, and engineering facilities of the water environment. In addition, it is necessary to pay attention to the management of special periods such as flood seasons and drought periods. Only by improving the regional water environment and flood control conditions can the healthy flow of flood regulation service in the Hainan Island be guaranteed.

Through the analysis of SDM, the LUD of the Hainan Island is affected by social and economic development. Specifically, the growth of GDP contributed the most positive driving effect, and the IFA and POP also had a positive impact on LUD. However, land use also has a response to flood disasters. The flow network of flood regulation services, which reflects the regional interaction, plays an important role in the spatial spillover of land use. The direct spillover effect of related county units through the ecosystem service flow path limits the LUD, showing the spatial constraints of the network of ecosystem service flows. However, as a spatial medium, the network of ecosystem service flows can also act as a spatial dynamic, which shows that the indirect effect of related county units is still positive to the LUD.

The land use pattern is simultaneously affected by economic drivers and ecological constraints. It is necessary to integrate resources, environment, technology, society, economy, and policies to build a well-structured science-technology-management system. In order to prevent and control flood disasters, the Housing and Urban-Rural Development Department of Hainan Province (2017) issued the ‘Guidelines for Sponge City Planning and Design in Hainan Province’. The primary planning control target outlined in this document is to achieve an annual runoff control rate of not less than 60% within the built-up area. It imposes corresponding rigid controls on water bodies, green spaces, and construction land in various regions of Hainan Province. However, in addressing disturbances caused by flood disasters, resilient land management is equally crucial alongside rigid controls. Rigid management represents the safety baseline for regional disaster prevention, while resilient management emphasizes flexibility, adaptability, and sustainability, contributing to better responses to dynamic and uncertain environments. Therefore, considering the actual situation in Hainan Province and referencing practices from typical cities, this study proposes the following recommendations for enhancing the resilience of land use and management in Hainan Province:

• (1)

  Upgrading infrastructure. To address the challenges posed by flood disasters, some cities emphasize the upgrading and renovation of infrastructure. For instance, London's approach involves collaboration with the Open Data Institute, emphasizing the use of data and models as a foundation for policymaking, and conducting risk monitoring and evaluation of infrastructure (Greater London Authority City Hall 2020). The Paris City Hall (2018) has proposed predicting and analyzing potential risks to infrastructure, estimating future energy consumption, and simultaneously increasing research on foundations to mitigate the risks of building collapse and flooding. Taking into account the specific circumstances in Hainan Province, infrastructure upgrades can be pursued in the following ways. First, comprehensively revamp public spaces in flood channels, clearing and rectifying illegal constructions below the designated water level. Second, enhance and update facilities such as water, electricity, gas, communication, fire protection, and safety evacuation in high-risk areas. For areas currently not suitable for large-scale relocation, adjust the functions of low-rise buildings and improve emergency refuge facilities. Third, strengthen the development of information infrastructure and establish a space-air-ground integrated monitoring and perception system for flood disasters.

• (2)

  Advancing diverse collaborative governance. In the resilience strategic planning of cities like London and Paris, each strategy clearly identifies responsible parties and specifies the institutions and organizations to be involved in the collaboration (Paris City Hall 2018; Greater London Authority City Hall 2020). The rapid development of big data can enhance information sharing and the efficiency of collaborative governance. This entails the need for extensive collaboration among multiple sectors and the public. Cities within Hainan Province can collaborate to establish an ecological network across cities, aiming to enhance the overall resilience of ecosystems through the restoration and protection of natural ecosystems. Leveraging ecological compensation as a lever for achieving collaborative governance between cities is also a promising approach (Lu et al. 2023). This involves providing economic incentives to regions that offer flood control services, encouraging them to protect and restore ecosystems. Furthermore, to effectively integrate resources and expertise from various aspects, collaborative governance among government departments should be encouraged. It is recommended that the natural resources department, environmental protection department, water resources department, transportation department, emergency management department, and big data management department in Hainan Province collaborate, integrate their respective plans, and establish an effective mechanism for information sharing. Additionally, leveraging technological means such as mobile applications and online geographic information platforms can make it easier for the public to access real-time data and spatial information regarding flood disasters.

• (3)

  Improving land use planning. Given the significant differences in objectives and functions across various spatial scales in land planning in Hainan Province, it is crucial to enhance the coordination and flexibility of planning at all levels. Macro-level planning (such as provincial-level land planning) should prioritize forecasting and responding to significant regional disasters. By integrating historical disaster data, it should propose comprehensive disaster prevention and mitigation strategies for key areas, including urban clusters, coastal zones, and major river basins. Meso-level planning (such as city-level and county-level land planning) should systematically lay out various disaster avoidance spaces and essential disaster prevention facilities, taking into account the disaster prevention needs of both central urban areas and surrounding towns. Micro-level planning can draw inspiration from New York City's resilient development planning for communities (NYC Government 2019). Treating communities as the smallest planning units, tailored strategies for each community can be determined based on local conditions. This includes zoning and land use changes to support the vitality and resilience of flood-prone communities, preparing them to face future risks effectively.

Many studies use standards such as whether there are public boundaries and the distance between geographic centers to express the spatial relationship between economic entities, and then transform this spatial relationship into the form of a spatial weight matrix as an important variable for judging the intensity of spatial spillover effect (Wang & Wu 2016; Song & Son 2020). However, the spatial weight matrix should change as the interaction relationship changes. The spatial correlation model of LUD in county units evolves with changes in the interaction of flood regulation service flows, and the spatial weight matrix based on geographical proximity cannot adequately reflect this change. It also shows that the spillover effect is different through different network structures.

Spatial externality theory is often used to answer the question of the spatial spillover effect (Capello 2007; Zhao et al. 2021). Regional externalities are the positive or negative impacts of a region's economic activities on other regions that are not considered or undertaken by them, and further believe that externalities are the cause of uncoordinated development between regions. How to coordinate the relationship between spatial units has become another important research topic. Firstly, attention should be directed towards the positive spatial spillover effect of economic growth resulting from population and investment. It facilitated two processes: On the one hand, the increase in LUD in the local area has triggered a centripetal movement, which is manifested in the expansion and spread of construction land; on the other hand, the centrifugal movement caused by the crowded market of the local area has become the driving force of the core county's unipolar expansion into the common development of the region. From the perspective of land management, this study suggests that differentiated spatial control strategies for construction land should be formulated based on the differences in the current status of LUD. For high-intensity development units such as the coastal areas of the Hainan Island, it is important to strengthen the potential tapping of construction land, while for low-intensity development units, it is necessary to adhere to the eco-oriented smart growth model. It should be noted that under the restriction of flood regulation services, LUD shows a negative spatial spillover effect. This study proposes to identify key ecological units based on the spatial interaction strength of the network of ecosystem service flows at the county scale. According to the characteristics of ecological units, it is helpful to formulate differentiated fiscal transfer payment and market-based ecological compensation strategies. Moreover, it is essential to take ecological compensation as a lever to regulate the relationship between environmental protection and land development, to optimize the spatial pattern of LUD in the Hainan Island. These measures are conducive to improving the long-term mechanism of ecological compensation between county units on the Hainan Island.

Our contribution primarily lies in two aspects. On the one hand, this study provides a research perspective on the relationship between flood regulation services and land use. It is well known that there are mutual relationships and influences between ecosystem services and land use (Xiong et al. 2020). Many studies have demonstrated the impact of land use changes on flood regulation services (Ryffel et al. 2014; Arowolo et al. 2018; Luo & Zhang 2022). However, some studies have focused more on the supply and demand of flood regulation services, with less emphasis on the spatial impacts of land use (Shen et al. 2021). Therefore, a key issue arising is whether flood regulation services with directional flow characteristics also have a spatial impact on LUD. Our research aims to answer this question, which is our important starting point and research hypothesis. On the other hand, the spatial weight matrix of the ecosystem service flow and the geographic adjacency matrix are separately incorporated into the SDM for comparison. This approach unveils the spatial spillover effects within the flood regulation service flow network. In an increasingly networked environment, various forms of spatial interactions embedded in widely used SDM should not be overlooked. Spatial econometric models have a strong dependency on the spatial weight matrix and exhibit considerable uncertainty (Yu et al. 2022). Therefore, some studies have found that the spatial spillover effects of land use not only depend on geographic adjacency relationships but also on network relationships, such as transportation networks (Zeng et al. 2020), infrastructure networks (Zeng et al. 2019), carbon emission correlation networks (Yu et al. 2022), and ecological networks (Yang et al. 2020). The results of this study with the flood regulation service flow network as the media can provide a supplement for the relevant research.

Using the Ecosystem Service Flows as a starting point, a theoretical framework is proposed to elucidate the interaction and potential impact within the land–ecology–socioeconomic system. Based on this framework, using land-use data, social and economic data, natural environment, and ecological data, this study calculated the LUD in the Hainan Island, constructed the flow path of flood control services, set the spatial weight matrix of ecosystem service flows into the SDM, explored the influence mechanism of LUD under ecological constraints, and proposed the corresponding optimization path. Our results indicate that the development of the flood regulation service flow network affects LUD by population, GDP, and investment in fixed assets. Compared to the geographical proximity matrix, our findings also demonstrate that the network of flood regulation service flow serves as an important medium for generating spatial spillover effects on LUD.

Due to data availability and adjustments in administrative divisions, the analysis was conducted exclusively using data from 2015 and 2018, but the study on the spatial-temporal change of LUD and its spatial–temporal relationship with the spatial spillover effect of SEF requires a longer period and a more detailed time series segment to conduct a refined study. Therefore, there are some questions worthies of further study. First, consider different climate change scenarios. While specific historical climate scenarios have been considered, it is important to acknowledge the uncertainty associated with future climate change. Agent-based modeling can be utilized to integrate climate modules, land-use change modules, and flood regulation service flow modules. Setting various climate change scenarios allows for the analysis of their impacts on flood regulation service flows and land use intensity. Second, compare spatial spillover effects at different scales. This can involve examining regions at various scales, ranging from small-scale areas to large-scale watersheds. As spatial scales change, analyze whether the intensity, extent, and direction of spatial spillover effects undergo changes. This facilitates the exploration of the varying patterns and mechanisms of spatial spillover effects across different scales. Third, investigate the spatial spillover effects of other ecosystem services. Apart from flood regulation services, other ecosystem services may also exhibit spatial spillover effects. Further research can be expanded to include other ecosystem services such as carbon sequestration and water quality purification, enabling a comprehensive understanding of the impact of ecosystem service flows on land use.

This research is funded by the National Natural Science Foundation of China (No. 41871179).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.