Research on a cloud model intelligent computing platform for water resource management

As technologies such as big data and artificial intelligence continue to mature and be applied, the technology of smart water conservancy has undergone continuous upgrades, with its management scope also expanding significantly. Countries worldwide have fully leveraged the power of new-generation information and communication technologies (such as the Internet of Things, cloud computing, and artificial intelligence) to enhance water conservancy infrastructure, thereby improving capabilities in forecasting, early warning, simulation, and contingency planning (Yuan 2022). With the continuous advancement of cloud computing, deep learning, and other technologies, models will see marked improvements in data computation, deduction, and other capabilities. Additionally, information aggregation and integration, as well as resource integration and sharing, will continue to be refined, significantly accelerating the rapid development of cloud-based water resource models (Huang et al. 2017). However, current research on cloud computing and cloud models for water resource systems is still in its infancy, and the most relevant simulation numerical model software for water resource systems remains in the model programming phase (Xia et al. 2022).

With the development of information technology, several hydraulic cloud platforms have emerged in areas such as water disaster management, water information systems, water environment monitoring, hydrodynamics-based hydraulic models, hydraulic model management, water resource systems, and hydrological cloud models. For example, in the field of water disaster management, Wan et al. (2014) have developed a global flood disaster information cloud platform, He et al. (2012) have proposed a cloud platform for debris flow disaster simulation, Xu et al. (2024) have constructed a reservoir flood control and drought relief system. In the domain of water information management cloud platforms, Zhang et al. (2015) have adopted a service-oriented architecture to construct a metadata and model-driven water information-sharing platform. Li (2021) has also established a water data-sharing cloud platform based on cloud computing. In terms of water environment management cloud platforms, Xu et al. (2022) have proposed a refined management system based on watershed control units and developed a refined water environment management cloud platform for the Huangshui River Basin. Chen et al. (2023) created an integrated water environment management platform using data fusion technology and an air-land-water coupled model. In the area of hydrodynamics-based hydraulic model construction, Liu et al. (2014) developed a hydrodynamics simulation service platform (the HydroMP system) based on cloud computing. Regarding hydraulic model management cloud platforms, Xia et al. (2022) created a numerical simulation cloud service platform for hydraulic models based on standardization, Qin et al. (2024) proposed to build a unified intelligent operation and maintenance management portal based on BIM + GIS(Building Information Modeling+ Geographic Information System), cloud computing, Internet of Things technology, and operation and maintenance big data. In water resource systems, Liang et al. (2017) developed an intelligent water resource app system based on cloud computing, while Arango et al. (2014) proposed a water resource allocation model system. For hydrological cloud models, Kurtz et al. (2017) combined data assimilation technology with hydrological models in a cloud environment to establish a real-time underground forecasting and management system. Chen & Dong (2013) developed the Liuxi River model cloud service platform based on a distributed physical hydrological model for watershed flood forecasting. The aforementioned research on hydraulic application cloud platforms primarily focuses on infrastructure services such as data publication, data storage, and the provision of computing resources. There is comparatively less research on the development functions of hydraulic application software within cloud platforms. Additionally, these cloud platform systems mainly emphasize system applications and some systems have yet to fully utilize web and cloud computing technologies, with room for improvement in computational efficiency.

To overcome the limitations of traditional hydraulic application cloud platforms, this study utilizes cloud computing, topological theory, and other technologies to describe the topological structure of a digital water network (Peng 2024; Xie 2024). The aim is to establish a cloud model service platform that couples water resource allocation with hydrological cycle simulation. Considering model reuse and flexible assembly, the study uses an incidence matrix to store topological relationships, allowing graphical data and the topological water network to be stored collaboratively, with data interactions managed uniformly using a data flow model. Taking the Weihe River Basin as an example, this study constructs a water resource simulation and multi-objective allocation model based on the digital water network. Adopting a business-oriented modeling approach and using a B/S architecture, the study establishes a cloud model service platform for hydrological cycle simulation and water resource allocation. This platform enables integrated simulation of the hydrological cycle and water resource management on the cloud, providing technical reference and guidance for the construction of intelligent water conservancy systems and refined water resource allocation in the Weihe River digital basin.

The main innovations of the study include the following: (1) Distinguished from a spatial data-based digital water grid, the study introduces and systematically implements a topological digital water grid. The description of the topological water grid is carried out through the generalization of topological elements and the establishment of topological relationships. The topological relationships of the topological digital water grid are stored using a topological incidence matrix, coordinating the storage of topological water grid graphic data and topological relationships. On a comprehensive integration platform, achieve the flexible construction of the topological digital water grid. (2) With component technology at its core, the study granularizes complex calculations. It combines the process of converting water flow in the regional water resource system with the topological visualization of water resource business processes through mapping relationships. (3) The tight coupling of the water cycle model and the water resource allocation model has been established to build a cloud model service platform for watershed water resource planning. This platform achieves a dynamic configuration of regional water resources relying on a digital topological water grid.

The foundation of basin water resources planning involves accurately characterizing the water network system within the basin. This includes both natural entities (e.g., lakes and rivers) and artificial entities (e.g., reservoirs, inter-basin water transfer projects, and water diversion and drainage canals). The characterization of natural water systems can be achieved using algorithms like TauDEM, whereas the topology of artificial water systems often requires manual definition. This study classifies the topological elements of the artificial side of the water network system into two categories: nodes and arcs. The generalized topological elements represent the smallest units of mathematical expression within the water resource system at a given computational scale and their characteristics can be described by defining attributes. The basic topological elements and their corresponding system entities are summarized in Table 1.

For the nodes in the incidence matrix, a ‘node vector’ can represent the connection relationships between the node and the edges. For example, the node Z1 has a node vector of (0, −1, 1, 0, 0, 0, 1, 0, 0, 0). The 1st, 4th, 5th, 6th, 8th, 9th, and 10th components are 0, indicating that directed edges 1, 4, 5, 6, 8, 9, and 10 are not connected to node Z1, meaning there is no hydraulic connection. The second component is −1, indicating that directed edge 2 is connected to node Z1 and that node Z1 is the starting point. The third and seventh components are 1, indicating that directed edges 3 and 7 are connected to node Z1 and that node Z1 is the endpoint. Therefore, the digital water network can be converted into an incidence matrix, which in turn can describe the relationship between the water source and the user in the digital water network.

The overall hydrological cycle model is improved based on the Xin'anjiang model (Zhao & Wang 1988), using the Taudem algorithm for sub-basin division and independent modeling for each sub-basin. To account for the vertical distribution effects of soil moisture, the model employs a three-layer evaporation approach to describe the evapotranspiration process. The runoff generation part uses a combined infiltration-excess and saturation-excess hybrid runoff generation model (Tain et al. 2020). For the confluence part, each sub-basin's slope confluence is first calculated independently, and then the flow processes reaching the outlet section of each sub-basin are linearly superimposed to obtain the total flow process at the basin outlet section. The slope confluence of each sub-basin uses a linear reservoir or lag algorithm, while the river channel confluence calculation uses the Muskingum segmented continuous algorithm.

The simulation of hydraulic engineering and water diversion and drainage follows relevant references (Zhai 2012; Bi et al. 2020). The water use in various socio-economic sectors is based on the results of water resource allocation, actual survey statistics, and planning data.

The multi-objective water resource optimization allocation model, based on the digital water network topology, includes a water cycle module, a multi-source water supply module, a water demand module, and a multi-objective allocation module. The water cycle module is used for scientific calculations of available water supply, providing data support for the multi-source water supply module. After the multi-objective allocation module calculates the water resource allocation plan, the system will filter and recommend an optimal plan, and through iterative water cycle computations, the ideal result will be achieved.

• 1. The water supply module

Surface water supply (including external water diversion): The supply volume depends on multiple factors, including the available water volume from the water source (derived from the hydrological cycle model), the supply capacity of the water intake projects (such as diversion channels, supply pipelines, and wells), and constraints such as the total water intake control redline for the water source (including local surface water and external water diversion).

Reservoir water supply: The reservoir storage calculation is determined by the reservoir storage at the end of the previous period, the inflow from upstream during the current period, evaporation from the reservoir surface, reservoir seepage, and the reservoir's supply volume.

Groundwater supply: Analyze groundwater recharge and exploitable volume, with the extraction upper limit set by the allowable extraction volume for each region. The extraction volume can be rolled over within the year but cannot accumulate across years.

Reclaimed water supply: Currently, the utilization of unconventional water resources in China is still in its initial stages. Unconventional water resources are generally included in water resource allocation for ecological, agricultural, urban greening, and industrial uses, laying the foundation for quality-differentiated water supply (Wu et al. 2021). This paper sets the topological relationships between reclaimed water and water users, designating reclaimed water supply for industrial, urban greening, ecological, and agricultural users.

• 2. The water demand module

To predict annual water demands for agriculture, industry, construction and services, residential life, and livestock planning, a quota method is employed. For ecological water needs within rivers, the Tennant method is used, ensuring that the minimum ecological flow in the river is no less than 10–20% of the long-term average flow. From the supply perspective, the ecological base flow within rivers is directly deducted. For reservoirs, the available water supply is adjusted by subtracting the mandatory release volume.

• 3. The multi-objective module

This model incorporates three objectives: economic, social, and water quantity goals. The economic objective minimizes regional economic losses due to water scarcity, while the social objective minimizes food production reduction caused by water shortages. Details can be found in Zhao et al. (2023).

Constraints include: Priority allocation for domestic water use; groundwater supply should not exceed available groundwater resources; ecological protection requirements for important river ecosystems; flow limits at water transfer nodes; reservoir capacity constraints; non-negativity constraints: all decision variables in the model must be ≥0.

The water resource allocation module first divides the computational units according to sub-basins nested within administrative regions. It conceptualizes the entire water resource system into a network system consisting of numerous computational nodes and transmission lines. During the computation process, the water cycle module provides runoff resource quantities (including surface runoff processes, node inflow processes, and reservoir inflow processes) to the water resource allocation module. It uses a ‘space-time aggregation’ method to provide input processes suitable for the temporal and spatial scales required by the water resource allocation module. The water resource allocation module then spatially and temporally distributes the simulated processes of supply, demand, consumption, and discharge to meet the data input requirements of the water cycle module.

Specifically, based on establishing a ‘natural-social’ topological structure of a digital river network, the water cycle module simulates natural hydrological processes to obtain daily inflow processes at various hydraulic engineering nodes and daily runoff processes for various hydrological calculation units. These units are then aggregated in time and space to provide the temporal scale (monthly or 10 days) and spatial scale (water resource allocation units) needed by the water resource allocation model.

Meanwhile, the water resource allocation module utilizes inflow information from the water cycle module, along with external demand information, to conduct long-term basin/regional water resource supply–demand balance analyses using methods such as the long-series method. This analysis yields the supply, demand, consumption, and discharge processes for each allocation computational unit and the operational processes of lakes and reservoirs and other hydraulic engineering structures. These elements are then temporally and spatially distributed, expanding monthly-scale allocation plans to daily/hourly-scale hydrological calculation units. Simultaneously, monthly scheduling processes for lakes and reservoirs (including water demand, supply, and spillage) are expanded to daily/hourly processes.

The multi-objective optimization model for water resource allocation, solved using the NSGA-III algorithm, yields multiple sets of non-dominated solutions, each representing a water resource allocation scheme. To select a reasonable scheme from the non-dominated solution sets, this study constructs a group multi-attribute decision-making model (Zhang et al. 2021). This model employs the interactive Chebyshev method to choose a reasonable solution from the multiple sets of non-dominated solutions.

The workflow model of the digital water network is structured as follows:

• (1) Process control: Process control nodes are divided into two types: start/end nodes and computation control nodes.

• (2) The common attributes of process control nodes include: Status, upper-level node address list, and result buffer, where start/end nodes also include start/end flags.

• (3) Component node: When a component node in the control flow is triggered, the node first checks whether the state has been computed. If it has been computed, it directly displays the computation result and clears the cache thereafter. If it has not been computed, it recursively moves upward until it finds a node that has been computed or the start node. It then calls the service corresponding to the calculation address of the calling node, sends the input to the calculation server, and when the input server returns the result, the control node caches the result and stores it in the lower-level node, proceeding with iterative calculations. If an error occurs in any node during the calculation, the calculation is terminated, and the error message is displayed directly.

• (4) Digital water network node: Define the control logic between digital water network nodes and business component nodes using logical type sets, representing the input and output logic.

The IaaS layer includes spatial database systems such as rainfall, spatiotemporal runoff data, Digital Elevation Model (DEM), soil, vegetation, land use types, hydrological engineering datasets, river network datasets, and sector-specific water usage data. Through data exchange services and extract-transform-load components, it provides timely spatial and temporal information services to the PaaS layer.

The PaaS layer provides a user-oriented operating and runtime environment, allowing users to make cloud calls for applications, data samples, and models, utilizing diverse model platforms and efficient data processing systems.

The SaaS layer deploys cloud-based hydrological cycle models and water resources optimization configuration models, providing multi-mode access mechanisms. (It usually refers to the multiple ways or channels through which users can access models and services deployed on the cloud platform).

The Visualization Service Layer features a visual operational interface, capable of generating visual reports and model result graphs, allowing users to directly view model results and batch output Excel reports.

The system design includes the core components at the server side: the hydrological cycle model and water resources optimization configuration model (SaaS layer), which provide computational services and encompass solution methods, numerical models, file systems, and database components. The database supports the entire system with graphic data, modeling data, computational data, hydrological data, and more. On the browser side, there are primarily two components: (1) Modeling section and flow field display: This section facilitates numerical model building, parameter setting, simulation result querying, and flow field visualization. (2) Data exchange between browser and web server: Communication occurs via HTML using Python interfaces to transmit data. Requests are sent to the server for analysis and computation using relevant models. The model computation results are then displayed within the HTML framework through the server.

The front-end development utilizes mainstream web technologies within a B/S architecture, concentrating the core system functionalities on the server side to simplify development, maintenance, and usability. Users access the system via browsers without needing additional client software installations. The B/S architecture centralizes system functions on the server, making it easy to maintain and user-friendly. Logically, this architecture is divided into three layers: the browser side, server side, and middleware. This distributed structure enables relatively independent storage, processing, and utilization of data, facilitating its maintenance and updates. In the B/S architecture, the web server layer serves as the primary functional implementation part. It processes data sent from the browser side, executes corresponding programs, and returns results to the client browser. Such architecture enhances the flexibility and scalability of web applications.

The front-end interface framework is implemented in HTML, designed using CSS for styling the HTML structure, and implements transaction logic through JavaScript to add dynamic effects. The user experience is web-based and integrates with GIS, enabling complex modeling and computational simulations of hydrological cycle models and multi-objective water resource configuration models through browsers.

Backend development involves MySQL as the backend database, designed based on the data types needed, including vector data (such as hydraulic engineering, water systems, and hydrological stations), basic data (such as DEM, soil, vegetation, land use types), and model data (such as rainfall, spatiotemporal runoff data, water supply from multiple sources, and water use quotas). It calculates annual water demand levels based on industry development plans and water use quotas, sending this demand data to the server. Additionally, it calculates available water supply based on the layout of hydraulic engineering in the study area and water inflow conditions, storing the results in the server.

Backend logic uses Python to call relevant models and transfer data from the database to the front end. When users initiate requests on the front end, the Python backend code receives and processes these requests, conducting data analysis and computations using the appropriate models. The backend transfers the computed results or database data to the front-end HTML, where users can view and utilize this data or computational results, with graphical representation using Echarts. This approach ensures efficient data interaction and information delivery between the front end and back end, ensuring the system's effectiveness and practicality.

The Shaanxi section of the Weihe River Basin is located in the arid and semi-arid regions of Northwest China, characterized by a typical continental monsoon climate and a warm semi-humid climate. Rainfall is concentrated mainly in July, August, and September, with an average annual runoff of 75.7 billion m³. The basin's terrain and geology are complex, encompassing areas such as river valley plains, rocky mountains, loess hills and gullies, and loess terraces. Due to intensified human activities, ecological problems in the Shaanxi section of the Weihe River Basin have become increasingly severe in recent years. Upstream areas suffer from soil erosion, while the middle reaches face water pollution issues, and the downstream areas experience significant siltation in the river channels. Consequently, the runoff in the Shaanxi section of the Weihe River Basin has gradually decreased (Jiang et al. 2015). The per capita water resources in this region account for 1/8 of the national average, and the per mu (Chinese unit of area) water resources for farmland are 1/6 of the national average, categorizing it as a water-deficient area (Han et al. 2016).

With the continuous and rapid development of the national economy, the conflict between water supply and demand will become increasingly prominent. At the same time, under the guidance of smart water management policies, establishing a cloud-based model service platform for water cycle simulation and water resource allocation in the Weihe River Basin, summarizing the patterns of water cycle changes in the basin, and proposing optimized regional water resource allocation solutions will provide technical support for a digital Weihe River Basin. This is of great significance in promoting the high-quality development and ecological protection of the Yellow River Basin.

Meteorological data are sourced from the China Meteorological Data Service Center. Actual runoff data are obtained from annual hydrological yearbooks, while irrigation area, cropping structure, socioeconomic, and water use data are sourced from various relevant databases.

Table 2

The meaning expressed by each icon

Serial number .	Icon .	Knowledge element .
1		Unconventional water Resources
2		Groundwater
3		Irrigated area
4		Reservoir

Serial number .	Icon .	Knowledge element .
1		Unconventional water Resources
2		Groundwater
3		Irrigated area
4		Reservoir

View Large

The main parameter values for the hydrological cycle model in the Weihe River basin were referenced from existing research results (Tain et al. 2020; Li et al. 2021), as detailed in Table 3.

The water sensitivity coefficient values for different crops in the Jensen model are referenced from existing research (Xiao et al. 2008), as expressed in Table 4.

Parameter values for solving the target model: in the NSGA-III algorithm, the preset population size is 130; the number of iterations is 500; the crossover probability is 1; the mutation probability is 1/(Max-Size), where Max is the maximum fitness value in the population and Size is the average fitness value in the population. Additionally, the simulated binary crossover parameter is 20, and the polynomial mutation parameter is also set to 20.

Based on the economic and social development plans of various cities (districts) in the Weihe River Basin of Shaanxi Province, and considering urban master plans, the projected water demand for different industries in the study area for the year 2025 is forecasted using the quota method, as shown in Table 6. The water demand is predominantly concentrated in three allocation units: Baoji to Xianyang North, Xianyang to Tongguan North, and Xianyang to Tongguan South, accounting for 30.60, 27.49, and 28.74% of the total water demand in the study area, respectively. Agricultural water use constitutes 51.46% of the total water demand in the study area, followed by industrial use at 32.34%. Specifically, in Xianyang to Tongguan North, agriculture dominates with a proportion of 68.37% of the water demand, whereas in Xianyang to Tongguan South, industrial water use predominates with 48.04% of the total demand.

To apply NSGA-III for solving a multi-objective model and perform interactive Chebyshev decision-making involving five decision-makers (DM1–DM5) with specific goal weight values in a water resources allocation zone, refer to Table 7 for the target weight values set by each decision-maker.

After multiple iterations, the benefit gaps between water resource allocation schemes become increasingly smaller, allowing the calculation to be terminated and a reasonable water resource allocation scheme to be selected, as shown in Table 8. In 2025, the study area requires 6.335 billion m³ of water, with a supply of 5.835 billion m³, resulting in a shortage of 0.503 billion m³. The water shortage is mainly concentrated in the areas of South Bank from Baoji Gorge to Xianyang, North Bank from Xianyang to Tongguan, and South Bank from Baoji Gorge to Xianyang Tongguan, with shortages of 0.148 billion m³, 0.156 billion m³, and 0.142 billion m³, respectively, accounting for 89.92% of the total shortage. Agricultural water shortages amount to 0.438 billion m³, accounting for 87.04% of the total shortage, while industrial water shortages amount to 0.060 billion m³, accounting for 11.89% of the total shortage.

In the study area, the water shortage rate is 7.95%. In the North Luo River area above Nanchengli and the upper reaches of the Baiyunshan Jing River, there is a basic balance between supply and demand. However, in the Jing River area above Zhangjiashan, the southern area of the Weihe River, and the northern area of the Weihe River above Linjiacun, the water shortage rates are as high as 8.37, 4.47, and 6.68%, respectively. In these regions, local water resources are insufficient to meet the growing demand. The water shortage rates in Baoji Gorge to Xianyang South, Baoji Gorge to Xianyang North, Xianyang to Tongguan South, and Xianyang to Tongguan North are 7.69, 7.65, 5.20, and 8.83%, respectively. This will have a certain impact on grain yield and GDP. In terms of water distribution, priority is given to ensuring nonagricultural water supply. The water shortage rate for domestic use is minimal, the industrial water shortage rate is 2.92%, and the agricultural water shortage rate is 13.42%, with the majority of the water shortage concentrated in agriculture. This allocation plan results in a 1.23% reduction in grain yield due to agricultural water shortages, coupled with a 0.15% decrease in GDP due to water shortages in other industries.

Models, such as Mike Basin, GWAS, and WEAP, have long served as industry benchmarks due to their proven capabilities in areas such as optimal water resource allocation, water quality simulation, and supply–demand balance forecasting. These traditional models, whose detailed features are outlined in Table 9, are typically built on a Client/Server (C/S) architecture. While offering robust functionality, these models often require users to install complex software environments on local machines, which can be cumbersome and resource-intensive.

However, with the rapid evolution of technology, particularly cloud computing, a paradigm shift in water resource modeling has emerged. The limitations of traditional models, such as the need for local installations and intricate configurations, are being addressed through the development of more agile, cloud-based solutions. In this paper, we present the CMICP model, which integrates cloud computing and adopts a B/S architecture. This approach eliminates the need for local software setup, allowing users to access the full functionality of the CMICP model through a simple web browser. By transitioning to a cloud-as-a-service model, CMICP offers a more flexible and user-friendly platform, overcoming the constraints of traditional C/S-based models and enhancing accessibility and ease of use.

This shift to cloud-based modeling marks a significant step forward in the field of water resource management, as it not only reduces the technical burden on users but also leverages the scalability and efficiency of cloud computing to improve the performance and accessibility of water resource optimization tools.

In this study, on the basis of constructing a watershed digital water network, the coupling of the hydrological cycle model with a multi-objective water resources optimization model is achieved. Leveraging the digital topological water network, the dynamic allocation of regional water resources is facilitated, establishing the Weihe River Basin in Shaanxi Province Water Cycle Simulation and Water Resources Configuration Cloud Model Service Platform. This cloud service platform adopts a business modeling approach integrated with B/S architecture, enabling integrated simulation of the water cycle and resource allocation at the cloud end.

The hydrological cycle model simulation results show relative errors of −1.665 and −4.63% for the calibration and validation periods of monthly average runoff, with Nash–Sutcliffe efficiency coefficients of 0.815 and 0.795, respectively, effectively capturing runoff variations during these periods. The study area experiences an annual rainfall of 217.80 billion m³, evapotranspiration of 179.70 billion m³, and an average annual runoff of 56.30 billion m³. Inflows are measured at 33.90 billion m³, while outflows amount to 72.00 billion m³. The total economic and social water intake is 51.00 billion m³. The spatial distribution of runoff depth primarily depends on rainfall distribution, with significant reductions observed at various hydrological stations.

Using the cloud service platform for the Weihe River Basin in Shaanxi Province, a multi-objective optimization of water resource allocation is conducted involving 5 decision-makers. Through a group multi-attribute decision-making model, a water resource allocation plan for 2025 is selected. The plan indicates a regional water shortage rate of 7.95%, a 1.23% reduction in grain production, and a 0.15% decrease in GDP. Agricultural water shortage amounts to 4.38 billion m³, representing 87.04% of the total water deficit. Water scarcity is particularly concentrated in Baoji Gorge to Xianyang North, Xianyang to Tongguan South, and Xianyang to Tongguan North.

The hydrological cycle in this model is based on the Xin'anjiang model. However, due to the complex topography and climatic conditions of the Weihe River Basin, the model exhibits inadequacies in handling spatial heterogeneity and has limited consideration of the impacts of groundwater and human activities. As a result, the accuracy of the simulation results may be affected.

In the context of the rapid development of smart water management, cloud platform construction is advancing quickly and gradually becoming an essential tool for water resources management. As computer technology continues to mature, the functionality and performance of cloud models will also see further enhancement and optimization. However, there are still some shortcomings in the current cloud-based platforms for water cycle simulation and water resource allocation, particularly in the consideration of water quality issues. Since water quality is one of the key factors affecting the sustainable utilization of water resources, it is crucial to incorporate it into the simulation and analysis within cloud models. Therefore, future efforts will focus on expanding the capabilities of existing cloud platforms to fully integrate water quality parameters, providing more accurate and comprehensive decision support for water resources management.

The authors acknowledge the support from the National Key Research and Development Program of China (2023YFC3206802) and the National Natural Science Foundation of China (No. 42307114).

T.W. is the machine learning expert, contributed to the investigation, developed the methodology, supervised the study, developed the communication, resources, and rendered support in data curation. T.W., J.D., Jiq.Z., J.Z. contributed to the original draft preparation, rendered support in data collection formal data analysis, investigated the whole process, contributed to the critical review, and final revisions. Y. G., F. G., L. Z., Y. Z. contributed to supervision, validation, revision, discussion, resources, improvement, and advice. All authors have read and agreed to the published version of the manuscript.

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

The authors declare there is no conflict.