A natural runoff reconstruction method based on the spreading of hydrological parameters in similar watersheds Open Access

With the continuous growth of the population and the ongoing development of the socio-economy, the issue of water resource shortages has become increasingly severe (Elahi et al. 2024). At the same time, under the influence of climate change and human activities, the problem of uneven spatiotemporal distribution of water resources has intensified (Abbas et al. 2022). This raises higher demands for water conservation and efficient utilization of water resources. To alleviate the water resource crisis, it is necessary to develop reasonable water resource allocation plans based on the spatiotemporal distribution and evolution characteristics of water resources, thereby improving water resource utilization efficiency. As a major component of the hydrological cycle, runoff responds to the impacts of climate change and human activities, and its evolution can lead to changes in regional water resource quantities. Surface water resources constitute the main body of total water resources and can be represented by natural river runoff. Natural runoff refers to the dynamic amount of water that can be renewed year by year in rivers, lakes, glaciers, and other surface water bodies formed by local precipitation (Miao et al. 2022). Under the influence of human activities, due to socio-economic water use, water storage, and other factors, the measured runoff of many sections can no longer truly reflect the natural water resources of the basin (Zhang et al. 2021a; Elahi et al. 2022; Abbas et al. 2023). In order to achieve rational allocation and efficient utilization of regional water resources, it is necessary to reconstruct the natural runoff series to find out the runoff volume in its natural state (Wang et al. 2022).

The traditional method for reconstructing natural runoff is the itemized survey method (Razavi & Coulibaly 2013). According to the water balance principle, for a closed watershed, the natural runoff from a control section is equal to the measured runoff from that control section plus the amount of each reconstructed water above that control section (Ren et al. 2012). Therefore, on the basis of the measured runoff, the natural runoff is obtained by investigating items such as water consumption by each water-using sector, cross-basin diversions, and changes in reservoir storage and reconstructing them item by item. Although the itemized survey method is easy to understand, the collection of survey information brings about a huge workload, requires a lot of human and material resources, and the accuracy of the survey results is difficult to guarantee, which is defective in practical application. Take the research on the Jinsha River Basin (Feng et al. 2024), Yalong River Basin (Wang et al. 2022), Tongtian River Basin (Liu et al. 2024), and Yangtze River Basin (Zhang et al. 2021a) in China as an example. Since it involves numerous water-using departments, the data collection channels for each department are complex, and the statistical standards vary, making it difficult to ensure data accuracy. At the same time, it is extremely difficult to obtain long-time-series data. Some historical data are missing or have vague records, resulting in deviations in the restoration of past water resource conditions. When dealing with complex river basins, the workload of this method increases exponentially, with high costs and low efficiency, which seriously restricts its wide application in practice.

With the gradual deepening of the study of hydrological cycle mechanisms, natural runoff reconstruction methods based on rainfall–runoff relationships have emerged. These methods use rainfall to calculate natural runoff based on the relationship between rainfall and runoff under approximately the same surface conditions. Sajikumar & Thandaveswara (1999) modeled monthly rainfall–runoff relationships for watersheds with short data lengths using artificial neural network techniques commonly used to deal with nonlinear problems. Ozelkan & Duckstein (2001) solved the problem of parameter uncertainty in rainfall–runoff modeling by developing a fuzzy rainfall–runoff model. Whigham & Crapper (2001) introduced the genetic algorithm into a rainfall–runoff model for natural runoff reconstruction. Methods based on the rainfall–runoff relationship have been widely used in natural runoff reconstruction, but there are shortcomings such as incomplete consideration of influencing factors, and the calculation results are not easy to infer. For example, Zhang et al. (2018) merely considered the relationship between rainfall and runoff in a simplistic way, neglecting the crucial impact of topography and geomorphology on the formation of runoff and the confluence process. In mountainous river basins, where the terrain undulates significantly, the velocity and direction of water flow are remarkably controlled by the terrain. There are huge differences in runoff generation and concentration in areas with different slopes and aspects, yet such methods cannot accurately reflect these differences. Additionally, the impacts of soil types and vegetation cover on water infiltration and evapotranspiration have not been fully considered, making it difficult for the calculation results to accurately reflect the actual natural runoff conditions.

In order to fully consider the physical mechanisms in the hydrologic cycle process in natural runoff reconstruction, hydrologic modeling-based natural runoff reconstruction methods have been developed. The hydrologic modeling-based methods have the advantages of strong physical mechanisms, few constraints, and ease of use (Tran et al. 2022). Hydrologic modeling is a product of the combination of hydrology science and computer science, which is a physical or conceptual structure constructed according to the formation process of rainfall–runoff in a basin to meet the principle of water balance (Clark et al. 2015). The hydrologic models in common use today can be divided into two groups: lumped models and distributed models. Lumped models simulate the runoff formation process of a basin as a whole. They are also called conceptual models because they are formed by combining conceptual elements according to the runoff formation process. Famous lumped models include the TANK model in Japan, the Sacramento model in the United States, and the Xinanjiang model in China. Distributed models divide the basin into multiple spatial units, each with different inputs and outputs. This can reflect the heterogeneous spatial distribution of runoff formation conditions in watersheds, making the simulation results more realistic and accurate. The study of distributed hydrologic modeling originated from the original concept and framework proposed by Freeze & Harlan (1969). Since then, scholars have worked on a variety of distributed hydrologic models. Abbott et al. (1986) proposed the System Hydrologic European model, which synthesizes numerous hydrologic processes and is the first representative distributed model. In 1994, the Soil and Water Assessment Tool (SWAT) was jointly developed by the USDA Agricultural Research Service and Texas A&M AgriLife Research. The SWAT has a strong physical mechanism for simulating hydrologic processes in complex watersheds with a wide range of soil, land use, and management conditions (Sarrazin et al. 2016). Various distributed hydrologic models play an important role in revealing the relationship between runoff formation and topography, geomorphology, soil, vegetation, hydrogeology, and land use. The SWAT model, as a small watershed to river basin-scale model over long-time periods, has a wide range of applications around the world, with natural runoff reconstruction being a major area of application (Wang et al. 2023). Bouraoui et al. (2005) used SWAT to simulate daily and monthly runoff in the Medjerda Basin, and the results showed that the accuracy of the simulated runoff is high and the natural runoff reconstruction is reliable.

In this study, natural runoff reconstruction is based on the SWAT model. After collecting digital elevation model (DEM) data information, land use information, soil spatial distribution information, and meteorological information, the river basin-scale SWAT model is established. The river basin is divided into sub-basins (watersheds), and the runoff of the watershed is reconstructed to its natural state after calibrating the model parameters based on hydrologic data. The traditional approach to calibrating model parameters is to adjust the basin-wide parameters to make the simulated runoff close to the early measured runoff, reasoning that the early measured runoff is less influenced by human activities and can be approximated as natural runoff (Zhang et al. 2021b). However, this parameter calibration method has two notable drawbacks. First, there are challenges in data collection. The period when runoff is least affected by human activities often dates back to a distant past. As the time span extends, the impact of human activities diminishes. Nevertheless, this also means that acquiring measured runoff data becomes increasingly arduous due to issues like data scarcity over long time periods, limited data preservation, and potential inaccuracies in historical records. Second, the spatial variability within the basin is overlooked. Employing a uniform parameter calibration across the entire basin and assigning identical parameter values to each sub-basin fails to account for the diverse influences of different surface conditions on runoff. This approach contradicts the fundamental principle of the distributed hydrologic model, which aims to represent the spatial heterogeneity of hydrological processes (Xu et al. 2023). This study proposes an innovative approach to overcome the shortcomings. A case study is conducted using real-world data from the Upper Yangtze River Basin (UYRB) in China. The natural runoff reconstruction method proposed in this research differs greatly from traditional ones. Integrating advanced data-driven algorithms and considering multiple environmental variables allows for a more accurate analysis of the natural runoff process. Validating this method in the complex and data-rich UYRB has two main aims. First, it significantly improves the accuracy and reliability of natural runoff reconstruction, providing a more precise quantification of the natural hydrological cycle. Second, it aims to create a generalized model that can be easily applied to other basins, overcoming the limitations of a single-study area. The findings from this research are expected to be crucial for developing more scientific water resource management strategies. Given the changing water availability due to climate change and land-use change, these strategies are designed to enhance the adaptability of water resource management systems. They aim to ensure sustainable water use and protect the ecological balance of river basins, thus contributing to the long-term stability of aquatic ecosystems.

The establishment of river basin-scale SWAT requires generating rivers and sub-basins based on DEM data, then generating hydrologic response units based on the spatial distribution of land use types, soil types, and slope types, and finally inputting meteorological data to run the model (Xu et al. 2023; Lotfirad et al. 2025).

The SWAT model requires the support of a variety of data from different sources and in different forms, which need to be processed and transformed before being added to the model. A detailed description of the various types of basic data is shown in Table 1.

Based on the basin DEM and human-set watershed thresholds, the SWAT generates streams and a number of watershed inlets and outlets. The total outlet of the basin is determined after adding additional inlets and outlets artificially according to the actual needs. Sub-basins (watersheds) are divided based on hydraulic connections and topographic parameters, geometric parameters, and flow paths, are analyzed and calculated for each sub-basin (watershed).

After delineating the sub-basins (watersheds), each sub-basin is further delineated into hydrological response units (HRUs). HRUs are areas within a sub-basin that have the same land-use type, soil type, and slope type (within a certain threshold), and they are the smallest computational units in the model run. Loading spatially distributed land use, soil, and slope datasets into the SWAT model generates a number of HRUs based on their combination and distribution.

Various types of soil data and meteorological data are input into the model, and input files, such as basin input file (.bsn), sub-basin input file (.sub), HRU input file (.hru), management input file (.mgt), soil input file (.sol), and groundwater input file (.gw), are created in the model database, and then the model can be run. The results produced by the model run mainly include the summary information of each HRU, each sub-basin, and each calculated river segment in the basin, which are stored in the HRU output file (output.hru), sub-basin output file (output.sub), and main river channel output file (output.rch), respectively. The results of the model run are analyzed and calculated to produce the natural runoff for each part of the watershed.

Model parameter calibration is the process of adjusting model parameter values so that model calculations match measured data. Any model that mathematically represents a simulated process cannot be based entirely on the physical substance of the process, and therefore the accuracy and reliability of the model are limited. Parameter calibration is a critical step before applying the model to reveal and compensate for deficiencies in model design and implementation and is useful when real parameter values are not available or are difficult to obtain. Typically, calibrated model parameters can only be used to simulate hydrologic processes. After calibration of the model parameters is completed, validation with data other than the calibration needs to be applied to assess the applicability of the model parameters.

The SWAT model has a large number of parameters, 28 of which are related to runoff. In order to improve the efficiency of parameter calibration, after performing parameter sensitivity analysis, parameters with less impact on model results are excluded, and parameters with higher sensitivity are selected for calibration. The parameter sensitivity analysis is performed using the Latin-hypercube one-at-a-time method (Wang et al. 2023).

The traditional approach to calibrating model parameters is to adjust the basin-wide parameters to make the simulated runoff close to the early measured runoff, reasoning that the early measured runoff is less influenced by human activities and can be approximated as natural runoff. However, this parameter calibration method has two shortcomings. First, it is difficult to collect information: the period in which runoff is minimally affected by human activities generally goes back a very long time, and the longer the time, the smaller the impact of human activities, but the more difficult it is to obtain information on measured runoff. Second, the spatial differences of different locations within the basin are not considered: the uniform parameter calibration for the whole basin and the same parameter values for each sub-basin (watershed) do not reflect the influence of different surface conditions on runoff, which is contrary to the original intention of the distributed hydrologic model.

In order to overcome these two shortcomings, a new idea is proposed in this study. Several near-natural sub-basins (watersheds) in the watershed that are less affected by human activities are selected for parameter calibration, and the measured runoff of each near-natural sub-basin (watershed) can be approximated as its natural runoff. In this way, the parameter calibration can be completed by collecting the measured runoff data of these near-natural sub-basins in recent decades, which avoids the difficulty of collecting hydrological data of old age. At the same time, calibrating the parameters for different near-natural sub-basins can fully take into account the spatial variability of the parameters and provide corresponding parameter sets for sub-basins with different surface conditions.

Therefore, M near-natural sub-basins (watersheds) are selected as representative sub-basins before the parameter calibration. The selection is based on the extent to which runoff from each sub-basin is affected by human activities. Generally, sub-basins in the mountainous areas in the upper part of the basin can be selected as representative sub-basins, where the population density is very low and human activities are minimal.

A Differential Evolution Adaptive Metropolis (DREAM)-based parameter calibration algorithm proposed by Wang et al. (2023) is applied in this study. This algorithm takes into account the multifactorial uncertainty of input data, model parameters, measured data, and is superior in model parameter estimation. Based on Bayesian inference, a set of parameters is randomly generated by sampling from the prior distribution of the parameters, from which the SWAT model is set up and simulated, and the difference between the simulated and measured values is compared to prefer the parameter ranges. The algorithm reflects the uncertainty in the calibrated parameter ranges, which is visualized by plotting a 95% prediction uncertainty graph. The 95 Percent Prediction Uncertainty (PPU) interval after parameter calibration contains most of the measured data. Both sides of the cumulative distribution of the output variables at the 2.5 and 97.5% levels are eliminated as the 5% with highly unsatisfactory simulation results. The uncertainty range of the parameters is continuously reduced through multiple iterations so that the simulated values are constantly close to the measured values.

The coefficients of determination (⁠⁠) and Nash–Sutcliffe efficiencies (⁠⁠) are selected in this study to evaluate the applicability of SWAT model parameters (Moriasi et al. 2007).

After completing the parameter calibration for the M representative sub-basins, the M sets of parameters need to be spread out to the full basin scale. To account for spatial differences across the basin, parameter spreading is based on similarities between sub-basins.

We assume that the whole basin is divided into N sub-basins and analyzed the characteristics of the N–M non-representative sub-basins one by one. For the Kth non-representative sub-basin, by comparing its similarity with M representative sub-basins, the representative sub-basin with the highest similarity is selected, and the parameter set corresponding to this representative sub-basin is spread to the Kth non-representative sub-basin. The above steps are implemented for each non-representative sub-basin to complete the basin-wide parameter spread.

The spatial differences considered in the model parameters spread are mainly in the effect of different surface conditions on runoff. Therefore, in this study, indicators reflecting surface conditions, including land use type and soil type, are selected for the analysis of sub-basin characteristics.

The direct characterization indicator for each sub-basin is the composition of the different land use types and soil types, i.e., the percentage of area occupied by each land use type and the percentage of area occupied by each soil type. In order to consistently apply land use and soil indicators for the integrated evaluation of sub-basin characteristics, the indicators are normalized for each characteristic.

The percentage of area occupied by each land use type and the percentage of area occupied by each soil type in each sub-basin are used as sample data. These are normalized by substituting them into Equation (1) to obtain uniformly comparable dimensionless values between 0 and 1 to quantitatively characterize the N sub-basins in the basin.

After clarifying the characteristics of each sub-basin, the non-representative sub-basins are analyzed for similarity with each representative sub-basin, and parameters are spread according to the degree of similarity. The measure of similarity between sub-basins is to calculate the ‘distance’ between them.

For the Kth non-representative sub-basin, the Euclidean distances between it and the M representative sub-basins are compared to find the closest representative sub-basin. This means that this representative sub-basin has the highest degree of similarity with the Kth non-representative sub-basin, so the parameter set calibrated for this representative sub-basin is spread to the Kth non-representative sub-basin. The above steps are implemented for each non-representative sub-basin to complete the basin-wide parameter spread.

In this paper, a basin-scale SWAT model was developed for the UYRB as an example, and the proposed hydrological parameter spreading method was applied to reconstruct the natural runoff in the basin.

Meteorological data from 80 meteorological stations in the UYRB were selected for this modeling, and the location of each station is shown in Figure 3. Meteorological data include daily precipitation, average/maximum/minimum temperature, wind speed, humidity, and radiation data for the period 1951–2013.

The land use data and soil-type data were loaded. Land use, soil, and slope types were superimposed, appropriate thresholds were set, and different combinations of their distributions were obtained, from which HRUs were generated. A total of 702 HRUs were defined in 99 sub-basins of the UYRB for this modeling, and land use, soil and slope distribution reports and HRU distribution reports, were generated.

The prearranged meteorological data were input into the model, and various model database files were created to run the model. The Soil Conservation Service (SCS) runoff curve method was chosen to simulate surface runoff, the Hargreaves method was chosen to simulate evapotranspiration, and the Muskingum method was chosen to perform the river channel algorithm.

Based on the results of parameter sensitivity analyses, the top 10 parameters with high sensitivity were selected: ALPHA_BF, CANMX, EPCO, ESCO, GW_DELAY, GW_REVAP, GWQMN, CN2, SOL_AWC. The exact description of the parameters can be found in Zhang et al. (2021b). Wang et al. (2023) and Xu et al. (2023) showed that the runoff in the UYRB changed significantly after 1993 due to the influence of human activities. In order to minimize the error of natural runoff reconstruction caused by anthropogenic disturbance of runoff in the near-natural sub-basins, the measured monthly runoff series from 1960 to 1980 in sub-basins #1, #6, #17, #22, #33, #44, #45, #95, and #99 were selected as the parameter calibration data, and those from 1981 to 1993 were used as the parameter validation data in this study.

The use of Euclidean distance as a similarity metric allows for a quantitative assessment of the relationships between sub-basins. A smaller Euclidean distance implies a higher degree of similarity in terms of land-use type shares and soil-type shares, which are key surface-condition indicators. This means that sub-basins with similar surface characteristics are grouped together. For example, sub-basins within the same set in Figure 8 likely have comparable impacts on runoff due to their similar land use and soil compositions. This helps in identifying sub-basin clusters that may respond in a similar way to precipitation and other hydrological factors. Conversely, larger Euclidean distances highlight the differences between sub-basins. These differences can be attributed to variations in land-use patterns (such as differences between urban, agricultural, and forested areas) and soil types (which affect water infiltration and storage capacities). By quantifying these differences, we can better understand how specific surface-condition factors influence runoff generation and distribution across the basin. For instance, sub-basins with a high proportion of urban land-use (impervious surfaces) are likely to have different runoff characteristics compared to those dominated by forested areas, and the similarity metrics help in precisely identifying such disparities.

In the context of natural runoff reconstruction in the UYRB, similarity-based parameter spreading is a fundamental approach. Since the model parameters are calibrated for representative sub-basins, the ability to extend these parameters to non-representative sub-basins based on similarity is essential. By using the parameter sets of the most similar representative sub-basins for non-representative ones, we can more accurately simulate the runoff processes in different parts of the basin. This method takes into account the spatial heterogeneity of the basin, which is critical for natural runoff reconstruction. The surface-condition indicators, as reflected by the land-use and soil-type shares, are closely related to the hydrological behavior of sub-basins. For example, areas with different soil types will have different water-holding capacities and infiltration rates, which directly affect the amount and timing of runoff. By grouping sub-basins based on similarity and applying appropriate parameter sets, we can improve the accuracy of runoff simulations, thereby providing more reliable data for understanding the natural runoff regime in the UYRB. Moreover, these similarity metrics and the associated parameter – spreading method – can also assist in predicting the impact of future land-use changes or soil-related modifications on natural runoff. If there are planned changes in land use (such as urban expansion or reforestation projects) in a particular sub-basin, we can refer to the similar sub-basins and their runoff responses to estimate the potential changes in the target sub-basin's runoff. This has important implications for water resource management, flood control, and ecological protection in the UYRB.

The calibrated parameter sets were input into the model to simulate the runoff of the UYRB from 1960 to 2013 on a monthly scale. The average annual runoff depths for the 99 sub-basins are shown in Table 2. The average annual runoff of the UYRB was calculated to be 438.18 km³.

Upon further analysis, it was found that there are significant spatial variations in the average annual runoff among different sub-basins. For instance, sub-basin 67 has the highest runoff of 267.93 km³, while sub-basin 16 has the lowest runoff of only 0.28 km³. Such large differences can be attributed to multiple factors. Climatic factors play a crucial role, with sub-basins in areas with higher precipitation generally having higher runoff. For example, sub-basins located in the southeastern part of the UYRB, which are influenced by the monsoon climate and receive abundant rainfall, tend to have larger runoff volumes.

In addition, underlying surface conditions also contribute significantly to the runoff differences. Sub-basins with more forest cover and better-preserved natural landscapes usually have a more stable runoff, as vegetation can enhance water infiltration and storage, reducing surface runoff and maintaining a relatively constant flow. On the contrary, sub-basins with intensive human activities, such as large-scale urbanization or industrial development, often show abnormal runoff patterns. The construction of impervious surfaces in urban areas reduces water infiltration, leading to increased surface runoff during rainfall events.

Surface water resources constitute the main body of total water resources and can be characterized by natural river runoff. Their evolution is influenced by climatic conditions and underlying surface conditions, while the differences between measured runoff and natural runoff can reflect the impact of human water extraction. Based on the temporal and spatial evolution patterns of the obtained natural runoff and measured runoff, as well as their causes, it is possible to further identify changes in regional water resources against the backdrop of climate change and human activities. The study of the temporal and spatial evolution analysis methods of natural runoff in watersheds holds significant scientific importance and practical value for the rational planning and utilization of water resources.

In the context of climate change and land use change, accurately reconstructing the natural runoff in the Upper Yangtze River is of great significance. The average annual natural runoff reconstructed in this study for the Upper Yangtze River is 438.18 km³, while the measured runoff is 428.91 km³ (Wang et al. 2023). The difference between the two clearly demonstrates the impact of human activities on runoff.

Climate change has led to alterations in precipitation patterns, with an increasing frequency of extreme precipitation events. In the UYRB, changes in the precipitation pattern directly affect the total runoff volume and its distribution within a year. Accurately understanding natural runoff can provide a scientific basis for coping with water resource fluctuations caused by climate change. For example, advance planning of water resource storage and allocation strategies can help store water rationally during the wet season and ensure water security during the dry season, effectively alleviating the problem of uneven spatial and temporal distribution of water resources.

Land use change also significantly impacts runoff. The rapid urbanization process in the UYRB has led to the expansion of urban land, resulting in hardened underlying surfaces. This causes surface runoff to converge rapidly, reducing infiltration and evapotranspiration and altering the runoff formation mechanism. Large-scale agricultural development has changed the soil structure and vegetation cover, affecting the natural water cycle. By reconstructing natural runoff, it is possible to quantify the impact of these land use changes on runoff and thus develop targeted ecological protection and restoration strategies. For instance, increasing green infrastructure in urban areas can restore the hydrological regulation function of cities, and optimizing irrigation methods in agricultural areas can reduce interference with natural runoff.

The reconstruction of natural runoff provides crucial basic data for water resource planning. Based on the reconstructed natural runoff data, more scientific water resource allocation plans can be developed to ensure the rational allocation of domestic, industrial, and ecological water use, and to achieve the sustainable utilization of water resources. At the same time, it is also an important indicator for evaluating the health of the ecosystem. It helps to maintain the stability and balance of the ecosystem in the UYRB, protect biodiversity, and promote the sustainable development of the basin.

In the case study, nine sub-basins with low population and construction densities and minimal human activities were selected as near-natural sub-basins Besides the degree of human activity influence, factors like topographic and geomorphic features, vegetation and ecosystems, hydrological characteristics, and soil properties can all potentially serve as criteria for selecting natural sub-basins. In the future, it is possible to further study the evaluation index system for assessing whether a sub-basin is in a natural or near-natural state to make the selection criteria more scientific and reasonable.

This study investigates the spatial spreading of SWAT model parameter sets. It uses the similarity of sub-basins based on land use and soil types as a key reference. Land use types, including forest, agricultural, and urban areas, significantly impact runoff generation. For example, forested areas have more infiltration and less surface runoff compared to urban regions. Similarly, different soil types vary in water-holding capacity and permeability, which also affect hydrological processes. By analyzing these factors, we can group sub-basins with similar characteristics and spread the parameter sets accordingly. However, this approach might not cover all aspects influencing parameter distribution. Future research could further explore whether there are other more scientifically grounded bases for the distribution of model parameters. This could involve studying the role of topography at a more detailed level, like micro-relief features. These factors could provide a more comprehensive understanding and potentially lead to more accurate parameter distribution methods in the SWAT model.

The reconstruction of natural runoff in a river basin serves as the cornerstone of scientific water resources management, playing a crucial role in maintaining ecological balance and ensuring the sustainable development of the social economy. Focusing on the shortcomings of existing methods, this study proposes a novel natural runoff reconstruction method based on the spreading of hydrological parameters in similar watersheds. In-depth case analyses are carried out using the actual data of the UYRB, yielding a series of valuable results:

Innovative parameter calibration approach: Breaking through the limitations of traditional parameter calibration methods, this study innovatively selects near-natural sub-basins within the river basin, which are less affected by human activities and can better reflect the natural runoff conditions, for parameter calibration. In the research of the UYRB, nine such sub-basins were identified, and multiple sets of accurate model parameter sets were successfully obtained. This strategy effectively addresses the data acquisition difficulties in traditional methods and fully considers the spatial variability of parameters, laying a solid foundation for accurate runoff process simulation.

Establishment of parameter spreading system: Land use types and soil types are used as key indicators to quantify the characteristics of sub-basins. By scientifically calculating the Euclidean distance, the similarity between non-representative sub-basins and near-natural sub-basins is accurately analyzed. Based on the similarity, the parameter sets of near-natural sub-basins are reasonably spread to the entire basin. This method fully takes into account the impact of underlying surface conditions on runoff, enabling the model parameters to better adapt to the actual situations of different sub-basins and significantly improving the accuracy and reliability of model simulation.

High-precision runoff reconstruction: With the help of the established SWAT model at the basin scale of the UYRB and the innovative parameter calibration and spreading methods, the high-precision reconstruction of the basin's natural runoff is successfully achieved. The study reveals significant spatial differences in the average annual runoff of the 99 sub-basins in the basin, which are mainly caused by different climate conditions and underlying surface characteristics. By comparing the reconstructed average annual natural runoff (438.18 km³) with the measured runoff (428.91 km³), the impact of human activities on runoff is clearly demonstrated. In addition, the method proposed in this study provides new ideas and methods for the construction of runoff series in ungauged areas, playing an important role in the scientific planning and efficient management of water resources, as well as the protection and sustainable development of the river basin ecosystem.

Conceptualized by L.L., Y.Z., Y.M.; developed the methodology by L.L., X.C., R.C.; rendered support in formal analysis and investigated by L.L., X.C., R.C.; wrote the original draft preparation by L.L., Y.Z.; wrote and reviewed and edited by L.L., X.C., R.C., X.Z., Y.Z., Y.M.; rendered support in funding acquisition by X.Z., Y.Z.; contributed in resources R.C., X.Z.; supervised by Y.M.

This work is supported by the National Natural Science Foundation of China (Grant No. 52209032) and the China Three Gorges Corporation Project (Grant No. 0704226).

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

The authors declare there is no conflict.