Attribution Analysis of Runoff Change in a Changing Environment: A Case Study of the Dawen River Basin

Abstract

Surface runoff change is significantly influenced by both human activities and climate change. Decoupling their respective contributions to runoff change represents a critical frontier in hydrological research and a pressing challenge for water resource management. This study focuses on the Dawen River Basin, a strategic area for ecological conservation and high-quality development in the lower Yellow River region. By integrating three methodological approaches—empirical models (Precipitation–Runoff Double Mass Curve), conceptual models (elasticity coefficient methods), and hydrological models (Soil and Water Assessment Tool, SWAT)—we systematically quantify the impacts of climate change and human activities on runoff change. A correlation analysis was first applied to screen independent runoff drivers and basin characteristic factors, followed by a random forest algorithm to rank their relative importance. This process informed the establishment of a comprehensive framework for runoff attribution analysis. Results demonstrate that hydrological modeling (SWAT) is the most appropriate method for the Dawen River Basin, revealing human activities as the dominant driver of runoff changes, accounting for 70% to 82%. These findings provide critical insights for guiding sustainable water resource planning and management in anthropogenically stressed basins under a changing environment.

1. Introduction

Since the 1950s, accelerated human socio-economic development has triggered significant global climate changes, primarily characterized by rising temperatures. In many regions, alterations in ice/snow melt and precipitation patterns have profoundly impacted water resources in terms of both quantity and quality [1,2]. Beyond climate change, human activities such as irrigation, afforestation, deforestation, and urbanization have further modified hydrological processes. For instance, afforestation reduces runoff [3]; urbanization increases surface runoff [4], shortens the lag time between precipitation and runoff (accelerating concentration times) [5], and amplifies peak flows [6]; and land use/cover changes significantly alter infiltration and evapotranspiration dynamics [7]. These interactions amplify hydrological extremes—including floods and droughts—generating unprecedented cascading risks to ecosystem integrity and regional socio-economic resilience [8,9,10,11,12,13,14,15,16]. Human interventions—including reservoir construction, urban water management, agricultural expansion, and urbanization—reshape hydrological processes, alter spatiotemporal water resource distributions, and interact in a complex manner with climate regulation [17,18,19]. The U.S. Geological Survey (USGS) recently updated its water cycle diagram after two decades, explicitly incorporating human activities [20]. This revision underscores the growing recognition of anthropogenic influences in hydrological research. Decoupling the contributions of human activities and climate change to runoff change has become a pivotal research topic [21,22]. As a critical component of the water cycle, runoff evolution profoundly influences water resource utilization across agriculture, industry, hydropower, and navigation. Researchers employ long-term runoff data to quantify variability and attribute total runoff changes to climate and human drivers within specific periods [23].

Most studies on quantifying the relative contributions of climate change and human activities to runoff variation focus on Chinese basins. For example, Xia et al. (2014) integrated Budyko hypothesis-based water–energy balance equations with the Penman-Monteith equation, revealing that human activities accounted for 87.4–89.5% of runoff reduction, substantially exceeding climate change’s 10.5–12.6% contribution [24]. In the Miyun Reservoir Basin, Ma et al. (2010) applied distributed hydrological and climate elasticity models, identifying climate change as responsible for 51–55% of runoff variation [25]. Zheng et al. (2009) used modified climate elasticity estimators in the Yellow River source region, attributing > 70% of runoff changes to LUCC and <30% to climate factors [26]. Wang et al. (2013) analyzed four sub-basins in the Haihe River Basin using hydrological sensitivity analysis, models, and elasticity methods, concluding that human activities dominated runoff reduction in three sub-basins, while climate change prevailed in the fourth [27]. International studies also highlight regional variability: Poelmans et al. (2011) linked 30% of runoff changes in central Belgium to climate via rainfall–runoff models [28]; D’Agostino et al. (2010) projected a 16–23% decline in runoff by 2050 in southern Italy’s Canale d’Aiedo Basin using hydrological models [29]; and Saidi et al. (2018) attributed 85% of alpine river runoff reduction in northwestern Italy to climate change using sensitivity and elasticity methods [30]. These divergent outcomes underscore the challenge of disentangling climate–human interactions, necessitating localized investigations. Current research predominantly relies on single-method approaches (e.g., hydrological model) to assess relative impacts, while multi-method evaluations remain rare. Few studies systematically analyze methodological applicability across basins, particularly for immediate human drivers (e.g., water abstraction, reservoir operations) whose impacts vary regionally. A standardized framework for method selection based on basin-specific characteristics is urgently needed.

The term “changing environment” in this study refers to both climate change and human activities. Mounting evidence confirms their synergistic effects on hydrological cycles [31,32,33,34,35,36], though spatial heterogeneity in these impacts reflects geographic and socioeconomic disparities [37]. Human activities fall into three categories: LUCC, reservoir operations, and water abstraction. Unsustainable practices—such as irrational land use, unregulated water abstraction, and poorly managed hydraulic infrastructure—exacerbate soil erosion, land degradation, water scarcity, and flood risks. Thus, comprehending the anthropogenic impacts on runoff is central to sustainable watershed management and water resource planning.

2. Materials and Methods

2.1. Study Area

The Dawen River, the largest tributary in the lower reaches of the Yellow River and one of China’s rare east-to-west flowing rivers, primarily traverses Laiwu City and Tai’an City in Shandong Province (Figure 1). With a total length of 231 km and a drainage area of 8944 km², its upper reaches flow through mountainous and hilly terrain, while the middle-lower reaches form a plain before eventually converging into Dongping Lake. Dongping Lake, the sole crucial flood detention basin in the Yellow River Basin, functions as a key regulation hub for the South-to-North Water Diversion Project’s eastern route and the source of Shandong’s Jiaodong Water Transfer Project. The natural evolution of the Dawen River Basin has been profoundly altered by human activities including agricultural expansion, hydraulic engineering, and urbanization. With a land utilization rate approaching saturation, agriculture dominates the basin, encompassing wheat, corn, orchards, vegetable fields, and cash crops like peanuts and cotton. Since the 1950s, large-scale water conservancy projects have significantly altered the natural hydrological rhythms. By 2019, the basin accommodated 23 large and medium-sized reservoirs (mostly upstream) and over 100 small ones. These structures reduced downstream flood volumes by approximately 60%, drastically lowered sediment loads, diminished natural sediment transport capacity, and triggered elevated riverbed phenomena in certain reaches. The Dongping Lake Reservoir, constructed in 1958, plays a dual role in regulating floods from both the Dawen River and the Yellow River, accounting for 60% of the annual regulated runoff. In Tai’an’s water supply system, large-scale projects such as reservoirs, sluice gates, and pumping stations contribute 83.9% (872 million m³) of total supply, underscoring heavy reliance on surface water [38]. While modern irrigation infrastructure enhances efficiency, it exacerbates upstream–downstream conflicts—for instance, reduced ecological flow in plains due to upstream reservoir interception. The famed “Wenyang Farmland”, renowned for irrigation agriculture since antiquity, traces its water management legacy to the Yuan Dynasty’s water diversion projects and the Ming–Qing period’s “storage-regulation integration” philosophy embodied in the Daicunba. These historical strategies continue to influence contemporary flood control paradigms. The Dawen River Basin stands as an ideal region for studying runoff evolution under changing environments, offering insights into the interplay between anthropogenic interventions and hydrological dynamics.

2.2. Data

• DEM Data

The digital elevation model (DEM) data for the Dawen River Basin were obtained from the China Geospatial Data Cloud platform, with a 30 m spatial resolution in GeoTIFF format. The raw elevation data underwent standardized preprocessing by using ERDAS IMAGINE 2023 (Hexagon Geospatial, Switzerland), including coordinate system unification and outlier removal, before being converted to the hierarchical data format (HDF5) for subsequent hydrological modeling.

• Hydrometeorological Data

The Dawen River Basin runoff data are monthly runoff series covering the period from 1961 to 2019, obtained from four hydrological stations, Laiwu, Beiwang, Dawenkou and Daicunba. These data were acquired from the Shandong Provincial Hydrology Center. The meteorological dataset consists of daily observations spanning the years 1961 to 2019 from 16 meteorological stations within and around the Dawen River Basin, including precipitation, temperature (mean/maximum/minimum), wind speed at 2 m elevation, sunshine duration, and relative humidity. These data were obtained from the National Meteorological Information Center of the China Meteorological Administration (CNMA) and from the Climate Data from the Climate Information Center of the National Meteorological Information Center of the China Meteorological Administration (CMA). The data are primarily utilized for the analysis of the influencing factors of runoff change characteristics and model-driven raw data input. The dataset serves a dual purpose: ① diagnostic analysis of hydroclimatic drivers influencing water resource variability, and ② providing high-temporal-resolution forcing inputs for hydrological models.

• Soil Data

Soil data for the Dawen River Basin were obtained from the Harmonized World Soil Database (HWSD; https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v12 (accessed on 29 March 2025)), a global 1:1,000,000-scale spatial dataset, constructed by the Food and Agriculture Organization of the United Nations (FAO) and the International Institute for Applied Systems Analysis (IIASA). The soil type map of the Dawen River Basin was cropped using ArcGIS, with classification according to the FAO-90 system.

• Land Use Cover Data

The multi-temporal land use raster dataset (30 m spatial resolution) for the study watershed was obtained from the Resource and Environmental Science and Data Center (RESDC) (https://www.resdc.cn (accessed on 5 April 2024)) of the Chinese Academy of Sciences, covering seven epochs (1980, 1990, 2000, 2005, 2010, 2015, and 2018). This dataset originates from the 1:100,000 National Land Use/Cover Database of China, developed through systematic interpretation of Landsat TM/OLI imagery combined with ground-truth validation and historical geographic archives.

• Reservoir data

Due to data availability constraints, three reservoirs were selected for analysis: Guangming Reservoir (large-scale) and Huangqian and Dongzhou Reservoirs (medium-sized). The hydrological dataset of reservoirs in the Dawen River Basin primarily comprises fundamental parameters, monthly outflow records, operational data, and storage capacity of Dongzhou Reservoir, Guangming Reservoir, and Huangqian Reservoir. The fundamental parameters are from the compilation of three investigations and three determinations for large and medium-sized reservoirs in Shandong Province. The monthly outflow data of each reservoir was obtained from the Shandong Hydrological Center. The reservoir operation data and water storage volume were obtained from the reservoir management bureau, reflecting monthly water abstraction by users and reservoir regulation outcomes.

• Consumptive water

The consumptive water data from 1987 to 2019 were primarily sourced from the Shandong Province Water Resources Bulletin, Tai’an City Water Resources Bulletin, and Laiwu City Water Resources Bulletin. Data for the period 1961–1986 were derived from comprehensive data collection and investigative analysis. The dataset covers administrative divisions within the Dawen River Basin (Taishan District, Daiyue District, Xintai City, Feicheng City, Ningyang County, Dongping County, Pingyin County, and Laiwu City), categorized under two classification frameworks: ① Usage categories: Agricultural, industrial, domestic, and ecological–environmental water supplementation. ② Source classifications: Surface water and groundwater (subdivided into shallow and deep aquifers).

2.3. Methods

2.3.1. Runoff Change Attribution Methodology

(1)

Precipitation–Runoff Double Mass Curve

The Double Mass Curve (DMC) was initially derived to assess the consistency of meteorological data in the Susquehanna River Basin [39,40]. River runoff is predominantly influenced by both human activities and climate change. To isolate the contribution of human activities by eliminating climate-induced variations (with precipitation serving as the primary manifestation of climate change), the Precipitation–Runoff Double Mass Curve method has been widely adopted [41,42]. This method divides hydrological series into two distinct periods based on runoff abrupt points: the Natural Period, which is characterized by minimal anthropogenic interference, where runoff is primarily driven by climate variability; and the Human Activity Period, marked by significant anthropogenic impacts, during which runoff reflects combined effects of climate change and human activities. The DMC quantitatively evaluates these impacts by plotting cumulative precipitation (independent variable) against cumulative runoff (dependent variable). A stable linear relationship in the curve indicates negligible human influence on runoff, whereas a slope change or deviation from linearity signifies altered hydrological dynamics due to anthropogenic interventions. This approach enables the attribution of runoff variations to either climatic factors or human actions, providing critical insights for watershed management and ecological restoration. The specific steps are as follows:

• Establish Cumulative Precipitation–Runoff Linear Regression Equations

The hydrological series is partitioned into a baseline period and a human activity period based on runoff abrupt points. Linear regression equations are formulated for cumulative precipitation (independent variable) and cumulative runoff (dependent variable) before and after the abrupt point, expressed as follows:

Precipitation–Runoff Double Mass Curve (DMC):

$$\sum R_{orig}=a_1\sum P_{orig}+b_1$$

(1)

In the equation above, $R_{orig}$ and $P_{orig}$ denote measured runoff and precipitation in the natural period, respectively; a₁, b₁ represent constant terms.

A regression Equation (1) was established using the pre-abrupt-point-year double mass curve of measured precipitation–runoff relationships. The cumulative measured precipitation series from the post-abrupt-point period was then input into this equation to calculate the reconstructed theoretical cumulative runoff series. The difference between measured and theoretical runoff values during the human activity period was quantified as the contribution of human activities to runoff variation.

• Quantification of Runoff Change Attribution

The theoretical runoff values during the human activity period are calculated as follows:

$$\sum R_{var}^{\mathrm{R}\mathrm{e}c}=a_1\sum P_{var}+b_1$$

(2)

$$\begin{array}{c}\Delta R=\Delta R_C+\Delta R_H=R_{var}-R_{orig}\\ \Delta R_C=R_{var}^{\mathrm{R}\mathrm{e}c}-R_{orig}\\ \Delta R_H=\Delta R-\Delta R_C=R_{var}^{\mathrm{R}\mathrm{e}c}-R_{var}\\ {\zeta }_C={\displaystyle \frac{\Delta R_C}{\Delta R}}\times 100\%\\ {\zeta }_H=1-{\zeta }_C={\displaystyle \frac{\Delta R_H}{\Delta R}}\times 100\%\end{array}$$

(3)

In Equations (2) and (3), $R_{var}^{\mathrm{Re}c}$ and $R_{var}$ denote the reconstructed theoretical runoff and measured runoff during the human activity period, respectively; $\Delta R$ represents the total runoff variation induced by environmental changes; $\Delta R_C$, $\Delta R_H$ represents runoff variations attributed to climate change and human activities, respectively; $R_{orig}$ represents the measured runoff during the natural period (mm); ${\zeta }_C$, ${\zeta }_H$ represent the contribution rates (%) of climate change and human activities to runoff change, respectively.

• Evaluation Index

The coefficient of determination (R²) was employed to assess the performance of the Precipitation–Runoff Double Mass Curve. A threshold of R² > 0.6 indicates that the regression equation is representative, with values closer to 1 signifying stronger precipitation–runoff correlations.

(2)

Elasticity Coefficient Method

The elasticity coefficient method based on the Budyko framework was employed to disentangle the impacts of human activities and climate change on runoff change. This approach utilizes the water balance equation, with precipitation and potential evapotranspiration as climatic factors, to compute the sensitivity of runoff to climatic change and subsequently quantify their contributions.

Long-term water balance equation based on the Budyko hypothesis:

$$R=P-E-\Delta Q$$

(4)

In Equation (4), R denotes the runoff; P denotes the precipitation; E denotes the actual evapotranspiration; and $\Delta Q$ represents the watershed water storage change, assumed to be negligible over long-term periods.

Runoff can be expressed as a function of climatic factors (P, E₀) and watershed characteristics (V):

$$R=f\left(C,H\right)=f\left(P,E_0,V\right)$$

(5)

In Equation (5), C, H denote the contributions of climate change and human activities to runoff change; $E_0$ denotes the potential evapotranspiration (calculated via the Penman–Monteith equation); V represents the watershed characteristic parameter (integrating topography, land cover, and subsurface properties).

The climate-induced runoff change $\Delta R_C$ is defined as follows:

$$\Delta R_C=f_P^{\prime }\Delta P+f_{E_0}^{\prime }\Delta E_0={\displaystyle \frac{\partial R}{\partial P}}\Delta P+{\displaystyle \frac{\partial R}{\partial E_0}}\Delta E_0$$

(6)

Runoff Elasticity to Climatic Factors $\epsilon$:

$$\epsilon ={\displaystyle \frac{\partial R}{R}}/{\displaystyle \frac{\partial X}{X}}$$

(7)

In Equation (7), X represents climatic factors (P, E₀). Rewriting Equation (6) using elasticity coefficients gives the following:

$$\Delta R_C=\Delta R_P+\Delta R_{E_0}=\left({\epsilon }_P{\displaystyle \frac{\Delta P}{P}}+{\epsilon }_{E_0}{\displaystyle \frac{\Delta E_0}{E_0}}\right)R$$

(8)

The ratio of actual evapotranspiration to precipitation (E/P) is expressed as a function of the aridity index (Budyko hypothesis integration):

$$F\left(\phi \right)={\displaystyle \frac{E}{P}}$$

(9)

Combining Equations (4) and (9), the elasticity coefficients for P and E₀ are derived:

$$\begin{array}{c}{\epsilon }_P=1+{\displaystyle \frac{\phi F^{\prime }\left(\phi \right)}{1-F\left(\phi \right)}}\\ {\epsilon }_{E_0}={\displaystyle \frac{\phi F^{\prime }\left(\phi \right)}{1-F\left(\phi \right)}}\\ {\epsilon }_P+{\epsilon }_{E_0}=1\end{array}$$

(10)

The relative contribution of climate change and human activities to runoff change are calculated as follows:

$$\begin{array}{c}{\zeta }_C={\displaystyle \frac{\Delta R_C}{\Delta R}}\times 100\%\\ {\zeta }_H=1-{\zeta }_C\end{array}$$

(11)

The method hinges on determining the $F\left(\phi \right)$. This study adopted seven Budyko-type formulations (

Table 1. Function used in the Budyko hypothesis.

Function	Description	Instructions
Schreiber	$F(\phi )=1-e^{-\phi }$	No parameters
Ol’dekop	$F(\phi )=\phi tanh({\displaystyle \frac{1}{\phi }})$
Budyko	$F(\phi )={\left[\phi tanh({\displaystyle \frac{1}{\phi }})(1-e^{-\phi })\right]}^{1/2}$
Tirc–Pike	$F(\phi )={\displaystyle \frac{1}{\sqrt{1+{\phi }^{-2}}}}$
Fu	$F(\phi )=1+\phi -{\left[1+{\phi }^m\right]}^{1/m}$	One parameter
Zhang et al.	$F(\phi )={\displaystyle \frac{1+w\phi }{1+w\phi +{\phi }^{-1}}}$
Choudhury–Yang	$E={\displaystyle \frac{P{E}_0}{{\left(P^w+E_0^w\right)}^{1/w}}}$

) to quantify the contributions of climate change and human activities to runoff change. The first four methods are parameter-free, while the latter three are parameterized. Multiple methods from each category were selected for cross-comparison to validate the results.

(3)

Hydrological model

The Dawen River Basin was modeled using the Soil & Water Assessment Tool (SWAT), with an input database comprising DEM, soil data, meteorological data, LUCC data, reservoir data, and consumptive water use data within the basin. The modeling time scale was divided into two phases based on an abrupt-point year (mutated year): the natural period (pre-abrupt-point phase, human activity is minimal and the period is assumed to be natural) and the human activity period. Two models were constructed: the natural model S1 for the natural period and the abrupt-period model S2 (with scenarios A1, A2, A3, and A4) for the human activity period (

Table 2. Temporal division of hydrological model construction in Dawen River Basin.

Model		Period	Warm-Up Period	Validation Period	Calibration Period
Natural Period S1		1960~1976	1961~1962	1963~1969	1970~1977
Human Activity Period S2	A1	1977~1990	1977~1979	1980~1985	1986~1990
	A2	1991~2000	1989~1990	1991~1995	1996~2000
	A3	2001~2010	1999~2000	2001~2005	2006~2010
	A4	2011~2019	2009~2010	2011~2015	2016~2019

). Full model construction specifics are detailed in Appendix B.1. SWAT-CUP is a program for calibration of SWAT models, employed to perform bidirectional calibration and validation of model parameters across different periods. Time-partitioned modelling enhances performance, with detail results provided in Appendix B.2.

We then input the hydrometeorological data corresponding to the human activity period (post-abrupt-point years) into Model S1 to reconstruct the natural river runoff data not affected by human activities. The simulation scenarios are configured as detailed in

Table 3. Scenario-based quantitative separation of climate change and human activities.

Scenario Name	Hydrometeorological Data	Reservoir and Consumptive Water Use Data	LUCC Data
Natural Period S1	1961–1976	1961–1976	1980
Human Activity Period S2	1977–2019	1961–1976	1980
Human Activity Period S3	1977–1990	1961–1976	1980
Human Activity Period S4	1991–2000	1961–1976	1980
Human Activity Period S5	2001–2010	1961–1976	1980
Human Activity Period S6	2011–2019	1961–1976	1980

.

The difference in simulated runoff value between the pre- and post-abrupt-point years (i.e., the natural period and human activity period) is attributed to climate change. The discrepancy between the observed runoff data and reconstructed runoff data during the human activities period is attributed to human activities. Based on the above scenario settings, a quantitative separation of climate change and human activity impacts is performed.

Contribution of climate change to runoff change:

$$\begin{array}{c}\Delta R=\Delta R_C+\Delta R_H=R_{var}-R_{orig}\\ \Delta R_C=R_{var,s_i}^{\mathrm{R}\mathrm{e}c}-R_{orig}\\ {\zeta }_C={\displaystyle \frac{\Delta R_C}{\Delta R}}\times 100\%\end{array}$$

(12)

Contribution of human activities to runoff change:

$$\begin{array}{c}\Delta R_H=\Delta R-\Delta R_C=R_{var,s_i}^{\mathrm{R}\mathrm{e}c}-R_{var,s_i}\\ {\zeta }_H=1-{\zeta }_C={\displaystyle \frac{\Delta R_H}{\Delta R}}\times 100\%\end{array}$$

(13)

In the equation above, $R_{\mathit{var},s_i}^{\mathrm{Re}c}$ denotes reconstructed runoff during the human activity period S_i, i = 1,2,3,4.

This study utilizes hydrological models to quantitatively separate the contributions of human activities and climate change to runoff change. Building on this foundation, it further separates the amount of contribution of each human activity to runoff change, which are LUCC, reservoir operation, consumptive water, and other anthropogenic factors. The human activity period (1977–2019) is divided into four periods: 1977–1990, 1991–2000, 2001–2010, and 2011–2019. Corresponding baseline models (A1, A2, A3, and A4) are established for each period. By designing appropriate scenarios based on these baseline models, the impacts of LUCC, reservoir operations, and consumptive water use activities on runoff change can be separated.

Initially, the baseline model of four periods was established, with the input data comprising the corresponding hydrometeorological data, land use/cover data, reservoir operation data, and consumptive water use data of each era. Subsequently, the baseline model of each era was utilized as the benchmark, and the consumptive water and reservoir operation were sequentially stripped off, with the model being run. The specific separation of each human activity scenario setting is shown in

Table 4. Separate scenario settings of various human activities in each era.

Scenario	Period	LUCC	Reservoir Operation	Consumptive Water Activities
Baseline A1	1977–1990	1990	√ 1	√
A1,1			√	× 2
A1,2			×	×
A1,3	1977–1990	1980	×	×
Baseline A2	1991–2000	2000	√	√
A2,1			√	×
A2,2			×	×
A2,3	1991–2000	1980	×	×
Baseline A3	2001–2010	2010	√	√
A3,1			√	×
A3,2			×	×
A3,3	2001–2010	1980	×	×
Baseline A4	2011–2019	2018	√	√
A4,1			√	×
A4,2			×	×
A4,3	2011–2019	1980	×	×

Notes: ¹ √ indicates that data are available. ² × indicates that data are not available.

. The impact of each human activity component is calculated as follows:

$$\begin{array}{c}\Delta R_{user}=R\left(A_{m,1}\right)-R\left(A_m\right)\\ \Delta R_{res}=R\left(A_{m,2}\right)-R\left(A_{m,1}\right)\\ \Delta R_{LUCC}=R\left(A_{m,3}\right)-R\left(A_{m,2}\right)\\ \Delta R_{Other}=\Delta R_H-\Delta R_{user}-\Delta R_{res}-\Delta R_{LUCC}\end{array}$$

(14)

where $R\left(A_{m,i}\right)$ represent the simulated runoff volume for multiple scenarios corresponding to each era using A1–A4 as the baseline model; $\Delta R_{user}$ represents the impact of consumptive water use activities on the runoff change; $\Delta R_{res}$ indicates the impact of reservoir operation on the runoff change; and $\Delta R_{LUCC}$ represents the impact of land use/cover on the runoff change. $\Delta R_{Other}$ denotes the impact of other human activities on the runoff change.

Calculation of the contribution of each human activity to runoff change:

$$\begin{array}{c}{\zeta }_{user}={\displaystyle \frac{\Delta R_{user}}{\Delta R_H}}\times {\zeta }_H\times 100\%\\ {\zeta }_{res}={\displaystyle \frac{\Delta R_{res}}{\Delta R_H}}\times {\zeta }_H\times 100\%\\ {\zeta }_{LUCC}={\displaystyle \frac{\Delta R_{LUCC}}{\Delta R_H}}\times {\zeta }_H\times 100\%\\ {\zeta }_{Other}={\displaystyle \frac{\Delta R_{Other}}{\Delta R_H}}\times {\zeta }_H\times 100\%\end{array}$$

(15)

where ${\zeta }_{user}$, ${\zeta }_{res}$, ${\zeta }_{LUCC}$ and ${\zeta }_{Other}$ represent the contribution of consumptive water activities, reservoir operation, land use/cover change, and other human activities, respectively, to runoff change.

2.3.2. Framework for Adaptability Assessment of Runoff Change Attribution Methods

Runoff generation arises from complex hydrological processes influenced by both driving factors and basin-specific attributes. Existing runoff attribution methods diverge in their selection of driving factors and basin characteristics, often prioritizing limited parameters. Evaluating parameter significance thus serves as a critical metric for assessing the applicability of these methods across heterogeneous basins. The random forest algorithm, which partitions data into random subsets to identify optimal feature variables, exhibits superior integrative performance and is widely adopted for quantifying the relative importance of drivers (e.g., runoff, water quality).

This paper employs three types of approaches, namely the empirical statistical method (exemplified by the Precipitation–Runoff Double Mass Curve), the conceptual model method (represented by the elasticity coefficient), and the hydrological model method (represented by SWAT), to quantitatively separate whether the dominant factor influencing runoff change is climate change or human activities. To evaluate their basin-specific applicability, a framework is established based on parametric significance (Figure 2). First, a comprehensive set of runoff drivers and basin characteristics is selected. Pearson correlation and wavelet coherence analyses are applied to filter statistically independent variables. Subsequently, random forest is utilized to rank the relative importance of factors in explaining runoff change, thereby systematically assessing the suitability of each attribution method within the study basin.

3. Results and Discussion

3.1. Delineation of Baseline and Change Periods

The representative indicators for climate change were selected as precipitation and temperature. Time series of climate indices (including 27 extreme climate indices calculated using RClimDex, based on daily-scale mean precipitation, mean temperature, daily maximum and minimum temperatures) were subjected to abrupt change detection using multiple statistical methods (Mann–Kendall test, Pettitt test, cumulative anomaly analysis, and Regime Shift Detection; see the referenced literature [43] for details). The results indicated that the abrupt change in precipitation-related indices predominantly occurred during the mid-to-late 1970s, while the abrupt change in temperature-related indices clustered in the late 1980s. In the Dawen River Basin, most large-to-medium reservoirs were constructed between the 1960s and 1970s, with expansions or retrofits completed in the 1970s. Regional water consumption stabilized after the 1990s, driven by economic development and policy adjustments. Runoff data from four hydrological stations (Laiwu, Beiwang, Dawenkou, and Daicunba) along the upper to lower reaches exhibited abrupt transitions between 1975 and 1977 [44]. The S1 model simulations generated 1961–2019 runoff series for all sub-basins (Figure 3). Observed and simulated runoff exhibited strong consistency before the 1976 change point, while simulations systematically overestimated observed values post-1976. Synthesizing these findings, 1976 was identified as the abrupt year. Consequently, the baseline period was defined as 1961–1976, and the change period as 1977–2019.

3.2. Precipitation–Runoff Double Mass Curve

This study constructed double mass curves using observed precipitation and runoff data (1961–2019) from the Dawen River Basin. Four sub-basins—Laiwu, Beiwang, Dawenkou, and Daicun Dam—spanning the upper to lower reaches were selected. Area-weighted average precipitation within each station’s sub-basin was calculated via the Thiessen polygon method, and station-specific Precipitation–Runoff Double Mass Curves were established. The study period was divided into a baseline period (S1: 1961–1976) and a human activity period (S2: 1977–2019), with S2 further segmented into four sub-periods: A1 (1977–1990), A2 (1991–2000), A3 (2001–2010), and A4 (2011–2019).

As shown in Figure 4, regression equations for the baseline period exhibited high explanatory power (R² = 0.92–0.98), confirming robust linear relationships between cumulative precipitation and runoff. Quantitative separation of climate change and human activities impacts using the double mass curve method (

Table A1. Climate–human runoff attribution: Double Mass Curve analysis.

Sub-Basin	Period		Observed Precipitation (mm)		Observed Runoff (mm)		Reconstructed Runoff (mm)	${\mathit{\zeta }}_{\mathit{C}}$	${\mathit{\zeta }}_{\mathit{H}}$
			Mean	Change	Mean	Change
Laiwu	Baseline Period	1961–1976	748.66	/	293.98	/	/	/	/
	Change Periods	1977–1990	619.53	−129.13	174.17	−119.81	263.92	25%	75%
		1991–2000	675.79	−72.87	180.64	−113.34	266.33	24%	76%
		2001–2010	728.47	−20.19	226.53	−67.45	287.09	10%	90%
		2011–2019	658.72	−89.94	153.36	−140.62	259.60	24%	76%
		1977–2019	666.15	−82.51	183.49	−110.49	268.96	23%	77%
Beiwang	Baseline Period	1961–1976	702.49	/	248.39	/	/	/	/
	Change Periods	1977–1990	586.88	−115.61	119.21	−129.19	207.69	32%	68%
		1991–2000	640.77	−61.72	142.57	−105.83	221.13	26%	74%
		2001–2010	676.67	−25.82	208.96	−39.44	233.52	38%	62%
		2011–2019	663.73	−38.76	130.39	−118.00	229.05	16%	84%
		1977–2019	643.79	−58.70	153.28	−95.11	227.32	22%	78%
Dawenkou	Baseline Period	1961–1976	722.15	/	225.80	/	/	/	/
	Change Periods	1977–1990	609.75	−112.41	114.50	−111.30	189.31	33%	67%
		1991–2000	647.56	−74.59	145.82	−79.99	187.66	48%	52%
		2001–2010	703.69	−18.46	212.94	−12.86	203.93	170%	−70%
		2011–2019	659.72	−62.44	129.29	−96.51	191.19	36%	64%
		1977–2019	651.94	−70.21	147.65	−78.16	190.77	45%	55%
Daicunba	Baseline Period	1961–1976	724.97	/	174.44	/	/	/	/
	Change Periods	1977–1990	616.27	−108.70	66.07	−108.38	151.08	22%	78%
		1991–2000	653.64	−71.34	84.95	−89.49	145.17	33%	67%
		2001–2010	722.18	−2.80	141.71	−32.73	160.40	43%	57%
		2011–2019	642.42	−82.55	67.44	−107.00	142.68	30%	70%
		1977–2019	655.06	−69.91	88.34	−86.10	150.12	28%	72%

, Figure 5) revealed consistent runoff reductions across all sub-basins during S2 compared to S1. Human activities accounted for 77%, 78%, 53%, and 72% of runoff change (1977–2019) at the four sub-basins, respectively, underscoring their dominant role in the basin. Notably, the A3 period (2001–2010) exhibited anomalous contributions, likely due to extensive retrofitting of water conservancy projects—particularly in the Dawenkou sub-basin, where human activities contributed −70% to runoff changes, reflecting heightened engineering interventions. These results demonstrate that human activities are the primary driver of runoff change in the Dawen River Basin, highlighting significant human-induced hydrological modifications.

3.3. Elasticity Coefficient Method

This study utilized observed precipitation and runoff data (1961–2019) along with potential evapotranspiration (ET0) calculated via the Penman–Monteith formula in the Dawen River Basin. Seven Budyko hypothesis-based elasticity coefficient methods were employed to quantitatively assess the impacts of climate change and human activities on runoff change (

Table A2. Climate–human runoff attribution: Elasticity Coefficient Method.

Method	Sub-Basin	Period	Parameter	ɛP	ɛE0	∆RP/mm	∆RE0/mm	∆RC/mm	∆RH/mm	∆R	${\mathit{\zeta }}_{\mathit{C}}$	${\mathit{\zeta }}_{\mathit{H}}$
Schreiber	Laiwu	1977–1990	/	2.80	−1.80	−142.03	22.95	−119.08	−0.73	−119.81	99%	1%
		1991–2000	/	2.64	−1.64	−75.41	25.19	−50.22	−63.12	−113.34	44%	56%
		2001–2010	/	2.51	−1.51	−19.87	26.39	6.53	−73.98	−67.45	−10%	110%
		2011–2019	/	2.72	−1.72	−96.05	14.52	−81.53	−59.09	−140.62	58%	42%
		1977–2019	/	2.67	−1.67	−86.51	22.67	−63.84	−46.65	−110.49	58%	42%
	Beiwang	1977–1990	/	2.90	−1.90	−110.54	24.31	−86.23	−43.98	−130.21	66%	34%
		1991–2000	/	2.81	−1.81	−74.34	22.81	−51.53	−72.00	−123.53	42%	58%
		2001–2010	/	2.71	−1.71	−35.21	23.97	−11.25	−45.89	−57.14	20%	80%
		2011–2019	/	2.74	−1.74	−48.92	24.17	−24.75	−110.95	−135.70	18%	82%
		1977–2019	/	2.80	−1.80	−70.83	23.85	−46.98	−65.83	−112.81	42%	58%
	Dawenkou	1977–1990	/	2.86	−1.86	−100.69	17.76	−82.93	−28.37	−111.30	75%	25%
		1991–2000	/	2.75	−1.75	−64.26	16.96	−47.30	−32.69	−79.99	59%	41%
		2001–2010	/	2.60	−1.60	−15.03	18.09	3.06	−15.92	−12.86	−24%	124%
		2011–2019	/	2.73	−1.73	−53.34	14.60	−38.75	−57.76	−96.51	40%	60%
		1977–2019	/	2.75	−1.75	−61.21	16.99	−44.22	−33.81	−78.03	57%	43%
	Daicunba	1977–1990	/	2.83	−1.83	−75.60	15.46	−60.14	−51.90	−112.05	54%	46%
		1991–2000	/	2.70	−1.70	−47.35	18.50	−28.85	−64.31	−93.16	31%	69%
		2001–2010	/	2.54	−1.54	−1.74	17.47	15.73	−52.13	−36.40	−43%	143%
		2011–2019	/	2.73	−1.73	−55.30	19.53	−35.77	−74.90	−110.67	32%	68%
		1977–2019	/	2.70	−1.70	−46.44	17.53	−28.90	−60.87	−89.77	32%	68%
Ol’dekop	Laiwu	1977–1990	/	2.78	−1.78	−141.01	22.69	−118.32	−1.49	−119.81	99%	1%
		1991–2000	/	2.74	−1.74	−78.42	26.81	−51.61	−61.73	−113.34	46%	54%
		2001–2010	/	2.70	−1.70	−21.42	29.82	8.41	−75.86	−67.45	−12%	112%
		2011–2019	/	2.76	−1.76	−97.56	14.88	−82.68	−57.95	−140.62	59%	41%
		1977–2019	/	2.75	−1.75	−89.09	23.75	−65.34	−45.15	−110.49	59%	41%
	Beiwang	1977–1990	/	2.80	−1.80	−106.68	23.01	−83.66	−46.54	−130.21	64%	36%
		1991–2000	/	2.78	−1.78	−73.59	22.45	−51.14	−72.39	−123.53	41%	59%
		2001–2010	/	2.76	−1.76	−35.91	24.71	−11.19	−45.95	−57.14	20%	80%
		2011–2019	/	2.77	−1.77	−49.40	24.54	−24.86	−110.84	−135.70	18%	82%
		1977–2019	/	2.78	−1.78	−70.37	23.61	−46.76	−66.05	−112.81	41%	59%
	Dawenkou	1977–1990	/	2.79	−1.79	−98.21	17.09	−81.12	−30.18	−111.30	73%	27%
		1991–2000	/	2.77	−1.77	−64.62	17.11	−47.52	−32.47	−79.99	59%	41%
		2001–2010	/	2.73	−1.73	−15.77	19.54	3.77	−16.63	−12.86	−29%	129%
		2011–2019	/	2.77	−1.77	−53.99	14.88	−39.11	−57.40	−96.51	41%	59%
		1977–2019	/	2.77	−1.77	−61.73	17.22	−44.51	−33.52	−78.03	57%	43%
	Daicunba	1977–1990	/	2.79	−1.79	−74.44	15.09	−59.34	−52.70	−112.05	53%	47%
		1991–2000	/	2.76	−1.76	−48.34	19.11	−29.23	−63.93	−93.16	31%	69%
		2001–2010	/	2.71	−1.71	−1.86	19.47	17.61	−54.00	−36.40	−48%	148%
		2011–2019	/	2.76	−1.76	−56.06	19.96	−36.10	−74.57	−110.67	33%	67%
		1977–2019	/	2.76	−1.76	−47.38	18.10	−29.28	−60.49	−89.77	33%	67%
Budyko	Laiwu	1977–1990	/	2.81	−1.81	−142.32	23.02	−119.30	−0.52	−119.81	100%	0%
		1991–2000	/	2.68	−1.68	−76.83	25.95	−50.88	−62.47	−113.34	45%	55%
		2001–2010	/	2.59	−1.59	−20.51	27.81	7.30	−74.76	−67.45	−11%	111%
		2011–2019	/	2.75	−1.75	−97.03	14.76	−82.27	−58.35	−140.62	59%	41%
		1977–2019	/	2.71	−1.71	−87.82	23.22	−64.60	−45.89	−110.49	58%	42%
	Beiwang	1977–1990	/	2.88	−1.88	−109.65	24.01	−85.64	−44.57	−130.21	66%	34%
		1991–2000	/	2.81	−1.81	−74.42	22.85	−51.57	−71.96	−123.53	42%	58%
		2001–2010	/	2.74	−1.74	−35.62	24.41	−11.21	−45.92	−57.14	20%	80%
		2011–2019	/	2.76	−1.76	−49.32	24.47	−24.84	−110.86	−135.70	18%	82%
		1977–2019	/	2.80	−1.80	−70.99	23.94	−47.06	−65.76	−112.81	42%	58%
	Dawenkou	1977–1990	/	2.85	−1.85	−100.26	17.65	−82.61	−28.69	−111.30	74%	26%
		1991–2000	/	2.77	−1.77	−64.68	17.13	−47.55	−32.44	−79.99	59%	41%
		2001–2010	/	2.66	−1.66	−15.36	18.74	3.38	−16.24	−12.86	−26%	126%
		2011–2019	/	2.76	−1.76	−53.82	14.80	−39.02	−57.49	−96.51	40%	60%
		1977–2019	/	2.77	−1.77	−61.68	17.20	−44.48	−33.55	−78.03	57%	43%
	Daicunba	1977–1990	/	2.83	−1.83	−75.53	15.44	−60.09	−51.95	−112.05	54%	46%
		1991–2000	/	2.73	−1.73	−47.92	18.85	−29.07	−64.09	−93.16	31%	69%
		2001–2010	/	2.61	−1.61	−1.79	18.31	16.52	−52.92	−36.40	−45%	145%
		2011–2019	/	2.75	−1.75	−55.82	19.83	−36.00	−74.67	−110.67	33%	67%
		1977–2019	/	2.74	−1.74	−46.99	17.86	−29.12	−60.65	−89.77	32%	68%
Turc–Pike	Laiwu	1977–1990	/	2.64	−1.64	−133.79	20.88	−112.91	−6.90	−119.81	94%	6%
		1991–2000	/	2.58	−1.58	−73.86	24.35	−49.50	−63.84	−113.34	44%	56%
		2001–2010	/	2.53	−1.53	−20.04	26.77	6.73	−74.19	−67.45	−10%	110%
		2011–2019	/	2.61	−1.61	−92.24	13.61	−78.63	−61.99	−140.62	56%	44%
		1977–2019	/	2.59	−1.59	−84.04	21.64	−62.41	−48.08	−110.49	56%	44%
	Beiwang	1977–1990	/	2.67	−1.67	−101.62	21.32	−80.31	−49.90	−130.21	62%	38%
		1991–2000	/	2.64	−1.64	−69.85	20.67	−49.18	−74.35	−123.53	40%	60%
		2001–2010	/	2.61	−1.61	−33.93	22.58	−11.35	−45.79	−57.14	20%	80%
		2011–2019	/	2.62	−1.62	−46.75	22.48	−24.27	−111.43	−135.70	18%	82%
		1977–2019	/	2.64	−1.64	−66.76	21.72	−45.04	−67.77	−112.81	40%	60%
	Dawenkou	1977–1990	/	2.66	−1.66	−93.42	15.79	−77.63	−33.67	−111.30	70%	30%
		1991–2000	/	2.62	−1.62	−61.20	15.69	−45.51	−34.48	−79.99	57%	43%
		2001–2010	/	2.57	−1.57	−14.83	17.70	2.87	−15.73	−12.86	−22%	122%
		2011–2019	/	2.62	−1.62	−51.08	13.62	−37.46	−59.05	−96.51	39%	61%
		1977–2019	/	2.62	−1.62	−58.43	15.78	−42.65	−35.38	−78.03	55%	45%
	Daicunba	1977–1990	/	2.65	−1.65	−70.71	13.92	−56.80	−55.25	−112.05	51%	49%
		1991–2000	/	2.61	−1.61	−45.66	17.45	−28.21	−64.95	−93.16	30%	70%
		2001–2010	/	2.54	−1.54	−1.75	17.53	15.78	−52.18	−36.40	−43%	143%
		2011–2019	/	2.61	−1.61	−53.02	18.26	−34.76	−75.91	−110.67	31%	69%
		1977–2019	/	2.61	−1.61	−44.76	16.53	−28.23	−61.54	−89.77	31%	69%
Fu	Laiwu	1977–1990	1.93	1.80	−0.80	−91.50	10.25	−81.25	−38.56	−119.81	68%	32%
		1991–2000	2.05	1.90	−0.90	−54.40	13.88	−40.52	−72.82	−113.34	36%	64%
		2001–2010	1.97	1.80	−0.80	−14.31	14.10	−0.20	−67.25	−67.45	0%	100%
		2011–2019	2.14	2.00	−1.00	−70.57	8.43	−62.14	−78.48	−140.62	44%	56%
		1977–2019	2.00	1.86	−0.86	−60.31	11.69	−48.61	−61.87	−110.49	44%	56%
	Beiwang	1977–1990	2.09	1.97	−0.97	−75.06	12.41	−62.65	−67.55	−130.21	48%	52%
		1991–2000	2.14	2.00	−1.00	−53.00	12.65	−40.35	−83.18	−123.53	33%	67%
		2001–2010	1.88	1.75	−0.75	−22.78	10.55	−12.24	−44.90	−57.14	21%	79%
		2011–2019	2.30	2.15	−1.15	−38.42	16.00	−22.42	−113.28	−135.70	17%	83%
		1977–2019	2.08	1.95	−0.95	−49.32	12.58	−36.74	−76.07	−112.81	33%	67%
	Dawenkou	1977–1990	2.27	2.14	−1.14	−75.15	10.84	−64.31	−46.99	−111.30	58%	42%
		1991–2000	2.16	2.02	−1.02	−47.01	9.81	−37.20	−42.79	−79.99	47%	53%
		2001–2010	1.94	1.80	−0.80	−10.38	9.00	−1.38	−11.48	−12.86	11%	89%
		2011–2019	2.31	2.16	−1.16	−42.13	9.76	−32.38	−64.14	−96.51	34%	66%
		1977–2019	2.15	2.01	−1.01	−44.85	9.85	−35.00	−43.03	−78.03	45%	55%
	Daicunba	1977–1990	2.86	2.72	−1.72	−72.59	14.51	−58.08	−53.97	−112.05	52%	48%
		1991–2000	2.79	2.61	−1.61	−45.78	17.52	−28.26	−64.90	−93.16	30%	70%
		2001–2010	2.47	2.27	−1.27	−1.56	14.45	12.89	−49.29	−36.40	−35%	135%
		2011–2019	3.00	2.83	−1.83	−57.48	20.75	−36.73	−73.94	−110.67	33%	67%
		1977–2019	2.74	2.57	−1.57	−44.12	16.15	−27.98	−61.80	−89.77	31%	69%
Zhang et al.	Laiwu	1977–1990	0.23	1.93	−0.93	−97.93	11.87	−86.07	−33.75	−119.81	72%	28%
		1991–2000	0.41	2.03	−1.03	−58.04	15.84	−42.20	−71.14	−113.34	37%	63%
		2001–2010	0.31	1.91	−0.91	−15.14	15.95	0.81	−68.26	−67.45	−1%	101%
		2011–2019	0.53	2.13	−1.13	−75.37	9.58	−65.79	−74.83	−140.62	47%	53%
		1977–2019	0.34	1.99	−0.99	−64.45	13.43	−51.02	−59.47	−110.49	46%	54%
	Beiwang	1977–1990	0.44	2.13	−1.13	−81.11	14.43	−66.67	−63.54	−130.21	51%	49%
		1991–2000	0.51	2.15	−1.15	−56.89	14.50	−42.39	−81.14	−123.53	34%	66%
		2001–2010	0.18	1.86	−0.86	−24.15	12.02	−12.13	−45.01	−57.14	21%	79%
		2011–2019	0.78	2.27	−1.27	−40.44	17.57	−22.87	−112.83	−135.70	17%	83%
		1977–2019	0.43	2.10	−1.10	−53.04	14.53	−38.51	−74.30	−112.81	34%	66%
	Dawenkou	1977–1990	0.71	2.27	−1.27	−79.93	12.14	−67.79	−43.51	−111.30	61%	39%
		1991–2000	0.55	2.15	−1.15	−50.25	11.15	−39.10	−40.89	−79.99	49%	51%
		2001–2010	0.27	1.91	−0.91	−11.03	10.27	−0.76	−12.10	−12.86	6%	94%
		2011–2019	0.79	2.27	−1.27	−44.29	10.69	−33.60	−62.91	−96.51	35%	65%
		1977–2019	0.54	2.15	−1.15	−47.93	11.19	−36.74	−41.29	−78.03	47%	53%
	Daicunba	1977–1990	1.94	2.59	−1.59	−69.15	13.42	−55.73	−56.32	−112.05	50%	50%
		1991–2000	1.72	2.52	−1.52	−44.15	16.51	−27.64	−65.52	−93.16	30%	70%
		2001–2010	1.09	2.31	−1.31	−1.58	14.86	13.28	−49.68	−36.40	−36%	136%
		2011–2019	2.28	2.61	−1.61	−52.91	18.20	−34.71	−75.96	−110.67	31%	69%
		1977–2019	1.62	2.50	−1.50	−42.95	15.44	−27.50	−62.27	−89.77	31%	69%
Choudhury–Yang	Laiwu	1977–1990	1.21	1.84	−0.84	−93.39	10.73	−82.66	−37.15	−119.81	69%	31%
		1991–2000	1.34	1.93	−0.93	−55.35	14.39	−40.96	−72.38	−113.34	36%	64%
		2001–2010	1.26	1.83	−0.83	−14.50	14.53	0.03	−67.48	−67.45	0%	100%
		2011–2019	1.43	2.04	−1.04	−72.02	8.78	−63.24	−77.38	−140.62	45%	55%
		1977–2019	1.29	1.90	−0.90	−61.40	12.15	−49.25	−61.24	−110.49	45%	55%
	Beiwang	1977–1990	1.37	2.02	−1.02	−76.92	13.03	−63.89	−66.32	−130.21	49%	51%
		1991–2000	1.42	2.05	−1.05	−54.21	13.23	−40.99	−82.54	−123.53	33%	67%
		2001–2010	1.16	1.78	−0.78	−23.19	10.99	−12.20	−44.93	−57.14	21%	79%
		2011–2019	1.59	2.20	−1.20	−39.27	16.66	−22.61	−113.09	−135.70	17%	83%
		1977–2019	1.36	1.99	−0.99	−50.40	13.15	−37.26	−75.56	−112.81	33%	67%
	Dawenkou	1977–1990	1.55	2.19	−1.19	−77.05	11.36	−65.69	−45.61	−111.30	59%	41%
		1991–2000	1.44	2.06	−1.06	−48.02	10.23	−37.79	−42.19	−79.99	47%	53%
		2001–2010	1.23	1.83	−0.83	−10.55	9.33	−1.22	−11.64	−12.86	9%	91%
		2011–2019	1.60	2.21	−1.21	−43.06	10.16	−32.90	−63.61	−96.51	34%	66%
		1977–2019	1.44	2.05	−1.05	−45.80	10.26	−35.54	−42.49	−78.03	46%	54%
	Daicunba	1977–1990	2.14	2.79	−1.79	−74.59	15.14	−59.45	−52.60	−112.05	53%	47%
		1991–2000	2.07	2.67	−1.67	−46.86	18.19	−28.67	−64.49	−93.16	31%	69%
		2001–2010	1.76	2.31	−1.31	−1.59	14.89	13.31	−49.71	−36.40	−37%	137%
		2011–2019	2.28	2.91	−1.91	−58.92	21.55	−37.37	−73.30	−110.67	34%	66%
		1977–2019	2.02	2.63	−1.63	−45.16	16.77	−28.39	−61.38	−89.77	32%	68%

, Figure 6). Non-parametric elasticity methods revealed that a 1% decrease in precipitation would reduce runoff by 2.51–2.90%, while a 1% increase in potential evapotranspiration would decrease runoff by 1.51–1.90%. Parametric elasticity methods indicated that a 1% precipitation reduction led to runoff declines of 1.75–2.91%, whereas a 1% potential evapotranspiration increase resulted in runoff reductions of 0.75–1.91%.

For the Laiwu sub-basin, climate change dominated runoff change during the full study period (S2, approximately 58%). During period A1, climate change accounted for >94% of runoff changes using non-parametric methods and approximately 70% via parametric methods. Human activities became predominant in A2 (non-parametric: approximately 55%; parametric: approximately 64%) and overwhelmingly dominated A3. Divergent results emerged in A4, with non-parametric methods implicating climate change while parametric methods identified human activities as primary. In the Beiwang sub-basin, human activities dominated S2 runoff changes (58–67%). All seven methods demonstrated increasing human activities contribution from A1 (34–52%) to A4 (82–83%). For the downstream sub-basins (Dawenkou and Daicunba), the influence of human activities progressively intensified from A1 to A2 and A4, while completely dominating A3. In S2, non-parametric methods attributed Dawenkou’s runoff changes primarily to climate change (57%) versus parametric methods emphasizing the impact of human activities (54%). Daicunba’s S2 runoff changes were predominantly human-driven (approximately 68%).

Collectively, human activities constituted the primary driver of runoff change in the Dawen River Basin. While climate change dominated during A1, anthropogenic factors prevailed in A2–A4, with escalating influence of human activities over time. Non-parametric methods generally yielded higher climate change contributions than parametric approaches across identical spatiotemporal scales, reflecting parametric methods’ enhanced incorporation of underlying surface conditions.

3.4. Hydrological Model

(1)

Separation of Climate Change and Human Activity Contributions to Runoff Changes

The quantitative contributions of climate change and human activities to runoff changes during the change period are summarized in

Table A3. Quantitative separation of climate change and human activities results by the SWAT model.

Sub-Basin	Scenario Setup	Observed Runoff (mm)		Simulated Runoff (mm)	Climate Change		Human Activities
		Mean	Change		Change (mm)	Contribution	Change (mm)	Contribution
Laiwu	1961–1976 S1	293.13	/	310.49	/	/	/	/
	1977–2019 S2	183.49	−127.00	272.20	−38.29	30%	−88.71	70%
	1977–1990 S3	174.17	−136.32	235.19	−75.30	55%	−61.02	45%
	1991–2000 S4	180.64	−129.85	280.87	−29.62	23%	−100.23	77%
	2001–2010 S5	226.53	−83.96	313.77	3.28	−4%	−87.24	104%
	2011–2019 S6	153.36	−157.13	273.96	−36.53	23%	−120.60	77%
Beiwang	1961–1976 S1	356.77	/	394.05	/	/	/	/
	1977–2019 S2	217.15	−176.90	356.68	−37.37	21%	−139.52	79%
	1977–1990 S3	192.50	−201.54	317.56	−76.49	38%	−125.05	62%
	1991–2000 S4	201.97	−192.08	357.70	−36.34	19%	−155.74	81%
	2001–2010 S5	296.02	−98.03	389.24	−4.81	5%	−93.22	95%
	2011–2019 S6	184.72	−209.32	380.20	−13.85	7%	−195.48	93%
Dawenkou	1961–1976 S1	304.30	/	356.16	/	/	/	/
	1977–2019 S2	209.35	−146.81	320.44	−35.71	24%	−111.10	76%
	1977–1990 S3	162.21	−193.95	268.54	−87.61	45%	−106.33	55%
	1991–2000 S4	206.58	−149.58	318.42	−37.73	25%	−111.85	75%
	2001–2010 S5	301.67	−54.49	381.14	24.99	−46%	−79.47	146%
	2011–2019 S6	183.16	−172.99	335.97	−20.18	12%	−152.81	88%
Daicunba	1961–1976 S1	235.74	/	284.53	/	/	/	/
	1977–2019 S2	125.15	−159.39	255.46	−29.07	18%	−130.31	82%
	1977–1990 S3	93.59	−190.94	205.16	−79.38	42%	−111.56	58%
	1991–2000 S4	120.35	−164.18	257.50	−27.03	16%	−137.16	84%
	2001–2010 S5	200.76	−83.77	322.22	37.69	−45%	−121.46	145%
	2011–2019 S6	95.54	−188.99	257.26	−27.27	14%	−161.72	86%

and Figure 7. In the S2 period, human activities dominated runoff changes across all four sub-basins, accounting for 70–82%. Excluding the S5 period, the proportional impact of human activities progressively increased during the S3, S4, and S6 periods. Notably, runoff changes during S5 were almost entirely driven by human activities (contributing > 95%), attributed to intensified anthropogenic disturbances near the Dawenkou hydrological station, which is situated at the confluence of the southern tributary of the Dawen River Basin. These results collectively demonstrate a temporal escalation in the influence of human activities, particularly over the past two decades, during which their contributions consistently exceeded 77%.

(2)

Separation of Human Activity-Specific Contributions to Runoff Changes

As established, human activities dominate runoff changes in the Dawen River Basin. To quantify the contributions of specific anthropogenic drivers, scenario modeling (Table 4) was implemented to disentangle impacts from water abstraction and consumption, reservoir operation, LUCC, and other human activities, with results summarized in

Table A4. Quantitative separation of various human activities using the SWAT model.

Sub-Basin	Scenario Setup	Simulated Values (mm)	Contribution of Human Activities: Magnitudes and Rates										Climate Change (mm)
			Human Activities		Water Withdrawal and Use		Reservoir Operation		LUCC		Other Factors
			(mm)	(%)	(mm)	(%)	(mm)	(%)	(mm)	(%)	(mm)	(%)	(mm)	(%)
Laiwu	Baseline model A1	121.78	−61.02	45%	−70.54	52%	0.00	0%	66.75	−49%	−57.23	42%	−75.30	55%
	A1,1	192.32
	A1,2	192.32
	A1,3	125.57
	Baseline model A2	177.32	−100.23	77%	−116.50	90%	0.00	0%	102.83	−79%	−86.56	67%	−29.62	23%
	A2,1	293.83
	A2,2	293.83
	A2,3	191.00
	Baseline model A3	222.84	−87.24	104%	−197.71	235%	0.00	0%	162.49	−194%	−52.02	62%	3.28	-4%
	A3,1	420.55
	A3,2	420.55
	A3,3	258.06
	Baseline model A4	113.20	−120.60	77%	−78.90	50%	0.00	0%	25.46	−16%	−67.16	43%	−36.53	23%
	A4,1	192.10
	A4,2	192.10
	A4,3	166.64
Beiwang	Baseline model A1	115.44	−125.05	62%	−140.63	70%	0.00	0%	146.54	−73%	−130.96	65%	−76.49	38%
	A1,1	256.07
	A1,2	256.07
	A1,3	109.53
	Baseline model A2	139.22	−155.74	81%	−157.34	82%	0.28	0%	136.89	−71%	−135.56	71%	−36.34	19%
	A2,1	296.56
	A2,2	296.28
	A2,3	159.39
	Baseline model A3	207.60	−93.22	95%	−226.54	231%	6.09	−6%	224.87	−229%	−97.64	100%	−4.81	5%
	A3,1	434.13
	A3,2	428.05
	A3,3	203.18
	Baseline model A4	107.65	−195.48	93%	−97.74	47%	3.89	−2%	35.42	−17%	−137.04	65%	−13.85	7%
	A4,1	205.40
	A4,2	201.51
	A4,3	166.09
Dawenkou	Baseline model A1	102.80	−106.33	55%	−120.38	62%	3.45	−2%	104.03	−54%	−93.44	48%	−87.61	45%
	A1,1	223.17
	A1,2	219.72
	A1,3	115.69
	Baseline model A2	128.77	−111.85	75%	−121.89	81%	2.41	−2%	86.02	−58%	−78.40	52%	−37.73	25%
	A2,1	250.66
	A2,2	248.24
	A2,3	162.22
	Baseline model A3	215.13	−79.47	146%	−225.57	414%	22.94	−42%	180.76	−332%	−57.61	106%	24.99	−46%
	A3,1	440.69
	A3,2	417.75
	A3,3	236.99
	Baseline model A4	78.99	−152.81	88%	−102.75	59%	4.77	−3%	−0.77	0%	−54.06	31%	−20.18	12%
	A4,1	181.74
	A4,2	176.97
	A4,3	177.74
Daicunba	Baseline model A1	72.08	−111.56	58%	−117.41	61%	2.36	−1%	70.63	−37%	−67.15	35%	−79.38	42%
	A1,1	189.49
	A1,2	187.12
	A1,3	116.49
	Baseline model A2	102.14	−137.16	84%	−130.19	79%	1.66	−1%	61.02	−37%	−69.65	42%	−27.03	16%
	A2,1	232.33
	A2,2	230.67
	A2,3	169.65
	Baseline model A3	163.12	−121.46	145%	−228.24	272%	15.77	−19%	126.57	−151%	−35.56	42%	37.69	−45%
	A3,1	391.36
	A3,2	375.59
	A3,3	249.02
	Baseline model A4	41.27	−161.72	86%	−113.17	60%	3.25	−2%	−15.78	8%	−36.02	19%	−27.27	14%
	A4,1	154.45
	A4,2	151.19
	A4,3	166.97

and Figure 8. Among these drivers, water consumption activities and other human activities exerted positive contributions (amplifying runoff reduction), primarily driven by water consumption activities, while reservoir operation and LUCC showed negative contributions (mitigating runoff reduction), dominated by LUCC followed by reservoir operation. Temporally, all human activities drivers exhibited an initial increase followed by a decline in their contributions, peaking during 2001–2010 and reaching minimal levels in 2011–2019. Specifically, water abstraction contributed 52–70% (1977–1990), 79–90% (1991–2000), and 47–60% (2011–2019), but surged anomalously to 231–414% during 2001–2010. Reservoir operation showed modest contributions (−2% to −1%) in 1977–1990, 1991–2000, and 2011–2019, but reached −42% and −19% at Dawenkou and Daicunba sub-basins during 2001–2010. LUCC contributed −49% (1977–1990) and −79% (1991–2000) at Laiwu sub-basin, while other sub-basins displayed comparable values: [−71%, −73%], [−54%, −58%], and [−37%, −37%] for these periods, escalating to −332–−151% during 2001–2010, before declining to −16–8% post-2010. Other human activities exhibited contributions of 35–65% (1977–1990), 42–71% (1991–2000), 42–106% (2001–2010), and 19–65% (2011–2019). Temporal variations in anthropogenic contributions may be attributed to the following: (1) post-1990 shifts in water abstraction and consumption patterns due to enhanced public water conservation awareness; (2) stringent water resource management policies implemented since 2002; (3) limited regulatory capacity of reservoir operation under low precipitation conditions (2011–2019); and (4) cumulative impacts of small-scale hydraulic infrastructure, particularly dams along main channels.

3.5. Integrated Comparison of Tri-Method Results

This study employed three methodological approaches—an empirical statistical model (Precipitation–Runoff Double Mass Curve), a conceptual model (elasticity coefficient method), and a hydrological model (SWAT)—to quantitatively separate the contributions of climate change and human activities to runoff changes. A comprehensive comparison of results from these methods is illustrated in Figure 9. All approaches consistently identified human activities as the dominant driver of runoff changes across the Dawen River Basin, aligning with previous findings in this basin [45].

During the 1977–1990 period, both the double mass curve method and SWAT model unequivocally demonstrated that human activities contributed substantially more to runoff reduction (67–78% and 45–62%, respectively) than climate change (22–33% and 38–55%). However, the elasticity coefficient method yielded divergent results, attributing runoff decline primarily to climate change, with human activity contributions calculated via non-parametric Budyko-based methods (2–51%) markedly lower than those from parametric approaches. This discrepancy likely stems from differences in mechanistic assumptions and parameterization schemes among methods, underscoring the profound influence of methodological choices on conclusions.

From 1991 to 2000, all three methods converged in identifying human activities as the principal driver of runoff reduction. The double mass curve and SWAT model showed human activity contributions (52–76% and 75–84%) vastly exceeding those of climate change (24–48% and 16–25%), with slightly stronger anthropogenic impacts observed upstream. While the elasticity coefficient method also supported human activities as dominant, non-parametric Budyko-based calculations yielded lower human activity contributions (42–70%) than parametric methods, reaffirming systematic methodological disparities.

The 2001–2010 period revealed pronounced spatial heterogeneity in double mass curve results across the sub-basin. Upstream-to-downstream gradients in human activity contributions were observed at Laiwu (91%), Beiwang (62%), and Daicunba (57%), whereas Dawenkou exhibited negative contributions (−70%), indicating complex nonlinear regional dynamics. Elasticity coefficient methods showed near-identical human activity contributions (79% and 80%) at Beiwang, contrasting with other stations where anthropogenic impacts overwhelmingly dominated, even as climate change exerted positive effects in some areas. SWAT modeling confirmed human activities as primary drivers, though upstream contributions were smaller than downstream, and climate change enhanced runoff in the mid-to-lower sub-basin, likely reflecting regional variations in anthropogenic intensity and climatic conditions.

During 2011–2019, both the double mass curve and SWAT model continued to identify human activities as dominant, with upstream contributions (64–84% and 77–93%) marginally exceeding downstream values. Elasticity coefficient methods showed comparable human activity contributions at Beiwang (82% and 83%) and Daicunba (68% and 67%), while Laiwu exhibited near-equal impacts from human activities and climate change. These results reinforce the overarching anthropogenic influence while revealing nuanced spatial heterogeneity.

Over the entire 1977–2019 period, the double mass curve and SWAT model consistently prioritized human activities as the primary factor, with slightly greater upstream contributions. Elasticity coefficient methods aligned with this conclusion but systematically produced lower human activity estimates via non-parametric Budyko approaches than parametric ones. This consensus underscores the critical role of human activities in hydrological cycles, while methodological discrepancies highlight limitations in capturing complex processes.

Multi-method consensus on human activities dominance: All four methods identified human activities as the primary driver of runoff reduction (55–90% contribution). The double mass curve method attributed 72–78% of runoff changes to human activities at Laiwu, Beiwang, and Daicunba sub-basins during 1977–2019, except at Dawenkou (55%), likely due to its location at the confluence of main and southern tributaries. Temporal heterogeneity emerged: human activities dominated most periods (75–76%), peaking at 90% in Laiwu during 2001–2010—a dry phase when upstream headwater regions exhibited heightened sensitivity to water abstraction. Non-parametric elasticity methods attributed runoff declines to climate change (55–59%) in Laiwu and Dawenkou but emphasized human activities (58–69%) in Beiwang and Daicunba, with anthropogenic contributions increasing temporally across all sub-basins except 2001–2010. Parametric elasticity methods aligned with non-parametric results but uniformly prioritized human activities (54–69%). The SWAT model confirmed human activities as dominant (70–82%), with increasing anthropogenic influence over time and climate-driven runoff augmentation during 2001–2010. In summary, multi-method analyses robustly demonstrate the predominant role of human activities in driving runoff reduction, particularly over extended timescales and across most stations. However, inter-method variations emphasize the necessity of critically evaluating methodological assumptions and constraints to refine mechanistic understanding of climate change and human activity impacts on hydrological systems.

Overall, the results from the Precipitation–Runoff Double Mass Curve method exhibit discernible discrepancies compared to those derived from the Elasticity Coefficient Method and SWAT model, whereas the latter two methods demonstrate more consistent trend alignment. The double mass curve method exclusively considers precipitation and runoff variables, leveraging their statistical correlation for simplified calculations. While computationally efficient, its limited consideration of hydrological variables renders its outcomes primarily indicative. In contrast, the Elasticity Coefficient Method incorporates potential evapotranspiration (ET0) alongside precipitation and runoff, better approximating actual basin-scale hydrological processes. Notably, the Parametric Budyko-based Elasticity Coefficient Method introduces parameters representing underlying surface characteristics, yielding systematically higher estimated contributions of human activities to runoff reduction than the non-parametric Budyko-based variant. The SWAT model further advances this paradigm by explicitly simulating spatially distributed hydrological processes (e.g., infiltration, groundwater recharge, land-surface interactions), surpassing the simplified parameterization of underlying surface conditions in parametric elasticity approaches. Consequently, SWAT-derived estimates of human activity contributions to runoff reduction exceed those from elasticity-based methods. This methodological hierarchy reveals a critical pattern: the granularity of underlying surface representation directly correlates with the magnitude of attributed anthropogenic impacts, suggesting that more physiographically detailed models amplify the quantified role of human activities in driving hydrological change.

3.6. Methodological Adaptation Framework for Runoff Attribution in Specific Basins

3.6.1. Identification of Runoff Change Drivers

Climate factors influencing runoff dynamics include precipitation, air temperature (daily maximum and minimum temperatures), evapotranspiration, solar radiation, relative humidity, atmospheric pressure, and wind speed. Precipitation serves as the primary source of runoff, while temperature and evapotranspiration modulate intermediate processes of precipitation-to-runoff conversion, interacting with other meteorological variables. Extreme precipitation indices and extreme temperature indices, derived from daily precipitation, maximum temperature, and minimum temperature data using the RClimDex model (

Table A5. Climate indicators from the Expert Team on Climate Change Detection and Indices (ETCCDMI).

ID	Indicator Name	Definitions	Units
TN10p	Cool nights	Percentage of days when TN < 10th percentile	Days
TX10p	Cool days	Percentage of days when TX < 10th percentile	Days
TN90p	Warm nights	Percentage of days when TN > 90th percentile	Days
TX90p	Warm days	Percentage of days when TX > 90th percentile	Days
FD0	Frost days	Annual count when TN (daily minimum) < 0 °C	Days
SU25	Summer days	Annual count when TX (daily maximum) > 25 °C	Days
ID0	Ice days	Annual count when TX (daily maximum) < 0 °C	Days
TR20	Tropical nights	Annual count when TN (daily minimum) > 20 °C	Days
TXx	Max Tmax	Monthly maximum value of daily maximum temp	°C
TNx	Max Tmin	Monthly maximum value of daily minimum temp	°C
TXn	Min Tmax	Monthly minimum value of daily maximum temp	°C
TNn	Min Tmin	Monthly minimum value of daily minimum temp	°C
WSDI	Warm spell duration indicator	Annual count of days with at least 6 consecutive days when TX > 90th percentile	Days
CSDI	Cold spell duration indicator	Annual count of days with at least 6 consecutive days when TN < 10th percentile	Days
DTR	Diurnal temperature range	Monthly mean difference between TX and TN	°C
GSL	Growing season Length	Annual (1 January to 31 December in NH, 1 July to 30 June in SH) count between first span of at least 6 days with TG > 5 °C and first span after 1 July (1 January in SH) of 6 days with TG < 5 °C	Days
RX1day	Max 1-day precipitation amount	Monthly maximum 1-day precipitation	Mm
Rx5day	Max 5-day precipitation amount	Monthly maximum consecutive 5-day precipitation	Mm
R95p	Very wet days	Annual total PRCP when RR > 95th percentile	Mm
R99p	Extremely wet days	Annual total PRCP when RR > 99th percentile	mm
SDII	Simple daily intensity index	Annual total precipitation divided by the number of wet days (defined as PRCP ≥ 1.0 mm) in the year	Mm/day
PRCPTOT	Annual total wet-day precipitation	Annual total PRCP in wet days (RR ≥ 1 mm)	mm
CDD	Consecutive dry days	Maximum number of consecutive days with RR < 1 mm	Days
CWD	Consecutive wet days	Maximum number of consecutive days with RR ≥ 1 mm	Days
R10	Number of heavy precipitation days	Annual count of days when PRCP ≥ 10 mm	Days
R20	Number of very heavy precipitation days	Annual count of days when PRCP ≥ 20 mm	Days
Rnn	Number of days above nn mm	Annual count of days when PRCP ≥ nn mm, nn is a user-defined threshold	Days

lists 16 extreme temperature and 11 extreme precipitation indices), exhibit minimal direct correlation. Among 176 calculated correlation coefficients, only eight showed statistical significance (Figure 10). Given this independence assumption between extreme precipitation and temperature indices, precipitation and temperature were selected as the dominant climate change factors for runoff attribution analysis.

(1)

Precipitation Factors

Correlation analysis of extreme precipitation indices is illustrated in Figure 11. Results demonstrate strong correlations (coefficients > 0.77) between runoff and indices including R95p, Rx1day, Rx5day, SDII, PRCPTOT, and Rnn (number of heavy precipitation days). R99p exhibits moderate correlations with other indices (0.53–0.72). CDD and CWD show weak correlations: CDD ranges from −0.31 to −0.07, and CWD from −0.08 to 0.51. Notably, R20, R10, and Rnn display exceptionally high inter-correlations (>0.87). Based on these findings, five extreme precipitation indices (R95p, R99p, CDD, CWD, R20) and annual precipitation at Daicunba Station were selected to assess precipitation impacts on runoff evolution.

As demonstrated in Figure 12, runoff exhibits correlations of 0.77, 0.77, 0.64, 0.57, and 0.29 with annual precipitation, R20, R99p, R95p, and CWD, respectively, ranked in descending order. In contrast, CDD shows a weak negative correlation with runoff (−0.10). These results confirm that runoff dynamics are closely linked to annual precipitation totals and intense rainfall events.

Wavelet coherence analysis was conducted between annual runoff and four precipitation indices with relatively strong correlations: R95p, R99p, R20, and annual precipitation (Figure 13). A higher wavelet coherence coefficient indicates stronger correlation strength between variables. Phase relationships are denoted as follows: “→”: In-phase oscillation (positive correlation); “←”: Anti-phase oscillation (negative correlation); “↑”: Runoff leads precipitation indices by a 90° phase shift; “↓”: Runoff lags precipitation indices by a 90° phase shift. The results (Figure 6) demonstrate strong correlations between all four precipitation indices and runoff. Specifically, annual precipitation, R20, R99p, and R95p exhibit progressively weakening correlation strengths with runoff, consistent with Pearson correlation analysis. On longer timescales (>10 years), all precipitation indices show persistent and highly correlated relationships with runoff across the entire study period. In contrast, short-term correlations (interannual scales) lack temporal continuity but display heightened coherence during 1990–2000. Nearly all precipitation indices exhibit positive phase relationships with runoff, with runoff showing a slight lag relative to precipitation drivers.

The relative importance ranking of 11 precipitation characteristics to runoff (Figure 14) reveals that annual precipitation dominates as the most influential factor, followed by SDII and R10. In contrast, Rx1day exhibits the lowest relative importance.

(2)

Temperature Factors

Compared to inter-precipitation index correlations, temperature indices exhibit weaker mutual associations. As shown in Figure 15, TN10p demonstrates moderate correlations (absolute coefficients ≥ 0.54) with TX10p, TN90p, FD0, TR20, TNn, CSDI, GSL, and DTR. TX90p correlates notably (absolute coefficients ≥ 0.55) with TN90p, SU25, and WSDI. ID0 shows stronger correlations (≥0.61) with TX0p, TXn, and TNn. TNx only weakly correlates with TN90p (absolute coefficient 0.53), while TXx exhibits minimal correlations (absolute coefficients < 0.4) with other indices. Consequently, TN10p, TX90p, ID0, TXx, and TNx are identified as relatively independent temperature drivers. Most temperature indices show weak correlations with runoff, with only TX90p, SU25, and TXx passing statistical significance tests. This suggests that high-temperature extremes exert greater hydrological influence than low-temperature events, likely due to their amplified effects on evapotranspiration and subsequent suppression of precipitation-to-runoff conversion efficiency.

Wavelet coherence analysis was performed between annual runoff and temperature indices with relatively strong correlations: TX90p, SU25, TXx, and DTR (Figure 16). Results demonstrate alignment with Pearson correlation outcomes, wherein SU25 exhibits the closest linkage to runoff. On short-term scales, TXx shows significant coherence with runoff post-2000. Long-term scales reveal persistent correlations between SU25, DTR, and runoff across the entire temporal domain. Notably, TXx and SU25 display intensified phase synchronization during 1995–2010, while DTR–runoff interactions exhibit intermittent coherence modulated by decadal climatic oscillations.

The relative importance ranking of 16 extreme temperature indices to runoff (Figure 17) identifies SU25 as the most critical factor, followed by TN10p, ID0, GSL, and FD0. This hierarchy suggests that cold-temperature-related indices exhibit higher relative importance within the Dawen River Basin.

3.6.2. Basin Characteristic Factors

In methodologies for disentangling climate change and human activity impacts, three Budyko framework-based elasticity coefficient approaches (Fu, Zhang, and Choudhury–Yang equations) incorporate basin-specific parameters termed BasinFeFu, BasinFeZhang, and BasinFeCY. The SWAT model employs 16 parameters to characterize basin attributes: SOL_AWC (soil available water capacity), SOL_K (saturated hydraulic conductivity), CN2 (SCS runoff curve number), ALPHA_BF (baseflow alpha factor), GW_DELAY (groundwater delay time), GWQMN (threshold water level for baseflow), REVAPMN (threshold depth for re-evaporation from shallow aquifer), GW_REVAP (groundwater revap coefficient), CH_K2 (effective hydraulic conductivity), CH_N2 (Manning’s roughness coefficient), SLSUBBSN (average slope length), EPCO (plant uptake compensation factor), ESCO (soil evaporation compensation factor), CANMX (maximum canopy storage), SURLAG (surface runoff lag time), and SFTMP (snowfall temperature).

Correlation analysis of these 19 basin characteristic parameters (Figure 18) revealed strong interdependencies: CN2 correlated significantly (|r| ≥ 0.73) with SOL_K, GW_REVAP, CH_K2, CH_N2, CANMX, SURLAG, and SFTMP; ALPHA_BF showed high correlations (r ≥ 0.85) with EPCO and ESCO; GWQMN exhibited robust linkages (|r| ≥ 0.74) with SOL_AWC, GW_DELAY, and SLSUBBSN; and BasinFeZhang demonstrated near-perfect correlations (r ≥ 0.89) with BasinFeFu, REVAPMN, and BasinFeCY. Consequently, four relatively independent parameters—CN2, ALPHA_BF, GWQMN, and BasinFeZhang—were selected.

Random forest-based importance ranking (Figure 19) identified REVAPMN, CH_K2, SURLAG, BasinFeCY, ALPHA_BF, and CH_N2 as critical drivers of runoff changes, highlighting the dominance of groundwater–channel interactions and anthropogenic basin modifications in hydrological dynamics.

3.6.3. Adaptability Assessment of Runoff Attribution Methods

The complexity of basin-scale runoff dynamics stems from their multifaceted drivers, which are categorized into driving factors (climate elements and human activities) and basin characteristic factors (reflecting underlying surface attributes). In this study, runoff driving factors include 11 extreme precipitation indices (e.g., R95p, R99p), 16 extreme temperature indices (e.g., TN10p, TX90p), water abstraction in the Dawen River Basin (WatUseDai), and land use categories (cropland, forest, pasture, urban, bare land). Basin characteristic factors encompass SWAT model calibration parameters (e.g., CN2, ALPHA_BF) and Budyko framework-derived parameters (BasinFeFu, BasinFeZhang, BasinFeCY).

Random forest-based importance ranking of these factors reveals three key insights (Figure 20): (1) Annual precipitation (PRCPTOT) overwhelmingly dominates runoff variability, validating the use of the Precipitation–Runoff Double Mass Curve—a traditional statistical method relying solely on precipitation and runoff time series—to quantify contributions of climate change and human activities to runoff changes. During periods with near-normal precipitation and water abstraction, double mass curve results align closely with SWAT model outputs, offering a simplified yet reliable approach when human activity factors rank lower in importance. (2) Water abstraction ranks third in importance, exhibiting spatiotemporal heterogeneity across sub-basins. During dry periods (e.g., 2001–2010), intensified abstraction altered runoff directionality (augmentation or reduction), rendering non-parametric elasticity methods or double mass curve less reliable, necessitating parametric elasticity coefficient methods or hydrological modeling to capture nonlinear interactions. (3) Secondary drivers include ID0, R95p, R99p, TNx, TN10p, and CDD, highlighting the dual influence of precipitation extremes and thermal dynamics. BasinFeZhang, ranking ninth, underscores the integrative role of underlying surface properties, while land use types and human-modified parameters exhibit lower importance. This dichotomy illustrates that single climate change factors can directly alter runoff, whereas cumulative basin-scale adjustments are required to manifest hydrological impacts. Post-exclusion of precipitation/temperature factors, water abstraction and basin characteristics emerge as dominant residual drivers.

For the Dawen River Basin, the prominence of water abstraction (rank 3) and BasinFeZhang (rank 9) necessitates hydrological modeling (SWAT) for robust attribution. In other basins, preliminary analysis of local drivers and characteristics—combined with data availability—should guide method selection, prioritizing approaches aligned with the highest-ranked factors (e.g., elasticity methods for precipitation-dominated systems, SWAT for groundwater–channel interaction systems).

For the Dawen River Basin, the high rankings of water abstraction (rank 3) and basin characteristic parameter BasinFeZhang (rank 9) indicate that hydrological modeling (SWAT) is the most suitable method for runoff attribution analysis. In other basins, preliminary assessments of local runoff driving factors and basin characteristic factors should be conducted. Method selection must integrate parameter importance ranking results with data availability, prioritizing approaches that align with dominant drivers.

4. Conclusions

This study conducted relative importance analysis on precipitation factors, temperature factors, and basin characteristic factors within runoff drivers, establishing a methodological adaptability framework for runoff change attribution. Using the Dawen River Basin as a case study, three approaches—Precipitation–Runoff Double Mass Curve, non-parametric/parametric elasticity coefficient methods, and the SWAT hydrological model—were employed to quantify contributions of climate change and human activities to runoff changes. First, climate change and human activities were broadly separated, followed by SWAT-based disaggregation of specific human activities (e.g., land use change, reservoir operation, water abstraction). Key findings include the following: ① Multi-method consensus on human activities dominance: All four methods identified human activities as the primary driver of runoff reduction (55–90% contribution). ② Quantitative separation of the effects of various human activities on runoff change: SWAT-based analysis revealed divergent impacts: consumptive water use consistently reduced runoff (primary contributor), and LUCC increased runoff in the Dawenkou and Daicunba reservoirs by 19–42%. Limited reservoir data (3 of 23 major reservoirs modeled) likely underestimated these impacts. ③ Methodological adaptability framework: In other basins, the selection of runoff-related parameters (i.e., runoff drivers and basin characteristic factors) should first be based on the accessibility of basin-specific data. The relative importance of these parameters is then evaluated via random forest ranking, with method selection guided by the ranking results. For preliminary identification of dominant runoff change drivers, elasticity coefficient methods are prioritized, with a choice between parametric and non-parametric approaches depending on data availability. In basins with extremely limited data, the Precipitation–Runoff Double Mass Curve serves as a viable alternative. Where detailed understanding of runoff change factors—particularly the contribution of various human activitiesis required, priority should be given to hydrological modeling, though its practical implementation entails greater operational complexity.