Optimization of baseflow separation parameters and responses to environmental drivers in the Liuxi River basin

Accurate baseflow assessment is a critical technique for evaluating riverine ecological health and ensuring the sustainable use of water resources. Despite its importance, conventional baseflow separation methods, particularly non-tracer approaches such as digital filters, often involve considerable uncertainties. These uncertainties primarily arise from the empirical selection of parameters and the lack of a strong physical basis, which may lead to misinterpretation of hydrological processes. Although international studies have explored the use of tracer methods to calibrate parameters in non-tracer approaches, research in China has largely focused on optimizing individual methods and lacks a systematic framework based on readily available tracers. Furthermore, existing tracer-based parameter optimization studies are mainly concentrated in temperate continental or Mediterranean climates, leaving a notable research gap in subtropical monsoon regions characterized by highly seasonal rainfall. To address these challenges and establish a robust framework for high-precision baseflow estimation, this study focused on the catchment of the Taipingchangzhen Hydrological Station, located in the middle reaches of the Liuxi River basin. This area represents a typical subtropical monsoon climate, with rainfall concentrated during the flood season (April–September) and significant anthropogenic disturbances. We utilized a comprehensive dataset spanning 2020–2023, including daily streamflow, daily electrical conductivity (EC) observations, precipitation, monthly potential evapotranspiration (PET) derived from MODIS, and monthly Normalized Difference Vegetation Index (NDVI) extracted from Landsat imagery. Methodologically, we developed a coupled optimization framework based on the electrical conductivity mass balance (CMB) approach to systematically calibrate three widely used baseflow separation models: the Recursive Digital Filter (RDF), the Smooth Minima Method (SMM), and the Bump and Rise Method (BRM). EC was selected as a cost-effective and reliable tracer because preliminary analysis revealed a significant negative correlation between streamflow and EC (Spearman correlation coefficient = −0.54), confirming its suitability for distinguishing rapid surface runoff from delayed baseflow. To account for seasonal variability, we applied a dynamic monthly extreme-value interpolation method to generate EC time series for both baseflow and surface runoff components, thereby overcoming the physical limitations associated with static annual extreme-value methods. Parameter optimization was conducted using the surrogateopt algorithm, with the objective of maximizing the Kling–Gupta Efficiency (KGE) coefficient between simulated runoff EC and observed EC. Implementation of this EC-guided optimization framework resulted in substantial improvements in model performance. After calibration, the KGE coefficients of the RDF and SMM models increased from 0.74 and 0.79 to 0.80, respectively, while the BRM model also achieved a KGE of 0.80. The optimization process substantially modified key parameters; for example, the maximum baseflow index (BFImax) in the RDF model was revised from the empirical value of 0.25 to 0.12, accurately reflecting the rapid recession characteristics of interflow and shallow groundwater in the study basin. Consequently, the optimized parameters led to a 52.7% reduction in the estimated multi-year average baseflow volume, decreasing from 6.64 m³/s obtained using empirical parameters to a more physically realistic value of 3.14 m³/s. Furthermore, the study revealed pronounced seasonal and interannual variability in baseflow estimates. The empirical-parameter methods (RDF and SMM) exhibited a systematic tendency to overestimate baseflow during the flood season and showed high sensitivity to storm events, resulting in artificially elevated standard deviations ranging from 13.60 to 14.95 m³/s. In contrast, the optimized models, particularly the BRM, demonstrated remarkable stability and a stronger capacity to withstand the influence of intense monsoon rainfall. The BRM proved to be the most robust method under interannual climatic variability, exhibiting the lowest coefficient of variation (0.46) among all evaluated techniques. To further elucidate the physical mechanisms governing baseflow dynamics, we conducted a comprehensive environmental response analysis using an ordinary least squares (OLS) multiple linear regression model. The results identified NDVI as the dominant driving factor, accounting for 47.39% of the observed variation in baseflow, followed closely by precipitation, which contributed 43.59%. In contrast, PET exerted a relatively limited direct influence, contributing only 9.02%. These primary drivers exhibited distinct seasonal variations and temporal lag effects. Specifically, NDVI demonstrated an immediate and synchronous influence on baseflow generation, underscoring the critical role of vegetation in rapid water retention and soil moisture maintenance. Meanwhile, precipitation exhibited a pronounced one-month lagged effect on baseflow, reflecting the time required for rainfall to infiltrate and recharge groundwater flow paths. PET showed a six-month delayed negative effect, indicating gradual depletion of basin water storage through continuous evapotranspiration processes. In conclusion, this study presents optimized parameters and a robust methodological framework for high-precision baseflow separation in the Liuxi River Basin and in geographically similar regions. By reducing uncertainties inherent in empirically driven approaches through an EC-constrained mass balance framework, the study advances the understanding of hydrological processes in subtropical monsoon climates. The integrated findings on model performance and environmental drivers have important theoretical and practical implications for regional water resource management, accurate estimation of ecological instream flows, and the development of adaptive strategies for climate change resilience.