数字滤波法作为估计基流最常用的自动分割方法，计算简单、易于推广，但滤波参数常采用经验取值，且固定参数难以反映基流的季节性动态变化，影响模型分析精度。以伊河上游栾川流域为典型研究区，采用4种数字滤波模型，通过分析径流退水系数（k）确定数字滤波参数，讨论滤波参数的季节性变化对模型估算基流的影响，并评估数字滤波方法在研究区域的适用性。结果表明：栾川流域降水径流仍呈显著正相关关系，NDVI增加及土地利用变化对流域退水系数影响不明显，研究期内在夏、秋季平均径流退水系数k约在20~35 h，春、冬两季k值约在90~100 h，即流域径流退水在春冬两季较为缓慢，且考虑滤波参数季节性变化的基流分割结果比全年使用同一滤波参数的结果更符合流域基流实际特征；对4种数字滤波法所得的基流指数进行误差分析和精度评价，发现在栾川流域，Eckhardt双参数滤波模型比其他3种单参数滤波模型更为适用，所得流域平均基流占比约为33%。研究结果降低了滤波模型参数的不确定性，提高了数字滤波方法模拟基流的准确性，也为校准或验证水文模型提供参考。

Baseflow is an important recharge to river runoff and is critical for maintaining the health of river ecosystems as well as for watershed water resource regulation and management. However, baseflow is difficult to measure and tracer-based methods are time-consuming and expensive, so non-tracer methods are commonly used for estimation. Digital filtering is a commonly used non-tracer method for partitioning baseflow, and the computational process is simple and easy to implement. The applicability of the digital filter model in different basins varies depending on the flow characteristics of the basin, and the filter parameters are often difficult to determine directly. In the past, most scholars often used empirical parameters to separate baseflow using digital filter model, and seldom considered the seasonal dynamic change characteristics of the parameters, which made the results of baseflow separation in different watersheds have a large uncertainty. To improve the accuracy of baseflow separation, a digital filter baseflow separation method was proposed based on seasonal recession analysis.The Luanchuan basin of the Yi River was taken as a typical study area, and four digital filtering models were used for baseflow separation, including three single-parameter filtering models (Lyne-Hollick, Chapman and Chapman-Maxwell), and a two-parameter filtering model (Eckhardt). Based on the 47 measured sub-flood data and daily runoff data of the watershed from 1964 to1979 and 2001 to 2020, the recession coefficient (k) of the watershed in different seasons was calculated. To analyze the effects of considering seasonal variations of model parameters on baseflow separation results and to calculate baseflow indices for each season in the study watershed, three evaluation indexes, namely, Nash-Sutcliffe efficiency coefficient (ENS), Gray correlation coefficient (DGR), and mean relative error (EMR) were used to analyze the errors and evaluate the accuracy of baseflow results and to assess the applicability of the four filtering models. The results showed that there was little land use change in the watershed over the two study periods, and the increase in NDVI did not have a significant effect on the number of receding water systems in the watershed. The precipitation-runoff double cumulative curve shows that precipitation is still the dominant factor leading to changes in runoff. The seasonal variation of the receding coefficient k in the Luanchuan basin was obvious, and the receding rates were faster in summer and fall, with k about 20 to 35 h, and slower in fall and winter, with k about 90 to 100 h. Compared with using a single parameter throughout the year, using different filtering parameters in each season significantly reduces the base flow index, which is more in line with the base flow characteristics of the watershed; Taking the oblique straight line separation method as the evaluation standard, the error and accuracy of baseflow indices obtained from several digital filtering methods were evaluated, the baseflow indices calculated by single-parameter than two-parameter filtering methods were generally larger, and the results of baseflow separation were greatly affected by the parameters. Eckhardt two-parameter filtering method had the best fitting effect, and it was the best baseflow separation method in Luanchuan watershed, and the average proportion of groundwater runoff recharge in the basin is about 33%. The results of this research provide new ideas for improving the accuracy of the digital filtering model to separate baseflow.