Quantifying flow pathways within a larger catchment can help improve diffuse pollution management strategies across subcatchments. But, spatial quantification of flow pathway contributions to catchment stream flow is very limited, since it is challenging to physically separate water from different paths and very expensive to measure, especially for larger areas. To overcome this problem, a novel, combined data and modelling approach was employed to partition stream flow in the Piako catchment, New Zealand, which is a predominantly agricultural catchment with medium to high groundwater recharge potential. The approach comprised a digital filtering technique to separate baseflow from total stream flow, machine learning to predict a baseflow index (BFI) for all streams with Strahler 1st order and higher, and hydrological modelling to partition the flow into five flow components: surface runoff, interflow, tile drainage, shallow groundwater, and deep groundwater. The baseflow index scores corroborated the spatial distributions of the flow pathways modelled in 1st order catchments. Average depth to groundwater data matched well with BFI and Hydrological Predictions for the Environment (HYPE) modeled flow pathway partitioning results, with deeper water tables in areas of the catchment predicted to have greater baseflow or shallow and deep groundwater contributions to stream flow. Since direct quantification of flow pathways at catchment-scale is scarce, it is recommended to use soft data and expert knowledge to inform model parameterization and to constrain the model results. The approach developed here is applicable as a screening method in ungauged catchments.

量化更大流域内的流动路径可以帮助改进跨子流域的扩散污染管理策略。但是，流动路径对集水流流量贡献的空间量化非常有限，因为将水从不同路径物理分离是具有挑战性的并且测量非常昂贵，特别是对于较大的区域。为了克服这个问题，我们采用了一种新颖的、结合数据和建模的方法来划分新西兰 Piako 流域的河流流量，该流域主要是农业流域，具有中高地下水补给潜力。该方法包括将基流与总流流分离的数字过滤技术，机器学习以预测具有 Strahler 一阶及更高阶的所有流的基流指数 (BFI)，和水文建模将水流划分为五个水流分量：地表径流、互流、瓦片排水、浅层地下水和深层地下水。基流指数分数证实了在一阶流域中建模的流动路径的空间分布。地下水平均深度数据与 BFI 和环境水文预测 (HYPE) 模拟的流动路径划分结果非常匹配，流域区域内较深的地下水位预测具有更大的基流或浅层和深层地下水对溪流的贡献。由于流域尺度的流动路径的直接量化很少，建议使用软数据和专家知识为模型参数化提供信息并约束模型结果。