| 20 Jul 2026
Assessing the Effect of Data Resolution on Optimized Hydrograph Separation
Abstract. Hydrograph separation is centered on quantifying groundwater contributions to streamflow, yet the influence of monitoring data resolution remains insufficiently characterized. In this study, we assess how temporal resolution influences chemically constrained hydrograph separation. We apply HydrOHS, a multi-objective optimization framework that calibrates the physically-grounded Eckhardt Recursive Digital Filter using SC-centered mass balance, to four modeling scenarios capturing two spatial settings and two temporal resolutions. HydrOHS reconstructs SC end-members and simultaneously minimizes SC mismatch, enforces chemically consistent flow component ordering, and constrains peak baseflow index (BFI) behavior. The four modeling scenarios produce similar optimized BFImax parameter values (0.80–0.89) and consistent downstream increases in baseflow contribution, indicating that temporal resolution (daily- vs 15 min monitoring frequency) exerts a secondary influence relative to hydrologic variability and missing flow periods. Higher resolution datasets provide additional detail but are more sensitive to zero-flow gaps, which degrades chemical reconstruction. Importantly, all optimized BFImax estimates substantially exceed the suggested “default” BFImax value for hard-rock aquifer systems, demonstrating that uncalibrated parameters underestimate baseflow. These findings highlight the importance of site-specific, chemically constrained calibration and show that daily resolution datasets may be sufficient for regional groundwater-recharge assessments.
The authors present an interesting study that applies an optimization process to baseflow separation methods combining a recursive digital filter and specific conductance-based mixing constraints. The method is tested at two locations on the Tule River in California using daily and 15-minute resolution data.
The methodological approach is unique and looks to have potential as a way of building more robust baseflow estimates in situations where flow and electrical conductivity data (or potentially other chemical tracers) are available. The manuscript explains the method and application well and the plots generally support the key points. The title of the paper suggests a focus on the effects of data resolution and, while this is examined, some additional analysis and explanation would be appropriate if the intent for the paper to retain that focus. This and a few other points to consider in improving the manuscript are provided below.
General comments
My main concern is that, if the focus of the paper is indeed intended to be on the effect of data resolution on the outcomes of the baseflow separation method, the analysis should include more than just 15-minute and daily average data. Hourly data are not uncommon and could be aggregated from the 15-minute data as an intermediate set between the 15-minute and daily. From there the analysis can provide a more rigorous assessment of the pros and cons of using data at these resolutions.
Along these same lines, it would be helpful to provide a bit more context (perhaps even an explicit hypothesis) about why it would be important to consider sub-daily resolution data for baseflow separation. Why is it important to test 15-minute (vs daily or hourly) data? What additional value might one expect higher resolution data to provide for baseflow estimation? One would expect groundwater contributions to streamflow to change slowly, a signal not necessarily evident in 15-minute streamflow variability alone. It appears from Figure 3, Figure 6, and the results of the optimization in Table 3 that the 15-minute data introduces noise that may obscure the key baseflow behaviors and make achieving similar results (as the daily data) more difficult.
The method to convert precipitation SC to a quickflow SC signal is interesting. Given that runoff in this watershed will come from a mix of snowmelt and some direct rain, depending on elevation and time of year, how robust is this parameterization to potential issues introduced by a potentially variable quickflow signal? Some additional text addressing this would bolster the methods section.
The 3D plots in Figure 5 are rather difficult to interpret in their static form and patterns aren't immediately legible in the ways described in the text. I'd suggest exploring a small multi-panel plot that shows the combinations of selected objective axes to better illustrate the shape of the pareto solution and the selected TOPSIS solution.
Provide some additional comment or explanation as to how sensitive the TOPSIS selected outcome is to the weighting factors. It is somewhat difficult to discern the shape of the pareto surface across dimensions, but it would be helpful to know if the assumptions regarding the weighting matter much for the results.
Figure 7 provides a useful summary of the resulting estimated specific conductance components. The manuscript could be improved by adding some more explanation and assessment of these results. Do they seem reasonable given the location and setting? For example, the daily and 15-minute Cholollo Campground results show very similar patterns (suggesting the optimization process found similar solutions regardless of resolution) while the Reservation Boundary daily results have much more variable quickflow SC. What mechanism might explain this? Do the timings of seasonal peaks of quickflow and baseflow SC make sense, especially across the drought events of 2014-2015 and 2021-2022 (and following wet periods)?
The manuscript explores baseflow separation methods but provides little analysis of the actual separated baseflow time series results. The summary in Table 4 provides a start - what would explain the substantial difference in baseflow to precipitation ratio between the two sites? Are there losing reaches between the two stations? What other characteristics of the baseflow results align or differ across the sites and time resolutions? Do periods of estimated high baseflow fraction properly align with the times of year when one would expect it (e.g. mid-summer/early fall)? An additional figure illustrating baseflow (and total streamflow) characteristics could be helpful in bolstering this aspect of the analysis.
Did you explore the effects of excluding the periods of extended missing and zero flow data in 2021 (by limiting the dataset to times prior/after) on the results? Knowing whether one might get different results be excluding problematic time periods would be helpful in balancing dataset length versus value and would further improve the impact of the manuscript.
Specific comments
Abstract and fourth paragraph: Make sure that "SC" is defined as specific conductance before the initials/acronym SC are used.
Line 126: Typo - I assume "ensuing" was meant to be "ensuring"
Figure 1: Minor point of clarification - do the vertical (downward pointing) arrows indicate that each step is done in succession, or is it more accurate to say that the data processing steps may be done in order but the decision variables are adjusted at the same time as part of each optimization iteration? If this is the case, I'd suggest revising the figure arrangement and arrows to better reflect the dependence and order of the process. If not, some additional clarifying language may be helpful.
Line 529: Potentially extra instance of the word "peak" here?