Paper Review: Groundwater storage changes in the United States using baseflow recession method: Comparison with GRACE and well observations

[COBrogan]

https://www.sciencedirect.com/science/article/pii/S221458182500775X

Summary:

• Authors have created a baseflow event algorithm and used it to estimate groundwater storage

This event-based approach requires careful identification of primarily dry, late-time, continuous recession periods which exclude short-duration or early recession points to reduce non-linear influences and isolate pure baseflows originating from groundwater discharge.

• Estimations compared to GRACE and observed well data
• Data sources are NLDAS2 evapotranspiration and the global precipitation resolution data, i.e., from GPM-IMERG Final Precipitation 1 day with spatial resolution of 0.1°× 0.1° version 6
• Only analyzed events from May - October to avoid snowmelt interference in recession analysis and period of study was 2001 - 2020

Leading Research Questions:

1. What are the spatial and temporal patterns of the baseflow drainage timescale (K), and do they relate to different hydroclimatic regions?
2. Can baseflow recession analysis reliably estimate long-term groundwater storage changes in minimally disturbed watersheds?
3. How well do baseflow-derived storage trends align with GRACE-DA and well-based storage estimates?

Why use event-based recession?

Analyzing streamflow recessions collectively as single event instead of individual events ignores the temporal variability of storage depletion. Multiple independent recession events help us to understand how storage dynamics vary under different conditions, like, groundwater and soil moisture levels, prior rainfall, temperature, evapotranspiration, soil types, human influence

What is a baseflow recession event?

Although there is no general criterion for baseflow recession events (Dralle et al., 2015), a common approach is that streamflow and its first or second derivative decline monotonically for a normally unrestrictive selection process for baseflow events (Dralle et al., 2017, Santos et al., 2019, Tashie et al., 2020).

What is the baseflow event algorithm here?

In the proposed approach, the start and end of each recession event in storm hydrographs are identified from zero-rainfall days (here, zero rainfall is defined as daily average rainfall (<= 0.5mm) following the condition of non-increasing streamflow for minimum of 7 days.

• Used exponential relationship of Boussinesq that relates $y = y_0 * e^(-K/t)$ where y is baseflow per unit area

To choose a representative timescale value, K for a watershed, it is selected as an average value i.e., trimmed mean of K corresponding to lowest baseflow events of each year during a study period. Operationally, we compute a trimmed mean after removing the highest and lowest 5 % of K values across all baseflow events to limit outlier influence and obtain a stable measure of central tendency.

• Calculate the lowest summer (May - Oct) 7-day flow to estimate the lowest storage in each year and then regress to find the trend in these low flows over time. This is designed to represent the minimum groundwater storage
• At the lowest flows, the authors note that the storage is likely linearly related to the seven-day low flow and some "timescale" coefficient (recession coefficient) K $S = K*y_{L7}$
  Regarding K:

Physically, it reflects how quickly a watershed drains during dry periods, with larger K values indicating slower drainage or a deep storage system and smaller K indicating rapid drainage, often due to shallow subsurface storage or steep slopes.

Well data:

• Ran regression on the minimum well level storage recorded each year and converted to comparable units
• Used wells that were withing 1 km, 2 km, and 5 km of the watershed via a spatial buffer

GRACE;

• Found minimum storage of each year and conducted a regression, similar to the well data and streamflow data
• Noted that GRACE storage is often much higher than that estimated from baseflow events, likely because it is assumed the whole DA contributes to the baseflow instead of only a small portion around the stream and does not consider inactive water in the unsaturatged zone

Incorporation of human influence:
$$\Delta S_{human} = \Delta s_{wb} - \Delta S_{bflow}$$

Comparing Storage estimates with GRACE and Well Data:

• All data was spatially aggregated to watershed boundary
• All three (estimates, GRACE, and well data) compared against each other using percent difference and change in percent difference
• Focused on the "sign" of the change rather than the magnitude
• Found higher K values in areas with less permeable rock
• In Appalachian region, ET is beginning to exceed precip and result in minimal recharge
• Flashier watersheds tended to have smaller K
• Mountainous areas tended to have higher k
• K did not vary with DA, but with watershed elevation and slope
• K was lowest in spring and highest in summer due to antecedant moisture in the Spring and drying in the summer

The smaller initial K values in May are a manifestation of antecedent precipitation that recharged the aquifers. The increase in K values from mid-summer are due to the dry and hot conditions in peak summer months, which subsequently declines the water table and increases the travel time of water parcels to reach stream outlet. Thus, it is inferred that precipitation and evapotranspiration, which govern natural groundwater recharge, are among the key factors influencing baseflow recession.

• Noted that GRACE and baseflow estimates were very similar in pattern, but not magnitude (75% of waterhseds)
• 67 - 75% of watersheds had similar patterns in baseflow K vs well estimated K
• All three methods agreed in 60% of watersheds on the direction

Conclusions

• Precip, not ET, is the largest driver of GW storage
• K may change with season

The resulting recession parameter, known as the drainage characteristic timescale (K), reflects the responsiveness of a watershed’s storage system and exhibits both spatial and temporal variability across the watersheds. The analysis reveals that there is no common month with a consistent K trend across all regions, but late summer to early fall (especially September) shows increased K in many areas, indicating delayed streamflow recession likely driven by seasonal groundwater dynamics.

Overall, the results demonstrate that baseflow events extracted from streamflow observations, combined with concurrent precipitation data, provide a reliable means for estimating the baseflow recession constant and, consequently, groundwater storage changes. By systematically comparing baseflow-derived estimates with both satellite observations and ground-based data, the study demonstrates the viability of a scalable, streamflow-based approach to groundwater monitoring, particularly valuable where well data are sparse or satellite resolution is inadequate for small basins.

[rburghol]

@COBrogan My conclusions based on the review is that this approach is supportive but not duplicative of our concept:

• They demonstrate the effectiveness of using base flow events for coefficient estimation.
• Their focus on very broad scales is very different than our high resolution focus that ultimately drills down into variability in base flow coefficient as S increases or decreases.
• They DO demonstrate some overlap between the coefficient analysis and GRACE, though it didn't make a strong argument for either approach. Either:
  • GRACE is highly accurate and one can only assume a 60% accuracy in estimating whether base flow stores are increasing/decreasing from gage flows and Boissenesq type analyses, likely due to difficulty in baseflow isolation.
  • Boissenesq type analyses are highly accurate, and GRACE has some difficulty in handling errors induced by precipitation and separating surface from sub-surface water storage.
  • OR both are true at times and introduce some error.
• If I were to speculate I would say that the GRACE method is fine, but that the actual flow-based methods should be much preferred, with the rationale that agreeing about the direction of changes should be very straight-forward, and suggests to me that the GRACE could be a reasonable coarse estimate when flow gage values are not available.

[rburghol]

@COBrogan Following up on this with some more thoughts:

• To my previous hypotheses on GRACE vs gage: a 3rd possibility, that is, that the GRACE performs well in some areas and poorly in others, depending perhaps on how shallow/deep the groundwater is relative to the surface layer water.

• Precip, not ET, is the largest driver of GW storage

• No argument with precip >> ET in terms of GW storage, but only when precip >> ET of course :)

• K may change with season

• And to this seasonal variability (below), I suspect that this is not in fact seasonal, but rather, storage dependent K, as we suspect in some of our basins. Increased K at low S could definitely explain the trend that they are seeing.

The resulting recession parameter, known as the drainage characteristic timescale (K), reflects the responsiveness of a watershed’s storage system and exhibits both spatial and temporal variability across the watersheds. The analysis reveals that there is no common month with a consistent K trend across all regions, but late summer to early fall (especially September) shows increased K in many areas, indicating delayed streamflow recession likely driven by seasonal groundwater dynamics.

Overall, the results demonstrate that baseflow events extracted from streamflow observations, combined with concurrent precipitation data, provide a reliable means for estimating the baseflow recession constant and, consequently, groundwater storage changes. By systematically comparing baseflow-derived estimates with both satellite observations and ground-based data, the study demonstrates the viability of a scalable, streamflow-based approach to groundwater monitoring, particularly valuable where well data are sparse or satellite resolution is inadequate for small basins.

[rburghol]

@COBrogan And finally, some comments on some of their recharge conclusions:

Conversely, regions with minimal or managed abstraction, especially in parts of the Midwest and Eastern U.S., show mixed trends with some areas experiencing positive storage changes, possibly due to managed recharge or reduced groundwater use. High abstraction zones highlighted in (Fig. S7) largely overlap with areas of negative storage change in (Fig. 4d), underscoring the direct impact of human activities on groundwater resources. Positive storage changes from (Fig. 4d) correspond to regions with lower abstraction rates, suggesting that some areas may be benefiting from effective groundwater management or natural recharge processes.

• I am concerned that they are conflating deep aquifer storage, which I think GRACE is not supposed to be able to quantify, with shallow (surficial) aquifer storage, that GRACE is supposed to capture?

[julieshortridge]

Cool find Connor! I haven't had a chance to read through in detail, but Rob your most recent point about GRACE is spot on. The GRACE measurements are just indicative of changes in mass that influence the gravitational pull that the satellite can sense - so assuming all other things stay constant at a given location, changes in the water stored is probably the main driver of mass changes. Which is awesome and amazing (imo), but from what I understand, doesn't allow it to distinguish the depth at which those changes are occurring, and in fact most of the research I've seen using GRACE data is focused on changes in deep aquifers. So I imagine that could very much conflate the relationship with stream discharge.