Evaluating the influence of watershed moisture storage on variations in base flow recession rates during prolonged rain-free periods in medium-sized catchments in New York and Illinois, USA



[1] When dQ/dt-Q plots of stream recession are constructed for individual recession events, the slopes of the dQ/dt-Q curves are near constant in log space, but the intercepts vary seasonally. Because the intercepts increase during the summer (indicating an increase in the recession rate at a given discharge), it has often been assumed that increased evapotranspiration (ET) leads to increased recession rates. To test this assumption, we related the intercepts of dQ/dt-Q curves from 72 recession events to the concurrent ET and watershed moisture storage as determined from the National Center for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR) data set. The analysis suggests that at least for the nine watersheds from Illinois and New York we studied, shifts in recession rate during prolonged rain-free periods had little linkage to concurrent ET. Instead, we observe that the shifting has a moderately strong linkage to watershed moisture storage during the recession event. While this seeming lack of dependence on ET during these prolonged rain-free periods is not necessarily reflective of more normal conditions, we suggest it provides some insight into underlying subsurface controls at the watershed scale. In particular, we hypothesize that the shift in intercept in dQ/dt-Q curves results from spatial heterogeneities in watershed surficial geology; under dryer conditions near-stream subsurface zones with high hydraulic conductivities contribute most streamflow but under wetter conditions subsurface zones in upland areas with lower hydraulic conductivities also contribute.

1. Introduction

[2] Plots relating the rate of change in stream discharge (dQ/dt) versus the concurrent mean discharge (Q) have long been used to make inferences into the aquifer properties of watersheds [Brutsaert and Nieber, 1977]. New work has investigated the possibility of considering dQ/dt-Q plots more broadly as a means to infer the storage-discharge relationship [Kirchner, 2009] or changes in the contributing channel network of a watershed [Biswal and Marani, 2010]. Traditionally, a key limitation to interpreting dQ/dt-Q plots has been the large amount of scatter among the data points. To date, different investigators have considered different approaches for dealing with the scatter: looking at the envelope of the data cloud, fitting a best-fit line though the middle of the data cloud, or binning data with similar Q and fitting a line to the collection of mean dQ/dt values within each bin. As an alternative to these approaches, Shaw and Riha [2012] and Biswal and Marani [2010] examined individual recession events within the data cloud. When viewed in this way, it becomes apparent that the data cloud largely consists of multiple individual events that shift seasonally (Figure 1). Relative to the extrapolated curve of log(−dQ/dt) versus log(Q) during wet months (the bold black line in Figure 1), the log(−dQ/dt) versus log(Q) curves in the summer months shift upward. Thus, there appears to be more structure to the data cloud than previously thought, if the shifts can be explained.

Figure 1.

dQ/dt-Q plot for the (a) Chenango River, NY and (b) the Sangamon River, IL. Select recession events are highlighted with distinct symbols. Data periods from other recession periods are shown as gray circles, thus defining the data cloud more often used in recession analyses. The black line represents a curve with a slope of 2. During different times of the year, most recession events retain a slope close to 2 but shift upward from the black line.

[3] Shaw and Riha [2012] hypothesized that the shifts were due to variations in evapotranspiration (ET) during different recession events. Physically, increasing ET could be assumed to concurrently draw down the water available for recession flow and thus make the recession rate faster, as evident by a larger intercept value of event dQ/dt-Q curves during the summer time. This conclusion is consistent with much of the dominant thinking regarding controls on recession rates. One of the primary references cited to support the notion that ET can control recession rates is work done by Federer [1973] on small experimental watersheds in Hubbard Brook, NH. Using paired watershed data from forested and cut forest land, Federer [1973] observed that recession rates were much slower in the cut forest land (a proxy for low ET). Building on the conclusion of Federer [1973], Szilagyi et al. [2007] presented dQ/dt-Q recession data by month (but not by event) and demonstrated faster recession rates occur during the summer months. In using recession data to construct watershed storage-discharge relationships, Kirchner [2009] and Teuling et al. [2010] assumed ET directly influenced recession rates and selected night-time periods with low ET in which to analyze recession data.

[4] However, because ET is often strongly correlated in time with catchment scale moisture storage (i.e., annual variations in ET reach a maximum in the summer months at a similar time moisture storage may approach its lowest value), it is possible that variations in initial watershed-wide moisture storage (not ET) may actually be the primary control on variations in recession rate. Most past studies have not explicitly considered the control of moisture storage on recession rate (although people have often considered the relation between storage and discharge) and have thus not directly considered the possibility. In the case of Federer [1973], the cut forested land may have had a low ET, but the land also presumably had much greater stored moisture (as suggested by high pre-event base flow; see Federer [1973, Figure 2]) than the uncut forested land. In the case of Szilagyi et al. [2007], the shift in recession curves at a monthly scale was only assessed qualitatively and the shift could likely be related to moisture storage as much as ET intensity. In the case of the approach of Kirchner [2009] and Teuling et al. [2010], Teuling et al. [2010] present a graph of 2 day dQ/dt versus Julian Day at near fixed Q [Teuling et al., 2010, Figure 1] that indicates dQ/dt varies seasonally in accord with the annual sinusoidal pattern of ET. However, especially given that variation in dQ/dt does not diminish during nonsummer periods when ET variation does diminish, it would suggest the possibility that other factors may play a role in dQ/dt variations.

Figure 2.

Relationship between intercept of event dQ/dt-Q curves and mean evapotranspiration during recession period as determined from NCEP NARR data for watersheds in (a) New York and (b) Illinois. The light gray line indicates a power law function fit to all the data points in each region.

[5] There has been some limited work to formally assess whether the degree of moisture storage influences recession rates. Rupp et al. [2009] reported on individual recession curves that resulted from the simulation of the aquifer dynamics during periods when recharge from the vadose zone was negligible. The dQ/dt-Q curves for different simulated events were approximately parallel but offset horizontally due to differences in initial aquifer storage conditions at the start of the recession period. Additionally, careful examination of plots presented by Shaw and Riha [2012] indicate that individual recession curves from August to October are often shifted upward more than curves from May to July even though, on average, ET would be highest in early to midsummer in their region of study.

[6] The objective of this paper is to use a sample of watersheds from central New York and central Illinois, United States to directly assess whether streamflow recession rates are more closely related to concurrent ET intensity or concurrent mean watershed moisture storage during prolonger rain-free periods.

2. Methods

[7] The analysis assesses nine medium-sized watersheds that have been gaged by the US Geological Survey (USGS) (Table 1). Sites were selected in Illinois and New York in order to represent two different physiographic regions, albeit ones still in a humid climate. Additionally, within each state, sites were selected so as to be near a National Climate Data Center (NCDC) Climate Reference Network (CRN) meteorological station from which potential evapotranspiration could be estimated. The sites also needed to have gage records between 1980 and 2011 when National Center for Atmospheric Prediction North American Regional Reanalysis (NCEP NARR) data are available. Discharge was assessed at a daily time scale. The recession analyses were carried out similarly to Brutsaert and Nieber [1977]. In log-log space, the rate of change in discharge (dQ/dt) calculated as Qi − Qi−1 was plotted against mean discharge (Q) calculated as (Qi + Qi−1)/2 during the same period. Since many of the watersheds did not necessarily have precipitation gage measurements representative of the entire watershed area, nonprecipitation periods were inferred by using features of the dQ/dt values, similar to Brutsaert [2008] and Shaw and Riha [2012]. Specifically, recession events were identified by periods starting when dQ/dt was less than zero for four consecutive days, continuing as long as consecutive negative dQ/dt values were decreasing, and ending 2 days prior to dQ/dt values that were positive. As a check on the assumption of minimal precipitation, we evaluated precipitation at available National Weather Service Coop meteorological stations. In most cases, there was no precipitation measured at the meteorological stations. There were several instances of 1 day rainfall inputs of up to 5 mm, but these incidents were not apparent in the outflow hydrograph and were presumably on the same order of the initial abstraction necessary to initiate runoff or infiltration. Based on the meteorological station data, we also removed events subject to the possible influence of snow melt. Within the 9 watersheds, 72 recession events were identified between 1980 and 2011.

Table 1. Gage Locations, Basin Areas, and Summary Information of Event dQ/dt-Q Recession Curvesa
SiteStateUSGS Gage IDArea (km2)Number of EventsSlopeSurficial Geology
  1. a

    The “slope” column indicates the mean and standard deviation (in parenthesis) of the slopes of the event recession curves plotted as log(−dQ/dt) versus log(Q). The final column provides a brief description of each watershed's surficial geology. S and G is an abbreviation for sand and gravel.

Fall CreekNY0423400032382.36 (0.58)Near-stream S and G
Tioughnioga RiverNY01509000748112.26 (0.61)Near-stream S and G
Otselic RiverNY01510000376101.89 (0.64)Near-stream S and G
Chenango RiverNY0150500063782.32 (0.51)Near-stream S and G
Sangamon RiverIL0557091061472.21 (0.39)Near-stream S and G
Salt ForkIL0333690034332.12 (0.52)Near-stream S and G
Lake ForkIL0559080038181.51 (0.28)Uniform till
West Okaw RiverIL05591700287111.28 (0.91)Uniform till
Embarras RiverIL0334340047660.98 (0.25)Till and clay

[8] The comment is sometimes made that applying such restrictive criteria to identify base flow periods excludes the vast majority of the flow record. However, we suggest this is necessary to insure that one is truly looking at base flow. In humid climates where precipitation is frequent, extended periods without precipitation that come the closest to generating true base flow are rare. Thus, using sustained periods of no precipitation necessitates excluding much of the discharge record. There has been recent acknowledgment that base flow remains an ambiguous term [Price, 2011]; in looking only at sustained rain-free periods, we possibly apply a more narrow definition of base flow than often used in other contexts.

[9] Moisture storage and ET in the watersheds during the recession events were determined from the NCEP NARR data set provided by the National Oceanic and Atmospheric Administration Office of Atmospheric Research/Environmental Science Research Lab/Physical Sciences Division (OAR/ESRL PSD), Boulder, Colorado, United States, (http://www.esrl.noaa.gov/psd/, last accessed: 11 September 2012). In NCEP NARR, ET and soil moisture are calculated using the Noah model [Chen et al., 1996]. The average ET and average moisture storage during each recession event were determined from daily values of latent heat flux and daily values of the percentage of maximum storage in the top 100 cm of soil, respectively. NCEP NARR data are provided at a 32 km by 32 km square grid cell and thus provide a measurement representative of a relatively large spatial area. Since the grid cells do not match the watershed boundaries, we used the spatial average from the three to five NARR grid points that are in the approximate locale of each watershed. Soil moisture in the top 100 cm of soil does not directly give total watershed storage, but we are primarily interested in a qualitative measure of relative wetness, which the top 100 cm moisture storage presumably provides. As will be discussed later, we are interested specifically in watershed moisture storage in so much as it acts as a proxy for the extent of active contributing area to streamflow. NCEP NARR soil moisture has been previously used for watershed-scale analyses of recession by Krakauer and Temimi [2011].

[10] As a check on the reasonability of the NCEP NARR latent heat flux, we also calculated potential evapotranspiration (PET) from hourly net solar radiation and temperature available from 2004 to 2011 at NCDC CRN stations using the Priestly-Taylor equation. PET values approximated from the CRN stations were closely correlated to the ET determined from NCEP NARR data (Pearson's correlation coefficient of 0.95 in Illinois and 0.86 for New York). The absolute magnitude of the NCEP NARR ET was 37% lower than Priestly-Taylor PET for ET values >3 mm d−1 in Illinois and was 21% lower for ET values >5 mm d−1 in New York (see figure in supporting information). Values of PET and NCEP NARR ET are similar when below the stated thresholds. This moderate divergence between PET and NCEP NARR ET at higher PET is not unexpected and indicates that NCEP NARR ET can capture some physically realistic diminishment in ET due to root zone moisture limitations. In part, because the NCEP NARR data provide a longer record, in the remainder of the paper, we only focus on the ET estimates provided by the NCEP NARR data.

[11] For the nine watersheds, the slope of individual recession events are relatively constant across the events analyzed here (see Figure 1 and supporting information), similar to past analyses [Shaw and Riha, 2012; Biswal and Marani, 2010]. Because, we are primarily interested in the degree of shift and because the slope of individual dQ/dt-Q curves are relatively uniform, for each watershed, the slope is fixed and the intercept is determined by minimizing the sum of squared errors for a linear regression line. On the Lake Fork River, the slope is fixed at 1.5. On the Embarras River and West Okaw River, the slope is fixed at one. On the remaining rivers, the slope is fixed at two. The magnitude of the y intercept indicates the degree of shift in each individual recession curve. A larger y intercept value indicates a greater recession rate. On a plot of log(−dQ/dt) versus log(Q) such as in Figure 1, the y intercept can be visualized as the height of the recession curve where Q = 1 (i.e., log(Q)=0). Intercept values were assessed against NCEP NARR latent heat exchange and NCEP NARR percent maximum soil moisture as estimated in the top 100 cm of the land surface.

3. Results

[12] All four watersheds in New York exhibited shifts in dQ/dt-Q curves, consistent with prior observations [Shaw and Riha, 2012]. Of the five watersheds analyzed in Illinois, three (Sangamon River, Lake Fork River, and the Salt Fork) exhibited shifts similar in magnitude to those seen in the New York watersheds, and two exhibited virtually no shifting (West Okaw River and Embarras River). The very behavior of the watersheds in Illinois provides some initial indication that ET (which should be quite similar in magnitude and seasonal pattern among nearby watersheds) may not be a control on recession rates. However, a lack of apparent relationship between ET and recession rate in one watershed (as evident by a lack of shift) does not eliminate the possibility that the shifting in another watershed is due to ET variation. Thus, it is still necessary to assess the role of ET on recession rates in each individual basin. However, because of the lack of shifting on the West Okaw River and Embarras River, we have not included these rivers in the remainder of our analysis so as not to prejudice the results found in the other watersheds. An explanation for the differences in shifting among the watersheds will be offered in the discussion.

[13] In New York, the intercept versus ET plot (Figure 2a) indicates that the intercept gradually increases with ET but that there is a large amount of variability at high ET values. In Illinois, there is little apparent relationship between intercept and ET (Figure 2b). Power law regression lines fit to the collection of intercept and ET data from all the watersheds in each region had R2 values of 0.34 and <0.01 for New York and Illinois, respectively. If moisture storage is instead considered as a possible explanatory variable of the intercepts, one sees a reasonably strong relationship between moisture storage as predicted by NCEP NARR, especially in Illinois (Figure 3). Power law regression lines fit to the collection of intercept and moisture storage data from all the watersheds in each region had R2 values of 0.44 and 0.84 for New York and Illinois, respectively.

Figure 3.

Relationship between intercept of event dQ/dt-Q curves and % maximum soil moisture content in the top 100 cm of the soil profile during the recession period as determined from NCEP NARR data for watersheds in (a) New York and (b) Illinois. The light gray line indicates a power law function fit to all the data points in each region.

[14] The moderate relationship between ET and intercept in New York is not surprising given the fact that ET intensity is often correlated (albeit, inversely) in time with moisture storage. Using the ET and moisture data pairs from the recession events analyzed here, the Pearson's correlation coefficient was −0.49 when anomalous values (as discussed below) were neglected. Thus, instead of simply looking at the ability of the statistical model to explain variability across all the events, we suggest that the recession events that provide the most valuable insight into the role of ET and storage may occur during anomalous periods when ET and storage values depart from their standard relationship of higher ET and lower storage and lower ET and higher storage (as shown in supporting information). We have indicated two such anomalous recession events in both Figures 2 and 3. On 27 August 2010, there is only moderate ET and very low watershed moisture storage. On 7 July 1984, there is high ET and relatively high moisture storage. Especially during these anomalous periods, soil moisture storage demonstrates a greater ability than ET to correctly predict the degree of shift in the dQ/dt-Q intercept.

4. Discussion

[15] Reviewers raised the relevant question whether it is possible shifting of the intercept is actually due to some interaction between ET and Q. In particular, if Q is a function of watershed moisture storage (S), at high S values, Q would be the dominant flux depleting S, and dQ/dt would be primarily dependent on Q. At smaller S values, one would be more likely to find ET > Q and the depletion of S, reduction in Q, and magnitude of dQ/dt more dependent on ET. Thus, it is possible that only at high ET/Q ratios is the role of ET apparent. We tested this possibility by plotting ET versus intercept only for data points with high ET/Q as well as by plotting ET/Q versus intercept for all data points (similar to Figures 2 and 3). ET versus intercept for only high ET/Q ratios showed no explanatory relationship. ET/Q versus intercept indicated that ET/Q had some capacity to explain the change in intercept, but this was primarily controlled by 1/Q. A multivariate regression model applied to the two river systems where ET/Q seemed to have some explanatory capacity (R2 > 0.70 in Otselic and Chenango) indicated that 1/Q was statistically significant (p < 0.01) but ET was not (p > 0.80).

[16] The relationship between moisture storage and change in intercept was reasonably strong in the Illinois watersheds but weaker in the New York watersheds (Figure 3). Given that the NCEP NARR data set applies uniform soil properties over a 32 km by 32 km square grid cell, it is possible that this simplification of spatial heterogeneity in soil properties could not adequately represent moisture storage across New York. Central New York's undulating topography would be expected to have greater spatial variability compared to the flatter terrain of Illinois. However, there are alternative proxies for soil moisture that do not rely on reanalysis data and that may be able to better represent changes in watershed moisture storage. For instance, as seen in evaluating ET/Q, 1/Q seemed to provide some reasonable explanation of changes in intercept. In refining this use of Q as a predictive variable, prior work has suggested that a measurement of streamflow prior to a storm event peak in the hydrograph can be representative of watershed-scale moisture storage [Shaw and Walter, 2009; Tuttle and Salvucci, 2012]. While watershed moisture storage will in part be influenced by rainfall input during the storm event prior to recession, it is reasonable to assume that storm event precipitation (∼20 mm) is in most cases at least an order of magnitude smaller than watershed moisture storage (∼500 mm). Drawing on intercept data from the post-1980 recession events plus additional events dating back to the 1930s, Figure 4 indicates that the variation in intercept can be closely related to watershed moisture storage as approximated by using streamflow prior to the storm event initiating recession (Qprior). As evident from past analyses using Qprior [Shaw and Walter, 2009], storage scales nonlinearly with Qprior, with stored water more rapidly diminishing at low Qprior values. Due to this nonlinearity, to clearly illustrate the relationship between Qprior and intercept at low Qprior, we plot intercept versus 1/Qprior. Biswal and Kumar [2013] have similarly observed a relationship between changes in intercept value and the mean stream discharge 2–8 days before the recession period.

Figure 4.

Relationship between intercept of event dQ/dt-Q curves and the inverse of the base flow prior to the rainfall event leading to the recession (Qprior). Qprior is a proxy for watershed moisture storage. Because Qprior is not reliant on NCEP NARR data, it includes recession events dating back to the 1930s.

[17] As indicated earlier, not all the watersheds we analyzed exhibited shifting in the intercept of the individual recession curves. While this would seem to support the idea ET does not strongly influence shifting (for ET did vary in these watersheds), it could also suggest that moisture storage may not control shifting (since moisture storage also varied). However, we suggest that this lack of shifting in certain watersheds instead provides some insight into how variations in moisture storage control shifting. In particular, we hypothesize that this shifting may be due to spatial heterogeneities in watershed subsurface geology. It is possible that during periods of low watershed moisture storage, most flow originates from near-stream areas. If these near-stream areas were composed of material with higher hydraulic conductivity than more upland areas in the watershed, when streamflow primarily originated from near-stream zones, the rate of recession would be faster, as is seen. While it does not seem this explanation has been extensively studied previously, there are several references that support such a possibility. In a review of low flows, Smakhtin [2001] specifically notes that some river systems can be fed in part by “near-surface valley-bottom storages” composed of alluvial fill, and tracer studies have noted the spatially varied nature of base flow contributions in certain watersheds [Tetzlaff and Soulsby, 2008]. At a smaller scale in the Panola Mountain Research Watershed, Hwang et al. [2012] have recently demonstrated the possibility of such a change in near-stream subsurface properties relative to more upland areas. An obvious next step would be to use a three-dimensional hydrogeological model to evaluate how spatial variations in subsurface geology could lead to shifting in recession curves. Papers exploring bank-storage have constructed scenarios with variations between near-stream and upland hydraulic conductivity [e.g., Chen et al., 2006]. These scenarios have led to shifting in the recession curve, albeit not under the exact same conditions we propose here.

[18] This explanation of changes in recession rate is generally consistent with the observed surficial geology of the different watersheds. In the nine basins we evaluated, seven of them had distinct shifting in dQ/dt-Q curves. As determined from surficial geology maps, six of these seven watersheds had readily discernible spatial heterogeneities with upland areas composed of glacial till and extensive sand and gravel deposits in near-stream areas (Table 1 and supporting information). The two watersheds with no shifting in dQ/dt-Q intercept (West Okaw River and Embarras River) were situated in areas with relatively uniform surficial geology with no distinct change in geology in near-stream areas. The last watershed—Lake Fork—does not hold to this explanation; Lake Fork displayed moderate shifting in dQ/dt-Q intercepts but geologic maps indicated uniform surficial geology. While there is a discrepancy in regards to the surficial geology for Lake Fork, identifying heterogeneities is something like looking for a needle in a haystack. If one finds the needle, there is conclusive proof for its existence, but if one fails to find it, one cannot be sure that it does not exist. The inconsistency at Lake Fork is due to a lack of heterogeneity, thus, we would suggest there is some reasonable chance such heterogeneities do exist but have just not been identified and mapped.

5. Conclusions

[19] Variations in streamflow recession rates during prolonged rain-free periods were better explained by NCEP-NARR estimates of concurrent watershed moisture storage during recession than by NCEP-NARR estimates of concurrent ET. Repeating the analyses with alternate measures of ET (Priestly-Taylor PET) and a proxy for watershed moisture storage (prerecession event streamflow) resulted in a similar outcome. This analysis suggests that at least in some locales, the impact of ET during these prolonged rain-free periods is primarily through its cumulative effect on moisture storage rather than being a direct, instantaneous control on base flow recession rates.

[20] This does not seem surprising considering the strict definition of base flow applied in this analysis. Since we only analyzed precipitation-free recession periods at least 4 days after the most recent precipitation event, it would seem likely that the root zone across most watersheds had drained below field capacity. Thus, there would be little likelihood of the root zone contributing sizable amounts of water to base flow at the same time that ET would be diminishing root zone moisture stores. The fact that ET may still influence recession rates through its cumulative effect on moisture stores is consistent with findings that aggregate ET rate can be reasonably estimated at longer time scales (monthly or yearly) from recession data [Szilagyi et al., 2007; Palmroth et al., 2010]. It is likely that this result does not hold over all geographic areas; there have been observations of diel fluctuations in low streamflow in certain locales [e.g., Wondzell et al., 2007; Kirchner, 2009] that certainly suggests ET has a direct and concurrent influence on streamflow in some places. Furthermore, there remain variations in slope of the dQ/dt-Q recession line as well as unexplained variability in the intercept that may ultimately be found to be attributed to interactions between ET and moisture storage.

[21] A lack of direct, instantaneous control of ET on base flow recession rates opens several interesting possibilities for new insights into hydrologic function. First, it suggests that at least at the large watershed scale, the construction of storage-discharge curves from recession data may be less sensitive to ET than previously implied [e.g., Kirchner, 2009], although this remains to be tested. Second, it suggests that the degree of shift among individual recession events can possibly be used to characterize spatially distributed features within a watershed. Namely, it is possible that a shift in dQ/dt-Q curves occurs because of spatial variations in the dominant streamflow contributing zones at different magnitudes of flows. Analyzing shifts in dQ/dt-Q curves may thus be a useful tool to characterize subsurface spatial heterogeneity within watersheds.


[22] We thank Brian Belcher for his technical assistance in accessing NCEP NARR data. Additionally, we wish to thank the associate editor and three reviewers for their constructive comments.