Effect of water table dynamics on land surface hydrologic memory

Authors

  • Min-Hui Lo,

    1. Department of Earth System Science, University of California, Irvine, California, USA
    2. UC Center for Hydrologic Modeling, University of California, Irvine, California, USA
    Search for more papers by this author
  • James S. Famiglietti

    1. Department of Earth System Science, University of California, Irvine, California, USA
    2. UC Center for Hydrologic Modeling, University of California, Irvine, California, USA
    Search for more papers by this author

Abstract

[1] The representation of groundwater dynamics in land surface models has received considerable attention in recent years. Most studies have found that soil moisture increases after adding a groundwater component because of the additional supply of water to the root zone. However, the effect of groundwater on land surface hydrologic memory (persistence) has not been explored thoroughly. In this study we investigate the effect of water table dynamics on National Center for Atmospheric Research Community Land Model hydrologic simulations in terms of land surface hydrologic memory. Unlike soil water or evapotranspiration, results show that land surface hydrologic memory does not always increase after adding a groundwater component. In regions where the water table level is intermediate, land surface hydrologic memory can even decrease, which occurs when soil moisture and capillary rise from groundwater are not in phase with each other. Further, we explore the hypothesis that in addition to atmospheric forcing, groundwater variations may also play an important role in affecting land surface hydrologic memory. Analyses show that feedbacks of groundwater on land surface hydrologic memory can be positive, negative, or neutral, depending on water table dynamics. In regions where the water table is shallow, the damping process of soil moisture variations by groundwater is not significant, and soil moisture variations are mostly controlled by random noise from atmospheric forcing. In contrast, in regions where the water table is very deep, capillary fluxes from groundwater are small, having limited potential to affect soil moisture variations. Therefore, a positive feedback of groundwater to land surface hydrologic memory is observed in a transition zone between deep and shallow water tables, where capillary fluxes act as a buffer by reducing high-frequency soil moisture variations resulting in longer land surface hydrologic memory.

1. Introduction and Background

[2] Land hydrologic processes can record previous atmospheric forcing anomalies and then manifest their effects the following season or year. Enhanced knowledge of this memory process can improve weather forecasting and climate prediction on seasonal-to-interannual time scales [e.g., Dirmeyer, 2000; Koster et al., 2000a; Koster and Suarez, 2001]. Positive soil moisture–rainfall feedbacks have been observed in the Illinois hydrologic network data [Findell and Eltahir, 1997; D'Odorico and Porporato, 2004], in Kansas [Eltahir, 1998], and across the United States [Koster et al., 2003]. Under changing climatic conditions, improved understanding of the effect of land surface hydrologic memory processes becomes even more important [Seneviratne et al., 2006a].

[3] The relationship between soil moisture and precipitation depends on both precipitation frequency and the time scale of soil moisture retention [Wei et al., 2008]. Delworth and Manabe [1988, 1989] and Manabe and Delworth [1990] found that temporal variations of soil moisture yield seasonal-to-interannual variability in the evaporation fluxes. Amenu et al. [2005] showed that interannual (annual) modes of climate variation impact deeper (upper) layer soil moisture. The variance of soil moisture with respect to atmospheric precipitation shows a lag of 2–3 months [Vinnikov et al., 1996; Entin et al., 2000], and the long-term (9 years) precipitation signal has strong covariability with soil moisture in the central and western United States [Castro et al., 2009].

[4] Although previous studies have emphasized the importance of soil moisture for atmospheric processes [e.g., Koster et al., 2000a; Dirmeyer, 2001; Koster and Suarez, 2001, 2003; Koster et al., 2004; Seneviratne et al., 2006b; Teuling et al., 2006; Zhang et al., 2008], the effect of progressively deeper soil moisture and groundwater on land surface hydrologic memory processes is not well understood [Wu and Dickinson, 2004; Amenu et al., 2005].

[5] Several studies have shown that deeper soil layers retain longer memory of precipitation anomalies than surface layers [e.g., Liu and Avissar, 1999a, 1999b; Wu et al., 2002a; Wu and Dickinson, 2004; Amenu et al., 2005]. Chen and Kumar [2002] showed that soil moisture storage has a significant influence on modulating the effects of climate variability on terrestrial hydrologic processes, i.e., stream flow. Through model simulation, Güntner et al. [2007] indicate that groundwater tends to have a larger contribution to the interannual mode of total water storage variations than to seasonal variations because of longer memory in groundwater aquifers. In addition, in a case study of Illinois, Yeh et al. [1998] have shown that both groundwater storage change and groundwater runoff are significant terms in the monthly water balance for shallow water table areas in humid climates. In fact, groundwater storage change variations in Illinois have the same magnitude as the soil moisture storage changes [Yeh et al., 1998; Rodell and Famiglietti, 2001].

[6] Since many land surface hydrologic processes are highly dependent on soil moisture, the effect of a shallow water table on the soil moisture distribution must be represented in land surface models (LSMs). However, most traditional LSMs used in climate modeling lack any representation of groundwater aquifers. Only recently has the representation of groundwater dynamics in LSMs begun to receive considerable attention [e.g., Famiglietti and Wood, 1991, 1994; Koster et al., 2000b; Liang et al., 2003; Maxwell and Miller, 2005; Yeh and Eltahir, 2005a, 2005b; Fan et al., 2007; Maxwell et al., 2007; Miguez-Macho et al., 2007, 2008; Niu et al., 2007; Lo et al., 2008]. These studies have shown the importance of representing shallow groundwater and its interaction with soil moisture in land surface hydrologic simulations. For regions with shallow groundwater, the distribution of soil moisture in the vertical profile is strongly dependent on water table dynamics [e.g., Salvucci and Entekhabi, 1995; Liang et al., 2003; Vivoni et al., 2007; Lo et al., 2010].

[7] An analytical framework has also been used to study the importance of a shallow water table to bare soil evaporation [Ridolfi et al., 2008] and ecosystem function [Laio et al., 2009]. Yuan et al. [2008] and Jiang et al. [2009] have shown that incorporating groundwater dynamics into a regional atmospheric model can improve precipitation simulations in the East Asian monsoon area and the central United States, respectively. In addition, the high sensitivity of simulated terrestrial water storage in LSMs to selected parameter sets can be reduced with the inclusion of a groundwater representation [Gulden et al., 2007]. Bierkens and van den Hurk [2007] have also demonstrated that groundwater convergence can impact the persistence of multiyear precipitation anomalies. Thus, deeper soil moisture and groundwater may play an even more important role in affecting climate because of their longer time scale memory relative to surface soil moisture.

[8] A critical depth of the water table exists where the groundwater can significantly affect the surface energy budget and evapotranspiration [Famiglietti and Wood, 1994, 1995; Kollet and Maxwell, 2008]. However, the effect of groundwater on land surface hydrologic memory has not been explored thoroughly. Does a critical zone exist where water table dynamics affect land surface hydrologic memory? Understanding the effects of groundwater on land surface hydrologic memory processes can help better quantify land-atmosphere interactions and climate feedbacks.

[9] In this study we investigate the effect of water table dynamics on land surface hydrologic memory using the National Center for Atmospheric Research Community Land Model (CLM 3.5, Oleson et al. [2008]). Model experiments and sensitivity tests are conducted to identify impacts of groundwater on land surface hydrologic memory, while power spectrum analyses are utilized to demonstrate the effect of groundwater on high-frequency temporal soil moisture variations.

2. Model, Experiment Setup, and Land Surface Hydrologic Memory

2.1. Brief Description of the Model

[10] The model used in this study is the CLM 3.5 [Oleson et al., 2008] with an unconfined groundwater aquifer model [Niu et al., 2007] and the simple TOPMODEL-based (SIMTOP) runoff scheme developed by Niu et al. [2005]. For detailed descriptions of the physics in the CLM 3.5, the reader is referred to Oleson et al. [2004, 2008] and Niu et al. [2005, 2007]. In this paper, only the groundwater recharge flux is briefly described.

[11] As discussed in Oleson et al. [2004] and Niu et al. [2007], the groundwater recharge flux in CLM 3.5 is described by Darcy's law:

equation image

where q [mm/s] is the soil water flux (negative upward), k [mm/s] is the hydraulic conductivity, ψ [mm] is the hydraulic potential, and z [mm] is the depth between the water table and the soil layer. The hydraulic potential can be separated into soil water potential (ψm) and gravitational potential (ψg). The reference level is in the soil surface, so ψg is equal to the depth (−z). In the previous version of the CLM (CLM 3.0), the lower boundary condition at the bottom of the lowest soil layer was universally prescribed as the gravity drainage flux (qg, drainage equal to the unsaturated hydraulic conductivity), as shown below:

equation image

[12] Compared to soil water flux computations in the CLM 3.0, the most significant modification of adding an unconfined aquifer model to the CLM 3.5 is the extra source of water from the groundwater. The water table is interactively linked to soil moisture model through the exchange of groundwater recharge (i.e., soil drainage flux) and capillary rise at the bottom of the soil column. The capillary flux (qm) can be described as follows:

equation image

[13] To explore the impact of groundwater on land surface hydrologic memory, we conduct two experiments: first, the default CLM 3.5 is used (with the groundwater component, hereafter called GW); second, the capillary rise (qm) is turned off (only the gravity drainage flux (qg) is allowed, hereafter called NO-GW). The same atmospheric forcing drives the two simulations with the same vegetation and soil types. When the capillary flux is activated, soil moisture will be affected. The modified soil moisture will then change the values of k and ψm, which will further affect the capillary flux according to equation (3) and the associated soil moisture. These interactions are initiated by the activated capillary flux in the GW run, which does not exist in the NO-GW run. Therefore, when compared to results without these interactions, the difference between the two simulations is attributed to the presence of groundwater.

[14] Niu et al. [2007] showed that the CLM usually needs to spin up for more than 250 years in order to obtain an equilibrium water table depth in arid regions. We found such a long spin-up to be computationally taxing at the global scale and adopted an efficient regression approach instead. In this approach, exponential decay functions are used to fit the time series of the first 100 years of water table depths at each model grid. An equilibrium water table depth is defined as the asymptote of the exponential decay function, greatly shortening a potentially far longer spin-up period. The equilibrium water table depths are saved as the initial conditions and used to initialize the model groundwater for the two 27 year experiments (1980–2006). After the equilibrium water table depth is found by the fitting approach, vertical soil moisture profiles can be determined from the soil moisture characteristic relationships by Brooks and Corey [1964]:

equation image

where θ [ ] is the soil moisture content, θr [ ] is the residual moisture content, θs [ ] is the soil moisture content at saturation (i.e., porosity), ψcf (mm) is the depth of the capillary fringe, and B [ ] is the pore size distribution index. The characteristic relationship has been adopted by previous studies [e.g., Sivapalan et al., 1987; Famiglietti and Wood, 1991] to estimate soil moisture under equilibrium conditions. When using the above equilibrium states as the initial conditions, 30 spin-up years, all with climatological (seasonal average from 1980 to 2008) forcing, are superimposed before the beginning of the 27 year (1980–2006) simulation in order to adjust the uncertain vertical soil moisture profile by using the Brooks and Corey [1964] relationship. The model is then run at 1 × 1 degree for the continental United States using atmospheric forcing from the Global Land Data Assimilation System [Rodell et al., 2004].

2.2. Land Surface Hydrologic Memory

[15] Several approaches can be used to measure land surface hydrologic memory (persistence), e.g., the value of the one-month-lag autocorrelation coefficient [e.g., Delworth and Manabe, 1988; Liu and Avissar, 1999a; Wu and Dickinson, 2004] or using daily soil moisture data to estimate the time (in days) required for the lagged autocorrelation coefficients to decay to some certain criteria [Dirmeyer et al., 2009]. Since the decay time scales are more representative than the value of correlation coefficients (A. Schlosser, personal communication), in this study, daily soil moisture output from 1980 to 2006 is used to compute the land surface hydrologic memory at each grid. Memory is defined as the number of days (τ) that the lagged autocorrelation coefficient of a soil moisture anomaly takes to decay to 1/e at a model grid cell. The value, τ, provides a single parametric measurement of memory for soil moisture. Note that in this study, we are interested in the monthly to subseasonal variability: memory greater than 90 days is assumed equal to 90 days. A time series of soil moisture anomalies are generated for each model grid by removing its climatological (27 year) mean from the daily time series.

[16] In CLM 3.5 there are 10 soil layers with increasing thickness from 0 to 3.43 m below the surface. In wet regions, the water table depth can be as shallow as 1 m, e.g., in the lower Mississippi River basin [Fan et al., 2007], and hence located within the soil layers of the model. In order not to include the shallow groundwater while computing land surface hydrologic memory, only the near surface (0–50 cm) soil moisture data are used. The upper 50 cm is represented as layers 1 to 6 in the CLM 3.5, and the water table rarely reached this height in the regions selected for detailed analysis (see below). We tested shallower depths (30 cm, 15 cm, and 7.5 cm), and memory results were found to be insensitive to these depth changes.

3. Results

3.1. Asymmetric Responses of Land Surface Hydrologic Memory

[17] Figure 1 shows the differences between the GW and NO-GW runs for upper layer (0–50 cm) soil moisture (volumetric soil water content). It indicates that in the case of GW, soil moisture increases for the continental United States because of the additional groundwater supply to the root zone. The result is consistent with Fan et al. [2007] and Niu et al. [2007], who showed that the soil moisture increases more in the eastern than in the western United States. A clear transition zone in the southern central United States is shown inside the white outlined box in Figure 1. This region is located close to a “hot spot,” defined by Koster et al. [2004] as a transition zone between wet and dry climates, where the atmosphere and land surface are strongly coupled. Hence hot spot regions are of particular importance in land surface hydrologic memory processes and provide an important testing ground to study the impacts of groundwater on land surface hydrologic memory. Figure 2 shows the difference in land surface hydrologic memory as persistence in days for the boxed region in Figure 1. Regions with negative (positive) values mean that the presence of groundwater has decreased (increased) land surface hydrologic memory as shown in zone 2 (zone 3). Zone 1 represents the area where groundwater has not changed land surface hydrologic memory. Figures 1 and 2 show that although the soil moisture increases, the GW run does not always lead to systematic increases in land surface hydrologic memory relative to the NO-GW run. Rather, a clearly asymmetric response of land surface hydrologic memory can be seen in Figure 2 along the zonal gradient from 105°W to 90°W. In the following sections we will focus on this region to explore what causes the asymmetric response, including why soil moisture increases with groundwater but not land surface hydrologic memory.

Figure 1.

The difference between the GW and NO-GW simulations for the upper layer (0–50 cm) soil moisture (volumetric soil water content). The white outlined box (33°N to 39°N and 106°W to 92°W) is selected for detailed analysis of land surface hydrologic memory.

Figure 2.

The land surface hydrologic memory difference (in days) between the GW and NO-GW runs over the selected box in Figure 1. The three black outline boxes indicate zones 1, 2, and 3, which are 7° by 2° (33°N to 39°N and 105°W to 104°W), 7° by 3° (33°N to 39°N and 102°W to 100°W), and 7° by 3° (33°N to 39°N and 97°W to 95°W), respectively.

3.2. Relationships Between Water Table Depth and Land Surface Hydrologic Memory

[18] The only difference between the two experiments (GW and NO-GW) is that the capillary flux is activated in the GW run, which allows for exchanges and feedbacks between groundwater and soil moisture. Figure 3a is the monthly time series of simulated capillary flux anomalies (the average seasonal climatology is removed) for the GW run for the three black outlined boxes in Figure 2. The capillary fluxes of the three regions exhibit significantly different temporal behavior, e.g., successively higher temporal variations can be observed from zone 1 to zone 3. The regions where groundwater increases (decreases) land surface hydrologic memory displays higher (smaller) temporal variance (Figures 3a and 3b). Famiglietti [1992] showed that deep water tables yield smaller capillary fluxes. Figures 3b and 3c further indicate that water table depth can affect capillary fluxes in the frequency domain, e.g., a deep water table has fewer high-frequency signals of capillary flux. Therefore, not only does the mean water table depth affect the magnitude of capillary flux, but the water table can also affect the time scale of the flux, resulting in different responses of land surface hydrologic memory. Notice that the capillary flux anomalies have higher-frequency signals than the water table does (Figures 3a and 3c). The capillary flux is described in equation (3) in terms of three parameters: k, ψm, and z (the distance between the unsaturated and saturated layers). Hydraulic conductivity (k) and ψm are a function of soil moisture, resulting in a higher frequency of the capillary flux than that of the water table.

Figure 3.

(a) The 27 year (1980–2006) monthly time series of simulated capillary flux anomalies (the average seasonal climatology is removed) for the GW run for the three black outlined boxes in Figure 2, (b) power spectral analyses of capillary fluxes, and (c) 27 year mean seasonal cycle of water table depth.

[19] Figure 4a shows the spatial distribution of mean (1980–2006) water table depth for the boxed region in Figure 1. A strong contrast between water table depths in the eastern and western part of the region is evident. This gradient is also apparent when considering the meridional mean (Figure 4b). It is mostly due to climate, i.e., precipitation is lower in the west and progressively increases to the east (not shown here). Figure 4c shows the land surface hydrologic memory difference between the two experiments. The asymmetric responses can be seen quite clearly in the meridional mean as well (Figure 4d). In general, no significant difference of land surface hydrologic memory between the two experiments is observed in the very shallow water table region, i.e., at 90°W. However, moving west, groundwater has increased land surface hydrologic memory with an increase in water table depth until the water table is deeper than 2 m (at 95°W). In other words, groundwater provides a positive contribution to land surface hydrologic memory. When the water table is between 2 and 5 m (95°W–100°W), a negative feedback loop develops, meaning that an increase in water table depth actually produces less memory than the shallower groundwater does. After the water table depth exceeds 5 m, groundwater has progressively lost its influence on land surface hydrologic memory, and no significant difference of land surface hydrologic memory between the two runs is observed in the deep water table region (at 105°W). The different behaviors of the meridional mean in Figures 4b and 4d indicate a nonlinear relationship between groundwater and the associated changed land surface hydrologic memory.

Figure 4.

(a) The spatial distribution of mean water table depth from 1980 to 2006 for the boxed region in Figure 1, (b) the meridional mean of Figure 4a, (c) the land surface hydrologic memory difference between the two experiments for the boxed region in Figure 1, and (d) the meridional mean of Figure 4c.

[20] The more detailed features of this relationship can be seen in Figure 5, in which water table depth is plotted versus the land surface hydrologic memory differences at each grid in the boxed region in Figure 1. The results shown in Figure 5 reflect the nonlinear relationship between water table depth and memory. Greater memory (i.e., positive memory difference) and strong correlations with water table depth are observed when the water table is located between 1.6 and 3 m. In this range, groundwater has increased land surface hydrologic memory. Moreover, as water table depth continues to increase, groundwater actually has decreased land surface hydrologic memory, and memory is reduced the most at about 4–5 m, suggesting a negative feedback of groundwater. When water table depth is deeper than 6 m, no significant differences of land surface hydrologic memory are observed between the two experiments, since the groundwater and soil moisture are decoupled [Lo et al., 2008].

Figure 5.

Land surface hydrologic memory difference versus water table depth for each grid in the boxed region in Figure 1.

[21] In addition, the double triangle pattern in Figure 5 indicates that groundwater has both positive and negative impacts on land surface hydrologic memory when water table depth varies from near surface to deeper layers. In fact, several studies [e.g., Famiglietti and Wood, 1994, 1995; Anyah et al., 2008] have reported on the nonlinear relationships between the water table depth and hydrologic fluxes (e.g., runoff and transpiration). On the other hand, Maxwell and Kollet [2008] have indicated that a strong relationship between water table depth and latent heat flux can be found between water table depth of 2 and 5 m, which has been identified as the critical zone in which the groundwater and soil moisture are strongly coupled [Kollet and Maxwell, 2008]. Vervoort and Van der Zee [2008] have also shown that capillary fluxes from groundwater can only influence soil moisture within a certain range of water table depths. The results shown in Figure 5 further underscore the importance of nonlinear responses of groundwater in land surface hydrologic memory and thus have important applications for groundwater-surface water interactions.

[22] We have also conducted sensitivity tests to explore the impacts of different equilibrium water table depths on the results of Figure 5. Three more simulations are performed, i.e., with the equilibrium water table depth at 1 m, 2 m, and 5 m deeper than that of the GW run. Sensitivity tests (not shown here) indicate that the double triangle pattern of Figure 5 for the 1980–2006 simulations still exists although the exact values of the critical water table depths, where groundwater has significantly affected land surface hydrologic memory, are different. Because an extra 30 spin-up years are applied in the model run, when using the equilibrium states (water table depths and soil moisture profiles) as the initial conditions, the impact of setting different equilibrium states on the results is minimized.

3.3. Power Spectral Analysis: Damping Processes in Soil Moisture Variations

[23] Wu et al. [2002a] used power spectral analysis to show that deeper soil layers exhibit more low-frequency signals because the high-frequency (2–3 months) signals of surface moisture variations have been damped. In other words, progressively deeper soil layers act as low-pass filters, resulting in increasing memory with soil depth [Wu et al., 2002a, 2002b]. Malamud and Turcotte [1999] have also shown that certain time series, e.g., self-affine time series, have a power law dependence of spectral density on data frequency, which can be used to quantify memory. Similar to Wu et al. [2002a], power spectral analyses are utilized in this study to show changes of soil moisture variations in the frequency domain and to demonstrate that an increase of land surface hydrologic memory corresponds to a reduction in high-frequency signals. The spectral densities are normalized by their standard deviations over the entire time series. In order to emphasize spectral characteristics in the high-frequency domain, the mean seasonal cycle of the soil moisture time series has been removed. In this study, we will focus on the changes of high-frequency signals in soil moisture, which occurs between 30 and 90 days in the spectral domain.

[24] Figure 6 shows the power spectral analysis of spatially averaged soil moisture for each of the three outlined boxes in Figure 2, for the two experiments. The power densities of soil moisture have significantly different spectral characteristics in each zone. Zone 1 represents the deep water table region to the west of the study domain, where groundwater has no significant impacts on land surface hydrologic memory as can been seen in Figure 4d; zone 2 is where water table depth varies from 3 to 5 m, where land surface hydrologic memory is reduced by groundwater; and zone 3 is the region where groundwater has increased land surface hydrologic memory. In Figure 6a, the spectral power density shows no difference between the two experiments, indicating no significant impact from groundwater on soil moisture temporal variations. A decrease in spectral power density indicates that high-frequency signals of soil moisture are reduced in the GW run, i.e., in Figure 6c, causing higher land surface hydrologic memory as seen in Figure 4d. In contrast, Figure 6b shows that high-frequency signals of soil moisture increase, resulting in a decrease in land surface hydrologic memory after adding groundwater.

Figure 6.

Power spectral analysis of spatially averaged soil moisture for each of the three outlined boxes in Figure 2 for the two experiments. Only the high-frequency (30–90 days) signal is shown.

[25] Wu et al. [2002a] found that the redness of the soil moisture spectrum increases with soil depth, which means that for deeper soil layers, the lower-frequency (interannual) spectrum has a larger power density, and the higher-frequency (monthly to seasonal) spectrum has a lower power density. Figure 6 indicates that under certain circumstances, groundwater aquifers can also act as low-pass filters to reduce high-frequency signals in the soil moisture variations, resulting in a longer land surface hydrologic memory. On the other hand, the groundwater can also perturb soil moisture variations, resulting in less land surface hydrologic memory.

4. Sensitivity Tests for Water Table Dynamics and Memory Differences

[26] In Figure 5, it is evident that if the water table is too deep (deeper than 6 m), groundwater and soil moisture are decoupled, and groundwater has no influence on land surface hydrologic memory. When the water table is between 6 and 3 m, groundwater begins to decrease land surface hydrologic memory. If the water table is between 3 and 1.6 m, groundwater can increase land surface hydrologic memory. In addition, different soil parameters or land types may also cause asymmetric memory behavior like that shown in Figure 5. For example, Teuling and Troch [2005] noted the significant impact of vegetation and soil type on soil moisture patterns. It is important to note that results presented here depend primarily on water table dynamics with a nonlinear relationship. As such, sensitivity tests performed in each location with the same soil parameters and land types can reveal whether changes in water table depth alone can affect land surface hydrologic memory. Such tests can isolate the dependence of the changed land surface hydrologic memory to the vegetation and soil types since the tests are conducted at the same location. Moreover, while our focus in this study is the difference between the GW and NO-GW runs, effects from atmospheric forcing can be eliminated (they are identical for both runs).

[27] A series of sensitivity tests are conducted on the basis of Figure 7, a schematic plot that is modified from Figure 4d. Three locations are chosen for analysis: point A at 105°W, where groundwater and soil moisture are decoupled; and point B at 99°W (point C at 95°W), where groundwater has most decreased (increased) land surface hydrologic memory. We hypothesize that as the water table becomes shallower at the decoupled point, land surface hydrologic memory will decrease, as shown by the gray arrows in Figure 7. However, the memory will not be influenced if the water table depth becomes deeper. For the point in which groundwater has most decreased (increased) land surface hydrologic memory, the memory should increase (decrease) whether the water table depth becomes shallower or deeper.

Figure 7.

Schematic of memory difference variations with mean water table depth: point “A” represents the decoupled region, point “B” represents the region where groundwater has decreased land surface hydrologic memory the most, and point “C” represents the regions where groundwater has increased land surface hydrologic memory the most. Gray arrows indicate how land surface hydrologic memory changes (increased, decreased, or not changed) with water table dynamics.

[28] Water table dynamics are significantly influenced by climate, i.e., precipitation and groundwater recharge. Hence, in order to study the sensitivity of land surface hydrologic memory to changes in water table depth, we perturb the precipitation rate from 10% to 500% of the original forcing for each time step at each location and explore how the climate-driven water table depth impacts memory. Since for each point the soil parameters and land types are the same, changes of land surface hydrologic memory can only be attributed to water table dynamics. Figure 8 indicates that when the mean water table depth for each location is altered in the model, land surface hydrologic memory responds as previously discussed and shown in Figures 4d and 5. Therefore, it can be concluded that the asymmetric response of land surface hydrologic memory is primarily due to the differences in water table dynamics at each location. While soil and vegetation type also impact groundwater levels, compared to climate-driven water table depth, both the soil and vegetation types have less of an impact.

Figure 8.

Sensitivity tests of land surface hydrologic memory difference (days) to water table depths for points A, B, and C in Figure 7. Red crosses represent the original run without perturbing the precipitation rate.

5. Discussion

[29] In this section, we discuss mechanisms of changing land surface hydrologic memory in different zones. The asymmetric responses of memory over a larger extent (the whole continental United States) are also presented.

5.1. Mechanism of Changing Memory

[30] Figure 9 is the scatter plot of daily capillary flux anomalies versus soil moisture anomalies of the GW run from 1980 to 2006 for the three zones, which highlights the mechanisms of groundwater impacts on land surface hydrologic memory. In zone 3 (red dots), Figure 9 shows that soil moisture changes closely follow capillary flux changes with a positive relationship. In other words, soil moisture and capillary fluxes are in phase with each other, which can prolong the duration of soil moisture anomalies. The high (low) soil moisture with higher (lower) capillary fluxes causes soil moisture to remain at high (low) values, leading to a general increase of memory of the surface soil moisture. That is, groundwater has a positive impact on land surface hydrologic memory. However, this is not the case in zone 2, in which the capillary flux anomalies vary from negative to positive with either positive or negative anomalies of soil moisture. Capillary fluxes cannot prolong the soil moisture anomalies; instead, soil moisture is perturbed by capillary fluxes, which increase the high-frequency signal of the soil moisture anomalies and thus reduce land surface hydrologic memory. Therefore, a negative feedback from groundwater on soil moisture is shown.

Figure 9.

Capillary flux anomalies versus soil moisture anomalies for each of the three outlined boxes in Figure 2 for the GW run in a daily basis from 1980 to 2006.

5.2. Asymmetric Memory Responses Over a Larger Extent

[31] Figure 10 shows scatter plots in which water table depth is plotted versus the land surface hydrologic memory difference at each grid for the whole continental United States, the southern half of the United States, and the northern half of the United States. As seen from Figure 10a, a nonlinear relationship between water table depth and land surface hydrologic memory difference still exists for the whole continental United States, but with higher noise. In general, most of those noisy points are from the northern United States as can be seen from Figure 10c. In other words, this nonlinear relationship is not globally applicable.

Figure 10.

Land surface hydrologic memory difference versus water table depth for each grid in (a) the continental United States, (b) the southern half of the United States, and (c) the northern half of the United States. (d) The two outlined boxes for the southern half and northern half of the United States.

[32] Usually strong land-atmosphere interactions do not exist everywhere; instead, it is observed in some specific local regions, e.g., the hot-spots defined by Koster et al. [2004]. In some places where the land and atmosphere are not strongly coupled, the effects of groundwater aquifer on the atmosphere will also be eliminated or very small. Previous studies [Famiglietti and Wood, 1995; Kollet and Maxwell, 2008] have reported the nonlinear relationship between soil moisture and evapotranspiration and capillary fluxes over the transitional zones. Therefore, it is not surprising that we saw the dramatic changes of land surface hydrologic memory responses to groundwater over this tiny region (southern central United States). Land surface hydrologic memory significantly depends on soil water content. If soil moisture is significantly varied, it will have impacts on the associated memory as well. In this study, we are not attempting to demonstrate a universal linear or nonlinear relationship between groundwater and land surface hydrologic memory globally; instead, we want to show the impacts of groundwater on changing land surface hydrologic memory over the climatic transitional zone where groundwater can further affect the atmosphere and climate.

6. Summary

[33] Although several recent land surface modeling studies have demonstrated the importance of water table dynamics, and various groundwater parameterizations have been developed, the impact of groundwater on land surface hydrologic memory has received little attention. Previous studies [e.g., Yeh and Eltahir, 2005a, 2005b; Fan et al., 2007; Niu et al., 2007] suggested that LSMs with a groundwater representation should increase land surface hydrologic memory because the low-frequency signals from groundwater aquifers can propagate to the soil moisture in the upper layers. In this study, we find that the presence of a groundwater aquifer can either increase or decrease land surface hydrologic memory. Interactions between soil moisture and groundwater determine whether memory increases or decreases and are primarily a function of water table depth.

[34] This study reveals that when groundwater is added to the land surface model used here, if the water table depth is between 1.6 and 3 m, then capillary fluxes from the water table result in a decrease in high-frequency power density. In this depth range, capillary fluxes from the water table act as a buffer by reducing high-frequency soil moisture variations, resulting in higher land surface hydrologic memory. In contrast, an increase in high-frequency soil moisture signals (relative to the simulation without groundwater) is observed when the water table is between 3 and 5 m. Thus, an asymmetric response of land surface hydrologic memory to the presence of groundwater is noted. When the water table is very deep (>6 m), groundwater and soil moisture are basically decoupled since the capillary fluxes are almost zero. No significant differences in land surface hydrologic memory for the two experiments can be seen in this situation, even though the deeper water tables generally exhibit lower-frequency temporal depth variations. When the water table is shallow (<1.5 m), there is no damping of high-frequency signals in the soil moisture variations. Hence, no significant effect of groundwater on land surface hydrologic memory is noted.

[35] Moreover, there is a transition depth at around 3 m, in which groundwater also has no impact on land surface hydrologic memory. The double triangle pattern in Figure 5 indicates that the changed land surface hydrologic memory with water table dynamics is a transitional process rather than a step function. At about 2.2 m (4 m) deep, groundwater has increased (decreased) land surface hydrologic memory the most. Therefore, the ability of increasing land surface hydrologic memory by groundwater becomes progressively less when the water table moves from 2.2 m to 4 m. As a result, 3 meters is not the threshold depth at which groundwater suddenly switches its role in land surface hydrologic memory from a positive to a negative effect or vice versa. The contour plot in Figure 4d can also reveal this transitional process.

[36] It is worth mentioning that, first, while the exact value of the critical water table depth and the exact locations of the three zones found in this study will vary with different LSMs and groundwater models, the major findings of the nonlinear relationship between water table dynamics and land surface hydrologic memory should not vary. Moreover, simulations in a fully coupled model land-ocean-atmosphere model could yield somewhat different results with further implications for land-atmosphere interactions. Second, during the winter season, snow can dominate land surface hydrologic memory [Liu and Avissar, 1999b]. In order to isolate the impacts of groundwater on soil moisture and memory, either the seasons must be separated during analysis, or regions with little or no snow impacts should be observed. The latter was followed here, while simultaneously exploring a memory hot spot region. Third, the nonlinear response of groundwater in land surface hydrologic memory not only existed in the southern central United States, but also in a much broader area of the southern half of the United States. This study, however, is not attempting to demonstrate a universal relationship between groundwater and land surface hydrologic memory; instead, it shows the impact of groundwater on changing land surface hydrologic memory over the climatic transitional zone where groundwater can further affect the atmosphere and climate in the hot spot regions.

[37] Finally, Wu et al. [2002a] demonstrated that low-frequency spectral power density increases in deep layer soil moisture because of damping processes: amplitude-damping results in a decrease in high-frequency soil moisture variations in progressively deeper layers. This is a top-down process, but the damping process can also occur from the bottom to the top. Downward damping is due to the decrease in signal intensity; however, groundwater acts as a buffer to modify the signal intensity. For example, here we show that the low-frequency groundwater signal propagates to the upper layers via the capillary flux, which is a bottom-up process that affects soil moisture variations. This study suggests that there are considerable nonlinear relationships regarding the interactions of land surface hydrologic memory and groundwater in LSMs. As seen in Figures 4, 5, and 6, the feedback can be positive, negative, or neutral, depending on water table dynamics. A region with a negative feedback will shorten the time scale of land surface hydrologic memory. Lower land surface hydrologic memory, as pointed by Mahanama and Koster [2005], tends to reduce the efficacy of soil moisture initialization and results in less accurate seasonal forecasts. Hence, it is important to incorporate the subsurface land hydrologic processes into LSMs and atmospheric general circulation models in order to better understand land surface hydrologic memory processes, which can help to improve weather forecasting and climate prediction on seasonal-to-interannual time scales.

Acknowledgments

[38] This research was sponsored by NOAA CPPA grant NA05OAR4310013 and the NASA Earth and Space Science Fellowship program NNX08AV06H.

Ancillary