Assessing the potential hydrological impacts of hydraulic redistribution in Amazonia using a numerical modeling approach



[1] Most evidence for hydraulic redistribution is from ecosystems in relatively dry regions. Recent data indicate that hydraulic redistribution (HR) may also exist in the central Amazon forest. Assuming that HR can take place in all plant types in the Amazon region, this numerical modeling study examines how the hydrological impact of HR varies spatially and temporally. HR influences transpiration and total evapotranspiration the most in places and during seasons of intermediate soil wetness. Although HR increases the long-term mean of dry season transpiration, it can reduce transpiration toward the end of the dry season in extremely dry years when the HR-induced acceleration of moisture depletion leaves less water available later in the dry season. Deep roots may, however, mitigate some of this negative impact. This HR-induced reduction of water availability is contrary to the general notion of HR increasing plant water availability; the spatial and temporal variation of the HR impact documented in this study may help interpret field observational data and locate future field experiment sites to evaluate the HR hypothesis in the Amazon region.

1. Introduction

[2] Hydraulic redistribution (HR) refers to the transport of moisture by plant roots from wetter to drier portions of the root zone. Field measurements of sap flow in taproots and lateral roots of trees have demonstrated that roots can move water upward, downward, or laterally from moist areas to drier areas [e.g., Burgess et al., 1998, 2001; Scott et al., 2008]. Past studies have documented the existence of HR in various climate regimes and ecosystems, mostly in relatively dry regions [e.g., Richards and Caldwell, 1987; Dawson, 1993; Caldwell et al., 1998; Jackson et al., 2000; Burgess et al., 1998, 2001; Brooks et al., 2002; Scott et al., 2008]. Species in the wet tropics, including three tree species common to the Tapajos forest in the middle of Amazonia, have recently been added to the list based on sap flow measurement [Oliveira et al., 2005]. However, soil moisture data at a nearby site show no clear evidence for HR [Romeo-Saltos et al., 2005; Bruno et al., 2006]. The existence of HR in the Amazon region and how widespread a phenomenon it might be at the continental scale remains an open question.

[3] HR occurs when and where there is a gradient of soil water potential within the rooting zone and there are conductive roots across this gradient. It is generally considered that HR helps buffer plants against seasonal drought when (and where) water is not freely available. It appears that HR typically increases dry season transpiration by 20–40% [e.g., Dawson, 1993; Jackson et al., 2000; Ryel et al., 2002; Ren et al., 2004], but this magnitude of increase can exceed 100% under extreme circumstances [Amenu and Kumar, 2008]. In addition to enhancing plant water availability, HR may also influence nutrient availability [Caldwell et al., 1998], as upward HR enhances root absorption from shallow soils where nitrogen is more abundant.

[4] Although HR is not yet routinely included in land surface or climate models, several studies have documented the significant impact of HR in such models. Hypothesizing that HR could occur in all plant types over the globe, Lee et al. [2005] incorporated the HR scheme of Ryel et al. [2002] into the NCAR Community Atmosphere Model (CAM3) and found that it significantly increases photosynthesis and transpiration at the global scale, and lowers surface temperature. Such effects are especially noticeable in tropical forest regions. Zheng and Wang [2007] incorporated the same HR scheme into two land surface models, the NCAR Community Land Model (CLM3) and the Integrated Biosphere Simulator (IBIS2), and tested the models against observations at the ABRACOS Reserva Jaru site (at 10°05′S, 61°55′W) in Amazonia. They found that HR improved the performance of land surface models in reproducing the observed latent heat fluxes during the dry seasons of 1992 and 1993, reducing a dry bias commonly seen in land surface models over Amazonia. Baker et al. [2008] found that the combination of several factors including hydraulic redistribution is needed for the Simple Biosphere model (SiB3) to reproduce the observed seasonal cycle of water and carbon flux exchanges in the Amazon forest.

[5] Beyond the general notion of HR increasing dry season transpiration, little is known regarding how the impact of HR may depend on climate characteristics. For example, where and when do we expect to see the most substantial impact of HR on the hydrological cycle? Does HR always increase plant water availability? Addressing such questions may help understand results from previous field experiments related to the existence of HR in the Amazon region [e.g., Oliveira et al., 2005; Romeo-Saltos et al., 2005; Bruno et al., 2006], and help select optimal sites for field experiments to further evaluate the HR hypothesis. This study examines the potential ecohydrological effects of HR at the regional scale in the Amazon region using a numerical modeling approach, assuming that all plants can employ HR.

2. Model, Data, and Experimental Design

[6] In this study the hydrological impacts of HR are investigated based on numerical experiments using the NCAR Community Land Model version 3 [CLM3, Dai et al., 2003; Oleson et al., 2004]. The model domain covers most of South America, including the Amazon region: 30°W–85°W and 35°S–10°N. The spatial resolution is 1 × 1 degree, and the time step in all simulations is 30 min.

[7] Driven with atmospheric forcing data, CLM3 simulates the land surface biogeophysical and physiological processes, and provides results on the state of the land surface (e.g., soil moisture and temperature, vegetation temperature) and fluxes from the land surface (e.g., evaporation, transpiration, sensible heat flux). CLM3 has 1 vegetation layer, 10 unevenly spaced soil layers summing up to the total soil depth (for which 3.4 m is the default), and up to 5 snow layers depending on snow depth. Data for soil colors (which influence albedo) are from Zeng et al. [2002], which in turn are derived from Dickinson et al. [1993] with adjustments based on satellite data; and soil texture data are based on the International Geosphere-Biosphere Programme (IGBP) soil data set of soil mapping units and their sand and clay contents for each soil layer [Bonan et al., 2002]. Vegetation is represented by a combination of different plant functional types (PFTs) to account for differences in plant physiognomy, leaf shape and longevity, and photosynthetic pathway. Root profiles for different PFTs follow the work of Zeng [2001], which was based on a comprehensive global field survey data set. CLM3 considers a total of 16 PFTs, including both natural vegetation and crops; up to 4 PFTs can be included in each grid cell. Fractional cover and leaf area index (LAI) for each PFT are prescribed according to observational data from Moderate Resolution Imaging Spectroradiometer (MODIS) on EOS [Tian et al., 2004].

[8] Two versions of CLM3 are used in this study, the default version and the HR version that includes representation of HR. The default version is otherwise the same as the public release except for two modifications [Wang and Wang, 2007]: First, all drainage is set to derive from soil bottom drainage only, which reduces a known dry bias in the region [Bonan and Levis, 2006]. Second, the canopy interception scheme of Wang and Wang [2007] is adopted to more realistically estimate the canopy interception loss. The version of CLM3 including these modifications has been validated against observations [Wang et al., 2007] and has been used in several previous studies [e.g., Zheng and Wang, 2007; Alo and Wang, 2008a, 2008b].

[9] In addition to the vertical moisture transport via soil pore space (which operates both day and night as in the default model), the HR version of the models includes a parameterization for vertical moisture transport via plant roots at night (i.e., HR). The movement of water upward and downward through the root conduit follows water potential gradients [Richards and Caldwell, 1987; Caldwell et al., 1998; Burgess et al., 1998], which form the basis for the HR scheme of Becker et al. [1999] and Ryel et al. [2002] adapted here. This HR scheme was also used by Lee et al. [2005]. In this scheme, the HR-induced soil water flux between a giving soil layer (j) and a receiving layer (i) is estimated as a function of water potential difference between the two layers equation image, the maximum radial soil-root conductance (CRT, which is set to a constant value of 0.097 cm MPa−1 hr−1 with a sensitivity test described in section 3), the relative soil-root conductance for water in the giving layer (cj, which depends on water potential in the giving layer and ranges from 0 to 1), and root abundance Froot in the two layers:

equation image

[10] HR does not transport water to the soil immediately near the land surface because many shallow roots die in very dry soils [Ryel et al., 2002; Lee et al., 2005]. Therefore the HR-induced moisture flux to the topsoil layer (0–1.5 cm) is set to 0. Excluding the topsoil layer from participating in HR partially prevents the loss of redistributed moisture through soil evaporation.

[11] The parameter Stime in equation (1) controls the timing of HR, which is set to 1 at night and 0 during the day. Being a passive process driven by the water potential difference between plant roots and the surrounding soil, moisture effluence from roots can take place whenever the direction of the pressure gradient permits, although it is more prevalent at night. However, equation (1) parameterizes the flux through roots without explicitly simulating the root water potential. As a result, the modeled flux cannot automatically shut off or turn on according to the root water potential gradient. This necessitates the use of Stime to limit HR to nighttime only, which may lead to underestimation of the HR effects.

[12] On the basis of comparison with measured data, Ryel et al. [2002] found that the scheme in equation (1) realistically simulates the HR processes at a stand of A. tridentate in Utah, North America. For the Reserva Jaru ABRACOS site in Amazonia, Zheng and Wang [2007] found that including the representation of HR processes in CLM3 improved the model performance in simulating site-scale evapotranspiration (ET) during dry seasons. Details about the implementation of HR in CLM3 and the model performance can be found in the work of Zheng and Wang [2007].

[13] To investigate the potential hydrological impact of hydraulic redistribution in the Amazon region, two simulations are carried out: “noHR” using the default land surface model CLM3, and “HR” using the version of the model including representation of HR processes. Both simulations are 20 years long, driven with atmospheric forcing during the period 1985–2004 from Qian et al. [2006]. This data set is a combination of observation-based analyses and the NCEP–NCAR reanalysis data. The two simulations are initialized using the same soil moisture conditions that result from running the “noHR” simulation under the 1984 forcing for two years.

[14] Several additional pairs of “HR” versus “noHR” experiments are designed to test the sensitivity of the model results to model parameters, including the rooting depth, soil water potentials at saturation and wilting point, and the maximum radial soil-root conductance.

3. Results

3.1. Spatial and Seasonal Variations

[15] The hydrological impact of HR is assessed based on the difference between the “noHR” and “HR” simulations. Averaged at the annual time scale, HR increases the total ET almost everywhere within the model domain. This increase is largely balanced by a runoff reduction of similar magnitude (see Figure 1), with a slight difference between the two because of HR-induced differences in soil moisture storage. Relative to the surface water fluxes in the “noHR” simulation, the magnitude of HR-induced changes is rather small at the yearly average. However, with the seasonal migration of precipitation, the location of a strong HR signal varies. In any specific season, the HR-induced ET changes are larger in magnitude but smaller in spatial extension than the annual average (see Figure 2). During July–September (which is the dry season for most of the Amazon region) the impact of HR is the largest both in spatial extent and in magnitude, covering most of the southern Amazon basin and extending into the savanna region (cerrado).

Figure 1.

Annual averages of ET and runoff (a and b) based on the “noHR” simulation and the differences caused by (c and d) hydraulic redistribution, in mm/d, based on the 20-year average over 1985–2004.

Figure 2.

Seasonality of (left) total ET from the noHR simulation, (middle) HR-induced differences in ET, and (right) the wetness index from the control simulation. Units: mm/d.

[16] Since the gradient of soil water potential is the driving variable for HR and because of the lag between soil moisture and precipitation, the lack of precipitation is not a good indicator for where and when the impact of HR is the most substantial. Instead, we define as an indicator a wetness index using the average of root water availability among all soil layers, weighted by the fraction of root in each soil layer. The root water availability in each soil layer is parameterized as the difference between the actual soil matric potential and the wilting point potential, normalized by the soil matric potential difference between saturation and wilting point. The greatest impact of HR on ET occurs at a medium wetness level, as evident from the comparison between Figures 2 (middle) and 2 (right). The impact is negligible at both moisture extremes. Under very wet conditions, the soil water potential gradient (which drives HR) is usually low, and moisture depletion is needed to create a strong gradient to drive HR. So the impact on ET increases as the wetness index decreases from 1. However, severe desiccation will lead to a very dry condition with little water available to be hydraulically redistributed, and soil water potential gradient will diminish again as moisture in the whole soil column is depleted. Correspondingly, the impact of HR on ET decreases as the wetness index approaches 0. This relationship is further illustrated in Figure 3. The peak impact of HR on ET appears to occur at the wetness index level of approximately 0.4.

Figure 3.

Scatter plot of HR-induced latent heat flux differences (equation image, in W/m2) versus the wetness index during (a) January and February and (b) July and August.

3.2. Effects on Plant Water Availability

[17] HR influences the surface water budget through its impact on soil moisture dynamics. Not surprisingly, the impact of HR on canopy interception loss is negligible throughout the model domain. The impact on soil evaporation, however, can be substantial even though direct redistribution of water to the surface layer is not allowed. Water transport via roots to the soil layer immediately below the surface layer (1.5–4.1cm) influences the surface layer soil moisture through transport via soil pores and therefore influences soil evaporation. In the forest region, the HR-induced ET change is dominated by changes in transpiration, although changes in soil evaporation (usually of the same sign as transpiration changes but much smaller in magnitude) also contribute to the ET change. In the grassland region, however, the increase in soil evaporation is so substantial that it slightly reduces transpiration. The HR-induced increase of soil evaporation and decrease of transpiration in the grassland region is likely an artificial effect of the model deficiency related to an overly strong soil evaporation process in CLM3 [Oleson et al., 2007]. The negative impact of this model deficiency on transpiration is negligible in forest regions where, because of the thick vegetation cover, soil evaporation is a much smaller component of the surface water budget. The following analysis will therefore focus on forest regions.

[18] In the Amazon forest region, HR enhances transpiration during the dry season under most circumstances, but not always. Exceptions occur during extremely dry years. For example, Figure 4 compares the HR-induced transpiration changes in a drought year, 1998, with those based on the 19 year average (1985–2004 excluding 1998) during July–September, the season when the hydrological impact of HR is the most extensive (Figure 2). Judging by the long-term average, HR causes transpiration to increase during the dry season, consistent with the general notion that HR increases plant water availability. However, during 1998 over some areas in the southern part of the Amazon forest, HR caused transpiration to decrease.

Figure 4.

HR-induced differences in transpiration (mm/d) during July, August, and September, averaged (top) over the period 1985–2004 excluding 1998 and (bottom) during the 1998 extreme drought.

[19] The year 1998 was characterized by a severe drought across the Amazon basin, as a result of the strong 1997–1998 El Niño. The dry season in 1998 started with anomalously low soil water storage, because of the rainfall deficit during the 1997–1998 rainy season. In the Amazon forest boundary areas where water is not enough to sustain growth through the dry season, the HR-induced acceleration of soil moisture depletion leads to more water stress (and a decrease of transpiration) later in the dry season (Figure 4). In the relatively wetter areas in the interiors of the Amazon, the water storage at the beginning of the dry season together with rainfall during the dry season of 1998, although lower than the long-term mean, was still relatively abundant. In such areas, the HR-induced aggressive exploitation of root zone soil moisture still enhanced transpiration during the El Niño–induced 1998 drought.

[20] This contrast between areas of different transpiration response is further illustrated in Figure 5 using two grid points as examples. In Figure 5, although the grid point at (60°W, 10°S) receives more precipitation than the grid point at (70°W, 10°S), both are quite wet under the long-term mean hydrological conditions (2197 mm/y versus 2015 mm/y), and both are covered by evergreen forest. Mean transpiration peaks during the dry season at both sites, and the rainy season transpiration is lower at the site with more rainfall. This indicates that light and/or vapor pressure deficit, rather than water, is the limiting factor for transpiration in the wet season. It is therefore understandable that rainy season transpiration at both sites increases during the 1998 drought year relative to the long-term mean, since drought would increase both light availability and vapor pressure deficit (VPD). Note that although the precipitation difference between the two grid points is fairly small (approximately 10% during the rainy season), the difference in transpiration is much larger. Transpiration at the drier grid cell is approximately 70% higher than at the wetter grid cell, leading to hydrological regime differences that are larger than what the precipitation amount alone might indicate.

Figure 5.

Rainfall (bars), transpiration (lines), and HR-induced transpiration differences (bars) at two representative grid points in Amazonia, based on long-term mean (1985–2004) and based on the dry year 1998: (a) for a wet grid point at (60°W, 10°S); (b) for a somewhat drier grid point at (70°W, 10°S). Blue bars are for the mean, and red bars are for the dry year 1998.

[21] Despite the drought-induced increase of wet-season transpiration, transpiration in 1998 at the wetter grid point (60°W, 10°S) still peaks in the dry season, although the peak occurs slightly earlier than the long-term mean. In other words, the hydrological regime is still light-limited or VPD-limited during the rainy season and early part of the dry season in 1998. As the dry season progresses and water stress develops, transpiration increases in response to HR, and such increase is further enhanced by the 1998 drought. In contrast, at the drier grid point (70°W, 10°S), transpiration in 1998 is higher than the long-term mean during the rainy season and lower than the long-term mean during most of the dry season, as a result of increased sunshine (because of the decrease of precipitation, therefore cloudiness) in the rainy season and reduced water availability in the dry season of 1998. Such changes caused transpiration in 1998 to peak in the rainy season and dip in the dry season, leading to a more water-limited hydrological regime. Therefore, while HR causes transpiration to increase in the early stage of the dry season, such increase occurs at the expense of water availability later in the dry season. As a result, the HR-induced acceleration of moisture depletion reduces transpiration later in the dry season.

3.3. Effects on Soil Moisture Dynamics and Runoff

[22] Driven by the difference in soil water potential between the giving and receiving layers, both downward HR and upward HR can develop depending on the soil moisture distribution. During the rainy season, shallow soil layers are wetter, which favors downward HR; as the shallow soil dries up going into the dry season, upward HR develops. The HR-induced difference, therefore, can potentially vary from season to season. Still using the relatively wet grid point at (60°W, 10°S) and the somewhat drier grid point at (70°W, 10°S) as examples, Figure 6 examines the HR-induced changes of water depth in shallow soil layers (within the top 0.5 m) and in deeper soil layers (in the 0.5–3.4 m depth interval) as well as the resulting changes in surface water budget. At both sites during the dry season, shallow soil becomes clearly wetter because of upward HR. During the wet season, changes in shallow soil moisture are negligible at the wetter site and in most years at the drier site too, as soil stays at or close to saturation most of the time. Downward HR causes a significant decrease of moisture in shallow soil at the dry site only and during several years (1992–1999) of relatively low rain only. In the deeper soil at both sites, water availability is lower because of HR most of the time regardless of season. This decrease results from the competing effect of downward and upward HR accumulated over time. That is, the impact of HR on deep soil moisture is dominated by the effect of dry-season upward HR transporting water away from the deep soil. The effect of downward HR during the rainy season only partially offsets some of the reduction in soil moisture, causing a decrease of the magnitude of HR-induced soil moisture reduction as the rainy season progresses. Therefore, for the HR-induced deep soil water changes (in Figure 6b), the upward trend in the wet season and the downward trend in the dry season respectively reflect the effects of downward HR and upward HR.

Figure 6.

HR-induced differences in soil water and surface water fluxes at two representative grid cells in Amazonia (a wet grid point at (60°W, 10°S), and a somewhat drier grid point at (70°W, 10°S)), during 1985–2004. The y axis variables are equation image soil water equivalent depth; R and equation image surface runoff and its HR-induced difference; D and equation image—Soil bottom drainage and its HR-induced difference; Tr and equation image Plant transpiration and its HR-induced change.

[23] Due to the wet-season downward HR, surface runoff decreases at both sites, but the magnitude of change is very small (Figure 6c). Due to the large magnitude of soil water decrease in deeper layers, soil bottom drainage decreases (Figure 6d). Although surface runoff is the main contributor to the total runoff, the response of total runoff to HR is dominated by the response of soil bottom drainage. The response of transpiration and, consequently, of total ET to HR (Figure 6e) primarily follows the response of water availability in shallow soils, as roots are more abundant in shallow soils.

3.4. Sensitivity to Model Parameters

[24] Obviously, results from this numerical modeling study are subject to model uncertainties. Of the model parameters tested, none can change the spatial and temporal patterns of the HR impact. However, two parameters, the rooting depth and the maximum radial soil-root conductance (i.e., CRT in equation (1)), are found to substantially influence the model results.

[25] The HR-induced moisture flux is parameterized to be proportional to the radial soil-root conductance CRT, for which a default value of 0.097 was adopted from the Ryel et al. [2002] study. As the value of CRT increases from 0 to 0.2, the magnitude of the HR-induced hydrological changes increases but shows a tendency to level off as CRT further increases. For example, Figure 7a shows how the HR-induced transpiration impacts in two representative areas vary with CRT. The increase of CRT enhances both the negative and positive impacts of HR on plant water availability.

Figure 7.

HR-induced transpiration changes (in mm/d) in September, averaged over two representative areas, (69°W–72°W, 8°S–10°S) for the drought year 1998 and (59°W–61°W, 9°S–12°S) for the long-term average: (a) simulated with different values of CRT (the maximum radial soil-root conductance) and the default rooting depth 3.4 m; (b) simulated with the default value of CRT (0.097) and different rooting depths.

[26] The default rooting zone depth is 3.4 m. Two additional rooting depths, 6 m and 10 m, are experimented on, with a rooting profile resembling the deep root found at one Amazonian site [Nepstad et al., 1994]. Over places where the impact of HR on transpiration is positive, results are not sensitive to changes of the rooting depth (Figure 7b). However, the negative impact of HR weakens as the plant rooting depth increases, and even becomes slightly positive when the rooting depth is increased to beyond 10 m (with ˜20% of the fine roots located at 6–10 m depth) (Figure 7b). With limited data availability, it is not clear how widespread the deep roots are in the Amazonian forests. Although the amount of deep roots prescribed in this sensitivity experiment probably represents an overestimation [e.g., Nepstad et al., 2002; Belk et al., 2007], it shows that deep roots can at least partially mitigate the negative impact of HR.

4. Summary and Discussion

[27] Assuming that all plant types can conduct HR, this study assesses the potential hydrological impact of HR in the Amazon region based on numerical experiments using a land surface model. The main finding is that the general notion of HR enhancing plant water availability, therefore plant growth, does not always hold. As seen from long-term means, HR increases ET (therefore, increases latent heat fluxes and reduces sensible heat fluxes), and reduces runoff. Such HR-induced hydrological changes are the most substantial where and when the rooting zone soil wetness is at a medium level. The impact of HR in specific years, however, can deviate significantly from the long-term mean. Under the influence of extreme events such as the 1998 El Niño drought, in some areas of the Amazon, HR can cause an earlier onset of soil drought thus reducing dry season transpiration. Such negative impact of HR on transpiration can be potentially mitigated if a substantial portion of the plant fine roots reside at great depth (e.g., 8–10 m).

[28] This study examines the potential impact of hydraulic redistribution, assuming that HR can occur in all plant functional types across the South American continent. Discrepancy between different studies on HR in the Amazon forest [e.g., Oliveira et al., 2005; Romero-Saltos et al., 2005; Bruno et al., 2006] calls for comprehensive field experimental studies to test the hypothesis about the common existence of HR. Results from this study can help locate future field experiment sites that optimize the detection of HR signal in Amazonia. For example, the sites of the Oliveira et al. [2005] study, the Romero-Saltos et al. [2005] study, and the Bruno et al. [2006] study were all located in East-central Amazonia, near (55°W, 3°S), an area where hydrometeorological conditions are conducive to a relatively weak HR signal. The signal might have been much stronger if the experiments were carried out in the southern part of Amazonia, for example (Figure 2).

[29] Hydraulic redistribution together with a deeper rooting zone represents one potential mechanism that contributes to the high level of transpiration in the Amazon forests during regular (nonextreme) seasonal droughts. Groundwater capillary rise is another mechanism that can potentially contribute to the water supply for dry season transpiration in the Amazon forest [Fan and Miguez-Macho, 2010]. Without clear observational evidence from the field, however, the extent to which these mechanisms do contribute to dry season transpiration remains hypothetical. Focused field studies are critically needed to determine the source of water for dry season transpiration in the Amazon region. This is important for improving our understanding of the Amazon hydrological cycle and for making reliable predictions about its response to future climate changes.

[30] The HR-induced changes in the timing and severity of seasonal drought found in this study can potentially have significant impact on water-regulated plant competition. In most years, HR can support vegetation growth into the early dry season, leading to a prolonged growing season for drought deciduous trees. Without completely eliminating the soil drought, the HR-induced exploitation of moisture would leave less water available for the rest of the dry season during which evergreen trees have to survive while drought deciduous trees are dormant. As a result, drought deciduous trees may benefit more from HR than evergreen trees. The potential ecological impact of the HR-induced hydrological changes is tackled in a follow-up study, Wang et al. [2011].


[31] This study was supported by funding from NASA (NNG04GQ01G), NSF (ATM 0531485), and the University of Connecticut Center for Environmental Sciences and Engineering. Zhe Zheng helped to set the model up. Constructive comments from four anonymous reviewers are greatly appreciated.