Contrasting effects of soil development on hydrological properties and flow paths



[1] Runoff pathways strongly influence hydrologic and biogeochemical losses and landscape evolution. On an evolving landscape, soil development may alter hydrologic properties and thereby change through time the relative importance of various pathways. Here we report in situ soil water retention, unsaturated and saturated hydraulic conductivity, and flow path characteristics of a 300 year old Andisol and a 4.1 million year old Oxisol, located at the extreme ends of a soil substrate age gradient across the Hawaiian Islands. The two soils contrasted in depth and texture; the young soil was shallow and coarse textured, while the old soil was deep and highly weathered with a near-surface plinthite horizon overlying numerous clay-rich subsurface horizons. The young soil drained freely under modest suction, whereas subsurface clay horizons at the old site required significantly more suction to start to drain than the upper horizons. Similarly, saturated hydraulic conductivity (Ks) was high throughout the soil profile at the young site, whereas Ks was two to three orders of magnitude lower through the subsurface clay horizons than the upper ones at the old site. Irrigation experiments with deuterium tracer demonstrated that water was downward advecting at the young site, while water at the old site moved both laterally along the subsurface clay horizon contact and slowly downward through it. Rainfall frequency distributions indicated a high probability of rainfall events exceeding subsurface Ks values in old soil. In Hawaii the addition of dust influences the time evolution of soil, but the tendency for subsoil clay accumulation in older soils leading to alteration in hydrologic flow paths has been proposed in other environments. Our findings together suggest that as soils develop with time, subsurface horizon Ks values decline, impeding rates of vertical water flow but also increasing the importance of shallow subsurface lateral flow.

1. Introduction

[2] Soil development on stable landscapes leads to soil profile differentiation and changes in soil physical properties that may alter the rate and direction of water flow. The effects of this change in runoff magnitude and path may in turn have cascading consequences for coupled weathering, erosion and biotic processes that shape the landscape and its mantling soil and ecosystems [Wells et al., 1987; McAuliffe, 1994; Eppes et al., 2002]. In semiarid to humid environments, soil development, in the absence of significant erosion, typically leads to progressive increases in subsurface clay content [Birkeland, 1999]. Increased subsurface clay content with pedogenesis should lead to reduced saturated hydraulic conductivity with depth [Brooks and Richards, 1993] and the potential for impeded vertical flow and increased shallow subsurface lateral flow. Numerous field studies have reported impeded drainage, shallow seasonal water tables, and in some instances, vernal pools in more advanced stages of soil development and associated these changes with the formation of iron hardpans [Arkley, 1964; Gardner, 1967; Jenny et al., 1969; Jenny, 1980; Kitayama et al., 1997]. Other studies have related changes in runoff processes to clay dust accumulation or calcic horizon development with time [McDonald et al., 1996; McFadden et al., 1998; Eppes et al., 2002]. These observations have lead to several conceptual models that explicitly propose a shift from free to impeded drainage to explain pedogenesis [Sherman, 1950; Mohr et al., 1972; van Breemen and Buurman, 1998]. Development of soil physical structure, however, may lead to preferential flow and thereby maintain rapid vertical water movement in older soils [Moniz and Buol, 1982; Seyfried and Rao, 1987; Sollins and Radulovich, 1988; Southard and Buol, 1988; Vervoort et al., 1999].

[3] These studies linking soil development and flow paths have mostly relied on inferential analysis about the hydrologic controls. As coupled soil development and transport models begin to include geochemical transfer processes [e.g., Mudd and Furbish, 2004], it will be important to develop quantitative linkages between soil properties and hydrology as they evolve through time. Deterministic models will require measurements of hydraulic conductivity and water retention in field-structured soils, preferably in situ, for specification of soil hydraulic functions [Green et al., 1982]. Such coupling will also be crucial in understanding biogeochemical cycling and nutrient loss pathways, particularly in the humid tropics where soils have evolved on relatively stable geomorphic surfaces over long periods of time [e.g., Bonell and Balek, 1993; Lesack, 1993; Elsenbeer et al., 1994; Bonell, 1999; Lohse and Matson, 2005].

[4] A well-studied soil chronosequence across the Hawaiian Islands provides an opportunity to quantify the mechanisms linking soil evolution to runoff processes. Across this chronosequence, studies have documented systematic changes in soil characteristics (see Vitousek [2004] for extensive details), including dramatic increases in subsurface clay content with soil age [Chorover et al., 1999]. Initial field studies directed at evaluating solute fluxes from tropical forests located at the extreme ends of this chronosequence indicated potentially significant differences in vertical and lateral flow dynamics. Field-scale unsaturated and saturated zone hydrology and vertical solute fluxes from these forests receiving experimental N additions are reported by Lohse and Matson [2005]. Here we contrast the differences in the soil water retention and unsaturated and saturated hydraulic conductivity characteristics between the soil end-members of this chronosequence, a 300 year old Andisol and a 4.1 million year old Oxisol. We then use irrigation rainfall experiments with deuterium tracer to illuminate the effects of these differences on hydrological flow paths. Finally, we use 10 year rainfall data to determine the probability of rainfall rates exceeding soil surface and subsurface Ks values across these sites to predict changes in the relative dominance of Horton overland (HOF), saturation overland (SOF), and subsurface storm flow (SSF). Collectively, these observations quantify how development of subsurface clay-rich horizons leads to impeded drainage and to lateral flow over time.

2. Field Sites

[5] The two sites used in this study are apart of a well-characterized chronosequence in the Hawaiian Islands [Crews et al., 1995; Vitousek et al., 1997; Vitousek, 2004]. This developmental sequence consists of six sites with the age of the soil based on bedrock age, which ranges from 300 years to 4.1 million years [Clague and Dalrymple, 1987; Wolfe and Morris, 1996], and the corresponding soils vary from Andisols to Oxisols [Vitousek, 2004] (Figure 1).

Figure 1.

Study sites along the soil chronosequence in the Hawaiian Islands. Boxes indicate approximate locations of sites (not to scale). Inset topographic maps show the locations of the 300 year old soil site on the Big Island of Hawaii and the 4.1 million year old (Myr) soil site on island of Kauai, Hawaii. Topographic map scale is in kilometers. Elevation contours are in meters. The double inset boxes show approximate experimental plot locations with scale increased by 50%.

[6] Variation in other state factors [Jenny, 1941] such as parent material, topography, organisms, and climate that could affect soil development was minimized in site selection [Vitousek, 2004]. Soils are derived primarily from basaltic tephra overlying lava or a lava-ash mixture [Torn et al., 1997]. All sites are located on slopes less than 6%, and on topographic positions estimated to be the original constructional surfaces of the volcano [Macdonald et al., 1983], hence physical erosion and depositional influences are considered minimal [Crews et al., 1995]. Each site supports native, wet tropical rain forest dominated by Metrosideros polymorpha (80–90% of the basal area of species within plots at each site) [Kitayama and Mueller-Dombois, 1995]. None is known to have been cleared by humans, but as acknowledged by Crews et al. [1995], the possibility of human and natural disturbances like hurricanes cannot be dismissed. Finally, climate conditions are similar at the sites; modern mean annual temperature and precipitation are 16°C and 2500 mm, respectively. Over geological time periods, however, the older sites have experienced some differences in climate and vegetation [Gavenda, 1992; Hotchkiss and Juvik, 1993], probably subtle effects of surface erosion [Hotchkiss et al., 2000; Vitousek, 2004], and varying inputs of atmospheric dust [Jackson et al., 1971; Fox et al., 1991; Kurtz et al., 2001]. Despite these long-term fluctuations, environmental variation across this chronosequence has been controlled to an extent that is difficult to find elsewhere. Further details on the chronosequence are provided by Vitousek [2004], and a full discussion of the various violations of state factor assumptions at all the sites is available on the Web (

[7] The detailed hydrological studies, performed as a part of a larger project examining hydrologic N losses after N additions [Lohse and Matson, 2005], were conducted at the extreme ends of this developmental sequence, at the youngest, 300 year old site on the island of Hawaii and the oldest, 4.1 million year old site on the island of Kauai (Figure 1, inset maps) referred to as the young and old site. The young site is located on a gently sloping surface at an elevation of 1190 m near Hawaii Volcanoes National Park on Hawaii (19°25′N, 155°15′W) (Figure 1). The soil is classified as a Lithic Hapludand (Andisol) consisting of 200–400 year old tephra deposits composed of volcanic glass and short-range order minerals [Chorover et al., 1999]. An organic matter (O) horizon of 10 cm thickness overlies a 10 cm mineral (A) horizon that consists of cinder (gravel content of 59%) (Figure 2) (Soil Survey Staff, National Soil Survey Characterization Data,, hereinafter referred to as Soil Survey Staff, 2004). Below this layer, a Bw horizon transitions into several buried horizons (2-3Bwb) that extend down to a relatively unweathered basalt boundary (4R) at ~38 cm depth. The 4.1 million year old site is situated on a gentle sloping ridge top surface (<2%) at 1134 m elevation in the Na Pali-Kona State Forest Reserve on Kauai (22°08′N, 159°38′W). This ridge top surface is thought to be the remnant shield volcano surface [Macdonald et al., 1983] and is bordered by two deep valleys and their tributaries (Figure 1). Determination of whether the soils were derived from tephra or lava at this site has been difficult because of the weathered condition of the substrate, but the parent material is estimated to be 4.1 million years old. It is currently thought that this site (and the three other older sites in the chronosequence) developed primarily on late stage caps of postshield alkalic rocks [Kurtz et al., 2001]. The soil at this site is classified as a Plinthic Kandiudox (Oxisol) and composed primarily of secondary crystalline secondary minerals including halloysite, kaolinite, and crystalline Fe and Al (hydr)oxides [Chorover et al., 1999, 2004]. An O horizon overlies a weak, often absent, mineral horizon (A) to approximately 10 cm depth. The majority of plant roots are in these two surface horizons [Ostertag, 2001]. Below these layers, a Bc horizon, composed of plinthite or indurated peds with a strong medium subangular blocky structure (40% gravel content), overlies a reduced sandy-clay-loam oxic horizon (Bo1) at 20 cm depth. Cracks and macropores have been observed in the Bc horizon but not in the Bo horizons (n = 22 soil pits). Gleying at the Bo1 horizon contact was indicative of a fluctuating or perched water table, a key observation that prompted us to examine differences in hydrological processes between the two sites. More oxidized silty-clay-loam Bo (2–5) to silty-clay Bg horizons extend downward to the underlying weathered bedrock at a depth of 1.2 m (Crg). The bedrock is weathered to at least 5 m depth (possibly 10–20 m, Oliver Chadwick, personal communication).

Figure 2.

Soil profiles of 300 year old andisol and 4.1 Myr old oxisol. Soil horizon classifications and field texture data are from National Soil Survey Center (NSSC) Soil Survey Laboratory Research Database (Soil Survey Staff, 2004).

[8] Compared to the young site, the old site has obviously had a much more complex and long history of soil development. The soil at the old site has developed under varying climates and vegetation due to Pleistocene climate change and subsidence [Hotchkiss and Juvik, 1999; Hotchkiss et al., 2000]. On the basis of vegetation records, however, Hotchkiss and Juvik [1999] suggest that climate fluctuations in Hawaii may have been dampened relative to those experienced by the continents. Similarly, the old site has evolved under varying inputs of atmospheric dust due to climatic fluctuations [Rea, 1994], compared to the young site that has experienced only modern day climate and dust conditions. Dust accretion in soils in Hawaii has been shown to vary as a function of topography, rainfall, and time [Jackson et al., 1971], with dust inputs estimated at 1.25 g m−2 yr−1 along the study chronosequence [Kurtz et al., 2001]. While the young site is relatively dust free, dust accumulation can be found at the old site, predominately in the surface horizons. Dust accretion at this site, however, makes up a small portion of the total dust inputs (33,000–34,000 g m−2 quartz from dust); chemical weathering, biotic cycling, and some physical erosion, probably wind erosion during drier, full glacial periods, are thought to have limited its accumulation [Kurtz et al., 2001; Derry et al. 2005; Hotchkiss et al., 2000; Vitousek, 2004]. While some surface erosion at the older site may have occurred, the physical erosion rate has not been sufficient to prevent strong soil development at this site, and geochemical studies indicate progressive decay of the underlying bedrock, incorporation into the overlying soil, and collapse of the soil due to chemical losses [Kurtz et al., 2001; Vitousek, 2004]. Thus the attributes of the soil at the old site reflect the consequence of pedogenic processes operating over a long period of time, and hence we refer to it as the old site and attribute its character to be in large part associated with its long period of development. For this study, soils were grouped into the following three distinct soil genetic horizons: (1) O, (2) A, and (3) Bw and buried horizons at the young site and (1) O, (2) Bc, and (3) Bo horizons at the old site.

3. Experimental Design and Instrumentation

[9] At the young and old sites, we established four 10 m × 10 m plots within 0.25–0.5 km of each other and located near the main experimental plots used to characterize this chronosequence [Crews et al., 1995; Vitousek et al., 1997] (Figure 1, double inset maps). This design was motivated by the larger study evaluating the solute fluxes from these forests following experimental nitrogen (N) additions. Within each plot, we installed two tensiometers, one piezometer, and one time domain reflectrometry (TDR) probe at three different depths (representing the distinct soil horizons described above) (n = 24 tensiometers, 12 piezometers, and 12 TDR probes per site) (Figure 3). Tensiometers were aligned along the maximum topographic slope, in order to document lateral flow should it occur, and the TDR probes and piezometers were placed 0.5 m from the tensionmeters. TDR probes were positioned vertically to minimize disturbance to the plots. At the young site, tensiometers and piezometers were installed at mean depths of 10 cm (O horizon), 20 cm (A horizon), and 30 cm (Bw horizon), and TDR probes were installed vertically to 20 cm and 30 cm depths. At the old site, we established tensiometers and piezometers to 9 cm (O horizon), 22 cm (Bc horizon), and 49 cm (Bo horizon) mean depths and TDR probes to 20 cm, 30 cm, and 50 cm depths. We also established one well per plot, at 30 and 50 cm depth at the young and old site, respectively, to monitor the water table level (4 wells per site).

Figure 3.

Design for each nested set of hydrological instruments including tensiometers, piezometers, wells, and time domain reflectrometry (TDR) waveguides at three different depths (horizons) within each of the four plot locations. Note differences in soil depth scales between two soil sites.

[10] Tensiometers, piezometers, and wells were custom made according to methods described by Cassel and Klute [1986] and Reeve [1986]. For the measurement of pressure head (h), the energy status of the soil water in units of length (cm), we used a digital read-out pressure transducer [Marthaler et al., 1983] that had a short response time of approximately 5s [Torres et al., 1998]. The pressure transducer was monitored for drift by reading a laboratory-standardized tensiometer before and after each set of tensiometer readings. Pressure transducer measurements were found to be accurate within 1 cm. In order to determine hydraulic head (H), i.e., the sum of gravitational (z) and pressure (h) components, tensiometers and piezometers were surveyed using a Leica TC-1100 total station surveyor (Leica Geosystems, Heerbrugg, Switzerland) to account for within-plot elevational differences (z). More detailed descriptions on construction and installation of tensiometers and piezometers are provided by Lohse [2002].

[11] Volumetric soil water content (θv), the volume of water per volume of soil (m3 water m−3 soil), was manually measured with commercially made three channel waveguides from Dynamax® (Dynamax, Houston, Texas) connected to a Tektronix 1502B time domain refectrometry metallic cable tester (Tektronix, Wilsonwille, Oregon). We collected and interpreted data using Dynamax TACQ software which uses algorithms by Topp et al. [1980] and automated waveform interpretation by Baker and Allmaras [1990]. We tested the calibration of the algorithms by comparing TDR-measured θv at different soil depths (and therefore different textures) to θv estimated from the product of gravimetric soil water content (θg) and soil bulk density. Specifically, we measured θv by TDR to 0–20 cm, 0–30 cm, and 0–50 cm depths (at the old site) and then collected soils of known volume to these depths, approximately 5 cm away from the TDR probes, weighed and dried them at 105°C to determine bulk density and θg. We then wetted the soils and performed these same measurements. Across sites, soil types, and depths, the TDR measured θv agreed well with gravimetrically obtained θv (r2 = 0.95 and θv = [(1.003)(θg) + 0.00094] [Lohse, 2002].

4. In Situ Soil Water Retention Properties

[12] At each site, in situ soil water retention properties were measured by simultaneously monitoring volumetric soil water content (θv) and pressure head (h) responses under field conditions. We constructed soil water retention curves for different soil depths using selected dates where soils were initially near field saturated conditions (i.e., pressure head through soil profile was near zero) and where both pressure head and θv data were available. Pressure heads were averaged by depth within each plot and then across plots. To construct the dry end of the retention curves, we incorporated estimates of θv at −20 and −150 m of pressure from the National Soil Survey Center (NSSC) Soil Survey Laboratory Research Database (Soil Survey Staff, 2004). Where appropriate, we estimated θv for depth increments using the following general formula [Topp and Davis, 1985; Parkin et al., 1995]: θv(b−a) = [(a)(θv(a)) − (b)(θv(b))]/(a − b) where a and b are different soil depths and θv(a) and θv(b) are the corresponding integrated θv from 0-a cm and 0-b cm, and θv(b−a) is the θv for the interval depth b−a cm (e.g., θv(20–30)).

[13] Because of the minimum TDR probe length of 20 cm, our experimental design was not able to capture differences in soil water retention between soil genetic horizons in the upper soil profile. Therefore here we define the surface horizon (0–20 cm depth) to include the O and A horizons at the young site and O, A, and Bc horizons at the old site. The subsurface horizon includes the Bw and buried horizons from 20 to 30 cm depth at the young site and the Bo horizons from 20 to 50 cm depth at the old site. We refer to the surface and subsurface horizons hereafter.

[14] We derived soil water retention equations following methods by van Genuchten [1980]. We used the Mualem–van Genuchten [van Genuchten, 1980] model: θv = θr + (θs − θr)/[1 + (αh)n]m where θv is volumetric soil water content, pressure head (h) is understood to be positive, θr is the residual soil water content, θs is the saturated soil water content, α and n are curve fitting parameters describing the inflection point and slope factor respectively, and m = 1− 1/n. The reciprocal of α is thought to represent the air entry value, i.e., the suction at which the first pores empty [Mualem, 1976b, 1976a; van Genuchten, 1980]. Using a nonlinear least squares curve fitting approach with the statistical package, Sigmaplot® (SPSS, Chicago, IL, USA), we simultaneously estimated four independent parameters (θr, θs, α, and n) from the observed soil water retention data and used remedial measures to evaluate the precision and significance of the parameter estimates [Neter et al., 1996]. Results were considered significant at the P < 0.05 level.

[15] Soil water retention in the subsurface horizons differed markedly between sites of contrasting soil age (Figure 4). At the young site, surface and subsurface horizons appeared to drain under modest suction (Figure 4a). Nonlinear curve fitting indicated that soil water retention parameters, air entry values (1/α) and slope factors (n), were similar between these horizons (Table 1). Only residual soil water content, θr, appeared to be different between depths, probably due to differences in gravel content between these horizons, but high leverage associated with the NSSC Soil Survey −150 m pressure head data point indicated large uncertainty in this difference.

Figure 4.

Soil water retention curves for surface and subsurface horizons at (a) the 300 year old and (b) the 4.1 Myr old sites. Symbols display mean, and bars indicate standard error (n = 4). Solid lines display nonlinear regression curve fits with Mualem-van Genuchten retention equation [van Genuchten, 1980].

Table 1. Parameter Estimates for Soil Water Retention Curve Equation and Closed Form Equation for Hydraulic Conductivity of Unsaturated Soils From van Genuchten [1980]a
SiteSoil Depth, cmSEEr21/αnmθrθs
  • a

    Air entry value, 1/α, slope (n), residual soil water content (θr), and saturated soil water content (θs) were estimated using a least squares curve fitting approach [Marquardt, 1963]. Coefficient of determination (r2), standard error of estimate (SEE), and standard error of parameters are displayed (±1 SE). Note that m = 1 − 1/n. Letters in parentheses represent significant differences between depths determined by remedial measures.

300 years0–200.0170.8411.11 ± 10.03 (a)1.32 ± 0.18 (a)0.2390.23 ± 0.059 (a)0.49 ± 0.073
300 years0–300.0080.9715.61 ± 8.20 (a)1.21 ± 0.12 (a)0.1750.042 ± 0.094 (b)0.47 ± 0.14
4.1 Myr0–200.010.965.23 ± 2.55(a)1.40 ± 0.12 (a)0.2870.151 ± 0.03 (a)0.49 ± 0.056
4.1 Myr20–500.0070.99343.17 ± 69.31 (b)1.59 ± 0.16 (b)0.3740.332 ± 0.019 (b)0.560 ± 0.0284

[16] The surface horizon at the old site also drained under relatively low pressure head, but as expected, the subsurface horizon required significantly more pressure to drain and it retained significantly more water than the surface horizon at low and high pressures (Figure 4b and Table 1). Bootstrap methods confirmed that there were significant differences between depths in air entry value (1/α), saturated volumetric water content (θs), and residual soil water content (θr) (Table 2). The air entry value was very high in the subsurface horizon but also highly variable compared to other horizons at both sites (Table 1 and Figure 4). Complex intra-aggregate soil structure could probably explain this high entry value and variability [Tsuji et al., 1975]. Nonuniform wet-up in the natural field setting, error in the dynamic nature of air entrapment, inherent difficulties in measuring the wet region in the field [Green et al., 1982], and propagated error associated with the soil water content interval estimate may also have contributed to this variation.

Table 2. Comparison of Analytical and Bootstrap Means (n = 100) and 95% Confidence Intervals of Parameter Estimates for the van Genuchten [1980] Equation for Near-Surface (0–20 cm) and Subsurface (20–50 cm) Horizons at the 4.1 Myr Old Sitea
ParameterDepth, cmAnalytical MeanBootstrap Mean95% Confidence Intervals
  • a

    The mean is represented by x in the 95% confidence interval.

n0–201.401.411.31 < x < 1.50
n20–501.601.601.57 < x < 1.62
1/α0–205.235.477.48 < x < 3.52
1/α20–50343.2343.7332.1 < x < 356.9
θr0–200.1510.1510.118 < x < 0.176
θr20–500.3320.3320.336 < x < 0.329
θs − θr0–200.3350.3330.301 < x < 0.386
θs − θr20–500.2280.2280.225 < x < 0.231

5. Unsaturated and Saturated Hydraulic Conductivity

[17] At both sites we conducted a modified, unsteady unsaturated hydraulic conductivity experiment to obtain field estimates of unsaturated hydraulic conductivity (K) [Green et al., 1986]. We irrigated a 1 × 2 m area including a nest of tensiometers, piezometers, well, and TDR probes to a desired depth and then simultaneously measured θv and pressure head. These experiments were conducted for three hours or until there was no distinguishable change in head pressure or θv. Evapotranspiration was assumed to be insignificant compared to the downward flux of water over the short duration of the experiment.

[18] Measured values of saturated hydraulic conductivity (Ks) are required to predict the unsaturated hydraulic conductivity function, K(h), from soil water retention characteristics [van Genuchten, 1980]. Given that Ks estimates were not available for these sites, and that Ks estimates representative of field conditions were difficult [Arya et al., 1998] and unfeasible to obtain at our remote sites, we used an alternative field approach. For each soil horizon, we derived a van Genuchten [1980] relative unsaturated hydraulic conductivity function model, Kr(h), from the previously described soil water retention parameters and used field measurements of K to constrain the actual K(h). We then extrapolated to estimate Ks. Such field estimated saturated hydraulic conductivity (Kfs) values are typically higher than Ks values determined in a laboratory [Bouwer, 1966]. These Kfs estimates were compared to literature Ks values and other indirect approaches of estimating Ks, including those Clapp and Hornberger [1978] and Cosby et al. [1984] and field texture from Rawls et al. [1981].

[19] Figures 5a and 5b show that K drops rapidly with small changes in pressure head in both surface and subsurface horizons in the young soil. K in the surface horizon of the old soil also drops rapidly (Figure 5c), but in the subsurface horizon K was low and varied little with pressure head (Figure 5d).

Figure 5.

Field measured and predicted estimates of K(h) functions (cm h−1) for the (a) surface and (b) subsurface horizons at the 300 year old site and the (c) surface and (d) subsurface horizons at the 4.1 million year old site. Predicted estimates of K were determined from the van Genuchten [1980] equation and knowledge of soil water retention (Table 1) and constrained with field K and pressure head (h) measures.

[20] At the young site, field saturated water flow (Kfs) was high in both horizons, while Kfs at the old site was almost three orders of magnitude lower in the subsurface horizon than the surface horizon (Table 3). Table 4 shows a comparison of the Kfs estimated using the procedure described above with results from other indirect methods. In general, the van Genuchten Kfs estimates agreed well with these indirect estimates for both horizons at the young site while the estimate of Kfs for the surface horizon at the old site was higher than the other estimates. Our Kfs estimate of 36 cm h−1 seemed reasonable given the high root density of the O horizon as well as the high gravel content (40%) and subangular blocky structure of the Bc horizon. For the subsurface horizon at the old site, all Kfs agreed reasonably well with each other, and they were consistently two to three orders of magnitude lower than those for the surface horizon. Compared to other field-measured Kfs estimates of Hawaiian oxic soils [Green et al., 1982], the subsurface horizon at the old site drained one to two orders of magnitude slower. Higher observed conductivities in these other Hawaiian oxic soils may be due in part to alteration of soil physical structure or aggregation with agricultural activity.

Table 3. Field Saturated Hydraulic Conductivity Estimates (Kfs) for Different Soil Depths at the 300 year and 4.1 Myr Old Soil Sitesa
SiteSoil Depth, cmKfs, cm h−1
300 year old0–2010
300 year old20–3014
4.1 Myr old0–2036
4.1 Myr old20–500.08
Table 4. Comparison of van Genuchten [1980] Estimates of Saturated Hydraulic Conductivity (Kfs) to Other Indirect Methods for the 300 year and 4.1 Myr Old Sitesa
SiteSoil Depth, cmApproachKs, cm h−1
  • a

    Indirect methods include those of Clapp and Hornberger [1978] and Cosby et al. [1984] and field texture from Rawls et al. [1981].

  • b

    Field unsaturated hydraulic conductivity experiment used to constrain Ks, Kr = K/Ks, where Kr(h) = {1 − (αh)n−1[1 + (αh)n]−m}2/[1 + (αh)n]m/2 with a, n, and m estimates found in Table 1.

  • c

    Ks solved given unsaturated hydraulic conductivity estimates from field experiment and corresponding volumetric water content and pressure head (cm), where Ki = Kssi(2b+3) and where pressure head (si) and unsaturated hydraulic conductivity (Ki) are from field unsaturated hydraulic conductivity experiment and clay% = (b − 2.91)/0.159 × 100. Clay% are from Soil Survey Staff (2004).

  • d

    Ks = 0.0070556 × 10(−0.884+0.0153(%sand), where sand estimates are based on 100 − clay% derived from Clapp and Hornberger (b coefficient).

300 year old0–20van Genuchtenb10.0
300 year old0–20Clappc6.11
300 year old0–20Cosbyd4.77
300 year old0–20field texture7.95
300 year old0–30van Genuchtenb14.0
300 year old0–30Clappc2.59
300 year old0–30Cosbyd7.7–16.6
300 year old0–30field texture5.07
4.1 Myr old0–20van Genuchtenb36.0
4.1 Myr old0–20Clappc6.11
4.1 Myr old0–20Cosbyd6.6–8.82
4.1 Myr old0–20field texture4.8
4.1 Myr old20–50van Genuchtenb0.08
4.1 Myr old20–50Clappc0.09–0.15
4.1 Myr old20–50Cosbyd0.06–0.09
4.1 Myr old20–50field texture0.56

[21] Differences in conductivities between the surface and subsurface horizons at the old site suggested that a perched water table might develop at this subsurface permeability contrast when rainfall rates exceeded the drainage rate, an influence consistent with gleying at the Bo1 horizon. Field data collected under natural rainfall conditions, as a part of a larger field study evaluating solute fluxes [Lohse and Matson, 2005], showed that during and after large storm events, the soil profile approached soil water saturation with pressure heads near zero, and even slightly positive pore pressures were observed in the subsurface Bo horizon. Moreover, 20–25 cm of water collected in all four wells but not in the piezometers. These data indicated that water was draining rapidly through the surface layer, ponding at the clay contact between the Bc and Bo horizon and collecting in the wells, and then slowly draining through the clay horizon. Because the plots were widely distributed at the old site (Figure 1), these data suggested that a perched water table was potentially widespread across the ridge top flat. They also indicated potentially significant differences in vertical to lateral flow dynamics between the two sites.

6. Water Flow Paths

[22] We conducted field-scale irrigation experiments with deuterium tracer over previously installed soil solution lysimeters to quantify the effects of soil age on hydrological flow paths. The overall design consisted of irrigating at a constant rate, first for two days to attain quasi-steady flow conditions, then for six days with enriched in deuterium (and nitrate), and then for three more days with water to track the movement and fate of the enriched water (11 days total, 132 cm water). We performed these experiments on only one lysimeter per site due water limitation at these remote sites. Approximately 2000 L of water and supplies had to be helicoptered or otherwise backpacked into each site.

6.1. Lysimeter Installation and Experimental Design

[23] Prenart tension lysimeters (PRENART® Corp, Denmark) were established below the majority of the rooting zone [Ostertag, 2001]. Lohse and Matson [2005] discuss the details of the installation. In brief, the narrow cylindrical lysimeters (20 cm diameter, 120 mm total length, 50 mm length of collection area) consist of a Teflon frame covered by 0.2 um pore size quartz gel. The lysimeter body is connected by a 3 mm Teflon tube to a 1.5 L clean high-density polyethylene (HDPE) vacuum bottle situated in a covered access hole at the soil surface. The specific depths of the lysimeters used in these experiments were 29 cm and 50 cm for the young and old site, respectively.

[24] Over the center of the lysimeter at each site, we established a 1.3 m diameter circle with rainfall drip system shown in Figure 6. This drip system consisted of a HDPE plastic 40 L water container connected to a tygon tube (1.4 cm diameter) pierced with hypodermic needles every 8–10 cm along its length. Within this circular plot, we installed time domain reflectrometry (TDR) probes at 20 cm and 30 cm depths for measurement of volumetric water content (θv). We also installed a well to 50 cm depth at the old site to collect water if ponding occurred at the subsurface Bo1 horizon contact. Finally, we constructed a tarp tent over this plot and extended it 10 m beyond the plot to minimize evaporation and to eliminate other precipitation inputs.

Figure 6.

Rainfall drip system, which was centered over one lysimeter and delivered rainfall to a 1.3 m diameter circular experimental plot. A rainfall rate of 0.5 cm h−1 was obtained by maintaining a constant head of water in a carboy and having a specified number of needles spaced along a tygon tube. A tarp was established over the experimental plot and extended 10 m outside of the plot to minimize evaporation and eliminate other precipitation inputs. The crosses indicate the radial positions where soils were sampled 20 and 40 cm outside of the plot to determine the extent and degree of lateral flow.

[25] Prior to the tracer application, we applied two days of deionized water at a drip rate of 0.5 cm h−1 to attain quasi-steady state flow conditions. The water applied (24 cm water) was based on the amount of water necessary to displace several pore volumes (V = A × L × θv) where L is the depth of the lysimeter (cm) and A is the area of the plot (cm2). Soil volumetric water content (θv) was measured every 15 minutes until steady state was achieved [Lohse, 2002], and then θv was monitored every 3 hours over the course of the experiment. We then applied 6 days of water enriched with deuterium (D) and 15N-nitrate at the same drip rate (72 cm water). The drip rate and duration of the tracer application were determined from a preliminary bromide study at the old site [Hedin et al., 2003]. Bromide applied at a 2 cm h−1 rainfall rate took approximately 36 hours to reach the saturating limb of a breakthrough curve, equivalent to six days at a 0.5 cm h−1 rate. After the tracer application, deionized water was applied for three more days at the same rate to recover residual tracer in the soil (36 cm water).

[26] Deuterium rather than bromide was used as a primary tracer of water for these experiments because bromide was not considered conservative in the variable-charge soils found at the old site [Chorover et al., 2004; Lohse and Matson, 2005]. Deuterium concentrations are expressed in delta (δ) notation as per mil (0/00) differences relative to the international standard SMOW [Craig, 1961] where

equation image

A δD target value of 1000/00 enrichment was based on the following criteria: 1) it was considered a modest enrichment and therefore could be run on stable isotope mass spectrophotometers and 2) δD values in meteoric waters, precipitation and groundwater, that have not been subjected to evaporation are typically negative [Sklash, 1990], hence an enriched δD label would be a clear signal. The δD in deionized water at the Hawaii field laboratory was 0–20/00, while background lysimeter water values varied between −4 and −60/00 at the young site and −8 to −120/00 at the old site. We calculated the appropriate enrichments of D according to Craig [1953].

[27] In addition to D, we added 6.81 mg L−1 nitrate-15N at a final enrichment of 53,6000/00 (50 kg N ha−1 total). Enrichments of δ15N are expressed as per mil (0/00) differences relative to a standard, as described above for D. In this case, the standard was atmospheric N2 with a 15N:14N ratio of 1:272 [Hauck et al., 1994]. The total N added was the quantity of 15N needed to label the soil N pool. This tracer experiment was a part of a larger study used to evaluate the relative importance of hydrological and biological processes in retaining nitrate additions (K. A. Lohse, unpublished data, 2002); only deuterium soil solution and soil 15N data are reported here.

6.2. Field Sampling and Laboratory Analyses

[28] Over the entire 11 day experiment, we applied a constant 34 cm Hg vacuum to the lysimeter collection containers and sampled input water and lysimeter output water every 3–6 hours. We used a preleached, disposable nylon syringe to withdraw soil solution from the collection containers. Similarly, we used a clean syringe attached to a preleached tube to draw solution from the well at the old site. We collected these samples in copiously washed, preleached HDPE Nalgene bottles. Sample bottles were triple rinsed with sample solution through preleached disposable Gelman acrodisc filter (<1 um normal pore size) and kept on ice until they were removed from the field for analysis (according to Lohse and Matson [2005].

[29] Samples for deuterium analysis were sent to the Stable Isotope Ratio Facility for Environmental Research (SIRFER) at University of Utah. Water samples were prepared for hydrogen analysis using the zinc reduction method [Coleman et al., 1982]. D/H ratios were determined on a Finnagan Delta S isotope ratio mass spectrometer with dual inlet. Analytical precision for δD was better than 20/00.

6.3. Estimating Lateral Flow and Soil Anisotropy

[30] To evaluate the extent and degree of lateral movement at the two sites, we collected deep soil cores, 30 cm and 50 cm depth at the young and old site, respectively, in a radial design in 20 cm distance increments away from the circular experimental tracer plot (Figure 6). Because of a 2 month delay in soil sampling, we used 15N as a surrogate for deuterium movement. We hypothesized the dominant direction of water flow based on microtopography (local topography indicating maximum fall direction) and sampled first at this position, 0°, and then took subsequent soil cores at 120° and 240°.

[31] Soil core samples were stratified by depth every 2.5 cm, dried at 70°C, ground using a grinding mill, and packed for 15N analysis. These samples were analyzed for total N and δ15N on a Finnigan Delta S isotope ratio mass spectrometer with a front-end Carlo Erba autoanalyzer at Stanford University. Analytical precision for δ15N was better than 0.20/00 for enriched USGS 530/00 standards. Estimates of 15N recovery as percentage of the total were calculated according to Hauck et al. [1994], and percent recoveries were summed within genetic horizons.

6.4. Parameter Estimation of Breakthrough Curves

[32] We constructed breakthrough curves of the tracer data in relative concentrations (C/Co), where C is the deuterium outflow concentration and Co is the input concentration, against cumulative pore volume (V/Vo) and time (hours) [Tindall et al., 1999]. We then used a one-dimensional transport model called CXTFIT version 2.1 [Toride et al., 1999] to estimate field-scale transport parameters including pore velocity, ν, the average velocity of water inside of soil pores, the dispersion coefficient, D, and diffusivity, λ, from the field data. We selected a stream tube model approach (mode 3 in CXTFIT for nonreactive solute transport) to estimate these parameters because it did not assume uniform flow or homogeneous soils. In brief, a series of independent vertical soil columns represents a field, and transport in each stream tube is described deterministically assuming a convective or convective-dispersion model [Toride et al., 1999]. Transport parameters were estimated using a nonlinear least squares inversion method according to Marquardt [1963], and 95% confidence intervals were calculated for the parameter estimates [Toride and Leij, 1996].

6.5. Rates and Mechanism of Water Flow

[33] Significant differences in the rates and mechanisms of flow were observed between sites of contrasting soil age (Figure 7). At the young site, deuterium breakthrough took place after 1.5 pore volumes of water displacement which occurred 23.7 hours after the start of the tracer application (Figure 7a). The breakthrough curve approximated piston flow where half the original solution is displaced by the incoming solution at 1 pore volume [Tindall et al., 1999]. The average water flux (Jw) estimated from the product of θv and pore water velocity (ν) (Table 5), was 0.48 ± 0.03 cm h−1, closely approximating the rainfall application rate of 0.5 cm h−1. Results from nonlinear curve fitting with CXTFIT version 2.1 also showed negligible dispersion, D (Table 5). Together, these findings suggest that little mixing of the tracer with the original solution took place and that these soils were relatively uniform in soil texture and pore geometry [Tindall et al., 1999].

Figure 7.

Breakthrough curves showing changes in relative concentrations (C/Co), where C is the tracer outflow concentration and Co is the input concentration, with cumulative pore volumes (V/Vo) for (a) 0–29 cm depth at the 300 year old site and (b) 0–20 cm (well) and (c) 0–50 cm depth at the 4.1 Myr old site. A well at the old site was installed to 50 cm depth and collected water that ponded at the Bo1 horizon contact at 20 cm depth. Arrows indicate end of tracer application.

Table 5. Hydraulic Property Parameter Estimates Derived From Nonlinear Least Squares Curve Fitting to Measured Field Data Using CXTFIT2 for the 300 Year Old Site From 0 to 30 cm Depth and for the 4.1 Myr Old Site From 0 to 20 cm Depth (to Bo Horizon Contact) and 0 to 50 cm Depth (Lysimeter)a
SiteDepth, cmr2MSEν, cm h−1D, cm2 h−1λ, cm
  • a

    Mean square error (MSE), coefficient of determination (r2), parameter estimates for pore water velocity, ν, dispersion coefficient, D, dispersivity (λ), and associated standard errors are shown (λ = D/ν). Letters in parentheses identify statistically significant differences among sites (and soil horizons) at the 95% level.

300 year old0–300.9240.0891.08 ± 0.031 (a)0.072 ± 0.028 (a)0.067
4.1 Myr old0–200.9620.0458.42 ± 1.65 (b)1.87 ± 0.19 (b)0.22
4.1 Myr old0–500.9950.00670.393 ± 0.002 (c)0.002 ± 0.0005 (a)0.006

[34] At the old site, water collected in the well indicating ponding at the subsurface Bo1 horizon at approximately 20 cm depth. Water from the well showed that tracer breakthrough took place before one pore volume of water was displaced (Figure 7b), indicative of incomplete mixing or macropore flow [Tindall et al., 1999]. The ν of 8.42 ± 1.65 cm h−1 was eight times faster than that observed through the surface horizon at the young site, and dispersion was 26 times higher (Table 5). Cracks and high gravel content (40%) in the Bc horizon could account for macropore flow, although preferential movement along the well could not be totally discounted.

[35] In contrast to rapid macropore flow through the upper horizons, transport of deuterium-enriched water was very slow through the oxidized Bo horizons to the depth of the lysimeter at 50 cm depth (Figure 7c). Deuterium concentrations only reached half the relative concentration after 7.4 pore volumes were displaced which occurred 120 hours after tracer introduction. Average ν through this horizon was 0.393 cm h−1, 21 times slower than the surface horizon. Dispersion was negligible in this horizon (Table 5). In contrast to the young site, relative concentrations did not equilibrate; rather they continued to increase to a maximum of 86.3% after the tracer application had ended, and the posttracer deionized water application had commenced, suggesting that previously stored water was being displaced downward through this horizon. At the end of the experiment, the tracer had not returned to pretracer deuterium levels suggesting that some tracer still remained stored in the soil. These results are similar to Kline and Jordan [1968] who found that their soil profile remained tagged with tritium-labeled water in a tropical forest in Costa Rica and attributed this to the tortuous pore spaces in the clay soils. In a temperate catchment in the Oregon Coast Range, Anderson et al. [1997] also found that lysimeter solution remained tagged with deuterium 8–10 days after the end of their initial sprinkling experiment.

6.6. Relative Importance of Flow Paths

[36] Tracer recoveries were used to quantify total vertical and lateral flow. The input flow rate (Qi, cm3 h−1) was the irrigation rate (0.5 cm h−1) times the plot area (A, 13273 cm2), and the total tracer input volume (Vi, cm3) was the product of the input flow rate (Qi) and the duration of the tracer addition (150 hours at the young site and 147 hours at the old site). Under steady state conditions, the mass delivery rate of water with tracer to the lysimeter or well was equal to the product of the plot area, A, the density of water (ρ, g cm−3), the vertical flow rate (J, cm h−1, which equals the product of the soil moisture content (θv) times the pore velocity (Table 5)), and the measured δD concentration, C. The equivalent new water arrival rate, Qn, (cm3 h−1) times the initial tracer concentration, C0, and the water density (ρ) (g cm−3) will equal the observed rate, thus

equation image

The density of water (ρ) cancels, and solving for the new arrival rate at any instant in time gives

equation image

[37] Integration of the right hand side of (2) over the period of the measurement gives the total volume of new water (VnT, cm3) applied with the original tracer concentration that has arrived at the sampling site (3).

equation image

At the young site, we used θv of 0.448 ± 0.006 (n = 73) measured with TDR to 30 cm depth during steady conditions and then used 0.4616 ± 0.011 (n = 73) to 20 cm depth for the well and 0.528 ± 0.015 (n = 73) to 50 cm depth for the lysimeter at the old site.

[38] Total lateral flow (Vl) was calculated by two methods. First,

equation image

where Vi is the total input volume over the period of the experiment, VnT is the new water (tracer enriched) that reached the sampling point from (3), and Vs is soil storage, the new water that remained stored in the soil at the end of the experiment.

[39] At the young site, relative concentrations returned to baseline concentrations during the 3 day posttracer water application indicating Vs was zero (Figure 7a). At the old site, relative concentrations also returned to baseline concentrations in the well (Figure 7b) indicating no soil water storage to the depth of the clay contact at 20 cm. However, concentrations did not return to baseline at the end of the 3 day posttracer application in the lysimeter at 50 cm depth indicating residual storage of the new water in the Bo horizons from 20 to 50 cm depth (Figure 7c). We estimated the maximum residual soil storage using the following equation:

equation image

where z is the soil thickness (30 cm), A is the plot area (cm2), θs is the saturated volumetric water content of the soil from 20 to 50 cm depth (Table 1), and equation image is the relative concentration at the end of the experiment (0.743) (Figure 7).

[40] We also estimated total lateral flow from the total amount of soil 15N label that we recovered outside of the experimental plot relative to the total 15N added (see Section 6.3 for details on sampling and analytical methods). Recovery of total soil 15N outside of the plot was area-weighted based on the radial measurements of soil 15N at 20 and 40 cm distance increments away from the circular plot.

[41] At the young site, we recovered almost all of the deuterium (95.9 ± 6.4%) as lysimeter soil solution loss indicating that water was dominantly downward advecting under steady state flow conditions. Lateral flow was not detected except at the basalt parent material boundary at 34–38 cm depth. Here we found relatively small recovery of 15N, 4%, which was probably an artifact of the experimental design. A lateral head gradient produced from spot wetting may have maximized the potential for lateral flow. Alternatively, these results indicated that the boundary was less permeable than previously assumed. While the degree of fracture and permeability of this underlying basalt remains unknown, studies examining fractured flow suggest that decreased permeability in fractured basalt can occur due to air entrapment, surface or fracture sealing, redistribution of sealing, or microbial clogging [e.g., National Research Council Committee on Fracture Characterization and Fluid Flow, 1996; Faybishensko et al., 2000].

[42] At the old site, we found that nearly all the tracer was recovered in the well (94.4 ± 4%), indicating vertical soil solution loss to the depth of the Bo1 horizon contact, whereas we recovered only 28.9 ± 3.6% to a depth of the lysimeter at 50 cm (Table 6). These recovery differences suggested that the remaining deuterium was (1) moving laterally along the Bo1 horizon contact, (2) stored in the Bo horizons, and/or (3) moving along preferential flow paths not detected with this experimental design.

Table 6. Relative Importance of Vertical (VnT) and Lateral (Vl) Flow at the 300 Year and 4.1 Myr Old Site (as Percentage of Total Tracer Input Volume, Vi)
Flow Volume ComponentsSites
300 year old4.1 Myr old
  • a

    Estimate of lateral flow from soil 15N recovered outside of experimental plot.

Vertical flow (VnT)95.9 ± 6.428.9 ± 3.6
Soil storage (Vs)017.2 ± 0.6
Lateral flow (Vl)4.1 ± 6.4 (4.0 ✝)53.2 ± 4.1 (22.5 ± 3.1a)

[43] Soils taken in radial positions, 20 and 40 cm outside of the experimental plot, confirmed that lateral flow was taking place along the reduced Bo1 horizon contact; peaks in soil 15N recovery coincided with this horizon in all radial positions (Figure 8). Moreover, lateral flow followed the hypothesized direction of movement (in the 0° direction). Large lateral variation in 15N soil recoveries indicated possibly important soil anisotropy and microtopography effects on flow direction.

Figure 8.

Tracer recovery of 15N in the soils as a percentage of the total N added. Soils were taken in a radial design at 20 and 40 cm distance outside of the tracer experimental plot. Zero-degree coordinate represents the expected direction of lateral flow.

[44] Total soil 15N recovery outside of the plot accounted for 19.4–25.6% of the applied N indicating that nearly as much water was moving laterally along the Bo horizon contact as it was moving vertically through it. We acknowledge that this radial sampling design (starting at 20 cm outside the plot) and the timing of the sampling (2 months after tracer additions) probably underestimated total lateral movement; however, these data provided direct evidence of lateral movement along the Bo1 horizon contact. By difference from equation (4), we estimated that the lateral component (Vl) along the Bo1 horizon contact was likely as large as 53% (Table 6). Soil storage of new water from 20 to 50 cm depth could account for an additional 16–18% of the tracer. Despite the uncertainties in absolute magnitude of lateral flow, these data together demonstrate that significant movement of water occurred along this impeding subsurface horizon. While preferential pathways were shown to be significant in the surface horizon at the old site, they were not considered likely flow paths through the subsurface Bo horizons based on our observations of 22 soil pits. However, it is possible that we did not detect some preferential flow with our design.

7. Rainfall Frequency and Duration Distribution

[45] Rainfall data from the young and old site were used to determine 1) the probability of hourly rainfall rates exceeding surface and subsurface Ks values and 2) the likelihood of Horton overland (HOF), saturation overland (SOF), and subsurface storm flow (SSF). Hourly precipitation data were obtained from automated tipping bucket precipitation collectors located within close proximity (<2 km) to each site. Data for the young site, obtained from Volcano National Parks Headquarters, included years, 1986–2000, while data for the old site were collected as a joint effort for the Hawaii Ecosystems Project and included years, 1991–2000, except 1995. The frequency and duration of storm size events (nonzeros) were summarized and compared to soil Ks values at these sites.

[46] Results from these analyses showed that the rainfall rate at the young site exceeded the surface horizon Kfs value of 10 cm h−1 on five different 1 hour storm events during the years 1986–2000. At the old site, the rainfall rate never exceeded 36 cm h−1 during the years 1991–2000. In contrast, the probability of the rainfall rate exceeding the subsurface horizon Ks value at the old site was extremely high; 37% of all rainfall events from 1991 to 2000 exceeded 0.08 cm hr−1 suggesting the frequent development of a perched water table and probably SSF during these storm events.

[47] We estimated the surface horizon storage capacity (0–20 cm depth) to evaluate the probability of storm sizes exceeding this value. We calculated a storage capacity of 0.077 ± 0.057 m3 m−3 based on the difference between the saturated volumetric water content of 0.49 ± 0.056 m3 m−3 (Table 1) and the field water holding capacity of 0.413 ± 0.01 m3 m−3 (Soil Survey Staff, 2004). Rainfall data showed that 58 storms exceeded this mean storage value in a 10 year period. The maximum storm size, observed in a 10 year period, was 229 mm of rainfall in 43 hours. Another storm delivered 227 mm of rain in 34 hours, or approximately one tenth of the mean annual rainfall (2.5 m yr−1). These data point to relatively infrequent but large rainfall events that may exceed the storage capacity of the near-surface horizons and possibly produce SOF at this site.

8. Discussion

[48] Results from our study clearly showed dramatic differences in rates and pathways of water flow between the two sites of contrasting degrees of soil development. At one extreme, the poorly developed soil at the young site drained under relatively low suction and had relatively high Ks in both near-surface and subsurface horizons resulting in dominantly vertical water movement. Under steady flow conditions, tracer movement approximated piston flow where the incoming tracer solution displaced the original solution. At the other extreme, the subsurface horizon at the 4.1 Myr old soil site required much more suction to start to drain than the near-surface horizons and drained three orders of magnitude slower than these horizons. In response to a relatively high irrigation rate, water moved rapidly as macropore flow through the upper horizons, ponded at the impeding subsurface Bo horizon, and then simultaneously moved laterally as SSF and slowly downward. These results are similar to earlier findings by Bonell et al. [1983] who showed lateral movement of tritium water along an impeding clay-rich subsurface horizon as well as slow vertical recharge of labeled water through it in a humid tropical forest catchment in Queensland, Australia. Our findings support Bonell et al.'s [1983] conclusions that flow as SSF along these impeding subsurface horizons may be significant under high rainfall conditions but recharge as slow piston flow may also be important in determining storm runoff. While slow vertical displacement of previously stored water has been observed as an important mechanism for recharge (see Anderson and Burt [1990] for a review), to our knowledge, our study is the first to demonstrate in situ differences in unsaturated flow mechanisms due to soil development with time (and quasi-steady state conditions).

[49] Differences in hydrological properties and flow paths due to soil aging help to explain the timing and magnitude of vertical solute fluxes from these tropical forests under different nitrogen addition treatments [Lohse and Matson, 2005]. At the young site, we observed large nitrate (NO3) solution losses immediately following experimental N additions that were associated with relatively large storm events and high soil water fluxes. Rapid soil water drainage and short contact time between the incoming water and the coarse-textured soil could explain this pattern of solute transport. At the old site, NO3 solution losses were significantly elevated immediately following first time N additions but more delayed than those measured at the young site. Slow downward movement of solution through the impeding subsurface Bo horizon could help to explain this delayed response.

[50] Results from this study also documented differences in lateral flow between the two sites. Whereas we found negligible lateral flow at the young site in a relatively high irrigation event (0.5 cm h−1), our data showed that a significant amount of the applied water at the old site moved laterally along the impeding layer (19–57%). These data supported our field measurements showing potentially significant horizontal water fluxes along this subsurface horizon during natural high rainfall events [Lohse and Matson, 2005]. These findings together suggest that shallow subsurface lateral flow likely short-circuits movement of water and nutrients through the impeding subsurface clay horizon during large storm events. Analysis of the long-term rainfall data indicated that 37% of all rainfall events exceeded the drainage rate of the subsurface horizon suggesting frequent development of a perched water table and potentially large lateral solution losses. Thus discrepancies between experimental N additions and vertical NO3 losses at this site may be in part explained by a significant lateral N loss pathway [Lohse and Matson, 2005]. These differential loss pathways of nitrate during storm and drainage events may have potentially significant implications for downstream ecosystems.

[51] In our study, higher subsurface clay content appeared to be the primary factor explaining hydrologic differences between the young and old site. Increases in subsurface clay content with soil age have been documented along this chronosequence [Chorover et al., 1999] and along other chronosequences in the temperate zone and in the humid tropics [e.g., Jenny, 1980; Mizota et al., 1988; Nieuwenhuyse et al., 1993; Birkeland, 1999; Nieuwenhuyse et al., 2000; Chadwick and Chorover, 2001]. These trends in subsurface clay content with time suggest that as soils develop, particularly for those soils evolving on basalt or silica poor parent materials, subsurface Ks values are likely to decline due to increases in subsurface clay content, impeding rates of vertical water flow but also potentially increasing the importance of SSF and possibly SOF. Though not observed in this study, pipe flow and seepage erosion are also likely to increase in older soils [e.g., Kochel and Piper, 1986; Dunne, 1990]. Alternatively, development of structural properties may interact to influence the rate and trajectory of water flow [e.g., Beven and Germann, 1982; Flury et al., 1994; Vervoort et al., 1999]. Further studies across this and other soil age gradients are warranted to evaluate these hypotheses.

[52] As seen in Table 7, the contrasting effects of soil age on hydrological processes that we observed in Hawaii can be found in humid tropical forests growing on highly weathered soils throughout the world. This comparison was done in part to illustrate that our results from Hawaii are not uniquely associated with the impacts of dust. In Brazil and Indonesia, numerous studies have found highly permeable old soils that result in the dominance of vertical water movement. In contrast, other studies have found soil permeability contrasts in older soils that result in SOF and significant SSF along the active storm flow layers. Similar to our findings at the old site, Bonell and Gilmour [1978] found that high intensity rainfall exceeded the average Ks value of subsurface soils and generated a perched water table in a tropical rain forest watershed in Queensland, Australia. In their study, they found that additional rainfall caused the perched water table to emerge at the surface and result in SOF [Bonell and Gilmour, 1978; Bonell et al., 1981].

Table 7. Comparison of Mean (Range) Saturated Hydraulic Conductivity (Ks) Contrasts Between Surface and Subsurface Soils and Runoff Mechanisms Among Tropical Forest Sites on Highly Weathered Soilsa
Soil Site LocationGeologyRainfall, mm yr−1Soil OrderKs Surface, cm h−1Ks Subsoil, cm h−1Runoff MechanismsSource
  • a

    Runoff mechanisms reported are Perched water table, (PWT), subsurface storm flow (SSF), saturation overland flow (SOF), and Horton overland flow (HOF). Geology (basalt (B), sedimentary (S), and metamorphic (M)), annual rainfall (mm yr−1) and soil order according to the Soil Taxonomy [Soil Survey Staff, 1999] are reported.

  • b

    This study with Lithic Hapludand at the 300 year old site on Big Island, Hawaii and Plinthic Kandiudox at the 4.1 Myr old site on Kauai.

  • c

    Geology: Kuamut formation, a mélange of siltstone, sandstone, cherts, and tuffs. Soils: Haplic Alisol (FAO).

  • d

    Soils: red podzolic. Also see Bonell et al. [1983] for more extensive survey.

  • e

    Geology: sedimentary shales. Soils: Acrisols (FAO).

  • f

    Soils: red podzolic. Also see Gilmour and Bonell [1979].

  • g

    Geology: Tertiary red beds consisting of sandstones, siltstones, and shales. Also see Elsenbeer et al. [1992] for spatial survey.

  • h

    Geology: Precambrian gneiss. Ks median values.

  • i

    Cambrisols (eastern half) (FAO).

  • j

    Soils: gleyic podzols. Ks 15.4 mean infiltration value ultisol (acrisol), 4.87 mean infiltration value.

  • k

    Geology: quartzite, quartz mica schist, graphitic schist. Ks average median value for surface and subsurface horizons.

  • l

    Soils: Plinthic Haplorthox. Also see Nortcliff et al. [1979].

  • m

    Geology: Neogene sedimentary.

  • n

    Geology: Precambrian gneiss. Soils: Kandiudult and Plinthic Hapludox. Ks median values, subsurface values for 90 cm depth.

  • o

    Soils: Allophane Latosolic (Oxisol), Kandoid (kaolin) clays.

  • p

    No surface data available, near surface (30 cm) and subsurface (50 and 80 cm).

  • q

    Geology: Siltstone and sandstone of Pliocene/Miocene origin.

Big Island, HawaiiB2500Andisol1010vertical drainagethis studyb
Kauai, HawaiiB2500Oxisol360.08PWT, SSF, possible SOFthis studyb
NE BorneoS2775Ultisol15.6–30.50–0.04PWT, SOF, SSFBidin et al. [1993]c
NE QueenslandM4239Ultisol-Inceptisol83.80.08PWT, SOF, SSFBonell et al. [1983]d
NW BorneoSN/AUltisol2.7–46.89.3vertical drainage, SSFDykes and Thornes [2000]e
NE QueenslandM4175Ultisol-Inceptisol135 (12–414)1.3 (0.22–2.60)PWT, SOF, SSFBonell and Gilmour [1978]f
La Cuenca, west Amazona, PeruS3300Ultisols26.1 (18.6–35.6)0.022 (0.01–0.047)HOF, widespread SOFElsenbeer and Cassel [1990]g
Rancho Grande, Rondonia, BrazilM2265Oxisol10.2PWT, SOF, SSFElsenbeer et al. [1999]h
Rancho Grande, Rondonia, BrazilM2246Ultisols27.5<0.2SSFGodsey and Elsenbeer [2002]
Barro Colorado Island, PanamaB and S2600Ultisol7.6–51.70–3.1HOF, SOFGodsey et al. [2004]i
Barro Colorado Island, PanamaB2600Oxisols25.4–51.20.2–1.2SOF, SSF 
Sabah, MalaysiaS3215Ultisols0.068, 0.0110.016, 0.003SSF, SOF, HOFMalmer [1996]j
Bukit Tarek, Peninsular IndonesiaMN/AN/A43.2–12614.4–32.4SSF, possible SOFNoguchi et al. [1997]k
Reserva Ducke, Manaus, BrazilS2500Oxisol655–13429.86–463.3vertical drainageNortcliff and Thornes [1981]l
Bukit Soeharto, East Kalimantan, IndonesiaS2450Ultisol10–990.1–0.99SSFOhta and Effendi [1992]m
Rancho Grande, Rondonia, BrazilM2265Ultisol oxisol7.31 8.560.41 2.35SSFSobieraj et al. [2002]n
DominicaB5432, 2500–3800Oxisol, Ultisols25, 2.54–25.40.6–3.0, 0.25–2.54PWT, SSF, local SOFWalsh [1980]o
Kiani Lestari, East Kalimantan, IndonesiaS2450Ultisol0.1–1.76p0.001–0.007SSF, possible SOFWenzel et al. [1998]q

[53] Determinants of these differences in hydrologic flow paths in humid tropical rain forests remain debated [Bonell and Balek, 1993; Elsenbeer, 2001]. Elsenbeer [2001] has suggested that in general, vertical flow paths dominate in Oxisols (Ferrasols, FAO) whereas near-surface lateral flow paths appear to play a larger role in runoff generation in Ultisols (Acrisols, FAO). However, findings from our study, Elsenbeer et al. [1999], and Godsey et al. [2004] documenting impeding subsurface horizons and near-surface lateral flow paths in oxisols run counter to this generalization. Bonell and Balek [1993] have posited that differences in hydrological properties in old soils likely reflect underlying variation in bedrock or parent material that controls how runoff leaves the hillside [Dietrich et al., 1982; Bonell and Balek, 1993]. For example, vertical flow paths observed by Nortcliff et al. [1979] may be in part attributed to permeable quartz-rich sandy soils found in Manaus, Brazil. In contrast, near-surface flow paths observed in Queensland by Bonell and Gilmour [1978] may be attributed to silica-poor and chemically unstable metamorphic rock that have weathered into clay-rich soil [Bonell and Gilmour, 1978; Nortcliff et al., 1979]. A recent study by Godsey et al. [2004] documented differences in subsurface Ks values that helped to explain contrasting patterns of runoff generation between two lithologically distinct catchments under similar rainfall conditions. These findings lend further support to the supposition that the underlying bedrock strongly controls soil hydrological properties and consequently runoff pathways.

[54] Results from our study collectively suggest that time as a soil forming factor is also a key determinant of hydrological properties and flow paths. By minimizing variation in other environmental factors that could influence soil development, we show that runoff pathways shift from dominantly vertical flow to a higher ratio of lateral to vertical flow with soil formation with time on basaltic rock. These findings have potentially important implications for processes shaping the landscape as well as ecosystem development. In particular, results from our study suggest that while leaching has remained strong in the upper soil profile throughout soil development, effective precipitation and leaching intensity has declined in the subsoil with soil age due to increases in the ratio of lateral to vertical flow. This implies changes in rate of weathering and soil and ecosystem development over time with consequences for nutrient limitation and release [Walker and Syers, 1976; Vitousek et al., 1997; Chadwick et al., 1999] and possibly new interpretation of long-term climate effects on ecosystem development [Hotchkiss et al., 2000]. If soil development leads to subsurface clay enrichment and eventually to reduction of vertical penetration of water and increased lateral flow, then nutrient loss pathways will progressively change with time [Lohse and Matson, 2005]. If sufficient saturation overland flow were to be generated due to clay development, this may, in turn, increase surface erosion and thereby increase the likelihood of channel development. Further analysis of the topography and erosional processes at the old site area needed to evaluate this hypothesis.

9. Conclusions

[55] Differences in water retention, unsaturated and saturated hydraulic conductivity, and flow path characteristics between the two soil end-members of a chronosequence in Hawaii suggest that soil development with time has a profound effect on hydrological properties and pathways. At the young site, surface and subsurface soils drained rapidly in a downward direction whereas the subsurface soil at the old site drained significantly slower than the surface soils resulting in a perched water table and lateral flow during and water irrigation experiment. These shifts in hydrologic processes suggest changes in weathering rates at the soil-rock interface with time as well as change in weathering in the surface and subsurface soils with consequences for nutrient loss pathways and availability. Casting these results into the traditional “state factors” approach of soil science suggests how understanding of soil development and hydrologic processes can be further united. Models of landscape evolution that incorporate the coupling of soil development and hydrologic processes may give greater insight into slowly evolving topography in which subsurface clay accumulation may shift runoff and, consequently, erosion mechanisms.


[56] We thank Pamela Matson at Stanford University for providing funding, assistance in the field and lab, and helping with manuscript preparation. Financial support for this project was provided by a USDA Terrestrial Ecosystem and Climate Change grant and Andrew Mellon Foundation training grant to P. Matson and a James P. Bennett Agricultural Fund Fellowship from University of California, Berkeley, to K. Lohse. We thank Jessica Moen, Heraldo Farrington, David Penn, David Alexander, Jamie Carter, and Jennifer Fox for assistance in the field, Craig Cook at University of Utah at SIRFER for analyzing the deuterium samples, David Mucciarone at Stanford University for analyzing the 15N soil samples, and Peter Jewett at Stanford University for assistance in the laboratory. We offer thanks to Robin Harrington and D. Herbert for providing the early rainfall records at Koke'e, Oliver Chadwick for kindly providing the NRCS-USDA soil profile data, Peter Vitousek for providing advice on experimental design and review of the manuscript, and Ronald Amundson for reviewing earlier manuscripts.