Published in Soil Sci. Soc. Am. J. 67:1672-1686 (2003).
© 2003 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
DIVISION S-1—SOIL PHYSICS
Deep Soil Water Dynamics and Depletion by Secondary Vegetation in the Eastern Amazon
Rolf Sommer*,a,d,
Horst Fölsterb,
Konrad Vielhauera,
Eduardo J. Maklouf Carvalhoc and
Paul L. G. Vleka
a ZEF, Center for Development Research, Univ. of Bonn, Walter-Flex-Str. 3, 53113 Bonn, Germany
b Institute of Soil Science and Forest Nutrition, Univ. of Göttingen, Büsgenweg 5, 37077 Göttingen, Germany
c EMBRAPA Amazônia Oriental, Trav. Dr. Enéas Pinheiro s/n, Cx. Postal 48, 66095-100 Belém–Pará, Brazil
d CIMMYT, Apdo. Postal 6-641, 06600 Mexico, D.F., Mexico
* Corresponding author (r.sommer{at}cgiar.org).
 |
ABSTRACT
|
|---|
Secondary/fallow vegetation is found extensively in the Eastern Amazon. The role of this deep-rooting vegetation in the hydrological cycle is unknown. We studied the water dynamics of this vegetation with emphasis on the deeper soil by means of a soil water model. The soil hydraulic properties were optimized in an inverse modeling procedure using the soil water model Hydrus-1D providing rainfall, actual evapotranspiration (Ea) determined with the Bowen ratio energy balance method (BREB), rooting depth, and distribution, as well as the in situ control measurements of soil water pressure head and soil moisture. In 1997, Ea according to BREB measurements amounted to 1174 mm, in 1998 this was 1475 mm. Modeled drainage at a 10-m depth in the 2 yr was 951 and 1016 mm, respectively. The model indicated that around 27% of the soil water was taken up below the 0.9-m depth in the 2-yr study period. During the severe 1997 dry season, according to the soil water model Ea was reduced drastically, as the soil water storage was depleted. According to micrometeorological measurements, however, Ea was not reduced as extremely. This difference might be due to general uncertainties of the soil water model as well as BREB measurements. On the other hand, formation and subsequent evaporation of early morning dew apparently contributed to Ea, which was not considered in the soil water model. In general, besides slightly lower interception, the water budget of a young secondary vegetation did not differ from that reported for Amazonian primary forest.
Abbreviations: a.s.l., above sea level • BREB, Bowen ratio energy balance • EMBRAPA, Empresa Brasileira de Pesquisa Agropecuária • LAI, leaf area index
|
INTRODUCTION
|
|---|
DEFORESTATION OF RAINFOREST due to the expansion of shifting cultivation accounts for roughly 30% of the ‘converted’ areas in the Amazon (Fearnside, 1993). As a result, secondary woody fallow vegetation nowadays covers more than 100000 km2 in this region. Although the negative impact of these anthropogenic disturbances with regard to species extinction, biodiversity loss, and CO2 release are unquestionable, consequences for the hydrological cycle are less obvious. It is commonly assumed that intensified drainage and surface runoff and thus severe flooding are possible consequences of forest conversion. This assumption, however, is assured only by studies on water dynamics after forest conversions to pastures (Nobre, 1998). Studies on the water balance of tropical secondary vegetation are rare or were performed in composite watersheds (Bruijnzeel, 1990).
Nepstad et al. (1994) were the first who proved the existence of a deep-root system and its importance for the water budget of a primary forest in the Brazilian Eastern Amazon. A similar deep root system was also found under young secondary vegetation (Sommer et al., 2000). In the present study we questioned, to what extent this root system is active in water uptake especially during the dry season of 3 mo. Deep soil-water uptake by this vegetation was suggested by Hölscher et al. (1997) on the basis of micrometeorological measurements. To check this suggestion, we measured soil water depletion directly to a depth of 7.35 m using tensiometers, applied a soil water model and performed micrometeorological measurements of Ea. The soil hydraulic parameters, which were obtained by the model, have never been elaborated before and are most valid for future studies on the soil water dynamics of primary forests or conversion plots.
The field studies were performed in 1997 and 1998 and were part of the second phase of the German-Brazilian cooperative research project on Studies of Human Impact on Forests and Floodplains in the Tropics (SHIFT), particularly dealing with ‘Secondary forests and fallow vegetation in the Eastern Amazon region—function and management.’
|
MATERIALS AND METHODS
|
|---|
Study Site
The study was conducted in the municipality of Igarapé-Açu located in the Bragantina region in the Eastern Amazon. The climate of the region is humid with a dry season from September to December. It has a mean annual precipitation between 1700 and 2700 mm and a mean annual temperature of 25 to 27°C. The landscape has a flat to slightly undulating relief at 30 to 70 m above sea level (a.s.l.). The predominating terra firme (upland) soils are strongly and deeply weathered typic Kandiudults of loamy sand to sandy clay loam texture (Fig. 1)
. The natural vegetation of this region is an evergreen to semi-deciduous tropical rainforest. After 100 yr of smallholder shifting cultivation, most of this forest has disappeared and 1- to 10-yr-old secondary woody fallow vegetation covers 43% of the region (IBGE, 1997).
View larger version (19K):
[in this window]
[in a new window]
|
Fig. 1. Textural distribution and percentage of clay-flocculation (Brazilian classification) of the soil of the study site, as well as the mean soil bulk density of the soil of a neighbored 5-yr-old fallow (according to earlier studies; n = 16, bars denote the SE).
|
|
We chose a fallow vegetation located 1°11'6'' S lat. and 47°34'15'' W long. (altitude 50 m a.s.l.), covering about 2.6 ha and totally surrounded by fallows of different ages. Beginning 1997 the vegetation was 3.5 yr old and had an average height of 2.3 m. It grew to about 3.8 m at the end of 1998. Beginning 1997 Schmitt (1997) determined a leaf area index (LAI) of 4.2 (range: 3.9–4.6, n = 5), and a dry matter biomass of 22.2 Mg ha-1 (range: 13.2–32.0 Mg ha-1, n = 5 subplots plots of approximately 9 m2).
Methods
The assessment of the deep soil water dynamics and drying under the fallow vegetation required a monitoring of the aboveground processes of precipitation and transpiration. The belowground processes of soil water storage and drainage were obtained with a soil water model, which was validated with soil water pressure head readings and water content measurements.
Net Precipitation
Gross precipitation (P) was measured continuously above the fallow vegetation with a rain gauge and recorded from 1 Apr. 1997 to December 1998 in 15-min-intervals with a data-logger system. Gross precipitation data for the remaining time were taken from a nearby (3 km) weather station maintained by the EMBRAPA-Belém meteorological department.
Net precipitation (PN) is required for the soil water model. Net precipitation can be described as:
 | [1] |
where I is the canopy interception, PT throughfall and PS stemflow.
Following the methodology of Lloyd and Marques (1988), the accumulated amount of throughfall of the fallow vegetation was usually measured weekly or, in times of less rainfall, biweekly until the end of the 2-yr study period (Sá, personal communication, 1999). Fifty gauges consisting of a funnel with a diameter of 10 cm, and connected to a 2-L bottle (total gauge height: 30 cm), were relocated randomly every month between 300 discrete positions beneath the fallow vegetation. The 300 positions were fixed within a rectangular 1-m grid along a 50 m long and 6 m wide transect. Throughfall data were expressed relative to the mean value of gross precipitation obtained from four identical gauges placed in an open area on the side of the fallow, with the funnel-height at 1 m above ground.
Stemflow as well as throughfall of single storm events could not be measured due to the remoteness of the area and the lack of automatic measuring-devices. These measurements are required to apply existing rainfall interception models for instance that of Rutter et al. (1971)(1975). It was known from previous studies of Hölscher et al. (1998) that stemflow of secondary vegetation contributes an important fraction to net precipitation. To account for this, and to assess hourly net precipitation, we pursued the following steps:
- The systematic decline of the percentages of accumulated throughfall on gross precipitation over time was described with a linear regression. The regression equation was also applied to obtain throughfall of shorter intervals, that is, single storm events.
- A literature-derived canopy storage capacity (S) was introduced. Canopy storage capacity is the maximum amount of water that can be stored on the vegetation's surface during a single storm and is subsequently evaporated. The only information about the S of secondary forest found in literature is that of Jipp et al. (in review) of a 17-yr-old vegetation in Pará (Brazil). They determined S to be 1.1 mm (Table 1). Based on this study and the potential range of S presented in Table 1, we assigned S of the fallow vegetation to be 1.0 mm, which is slightly less than the S of the older—and presumably more dense—secondary vegetation of Jipp et al. (in review).
- Canopy storage was combined with throughfall data. At this, net precipitation of single storms was set to equal the calculated throughfall as long as the difference between gross precipitation and throughfall did not exceed the canopy storage capacity (Eq. [2]). In case this difference reached S, gross precipitation was reduced by this amount only (Eq. [3]).
 | [2] |
 | [3] |
The second assumption (Eq. [3]) was made to avoid unrealistic high interception. It was assumed that the intercepted rainwater of a single storm event would need at least 1 h to fully evaporate.
Transpiration
From 9 Apr. 1997 to 29 Mar. 1998 Ea of the fallow vegetation could be measured applying the Bowen ratio energy balance method (Sommer et al., 2002). For the remaining year 1998, Ea was determined with the Penman-Monteith method, with the aerodynamic and canopy conductance adjusted to the site conditions. The Ea for initial 1997 and the 1996 modeling pre-phase was derived from daily net-radiation data obtained by the EMBRAPA-Belém meteorological department. Actual evapotranspiration was proven to have a remarkable constant share on net radiation (Rn) according to (Sommer et al., 2002; R2 = 0.855):
 | [4] |
Evapotranspiration was assumed to occur only during daytime (0600–1800 h).
However, not Ea but transpiration alone, corresponding to root water uptake, is required as input for the soil water model. Therefore, Ea was reduced by the amount of rainwater of single storm events intercepted by the vegetation and subsequently evaporated, as it was calculated above. In case hourly interception was higher then hourly Ea, evaporation of intercepted water was assumed to continue also the following rainless hours until all intercepted water had completely evaporated. This meant that after a storm event transpiration was often close or equal to zero for one or several hours. As Ea was assumed to be zero during nighttime, it could not be reduced furthermore by the amount of intercepted rainfall. Though in the 2-yr observation period nighttime rainfall was rare, a small systematic error in the water balance was inevitable.
Soil evaporation was not considered in this study, as it is of negligible importance in a closed secondary vegetation stand.
Soil Water Model Settings
We assumed that soil water movement would obey the Richards equation and would be restricted to one vertical dimension. This was justified as the soil of the study site was homogeneously textured with a high sand content, non-layered, deeply weathered, and without inclination, all of which diminishes greatly the probability of a two-dimensional flow.
Among the high number of well-established SVAT modeling programs the Hydrus-1D soil water model (U.S. Salinity Laboratory, Riverside CA) was chosen, as it implements specific tools to (inversely) predict soil hydraulic parameters and root water uptake. This program numerically solves the Richards equation written in a mixed-form algorithm (Celia et al., 1990) with improved convergence criterion (Huang et al., 1996) using Galerkin-type linear finite element schemes (Vogel et al., 1996).
In times when Ea data according to the Bowen ratio energy balance were available (approximately 1 yr), the model was run with a data input of net precipitation and transpiration on an hourly basis. During the remaining times their daily values entered the model.
The upper boundary, that is, the soil–air interface was defined according to Neuman et al. (1974) with a surface reservoir allowed to build up during heavy storm events concurring with field observations. The lower boundary at a 10-m soil depth was defined to simulate a freely draining soil, where the soil water potential is a function of gravity only and ground water is without significant influence. This corresponds to measurements of the ground water level performed on 2 July 1997 (11.25 m) and 17 Mar. 1998 (15.25 m).
To describe the physical properties of the soil we used the so-called modified Van Genuchten type equation (Vogel and Císlerová, 1988). Vogel and Císlerová's (Vogel and Císlerová, 1988) model builds up on the original soil hydraulic parameters established by Van Genuchten (1980) including the pore-size distribution model of Mualem (1976). The soil water retention (curve) according to the modified Van Genuchten equation is given by
 | [5] |
where 
a (cm cm-1) is a fictitious (extrapolated) parameter defined to be equal or smaller the residual water content r (in our case taken at h = -15000 hPa), m is a fictitious (extrapolated) parameter equal or higher the saturated water content s, 
vG (cm-1), n (n > 1) and m are variables describing the shape of the curve and h is the soil water pressure head (cm 
hPa). The above equation allows for a non-zero minimum capillary height, hs, or the so-called air-entry value, up to which saturation of soil is still maintained.
The equation to predict unsaturated hydraulic conductivity according to the modified Mualem-Van-Genuchten approach (Vogel and Císlerová, 1988) reads as:
 | [6] |
where Ks is the saturated hydraulic conductivity (cm d-1),
 | [7] |
 | [8] |
 | [9] |
is the effective water content and
 | [10] |
Equation [6] matches the hydraulic conductivity function to a value Kk [ = K(k) respectively K(hk)], at some water content k (at hk) less than or equal to the saturated water content. In theory, Kk is a measured value at k still in the relatively wet water content range. This is considered to be a more accurate matching point than Ks, as it omits the very step slope of Eq. [6] close to saturation, where very small measurement errors in the water content lead to unacceptably large errors in the hydraulic conductivity (Luckner et al., 1989). The exponent 
is the so-called pore-connectivity parameter. It was originally found to be about 0.5 as the best estimate for many soils (Mualem, 1976), but was also variable in the present study.
Assuming m = 1 - 1/n, 10 independent parameters had to be provided as model inputs.
Soil water retention curves were determined in the EMBRAPA-Belém soil-laboratory with 100 cm3 undisturbed soil core samples using pressure plate procedure. Four repetitions of core samples were taken at 15-, 30-, 60-, 90-, 120-, 180-, 240-, 300-, 400-, 500-, and 600-cm soil depth. Water content was measured in terms of desorption at pressure heads (high air pressure) of 60, 100, 300, 1000, 5000, 10000, and 15000 hPa (analogous to Maklouf et al., 1997). Additionally, laboratory-saturated water content was determined. The soil water retention parameter of Eq. [5] then were fitted to this set of soil water retention data points assuming that m = s and a = r and using the least-square optimization technique of the RETC program (Van Genuchten et al., 1991). Reciprocal values of the standard deviation of the four repetitions entered as weighting coefficients of each data point.
The saturated soil hydraulic conductivity was determined with the ‘Rosetta’ program (Version 1.0, U.S. Salinity Laboratory). Using soil texture, bulk density, -330 and -15000, this neural network application calculates Van Genuchten's (1980) soil hydraulic parameters using the pedotransfer functions established by Schaap et al. (1998)(and 1999). The texture was determined at the above-mentioned soil depths at EMBRAPA soil laboratory according to standard EMBRAPA procedure (EMBRAPA, 1997). Bulk density data were taken from a previous study (Sommer, 1996; see Fig. 1). The water content at -15000 hPa (-15000) was directly available from the soil water retention curve, the value at a pressure head of -330 hPa was obtained by linearly interpolating the average values of -300 and -1000 hPa.
According to obvious differences in texture and/or soil hydraulic parameters (Fig. 1), the soil profile was divided into three strata: 0 to 5, 5 to 105, and 105 to 1000 cm. Soil hydraulic parameters for the first stratum were derived only from the Rosetta program, as no soil water retention curves were determined in the laboratory. The lower two strata were characterized by the soil hydraulic parameters of four respectively seven relevant soil depths, where soil water retention curves were available. The overall hydraulic parameters of each stratum were calculated using the scaling technique, which was developed by Miller and Miller (1956) and extended by Vogel et al. (1991). According to their studies, the variability of a soil profile can be approximated by means of reference characteristics, *(h) and K*(h) and a set of linear scaling transformation factors for each profile depth. The relationships between reference characteristics, scaling factor, and measured hydraulic characteristics at a certain depth of the soil profile are:
 | [11] |
 | [12] |
where K is the scaling factor for the hydraulic conductivity and that for the water content.
Finally, for the first (inverse) model run the following settings were made: hs = hk = -2 cm, a = r, k = s, and Kk = Ks. According to latest suggestions of Schaap et al. (2000) for sandy soils, the pore-connectivity parameter was initially set to -1.
Root water uptake, driven by transpiration, affects the soil water content over the whole rooting zone and enters as a sink-term in the Richards equation. Hydrus-1D incorporates the relationship between actual root water uptake (= actual transpiration, Tmodel) and potential transpiration (Tpot) as described by Feddes et al. (1978):
 | [13] |
where r is a dynamic reduction factor dependent on the soil water pressure head; r is supposed to account for water stress and reduced actual transpiration, when soil water availability drops. Basically, r is highly dependent on the ecophysiology of the plant community. Van Genuchten (1987) reconsidered the original approach to assess r of Feddes et al. (1978) and described its behavior with a single mathematical equation:
 | [14] |
where h50 is the pressure head at 50%-reduced root water uptake and p is an experimental constant positively determining the steepness of the decline of r(h).
In our case, however, not potential but actual transpiration was used. This meant that Eq. [13] basically had lost its importance. Consequently, h50 and p were set to arbitrarily high values of -5000 hPa and 6, respectively, to possibly keep further (unjustified) reduction of actual transpiration small.
Furthermore, the root distribution within the soil profile was introduced into the model. It affects the depth of direct soil water depletion due to root water uptake and the relative distribution of the depletion process. Therefore, average data on the root distribution of sites under fallow according to a previous study (Sommer et al., 2000) were used.
Above-described net-precipitation, actual transpiration, soil hydraulic parameters, as well as the root water uptake and root distribution were part of the initial model settings.
In the field, beginning of April 1997, high-resolution, automatic soil water pressure head records (pressure head transducers SILVAQ, Canada) were taken at time intervals of 15 min with 1.4-m long tensiometers. They were installed in a soil pit horizontally, slightly inclined, at soil depths of 30, 60, 90, 120, 180, 240, 300, 400, 500, and 600 cm. To avoid evaporation from the pit walls, these were painted with a concrete-solution and the pit itself was roofed. At the end of October 1997 an additional tensiometer was added to reach 735 cm. On 6 Nov. and 10 Dec. 1997 the soil water content was determined gravimetrically at the same depths (5-cm increments, manual drilling) to maximum 6 m and also at 0- to 5-cm soil depth. Automatic soil water content readings could not be performed, as the necessary equipment (TDR-Probes) were in bond of Brazilian customs authorities over the whole field-research period.
Soil water pressure head and gravimetric water content records were used to adjust the initial soil hydraulic model settings in an inverse modeling procedure. For this, the parameter-optimization module of Hydrus-1D was used. Discrepancy between observed and modeled values is brought to a minimum with soil hydraulic properties kept as dependent variables using the Levenberg-Marquardt nonlinear minimization method (weighted least-square approach; Marquardt, 1963).
Statistical Evaluation
Finally, the model performance was evaluated. All log-transformed measured and modeled pressure head data in a 15-min time-step were included. Only those data above -860 cm (= measuring limit of the tensiometers) were used. This was done to comply with the required normal distribution, and to avoid bias caused by extremely low (modeled) values. The goodness of the relation between observed and predicted soil water pressure head was statistically estimated by
• the coefficient of efficiency (E) defined by Nash and Sutcliffe (1970) as:
 | [15] |
• the root mean square error (RMSE), and
• the ratio of the systematic and the unsystematic square error (MSEs and MSEu, respectively) to the total MSE (as the sum of MSEs and MSEu). Here:
 | [16] |
and
 | [17] |
where Pei is the least square linear regression of predicted and observed pressured head according to Pei = a x observedi + b.
Temporarily on 22 Apr., 12 May, 20 May, 26 May, 17 June, 25 June, 30 June, and 9 July 1998, a neutron probe could be used to control the predicted water contents. A 6-m Al tube was installed vertically close to the soil pit. The neutron probe itself was calibrated with gravimetrically determined water content data (n = 105) and a linear regression between water content and counts of back-scattered neutrons was established (R2 = 0.56, SE of estimate = 0.018).
|
RESULTS
|
|---|
Net Precipitation and Actual Evapotranspiration
The annual rainfall was 2104 mm in 1997 and 2545 mm in 1998. Due to the influence of El Niño, the dry season of 1997 was extremely severe with little precipitation (Fig. 2)
.
Measurements of cumulative throughfall expressed as percentage of P, showed high spatial as well as temporal variations. No significant correlation could be found between percentages of throughfall and (i) P amounts (R = 0.147, n = 64), (ii) number of storm events within the collection period (R = 0.132, n = 64), and (iii) a combination of both variables (linear regression; R2 = 0.022, n = 64). However, throughfall declined significantly from about 81% in early 1997 to about 68% at the end of 1998, apparently related to the growing vegetation (Fig. 3)
. Ninety-five percent confidence intervals during this period were about 10% above or below.
View larger version (28K):
[in this window]
[in a new window]
|
Fig. 3. Relative mean throughfall (PT) under the fallow vegetation in 1997 and 1998 (Sá, personal communication, 1999); dotted straight line: linear regression (R2 = 0.212, n = 64), gray line: linear regression of 95% confidence intervals of PT, dotted upper line: calculated relative net precipitation.
|
|
The net precipitation calculated according to Eq. [2] and [3] with S = 1 mm most times exceeded the percentages of throughfall (dotted line in Fig. 3). On an annual basis, interception, as the difference between gross precipitation and calculated net precipitation, amounted to 6.8% of gross precipitation in 1997 and 8.0% in 1998. Taking the mean annual throughfall of 77.4 and 72.5%, stemflow must have contributed another 15.8 and 20.5% to net precipitation in 1997 and 1998, respectively.
Annual Ea in 1997 was 1383 and 1475 mm in 1998. Actual evapotranspiration was highest in August 1997 (average: 4.7 mm d-1) and September 1998 (4.8 mm d-1), which is the end of a transitional period/beginning dry season. It was generally lowest in the rainy season from January to April of both years (ranging between 3.3 and 4.2 mm d-1). February 1998 with on average 4.3 mm d-1 actual evapotranspiration was quite exceptional with altogether 13 rainless days, high net radiation, temperature, wind speed, and atmospheric vapor pressure deficit promoting high Ea (for more details see Sommer et al., 2002).
The annual evapotranspiration was reduced by the intercepted and subsequently evaporated amount of daytime rainfall (I) to 1294 and 1300 mm annual actual transpiration in 1997 and 1998, respectively. As nighttime rainfall interception was not discounted, actual transpiration over the 2-yr period was actually 84 mm higher than Ea - I.
Rooting depth and vertical distribution determines the amount and pattern of root water extraction in response to transpiration. Secondary fallow vegetation in the Bragantina region has a deeply penetrating root system to at least 6-m soil depth (Sommer et al., 2000). Though most of the root mass is located in the upper soil profile, still on average 28.2% is found below the 90-cm depth (Fig. 4)
.
View larger version (21K):
[in this window]
[in a new window]
|
Fig. 4. Mean root mass density of 1- to 12-yr-old fallow vegetation (according to Sommer et al., 2000; n = 60, bars denote the SE), its cumulative percentage distribution (secondary x axis) and the distribution as entering the soil water model.
|
|
In times of optimum water supply over the whole soil profile, the given vertical root distribution would obligatorily determine the pattern of root water uptake. This relative weighting of root water uptake over the soil profile changes in responds to proceeding drying of the soil.
Inverse Modeling–Optimization of Soil Hydraulic Parameters
The modified Van-Genuchten soil hydraulic parameters, as primarily determined based on the soil water retention curves and Rosetta neural network application, assuming an air-entry value of -2 cm were optimized by inverse modeling.
We restricted the inverse modeling to the period with hourly net precipitation and transpiration data input from April 1997 to March 1998. This period comprised the full range of measured soil water pressure heads maintaining a fairly appropriate relation of times with desiccated conditions to times with satiated (= field-saturated) soil conditions. The control data entering the objective function were the daily means of measured hydraulic pressure head and water contents instead of hourly or even 15 min data to decrease the number of data points to keep the model applicable. Altogether 2288 data points entered the objective function. Inverse modeling in this way required that at least three of the optimized parameters had to be kept constant. We chose r and a (= r) because test runs showed that this was the parameter with smallest alteration during optimization. Additionally, we left s as fitted by the RETC program. In this case prior optimization runs with the original Van-Genuchten parameters (i.e., m = k = s, a = r, Kk = Ks) showed that the optimization process had reduced s values by partly over 55%. This was obviously related to the shortcomings of the original Mualem-Van-Genuchten approach to explain K(h) close to saturation. With the modified Van-Genuchten equations of Vogel and Císlerová (1998) and the addition of m and k this has been largely overcome enabling s to be set to the measured value.
The parameter optimization process by inverse modeling only slightly changed the values initially set for m. With final values of 0.544, 0.435, and 0.410, the air-entry pressure head hs reached values of 1.8, 0, and 1.48 cm in the three strata, respectively (Table 2).
The empirical constant vG of the first stratum (0–5 cm) as well as the constant n of all three strata did not change during optimization. In the second and third stratum, however, vG—whose reciprocal value corresponds to the inflection point of the (h) curve—increased from 0.072 to 0.180 cm-1 and 0.052 to 0.160 cm-1 respectively. This modified the soil water retention curve, so that the water content values within the range of a pressure head of -10 to -100 cm dropped by about 12 to 20%.
The optimization process slightly increased the saturated hydraulic conductivity of the upper strata from 250 to 254 cm d-1. Ks of the second strata decreased from 178 to 160 cm d-1, while Ks of the 105- to 1000-cm stratum increased slightly from 152 to 160 cm d-1. Thus, Ks was reasonably predicted by the Rosetta neural network. However, Kk, the hydraulic conductivity at k (hk respectively), dropped quite noticeably during the optimization process to 199 cm d-1 at h(k) = 1.8 cm in the first stratum, 11 cm d-1 [h(k) = 4.2 cm] in the second stratum and 18 cm d-1 [h(k) = 24.2 cm] in the third stratum. This led to a sharp drop of the hydraulic conductivity curves at the particular values of h(k) (Fig. 5)
.
View this table:
[in this window]
[in a new window]
|
Table 4. Comparing log-transformed data sets of observed and predicted pressure heads of different soil depths measured/computed in 15-min. intervals; data below a pressure head of -860 cm were excluded from analysis (n = number of data; RMSE = root mean square error; MSE = mean square error; subscript s, u and t = systematic, unsystematic and total, respectively).
|
|
The pore-connectivity parameter of initially -1 seemed to be almost suitable for the studied soil. It decreased to -1.28 and -1.50 in the first and second stratum, respectively, and increased to -0.50 in the third stratum.
Individual vertical scaling of K(h) and (h) was done by applying the scaling factors K and (Table 3). Modifications of the soil water retention curve through scaling were almost negligible in the second, but somewhat more important in the bottom stratum (Fig. 5). Largest reduction of K(h) occurred at 30 cm (K of 0.63) and at 400 cm (K of 0.61), largest increases at 90 and 180 cm (K of 2.00 and 1.40, respectively).
The predicted soil water pressure head dynamics in comparison to the observed data are shown in Fig. 6
for 30-, 90-, 300-, 600-, and 735-cm depth. During the rainy season the pressure head at the 30-cm depth most times varied between 0 and -50 cm. Only major rainfall events caused the pressure head at 90 cm to approach values close to saturation. In the deeper soil profile pressure heads never exceeded -30 cm, indicating that the deeper layers rarely if ever reach saturated soil moisture conditions.
View larger version (35K):
[in this window]
[in a new window]
|
Fig. 6. Observed and predicted pressure head dynamics at 30, 90, 300, 600, and 735 cm over the 2-yr observation period.
|
|
The response to the intensive dry season in 1997 is obvious. Starting at the end of August, the soil desiccated and the pressure head rapidly fall below the lower measuring limit (-860 cm) of the tensiometers successively down to a depth of finally 500 cm. Temporary rewetting in response to major storm events on 20 to 21 November (39 mm) and 17 to 21 December (56 mm) was measurable down to 180 cm. The topsoil rewetting was predicted correctly, but, according to the model, the wetting front was not deeper than the 60-cm depth. This led to major deviation of observed and predicted pressure heads visible at 90 cm in Fig. 6 and noticeable still at the 400-cm soil depth (data not shown). Continued high transpiration/root water uptake in the model caused drying of these soil layers to an extent that pressure head dropped below -1000 cm from mid-October 1997 till mid-January 1998, when the rainy season finally began.
Exaggerated drying of the soil profile by the model became also apparent on 6 Nov. and 10 Dec. 1997, when the soil water content was measured. The measured soil water contents were slightly but significantly higher than those predicted (Wilcoxon Signed Rank test, n = 22, p = 0.014; Fig. 7)
.
View larger version (16K):
[in this window]
[in a new window]
|
Fig. 7. Observed and predicted soil water contents on 6 Nov. and 10 Dec. 1997 (bars denote SE of n = 2).
|
|
Pressure heads during moist soil conditions (h > -200 cm) in the rainy season and the transitional period were well described with the soil water model. In the second soil stratum, predicted values tend to lag behind the observed ones, especially in the lower section of this stratum at 90 cm. This was not the case in the third stratum, where predicted initial drying during the 1997 dry-season as well as, at 600 cm only, rewetting due to the beginning 1998-rainy season was slightly faster than observed.
The visible impression was confirmed by statistical analyses of the model performance. The coefficient of efficiency (E) ranged between 0.64 (60 cm) and 0.972 (600 cm; Table 4). Generally, the model efficiency increased with depth, as is also visible in the RMSE-values. The systematic deviation of predicted from observed pressured heads (MSEs/MSEt) was greatest at the 400-cm depth with 72.7%. This reflects the above-mentioned drying of the soil profile during the dry seasons, where predictions systematically deviated from observed data. Overall systematic deviations however were quite small (insignificant at the 500-cm depth) indicating only limited scope for improvement of optimized soil physical parameters.
Comparison of neutron probe measurements with model predictions of water content at a number of days from April to July 1998 showed that the model predictions slightly underestimated observed data of water contents of the subsoil. Contrarily, those measured in the topsoil, at 30 and 60 cm, were lower than the predicted (Fig. 8)
. Thus, neutron probe water content data were significantly different from those predicted (Wilcoxon Signed Rank test, n = 83, p = 0.001).
View larger version (22K):
[in this window]
[in a new window]
|
Fig. 8. Water contents measured by means of neutron probe in relation to model predictions of different times during the rainy season and transitional period of 1998 as well as those gravimetrically determined on 6 Nov. and 10 Dec. 1997 (values below 0.17) distinguished according to the measuring depth.
|
|
Soil Water Dynamics and Depletion
Results of the soil water model were calculated for the 2 yr of observation, and separately also for the extraordinary dry season of 1997 (Table 5).
View this table:
[in this window]
[in a new window]
|
Table 5. Water balance of 1997, 1998, and during the 1997 dry season (P = gross precipitation, I = interception, Ea = actual evapotranspiration, PN = net precipitation, Ta = actual transpiration, TModel = modeled transpiration, D10m = drainage at a 10-m depth).
|
|
In 1997 precipitation could not balance the sum of the model-predicted water losses by transpiration (TModel), interception (I), and drainage at 10-m depth (D10m) leading to a net deficit of 59 mm, which were taken out of the soil reservoir. Contrarily, precipitation of 1998 exceeded the sum of drainage and evapotranspiration by 188 mm, leading to an overall positive balance of 129 mm, which was stored in the soil profile.
Despite higher precipitation in 1998, the amount of water drained at 10-m depth (1016 mm), was less than the year before. This was on one hand related to the fact that soil was heavily dried at the end of 1997; a deficit that had to be first of all filled up again. On the other hand, precipitation in 1998 was more evenly distributed (see Fig. 2). Thus, momentary infiltration rates were generally not as high as the year before, and the secondary vegetation was less exposed to water stress permitting higher evapotranspiration rates.
During the 2 yr, the micrometeorologically determined Ea was not attained by the soil water model (I + TModel). In 1997 modeled transpiration was 290 mm less than the actual transpiration (Ta) entering the model. In 1998 this was merely 99 mm. During the dry season, in response to water stress, Ta was reduced by the root water uptake function (Eq. [13]). During most intensive water stress, the quotient out of TModel and Ta (which per definition equals r in Eq. [13]) temporarily even decreased to 0.1 (Fig. 9) .
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 9. Relation of modeled to micrometeorologically observed transpiration (TModel/Ta) during the 2-yr observation period.
|
|
From 22 August 1997 to 8 January 1998, TModel reached only 48.3% of what was observed by micrometeorological measurements, that is, 298 mm less (Table 5). In the moderate dry season of 1998 (22 Aug. 1998 till the end of the year) this gap was only 75 mm.
Compared with the moist soil water conditions in the rainy season (30 Mar. 1997), on 22 Aug. 1997 the soil water profile was already drained to a large extent (Fig. 10)
, and downward soil water fluxes over the whole profile to the 6-m depth were below 1.6 mm d-1. Considering the period of 22 Aug. 1997 to 8 Jan. 1998, the root water uptake amounted to 279 mm, of which 144 mm were taken out of the soil reservoir (TModel - PN). Root water extraction caused a slight upward movement of water from below the 6-m depth of altogether 3 mm. Twenty percent of the transpired water were withdrawn from the soil depth of 3 to 6 m (Table 6); 7% more than during the rainy season and the transition period (1 Jan.–22 Aug. 1997). In the moist Year 1998 merely 14% of root water uptake came out of this layer. As expected the topsoil (0–90 cm) with 67 to 73% of the root system contributed most water for transpiration. Still, 277 to 327 mm annually were taken up from below the 90-cm depth.
View larger version (25K):
[in this window]
[in a new window]
|
Fig. 10. Modeled soil water content over the 10-m profile on 30 Mar. 1997 (maximum soil water storage), 22 Aug. 1997 (beginning dry season), and 8 Jan. 1998 (minimum soil water storage).
|
|
View this table:
[in this window]
[in a new window]
|
Table 6. Root water uptake in the rainy season and transitional period (1 Jan.–22 Aug. 1997) and in the dry season (22 Aug. 1997–8 Jan. 1998) as well as in the Year 1997 and 1998 considering different soil layers
|
|
After 506 mm rainfall and 180 mm evapotranspiration since 8 Jan. 1998, the rewetting front reached the 6-m soil depth only on 24 Feb. 1998. The bottom boundary (10 m) finally was reached 16 d later on 12 Mar. 1998, after 782 mm of rainfall and 234 mm evapotranspiration since 8 January.
The overall maximum amount of water stored in the profile (0–10 m) was 2091 mm on 30 Mar. 1997 (Fig. 11)
. The minimum was reached 9 mo later on 8 Jan. 1998 (1308 mm). The soil water storage of 6- to 10-m depth varied annually between 572 and 805 mm. Though some of this store took part in upward fluxes during the dry season, the soil water below 6 m did not really contribute to root water uptake and was only subjected to replenishing and drainage. Annual fluctuation of the soil water storage of 0- to 6-m depth was stronger (727–1302 mm), as the roots of the fallow vegetation directly depleted this part of the soil profile. If we consider this maximum difference as potentially available water, this would correspond to an average of 96 mm m-1 of soil.
Root water uptake gradually reduced the amount of water, which percolated through the soil. From 4300 mm net precipitation within the 2 yr only 2631 and 1971 mm passed 0.9- and 6-m depth, respectively. This equals the cumulative drainage at the bottom boundary of 1967 mm, including a storage change of 4 mm.
|
DISCUSSION
|
|---|
Net Precipitation
Net precipitation of secondary vegetation is subject of only a few studies in literature (Table 7). Percentages of throughfall, stemflow, and interception of the present study lie between those found by Schroth et al. (1999) and Hölscher et al. (1998), which are the most suitable studies for comparison because of similar climatic and edaphic conditions as well as a comparable age of the vegetation. With increasing vegetation age, stemflow seems to contribute less to net precipitation. According to Jipp et al. (in revision) it amounted to only 1.7% of P in a 17-yr old secondary vegetation, which is close to PS of primary forests.
View this table:
[in this window]
[in a new window]
|
Table 7. Annual throughfall (PT), stemflow (PS) and interception (I) of primary and secondary tropical vegetation in relation to annual gross precipitation (P).
|
|
Higher stemflow in young secondary vegetation might at least partly be counterbalanced by lower throughfall as in our study in the second year and in Hölscher et al. (1998). However, Table 7 seems to indicate a trend toward higher interception (and lower percentage of PN) with increasing forest biomass.
Rainfall partitioning should best be based on the evaluation of single storm events. In the present study this was impossible due to lacking automatic devices and the remoteness of the area. Because of the great differences in habitus, height and stem diameter, measurement of stemflow would have required a high input of instrumentation, and would have caused great disturbance in this low and dense vegetation. In comparison to existing rainfall interception models (e.g., that of Rutter et al. 1971, 1975), therefore, our approach was simplified regarding some details: We considered free throughfall (without striking the canopy) negligible. Leaf and stem interception were merged, and evaporation from the wet canopy was set to be at the most 1 mm h-1 throughout. Evaporation of water intercepted by the canopy during storm events was not accounted for. This might however cause only minimal underestimation of rainfall interception, as in the tropical study region rainfall events are mostly intensive and short and humidity of the air rapidly reaches 100%, all of which is drastically diminishing quantities of canopy evaporation. A quite similar approach to calculate net-precipitation was suggested by Campbell and Diaz (1988) for model prediction of crop evaporation.
Model Optimization and Performance
Using the modified Van Genuchten type equation to describe (h) and K(h) dynamics in the model settings (Vogel and Císlerová, 1988), it was possible to utilize the saturated water content s as determined in the laboratory. In an earlier modeling exercise using the original Van Genuchten approach (data not shown), this was not possible, because these s values were considerably diminished during the optimization procedure. Thus, the approach given by Vogel and Císlerová (1988) introduced flexibility to describe water dynamics, especially near saturation, at the same time maintaining measured value of s. The sharp drop of hydraulic conductivity in the second stratum (5–105 cm) at a pressure head of -4 cm might indicates air-entry into the macropores. On the other hand, the domain of macroporous flow obviously was rarely if ever reached in the lower soil profile (third stratum) below 105-cm depth during the 2-yr observation period. This might be due to the fact that the studied soil has a high drainage capacity. Another explanation however might be that in the field air entrapment hampers saturation of the soil, whereas this is not the case in laboratory measurements using soil core samples. This has repeatedly been reported for Brazilian Oxisols (Klinge et al., 2001; Buttler and Riha, 1992).
The conductivity function of the second and third stratum differed markedly. Conductivity in the subsoil at a pressure head of -4 to -1000 cm was about a factor 10 higher than in the above stratum. The only plausible explanation for this might be the stable (100% flocculated) microaggregation of clay particles to a kind of pseudo-sand in the third stratum leading to an increase in conductivity (Bouma and Anderson, 1973). The drop of conductivity in the second stratum coincides with a partial (50%) deflocculation (Fig. 1). To describe water dynamics in these kind of soils correctly, modifications of primarily obtained K(h)-relationships seem to be more the rule than the exception (Johnson et al., 1999; Tomasella and Hodnett, 1994).
Appropriate values of remain subject of debate till these days. While earlier publications set at 0.5 as proposed by Mualem (1976) as best-estimate (Van Genuchten, 1980; Dane & Hruska, 1983; Van Genuchten & Nielsen, 1985; Kool et al., 1987), later studies noted that should be kept as an experimental unknown (Table 8). In an extensive recent study focusing on the pore-connectivity parameter Schaap et al. (2000) claimed an optimum -value of -1.0 for all their 235 considered soils, with the database comprising a wide range. Splitting up the database according to the textural groups, on the average all textural groups in their study had negative values, with lowest values for loam and sand. Only = -1 for sandy soils was statistically different from the original Mualem (1976) value of = 0.5. In terms of best-fitted hydraulic conductivity, the variability of the other textural groups was high, with mean values not significantly different from 0.5. The modifications on the Mualem-Van-Genuchten approach made by Vogel and Císlerová (1988) do not alter the absolute value of . Thus, the pore-connectivity values adjusted for the water model settings of our study are comparable with the results of Schaap et al. (2000) for sandy soils.
The model performance after optimization was first of all evaluated by comparing the observed and predicted pressure head dynamics at different soil depths, as well as observed and predicted soil moistures at two times in the dry season of 1997. Since the observed pressure heads and soil moisture contents also entered the objective function to optimize the model, the evaluation was not independent in a strict sense. Nevertheless, it demonstrates the power of the optimization process. Moreover, none of observed pressure head data of the last 9 mo of observations entered the objective function. Considering only this period for validation (data not shown) would have even improved the coefficient of efficiency, RMSE and the measure of systematic deviation, because the strongest deviations of observed and predicted data during the extensive dry season of 1997 were excluded.
The time lag between observed and predicted pressure heads in the top 5 to 105 cm displays scopes for further improvement. However, the systematic error was relatively low in this stratum, indicating that possibilities are limited for model improvement by correction of predicted pressure head values by a further increasing/decreasing Ks, Kk or water content in the model settings.
Model predictions of soil moisture slightly but systematically deviated from determination with the neutron probe at selected times during the rainy season and the transitional period in 1998. Deviations are acceptable given the error involved in the neutron probe measurements (SE of estimate = 0.018). They furthermore demonstrate the limits of the model optimization considering the fact that only a few in-field gravimetric measurements entered the objective function, and that laboratory soil water retention curves, as part of the basis of the initial model parameter settings, might be of limited applicability to field conditions with air-entrapment.
The pore-space of aggregated clay particles of the loamy sandy soil of the present study might have been excluded from quantitative important water movement (immobile water domains). This could have promoted macroporous structure and rapid, bypassing water movement (Young & Leeds-Harrison, 1990); a phenomenon not enclosed in our approach based on the classical Richards equation. Such an effect was shown to be important for instance by Arya et al. (1999) for an Indonesian clayey Kanhapludult. The clay content of 50 to 80% of their soil, however, is much higher than that found in our soils, which are homogeneously structured with a medium to coarse texture, where K- models generally fit best (Schuh & Cline, 1990).
Soil Water Depletion and Evapotranspiration
Despite the fact that the soil water model slightly overestimated water depletion during the dry season of 1997 (Fig. 7), evapotranspiration (as the sum of root water uptake and interception) was not as high as determined micrometeorologically with the BREB, but 298 mm less. This gap might partly be explained by early morning dew, which frequently caused complete wetting of the whole vegetation canopy, especially during the dry season. During this 140-d period, assuming a canopy storage capacity of 1 mm, up to 140 mm dew evaporation were not included in the soil water model but detected with the micrometeorological measurements. This amount seems realistic as already Hölscher et al. (1997) estimated dew evaporation of a 2-yr-old secondary vegetation in the Bragantina region to be about 167 mm yr-1. Uncertainties in actual evapotranspiration determinations with the BREB might add to explain differences between soil water model and BREB results. Foken et al. (1997) proposed a general accuracy of the BREB method of not better than ±10%, which in our case in 1997 could cause a difference of up to ±138 mm. Another explanation might be that root water uptake during the dry season from deeper soil layers, despite relatively low root mass densities, is proportionally higher accompanied by lower soil hydraulic conductivity rates than anticipated. Such an emphasized water uptake from certain soil layers currently cannot be considered in the Hydrus-1D soil water model.
The contribution of deep roots to the annual water balance of the fallow vegetation is most remarkable. Considering plant-available soil water of utmost 96 mm m-1 soil profile, the usage of deep soil water is necessary for the evergreen fallow vegetation to survive recurring extensive dry seasons such as 1997.
Applying the BREB method, Hölscher et al. (1997) determined an Ea (April 1992–April 1993) of a 2-yr-old fallow vegetation in the Bragantina region of 1364 mm, which was about 75% of the rainfall in this period (1819 mm). Deep soil-water uptake below 1-m soil depth was calculated to be 322 mm between June and December 1992. The absolute amount of deep soil water uptake and Ea thus matches well with our results. Since annual precipitation in our study, however, was 285 and 726 mm higher in 1997 and 1998, respectively, drainage rates exceeded those proposed by Hölscher et al. (1997). They assumed a balanced soil water storage (beginning the end of observation period) and thus claimed that 455 mm, that is, the difference between precipitation and actual evapotranspiration, must have been drained during the observation period. This is less than half of the amounts of the present study based on the soil water balance, questioning the assumption of a balanced 1992–1993 water storage made by the authors.
Klinge et al. (2001) detected deep soil-water uptake under a primary forest near Belém, Brazil using a soil water model. Modeled root water uptake amounted to 365 to 402 mm between 1.1 and 5 m, that is, 26 to 29% of modeled Ea, and thus was comparable with values of our study. Poels (1987) studied the water balance on a catchment-area scale under primary and disturbed forest in Suriname. In his hydrological model he assumed a maximum rooting depth of 4.5 m to explain water discharge through evapotranspiration. Evapotranspiration in the dry season was reduced by up to 50% like in our study. On the other hand, Poels (1987) also stressed the possibility that some of the primary forest trees might have roots extracting water directly from the ground water, which in the dry season fell below a 10-m depth. This cannot definitely be excluded in the present study, but seems unlikely, as ground water level was below 15 m in the dry season.
Deep soil water uptake, as detected for young secondary vegetation in our study, has recently been proven also to be important for primary forests. Hodnett et al. (1996a)(1996b) monitored the water storage by neutron probe measurements in a soil under primary forest near Manaus/Brazil, and could show that water uptake from below 2-m soil depth must have occurred to account for evaporation demands in the dry season. According to their long-term simulation of 27 yr, annual water uptake of the soil below 2 m was on average 72 mm, reaching at maximum 254 mm. Roots could be found to a depth of 6 m. Nepstad et al. (1994) even claimed a soil-water uptake from down to 8 m by a primary forest in the south of Pará. They assumed that more than 75% (380 mm) of transpired water was taken out of 2- to 8-m depth during the 5-mo dry season.
It is remarkable that evapotranspiration of young fallow vegetation hardly differs from that of mature primary forests. An annual evapotranspiration of 1319 mm is calculated by Shuttleworth (1988) for primary forest near Manaus/Brazil (annual precipitation: 2636 mm), Bruijnzeel (1990) gives 1430 mm yr-1 (n = 11; range: 1311–1498 mm yr-1, P: 1727–4073 mm yr-1) as an average for selected (worldwide) tropical lowland forests, and Klinge et al. (2001) obtained about 1378 mm yr-1 (P: 2669 mm yr-1) for a primary forest close to Belém/Brazil. Comparing the percentages of evapotranspiration to precipitation, young fallow (58–60%; our results) even exceeds primary forest (50–52%). This is likely related to the vigorous regrowth of the fallow from remaining stumps and roots that survived the fallow period. After about 2 to 3 yr, LAIs are already around 4 (Schmitt, 1997), and thus a fully installed canopy is present.
Consequently, as evapotranspiration of secondary vegetation is similar to that of primary forest and as interception is slightly lower, drainage nevertheless might be slightly higher under a comparable rainfall regime. This, however, still has to be verified by direct comparison.
|
CONCLUSION
|
|---|
Although the establishment and optimization of a soil water model is a complex and time-consuming exercise, it is the only way to obtain detailed insights into the water dynamics of deep soils. Model results showed that secondary vegetation maintains a water turnover comparable with that of a primary forest, with some evidence of a slightly higher throughfall and possibly also drainage. We proved that the deep rooting secondary vegetation in the Eastern Amazon is able to take up considerable amounts of water and thus sustain its evergreen canopy. About 27% (277–327 mm) of the annually transpired water was taken up from below the 0.9-m soil depth. It now has to be studied, whether the natural secondary/fallow vegetation with its deep-reaching root system is likewise able to take up nutrients, which are set free during the cropping period of shifting cultivation. Putting science into practice—in terms of sustainable agricultural management options—this is of great importance.
|
ACKNOWLEDGMENTS
|
|---|
From the cooperation between Center of Development Research (ZEF), University of Bonn, Germany, and EMBRAPA Amazônia Oriental, Belém-PA, Brazil, under the Governmental Agreement on Cooperation in the field of scientific research and technological development between Germany and Brazil. The study was financed by the German Federal Ministry of Education and Research. With special acknowledgment to EMBRAPA Amazônia Oriental.
Received for publication March 7, 2002.
TOP
ABSTRACT
INTRODUCTION
(in combination with 