SIMULATION OF MOISTURE PROFILES IN A LATOSSOL IN DOURADOS REGION

1 Researcher Ph.D., Embrapa Western Region Agriculture, PO BOX 661, Dourados MS, romulo@cpao.embrapa.br 2 Undergraduate Student of Computer Science, State University of Mato Grosso do Sul, UEMS, s.joaopaulo@gmail.com 3 Professor Ph.D., Entomology Department, Federal University of Lavras, UFLA, rigitano@ufla.br Recebido pelo Conselho Editorial em: 10-11-2008 Aprovado pelo Conselho Editorial em: 16-10-2009 Eng. Agríc., Jaboticabal, v.30, n.1, p.22-32, jan./fev. 2010 SIMULATION OF MOISTURE PROFILES IN A LATOSSOL IN DOURADOS REGION, IN THE STATE OF MATO GROSSO DO SUL, BRAZIL


INTRODUCTION
Unsaturated soil water flow is a key factor in the hydrological cycle because it determines the rate of soil evaporation, recharge to groundwater, plant transpiration and runoff.An accurate understanding of unsaturated soil water flow is essential to derive proper management conditions for environmental protection in agricultural systems.
However, describing the interaction of all processes that influence the soil water flow is rather complex.One way to overcome this complexity and to integrate all the processes is to use modeling and simulation tools (SCORZA JÚNIOR, 2006;MONTEIRO et al., 2009).BOESTEN (2000) mentioned two important advantages of modeling tools compared to experimental assessments, that are: i) possibility to investigate the effect of weather conditions using multiyear simulations and thus avoiding long-term experiments that are expensive and time consuming; ii) modeling enables the integration of the actual knowledge of the processes and thus can be useful in identifying gaps and research priorities.
An important tool for simulation of soil water flow that has been used is the SWAP model (KROES & DAM, 2003;DAM et al., 2008), which simulates transport of water, solutes, and heat in the unsaturated zone in interaction with vegetation development.SWAP has also been linked to other models with the aim to generate accurate soil water fluxes for detailed chemical transport, such as PEARL (LEISTRA et al., 2002) for pesticides and ANIMO (GROENENDIJK et al., 2005) for nutrients.SWAP has been used worldwide for many different applications (e.g.groundwater conservation, nutrient transport, water balance prediction etc).DAM (2000) tested SWAP to simulate water flow in a cracked clay soil in the Netherlands and concluded that the solution of Richards's equation considering the soil as only one domain could not mimic the average behavior of solute transport.Also in the Netherlands, SCORZA JÚNIOR & BOESTEN (2005) LOUZADA et al. (2008) used the SWAP model to simulate groundwater levels in a drained area with rice crop at the state of Rio Grande do Sul, Brazil, and found good agreement between measured and simulated groundwater levels.Until now, the SWAP model has not been tested to simulate soil moisture profiles in Brazilian soils.The aim of this work was to test and to calibrate the SWAP model to describe moisture profiles in a Brazilian very clayey Latossol in Dourados, in the state of Mato Grosso do Sul, Brazil.

MATERIAL AND METHODS
SWAP model is described in detail elsewhere (KROES & DAM, 2003;DAM et al., 2008) and will be briefly described here.It describes water flow using Darcy equation considering onedimensional, vertical and transient flow.The combination of Darcy equation with the conservation equation results in the well-known Richards equation: (1) in which, C -water capacity, cm -1 ; h -soil water pressure head, cm; t -time, day; z -depth in soil, cm; K -hydraulic conductivity, cm day -1 , and S -rate of water uptake by plant roots, day -1 .
The rate of water uptake is given by: (4) in which, red -reduction factor (-), and S P -potential rate of water uptake, day -1 .S P can be obtained from: (5) where, T P -potential transpiration rate, cm day -1 , and d r -root layer thickness, cm.
Based on the calculation of potential evapotranspiration rate (ET P ) using the Penman-Monteith method, T P is given by: (6) in which, f w -fraction of the day that canopy is wet (-), and E P -potential soil evaporation rate, cm day -1 .
The SWAP model was tested in an experimental field (30 m long and 30 m wide) cultivated with soybean located at Embrapa Western Region Agriculture (22°16'26''S, 54°48'50''W), in Dourados, state of Mato Grosso do Sul.The experimental field consisted of 100 subplots of 3 m long by 3 m wide.The soil is a very clayey typical dystroferric Red Latossol (EMBRAPA, 1999).Soil properties of the experimental field are shown on Table 1.Soil profiles were sampled eight times on the following dates: 18 December 2006, 26 January 2007, 23 March 2007, 17 May 2007, 27 June 2007, 31 July 2007, 12 September 2007and 30 October 2007.On each sampling date, four subplots were randomly selected and each subplot was sampled only once during the duration of the experiment.A pit (1 x 1 x 1m) was dig on each selected subplot and soil cores were collected from 0-10; 10-30; 30-50; 50-70, and 70-100 cm depth.All samples were put in a hermetic box and transported to the laboratory for determination of the moisture content by drying at 105 °C for 24 h.Three undisturbed soil cores (5.5 cm of inner diameter and 4 cm height) were taken at five different depths (0-10; 10-30; 30-50, and 70-100 cm) for determination of the soil water retention curve in only one pit of the experimental field.These undisturbed soil cores were saturated in the laboratory and submitted to pressure heads of 100; 330; 500; 1,000; 3,000; 5,000; 10,000, and 15,000 cm using Richards's pressure chamber.Soil water retention curves were described by fitting the Genuchten model [eqs.(2) and ( 3)] to the measured data using RETC (GENUCHTEN et al., 1991).Genuchten parameters used as input in SWAP are shown on Table 2. Saturated hydraulic conductivities in the field were measured at 0-10; 10-30; 30-50; 50-70; and 70-100 cm depth using a Guelph-type permeameter (VIEIRA, 1998).For these measurements, head ponded of 3 and 6 cm were used in the wells.Crop characteristics were obtained from measurements or taken from literature.Soybean was planted on October 30, 2006 and harvested on March 6, 2007.LAI values [eq.( 7)] were obtained from literature (VISSOTTO JÚNIOR, 2003).The estimation of d r [eq.( 5)] was based on visual observations at different sampling dates.The factors red [eq.( 4)] as a function of soil water pressure head were obtained from WESSELING (1991).The bottom boundary condition used in the simulations was free drainage of soil profile.Meteorological data as input for SWAP model were obtained from an automatic meteorological station at Embrapa Western Region Agriculture, which is located about 500 m from the experimental field.The potential evapotranspiration rate was estimated based on the Penman-Monteith method (ALLEN et al., 1998).
SWAP calibration of moisture profiles was carried out using the package PEST (DOHERTY, 2000) and the objective function ( ) to be minimized was: (8) where, b -vector with the fitting parameters; y meas -measured soil water content at time t and depth z, cm 3 cm -3 ; y sim -simulated soil water content at time t and depth z (cm 3 cm -3 ) for the corresponding set of b parameters, and w -weighting factors, (-).
Model performance was assessed using graphic display (i.e. visual

RESULTS AND DISCUSSION
Cumulative rainfall during the entire experimental period (i.e. between Dec. 18, 2006 andOct. 31, 2007) was 1,004 mm (Figure 1).The highest rainfall rates and frequencies did occur between December 2006 and March 2007 that corresponds to a period of 90 days after trials started.About 60% of total cumulative rainfall did occur within these 90 days.The average daily rainfall rate during the experimental period was 3.2 (±9.2) mm day -1 , with a maximum value of 66.4 mm day -1 on February 6, 2007 (i.e.51 days after experiment start).Calculated cumulative reference evapotranspiration during the experimental period was 1,118 mm.
Cumulative evapotranspiration was slightly greater than cumulative rainfall during the experimental period, resulting in a negative water balance of -114 mm.Average values of cumulative rainfall and evapotranspiration during the last 28 years (between 1979 and 2007) at the experimental area are 1,160 (±308) and 1,191 (±71) mm, respectively.These average values were calculated considering only the period between December 18, and October 31 for the last 28 years.Based on this information, the cumulative values of rainfall and evapotranspiration obtained during the experimental period correspond, in average, to 87% and 94%, respectively.Soil moisture profiles measured during the experimental period are shown in Figures 2 and 3.The greatest variations on soil moisture contents as a function of time were observed at 0-30 cm depth.These most pronounced variations of moisture content on the top soil are the result of high influence of evapotranspiration.On the other hand, small variations on soil moisture contents were observed for soil layers below 30 cm.The range of measured soil moisture contents was large for some sampling dates and depths with the average value plus and minus two times the standard deviation close to 0.2 m 3 m -3 (Figures 2 and 3).Uncalibrated simulations resulted in moisture profiles that were too wet, with the exception on January 26, 2007, July 31, 2007and October 30, 2007.The largest differences between measured and uncalibrated simulated values were found in the top 10 cm on December 18, 2006, March 23, 2007and September 12, 2007. Large deviations (i.e.simulated values outside the range of measured values plus and minus two times the standard deviation) between measured and uncalibrated simulation values were also found at 30-100 cm on May 17, 2007 and at 0-50 cm soil depth on June 27, 2007.ME values were negative for uncalibrated simulations for almost all sampling dates (Table 3).These negative values suggest that the average of the measurements is a better data representation than the values of uncalibrated simulations.
The fact that uncalibrated simulations of moisture contents resulted in higher values than the measured ones for almost all sampling dates indicate systematic differences.These differences suggest that the soil water retention curves measured in the laboratory were not representative of the experimental field.This lack of representativeness may have been caused by inadequate handling of spatial variability for the soil water retention in the experimental area.Soil cores for measurements of soil water retention curves at different depths were collected in a single pit of the experimental area (i.e. in only one location).2001) compared four different methods to derive soil hydraulic properties to be used as input for a simulation model by analyzing their effect on simulated soil moisture profiles at three different areas of the Netherlands.The applied methods were laboratory measurements, class pedotransfer functions, continuous pedotransfer functions and continuous pedotransfer functions combined with simple laboratory measurements.These four methods originated different input values for the soil water retention curves.ALPHEN et al. (2001) concluded that input values based on laboratory measurements using samples collected at two points in the experimental area did not result in accurate simulations of moisture profiles.This conclusion is in agreement with our findings.ALPHEN et al. (2001) also concluded that the best simulations of moisture profiles were obtained when the combination of pedotransfer functions and laboratory measurements were used as input values for soil water retention curve.Therefore, it is recommend that input values of soil water retention curves based on a few measurements in the field should be used only as initial guesses for a calibration procedure when simulating field moisture profiles.In compliance with the above discussion, the next step was to calibrate SWAP model using the measured moisture profiles.The optimizing parameters were n, e r.Calibration procedure did decrease n and r values for all depths (Table 2).With regard to parameter , there was an increase for the 0-30 cm and a decrease for the 30-100 cm soil layers.
Decision to optimize the parameters n and was based on VANCLOOSTER et al. (1992).These authors showed that simulated soil moisture profiles are very sensitive to parameters n and and concluded that any calibration of this variable should preferably include the optimization of these two parameters.VANCLOOSTER et al. (1992) concluded that simulations of soil moisture profiles were not sensitive to parameter r .However, in this study, r was optimized because it was not experimentally measured but estimated by fitting measured data of soil water retention to Van Genuchten model.This procedure to estimate r leads to a certain degree of uncertainty.In general, the optimization of n, and r parameters resulted in decreasing water retention capacity for all soil layers.Figure 4 shows this decrease in the soil water retention curve for 10-30 cm and 50-70 cm depth.

FIGURE 2 .FIGURE 3 .
FIGURE 2. Moisture profiles in Dourados, state of Mato Grosso do Sul , Brazil.The area within the red lines is the range of measured values plus or minus two times the standard deviation.
tested SWAP to simulate soil moisture profiles in a cracked clay soil and obtained good simulations after calibration of soil water retention curves.Studies showing SWAP performance in Brazil are scarce.More recently,

TABLE 1 .
Soil properties at the experimental field in Dourados, State of Mato Grosso do Sul, Brazil.

TABLE 2 .
Uncalibrated and calibrated Genuchten input parameters used as input in SWAP to describe soil water retention curves at different depths.
comparison between measured and simulated soil moisture contents at each sampling date and depth) and the statistical indices Normalized Root-Mean Squared Error (NRMSE) and Modeling Efficiency (ME).These indices are given as: NRMSE shows the spread around the average of the measurements and ME indicates if the model describes the data better than simply the average of the measurements.Optimal values of NRMSE and ME are zero and one, respectively.

TABLE 3 .
Normalized Root-Mean Squared Error (NRMSE) and Modeling Efficiency (ME) of simulated moisture profiles.