Published online 1 January 2007
Published in Soil Sci Soc Am J 71:43-50 (2007)
DOI: 10.2136/sssaj2006.0098
© 2007 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
SOIL PHYSICS
Excluding Organic Matter Content from Pedotransfer Predictors of Soil Water Retention
Steffen Zacharias*
Faculty of Civil Eng. and Geodetic Sci., Inst. of Water Resources Management, Hydrology and Agric.Hydraulic Engineering, Univ. of Hannover, Appelstr. 9A, D-30167, Hannover, Germany
Gerd Wessolek
Dep. of Soil Science, Inst. of Ecology, Technical Univ. of Berlin, Salzufer 11-12, D- 10587, Berlin, Germany
* Corresponding author (steffen.zacharias{at}googlemail.com).
 |
ABSTRACT
|
|---|
The majority of pedotransfer functions (PTFs) published for estimating water retention characteristics (WRC) use data on soil texture, bulk density, and organic matter content (OM) as predictors. For soil hydrological modeling on a regional scale, in particular the derivation of appropriate values for a PTF parameterization can be difficult where organic C data are missing. Assuming the indirect interdependency between OM and bulk density, a new PTF has been developed that estimates the WRC using only soil texture and bulk density data. To achieve a regression-based reproduction of the correlations, a calibration was chosen that connects the parameters of the van Genuchten equation with the data on bulk density and soil texture, using linear and nonlinear relationships. More than 90% of the variability in measured soil water contents was explained by the new model. The validity of the PTF was tested with a data set of 147 measured WRCs (r2 = 0.94). Compared with another frequently used PTF model, which uses the organic C content as an additional predictor, the new model provided comparable or slightly better predictions of the WRCs.
|
INTRODUCTION
|
|---|
Effective soil management requires a considerable understanding of the soil water balance. The use of models to answer soil hydrological questions plays a particularly important role. A substantial problem with this kind of modeling is the lack of suitable parameters, in particular data on the soil water retention or hydraulic conductivity. Using PTFs to predict water retention is a practical way of obtaining the required parameters. At a regional scale, up to the scale of the catchment, this is often the only way to apply hydrological models.
In the last few years and decades, many PTFs have been developed that allow water retention properties to be estimated. Generally, three different approaches can be distinguished for deriving the WRC from more easily available parameters:
- Point-based estimation methods: estimating the water content of selected matric potentials from predictors such as the percentage of sand, silt, or clay, the amount of organic matter, or the bulk density (e.g., Gupta and Larson, 1979; Rawls and Brakensiek, 1982)
- Semiphysical approach: deriving the WRC from information on the cumulative particle size distribution (Arya and Paris, 1981); theoretically, this approach is based on the similarity between cumulative particle size distribution and water retention curves. The water contents are derived from the soil's predicted pore volume and the hydraulic potentials are derived from capillarity relationships.
- Parameter estimation methods: using multiple regression to derive the parameters of an analytical closed-form equation for describing the WRC, using predictors such as the percentage of sand, silt, and clay, the amount of organic matter, or the bulk density (e.g., Vereecken et al., 1989; Wösten et al., 1999).
In the last few years, WRC optimization by artificial neural network modeling has been successfully used by several researchers (Pachepsky et al., 1996; Schaap et al., 1998; Minasny et al., 1999).
Extensive overviews of the recent development and use of PTFs have been given by Timlin et al. (1996), Wösten et al. (2001), Cornelis et al. (2001), Gijsman et al. (2002), and Nemes et al. (2003).
Approach 1 has the disadvantage that it uses a large number of regression parameters depending on the number of WRC sampling points, which makes its use in the mathematical modeling more difficult. For Approach 2, very detailed information about the particle size distribution is required. Furthermore, for modeling continuous scenarios of the soil water regime, these methods require interpolations between the WRC sampling points. The adequacy of the selected interpolation method cannot be easily guaranteed. The use of parameter-based estimations often is the preferred approach.
The majority of published PTFs use soil texture, bulk density, and OM as predictors. For hydrologic modeling on a larger scale, the determination of these input parameters often presents a problem. Normally, available soil mapping data is used. While the OM is often measured in soil surveys, such data are often missing in regional soil maps. On the other hand, in many regions data on soil texture or bulk density can often be derived or estimated using soil maps and information about land use. In Germany, for example, there are extensive current and historical soil mapping data that provide a nationwide overview of particle size distribution.
Furthermore, bulk density is a very dynamic soil property that can be strongly influenced by different factors such as land use or climatic effects and can show a distinctly higher temporal variability than the OM. The use of bulk density as the only predictor apart from the soil texture data can make it easier to replicate temporarily variable WRCs.
Pedotransfer functions that do not use the OM are rare. Hall et al. (1977) developed point-based regression equations using soil texture and bulk density (only for subsoils) for British soils. Oosterveld and Chang (1980) developed an exponential regression equation for Canadian soils for fitting the relationship between clay and sand content, depth of soil, and moisture content. For chernozems and meadow soils, regression equations have been developed by Pachepsky et al. (1982). Equations to estimate the WRC from mean particle diameter and bulk density have been proposed by Campbell and Shiozawa (1989). Williams et al. (1992) analyzed Australian data sets and developed regression equations for the relationship between soil moisture and soil texture, structure information, and bulk density including variants for both the case where there is available information on OM and where the OM is unknown. Rawls and Brakensiek (1989) reported regression equations to estimate soil water retention as a function of soil texture and bulk density. Canarache (1993) developed point-based regression equations using clay content and bulk density for Romanian soils. More recently, Nemes et al. (2003) developed different PTFs derived from different scales of soil data (Hungary, Europe, and international data) using artificial neural network modeling including a PTF that uses soil texture and bulk density only.
A specific problem in using PTFs is their limited applicability, as they depend on the data sets that were used for the PTF development. The use of the above-mentioned PTFs, especially for large-scale hydrologic modeling, may be limited due to the regional dependence of the data set used for the PTF development (e.g., Hall et al., 1977; Oosterveld and Chang, 1980; Williams et al., 1992; Canarache, 1993). Especially in hydrological modeling on a regional scale, a wide range of soil textures, bulk densities, and other soil parameters must often be taken into account. In many cases, the validation limits of the PTFs are exceeded (e.g., Rawls and Brakensiek, 1989, valid only for sand >5 and <70% and clay >5 and <60%). Different investigations have shown that extrapolating PTFs can lead to large errors (Schaap and Leij, 1998; Nemes et al., 2003). Other PTFs can often only be used to a limited extent due to the additional information that is needed (e.g., Williams et al., 1992, additional structure information). Another fact that may limit the applicability of some of the above-mentioned PTFs is that they provide only point-based estimations for selected matric potentials (e.g., Hall et al., 1977; Canarache, 1993) or are valid only between certain matric potential limits (e.g., Oosterveld and Chang, 1980).
Cornelis et al. (2001) evaluated different PTFs for the WRC with different numbers of predictors. The best results obtained were PTF forecasts including the OM. Tietje and Tapkenhinrichs (1993) evaluated 13 PTFs. The best results were obtained from the PTF by Vereecken et al. (1989), which includes the organic C content as a predictor. At the same time, when Gijsman et al. (2002) evaluated eight methods for estimating water content using field capacity, wilting point, and saturation water contents, they found that the method by Saxton et al. (1986), which makes a point-based estimation on the basis of soil texture and disregards the OM, had the best results. This did not apply, however, for very sandy soils.
The fundamental hypothesis of this work is that the influence of OM on soil water retention can be represented by bulk density. The influence of OM on soil structure and the associated effects on pore structure have been the subject of numerous investigations (e.g., Rawls and Brakensiek, 1982; Rawls et al., 2003; Six et al., 2004; Prevost, 2004). The extent of this influence depends on other soil structure-influencing factors such as texture, the composition of exchangeable ions in the soil solution, weathering influences, and clay mineralogy. The complexity of the interdependencies between the structure-influencing factors makes an exact determination of the effects of OM on the water retention difficult. One of the best-investigated interrelations is the connection between OM and bulk density (e.g., Rawls, 1983; Manrique and Jones, 1991; Heuscher et al., 2005). The strong negative correlation between these parameters has been demonstrated repeatedly. This strong interrelation suggests that in PTFs, the effects of the OM on water retention can be accounted for indirectly using bulk density only. Rawls et al. (2003) also expressed the assumption that the bulk density could be a suitable parameter to represent the effect of organic C on water retention if the main effect of the OM is changing bulk density.
Based on these considerations, the objective of this study was to develop a new PTF that: (i) abandons the use of the OM as a predictor and uses easily determinable predictors (soil texture, bulk density), (ii) estimates a continuous WRC allowing for a flexible use in unsaturated water flow models, and (iii) is valid across a wide range of soil textures and bulk densities.
|
MATERIALS AND METHODS
|
|---|
Calibration Data Set
The calibration took place on the basis of a data set containing 676 measured WRCs (353 topsoils, 323 subsoils), which were determined using undisturbed soil cores from 295 soil profiles. The data sets were obtained from two large international databases: the IGBP-DIS soil data set for pedotransfer function development (Tempel et al., 1996) and the UNSODA—Unsaturated Soil Hydraulic Database (Nemes et al., 1999a).
The selected soils represent a wide range of soil textures (see Fig. 1
). The definition of the particle-size fractions follows that of the USDA classification (clay <0.002 mm, silt 0.002–0.05 mm, sand 0.05–2 mm). A number of the selected UNSODA soils used the German particle-size fraction classification (particle-size diameter of 0.063 mm as a distinction between sand and silt). The values were converted to the USDA system using a log-linear interpolation (Nemes et al., 1999b; Nemes, 2003).
The selected WRC covered a matric potential range from 0 to –1500 kPa. For the calibration, eight matric potentials were used: 0, –1, –3, –10, –20, –50, –250, and –1500 kPa. Table 1 contains statistics of selected parameters that were used for the PTF development.
View this table:
[in this window]
[in a new window]
|
Table 1. Mean, standard deviation, median, maxima (Max.) and minima (Min.) of the soil properties of the calibration data set.
|
|
Figure 2
shows the distribution of the calibration data set across the soil textural classes. It indicates that they are not uniformly distributed, with the textural class of clay dominating the data set.
Validation Data Set
The PTF was validated from a data set of 147 WRCs measured on samples from 122 soil profiles (topsoils and subsoils) that were not included in the calibration data set. The following data were used: 49 WRCs (32 soils) from the UNSODA—Unsaturated Soil Hydraulic Database (Nemes et al., 1999a), 21 WRCs (21 soils) from the IGBP-DIS soil data set for pedotransfer function development (Tempel et al., 1996), 14 WRCs (six soils) measured by the Saxon State Institute for Agriculture (Leipzig, Germany), 36 WRCs (35 soils) measured by the Geological Survey of Lower Saxony (Hannover, Germany), and 27 WRCs (27 soils) measured by the Technical University of Berlin, Department of Soil Science.
The distribution of the validation samples within the particle-size triangle is shown in Fig. 3
. The validation data was selected depending on the availability of data on soil texture, bulk density, and OM. As the source of the data differed to some extent from the calibration data, the two data sets are not equally distributed. The relative proportions of the textural classes sandy loam, silt loam, and silt are clearly higher for the validation data than for the calibration data set. Table 2 contains statistics of selected validation data parameters. The validation data set contained several soils whose bulk densities lay outside of the range of the calibration data set. These were left in the data set to test the suitability of PTF for soils with low or very high bulk densities.
View this table:
[in this window]
[in a new window]
|
Table 2. Mean, standard deviation, median, maxima (Max.) and minima (Min.) of the soil properties of the validation data set.
|
|
Preliminary Analysis and Nonlinear Parameter Optimization
The function suggested by van Genuchten (1980) was used to generate the WRC. It describes the soil water content, 
, as a function of the matric potential, h:
 | [1] |
where r is the residual water content (m3 m–3), s is the water content at saturation (m3 m–3), and 
, n, and m are empirical parameters describing the shape of the curve. Following van Genuchten (1980), the parameter m was linked to the parameter n by the relationship m = 1 – 1/n.
Using preliminary data analysis, the linearity of the relationship between the model parameters and soil characteristics was examined. For this, Eq. [1] was adapted to the WRCs using a least-square analysis algorithm, done with the SOLVER routine in the software program Microsoft Excel (Microsoft Corp., Redmond, WA).
Plotting bulk density against OM for the calibration data clearly shows a negative correlation of the two soil properties (see Fig. 4
). This result complements the investigation by other researchers (Rawls, 1983; Manrique and Jones, 1991; Heuscher et al., 2005; Rawls et al., 2003). Heuscher et al. (2005) reported that 25% of bulk density variability could be explained by the OM values alone. In a similar investigation of the regression relationship between bulk density and the square root of organic C, Manrique and Jones (1991) found r2 values from 20 up to 53%. For the calibration data set tested here, the relationship was clearly weaker (r2 = 0.10). A reason may be attributed to the large variability in the bulk density values (see Table 1). When the topsoils were considered separately, the determination coefficient increased to a r2 value of 0.15. The comparison of the OM and the van Genuchten parameter s shows the expected positive correlation—increasing OM values are linked with increasing porosities or saturated water contents. Due to the large variability in the OM values, this relationship is also weak. The simultaneously strong dependence of the fitted s values on the bulk density values supports the hypothesis that it should be possible to take into account the effects of OM on water retention using only the bulk density. Here it must be pointed out that the bulk density can be considerably influenced by land use. This is especially true for topsoils, which are also subjected to weathering influences (wetting–drying cycles, freezing, thawing) to a far greater extent than subsoils. These influences can, of course, conceal the effect of the OM on bulk density and further repress the OM effects on the soil water retention.
View larger version (21K):
[in this window]
[in a new window]
|
Fig. 4. Box-whisker plots (median and 10, 25, 75, and 90% percentiles) of the van Genuchten saturated volumetric water content (s) and bulk density vs. organic matter content for the calibration data set.
|
|
A classification of the mass percentages of clay and sand and the bulk densities followed by a comparison of the class average with the model parameters (r, s, , and n) showed pronounced differences in the parameter dependence for soils with high sand contents (see Fig. 5
).
View larger version (29K):
[in this window]
[in a new window]
|
Fig. 5. Box-whisker plots (median and 10, 25, 75, and 90% percentiles) of the van Genuchten saturated volumetric water content (s) and fitting parameter n vs. sand content for the calibration data set.
|
|
To derive PTFs, a two-stage procedure is typically used. In the first step, the selected function is fitted to the measured WRCs. The estimated parameter values are then used in the second step to find dependencies between the parameter values and the selected soil properties. This procedure involves the risk of errors resulting from an overparameterization being transferred into the PTF. In this study, another procedure proposed by Scheinost et al. (1997) was used. In accordance with the preliminary data analysis, the parameters of Eq. [1] were connected with the soil properties using the following relations:
 | [2] |
where a, b, c, d, and e are the regression coefficients, "clay" and "sand" are the clay and sand content (% by weight) and Db is the bulk density (g cm–3). The model parameters (r, s, , and n) were optimized, now directly depending on the predictor variables.
Equation [2] was fitted to the measured data using nonlinear least-square optimization by minimizing the objective function F:
 | [3] |
where N is the number of measured soil water contents, M is the measured soil water content, and E the corresponding soil water content estimated using the van Genuchten parameters from Eq. [2]. It was assumed that with a calibration procedure of this type, correlations between the predictor and function parameter can be better demonstrated than with a calibration based on parameters that were determined independently. In many cases, the latter may reflect correlations between the individual parameters as well as from the soil properties. These intercorrelations influence the physical equivalence of the parameters and make it harder to analyze what the soil properties really depend on.
The minimization of Eq. [3] was done using the SOLVER routine in Excel. To exclude dependencies of the final results from the starting parameters, the SOLVER routine was coupled with a Monte-Carlo procedure. For this, 1000 optimization runs were performed where the starting parameters were chosen using a Monte-Carlo simulation. The probability distributions of starting parameters were predefined. For this, each German soil of the calibration data set was fitted using Eq. [2]. The following constraints for the r, , and n parameters were used: r 
0 m3 m–3, 0.00001 kPa–1, and n 1.01.
|
RESULTS AND DISCUSSION
|
|---|
Model Calibration
Using the procedure described, the regression coefficients of Eq. [2] were optimized. The first optimization (1000 runs) showed that the parameter r can be set to zero (in about 70% of the cases, the estimated r value turned out to be zero). This is in agreement with several researchers, who suggested that r could be omitted from fitting (Wösten et al., 1995; Rajkai et al., 1996; Schaap and Bouten, 1996). This reduced the number of independent parameters to three (s, , and n). To check the model prediction, the root mean squared deviation (RMSD) and the r2 value between the predicted and the measured water contents were calculated.
The preliminary data analysis showed different parameter dependencies for soils with a high sand content (see Fig. 5). A test of separate parameterization for these soils showed a notable improvement in the WRC prediction. To determine the threshold value for the sand content, the calibration data set was separated using a threshold value of 50% and Eq. [2] was fitted to both parts of the data set using the described procedure. Performing a stepwise increase of the threshold value followed by a recalculation of the regression coefficients and corresponding RMSD and r2 values showed that a threshold value of 66.5% leads to the most significant improvement in the WRC prediction.
The following regression coefficients were calculated:
sand content < 66.5%
 | [4] |
sand content 66.5%
 | [5] |
Table 3 gives an overview of the prediction quality arranged according to the individual soil textural classes. The determination coefficients were in the range of r2 = 0.92 to 0.96 for all textural classes. The lowest determination coefficients were achieved for the clay textural class, although even here, >90% of the variability could be explained.
View this table:
[in this window]
[in a new window]
|
Table 3. Means of the coefficient of determination (r2) and the root mean squared deviation (RMSD) for the single textural classes.
|
|
The preliminary analysis of the calibration data set indicates that in estimating s, bulk density is the most important predicting soil property. This close relationship was also found by Vereecken et al. (1989), using the same linear relationship. Remarkably, for the clay and silty soils, values almost identical with the regression coefficients of the Vereecken model were found (Vereecken s = 0.81 + 0.001clay– 0.283 Db). This is just as remarkable as the fact that the directly fitted van Genuchten parameters were not the starting point of the calibration, but were estimated during the model calibration as a function of the soil properties. Furthermore, in this study, different soil data were used from those in the study by Vereecken et al. (1989).
As indicated by the regression coefficient, bulk density explained most of the variation in the value for soils with a sand content <66.5%. The minus sign reflects the occurrence of an increase in pressure head at the air entry point (or a decrease in the volume of large pores) with increased bulk densities. As expected, this influence is smaller with the sandy soils. Sand and clay content have positive coefficients for all soils. For the clay content, a negative coefficient was also expected. An increase in clay content should be linked with an increase in the pressure head for the air entry point and therefore smaller value. At the same time, however, under comparable structural conditions, fine-textured soils often have more pore space and lower bulk densities than coarse-textured soils. This tendency can be found especially in German and central European soils.
Model Validation
With the validation data set, the performance of the PTF in this study was evaluated along with the performance of the Vereecken PTF (Vereecken et al., 1989). The Vereecken model uses sand and clay content, bulk density, and organic C content as predictors.
The validation indices were selected following Tietje and Tapkenhinrichs (1993). The following validation indices were chosen: the mean difference (MD, m3 m–3), the root mean squared deviation (RMSD, m3 m–3) between the measured and predicted water content, and the Pearson correlation coefficient (r, dimensionless):
 | [6] |
 | [7] |
 | [8] |
where the subscripts P and M indicate the predicted and the measured values. The upper (u) and lower (l) iteration boundaries were –0.01 and –1585 kPa, respectively. The value of MD can be positive or negative, indicating the bias of the estimation. The absolute value of MD should always be small. The RMSD is always positive and values near zero indicate a high model performance and are a measure of the overall error of the estimation. The Pearson correlation coefficient is a measure of the extent to which measured and predicted values are linearly related. Values near 1 indicate that the curve shapes have high comparability.
The measured and predicted functions in Eq. [6] through [8] are integrated between upper and lower boundaries. Therefore it is necessary to fit a continuous function to the measured data. In this study, the van Genuchten equation (Eq. [1]) was used. A similar procedure was selected by Cornelis et al. (2001). This method has the advantage that possible irregularities in the measured values are smoothed out. At the same time, it is possible to test the performance of the PTF across the entire range of hydrologically relevant matric potentials.
The van Genuchten parameters for measured WRCs were calculated by a least-square analysis using the SOLVER routine in Excel. The r, , and n parameters were estimated using the following constraints: r 0 m3 m–3, 0.00001 kPa–1, and n 1.01.
To complete the validation, the model performances were displayed graphically by plotting measured vs. predicted values. The following matric potentials and assigned soil water contents were used: –0.1, –1, –2.5, –6, –10, –15, –30, –60, –120, –500, –1000, and –1585 kPa.
Table 4 shows the mean values of the calculated validation indices for the new and Vereecken PTFs. From Table 4 it is clear that the performance of both PTF models is comparable. It was not possible to determine any statistically significant differences. The new model generally shows somewhat smaller RMSD and somewhat better correlations for the comparison between measured and estimated WRCs. This also becomes clear when the measured and estimated soil water retention figures are presented in a graph (Fig. 6
).
View this table:
[in this window]
[in a new window]
|
Table 4. Comparison of the mean validation indices (root mean squared deviation [RMSD], mean difference [MD], and Pearson correlation coefficient [r]) for the newly developed pedotransfer function (PTF) and the PTF by Vereecken et al. (1989) as calculated for the validation set.
|
|
Dependencies between MD, RMSD, or r and the soil texture or bulk density could not be found. The investigation into possible dependencies between the prediction quality and soil organic matter contents was of special interest. Here it can be determined that the quality of the prediction decreases somewhat as the C content increases, although the correlation between the OM and the estimation error is generally low (r2 for the correlation between OM and RMSD: new PTF = 0.023, Vereecken PTF = 0.017). Similar results were found by Tietje and Tapkenhinrichs (1993). It was expected that the Vereecken PTF, which uses the organic C content as a predictor, would result in better prediction for soils with high OM. There was no sign of this. Both PTFs tended to underestimate the soil water contents in the wet range of the WRC (mean difference at saturation for the new PTF = –0.017, for the Vereecken PTF = –0.022).
It can be assumed that in the new PTF the information loss due to neglecting OM is partly balanced by the selected calibration method. The intercorrelations between the soil properties (especially bulk density and organic C) can be more easily taken into account by directly fitting the regression coefficients to the measured WRC. This is also indicated by the comparison of the mean absolute prediction errors. Figure 7
shows that the new model results in comparable predictions and even slightly better predictions for the wet and the drier matric potential range than the Vereecken model, which takes into account the soil organic C.
View larger version (15K):
[in this window]
[in a new window]
|
Fig. 7. Mean of the absolute difference vs. matric potential and mean Pearson correlation coefficient of the two models.
|
|
The influence of the soil structure and the corresponding effects of the organic matter on the structure has the greatest effect in the wet range of the WRC. For drier conditions, the influence of the texture on the water retention increases (Hillel, 1998). The calculated values for the mean absolute difference of the new PTF (Fig. 7) indicate the largest error for the matric potential range of –10 kPa. But even at this matric potential, no difference can be found with the Vereecken PTF (with organic C content as a predictor).
In the discussion about the possible influences of OM on water retention, one general factor that must be pointed out is the often low organic C content in the data records used to develop PTF models. It should be noted that in large data sets, subsoil samples tend to be better represented than topsoil samples. Thus the proportion of the topsoil (samples from 0–30-cm depth) in both of the databases used in this work—UNSODA Unsaturated Soil Hydraulic Database (Nemes et al., 1999a) and IGBP-DIS soil data set (Tempel et al., 1996)—amounts only to approximately 30%. There is an intention to use extensive data records for the PTF development. This may lead to a skewed distribution of OM, with a majority of the samples having low C contents and an underrepresentation of the OM effects on the soil structure and the water-holding capacity. For PTFs that use the OM as a predictor, the corresponding effect on the water retention covered by the regression coefficients might therefore be smaller than if more samples with higher contents of organic C were used.
Hudson (1994) discussed the fact that a wide range of textures in data sets for PTF development could hide the effect of organic matter on soil water retention. Another aspect was discussed by Kay (1997). He described the different effects of varying forms of organic C in the soil on water retention. The larger the data sets become, the more varied the forms of organic C considered (due to agricultural management, landscape, type of vegetation, drainage, groundwater table depth, etc.) and therefore the smaller its influence on soil water retention becomes.
|
CONCLUSIONS
|
|---|
Pedotransfer functions are a basic tool for the parameterization of soil hydrological models. The majority of the published PTFs use information on soil texture, bulk density, and soil OM to predict the WRC. A further reduction of the input is particularly desirable for work on a large scale.
In this study, a new PTF is presented that forgoes the use of OM as a predictor. The new model allows the prediction of the parameters of the van Genuchten equation (Eq. [1]) using soil texture and bulk density only. A data set containing 676 measured WRCs from international databases was used for the PTF development. The PTF was calibrated by direct adjustment to measured WRC data and not to previously estimated van Genuchten parameters. For the calibration, the van Genuchten parameters were related to the soil properties by linear and non-linear equations. The substituted parameters were then fitted to the measured WRCs. Our investigations show that the bulk density can be a suitable parameter to indirectly represent the effects of the OM on water retention. The new PTF was successfully validated using an independent data set containing 147 measured WRCs. Additionally, the predictions were compared with the results of the PTF model proposed by Vereecken et al. (1989). It was possible to show that a comparable performance can be achieved with the new model despite smaller input. The prediction error is only marginally affected by including the OM in the list of input variables.
The bulk density is a dynamic soil property that can undergo rapid changes due to land use, weathering influences, or biological activities. Forgoing the OM as a predictor in PTFs can lead to a better reflection of this dynamic in WRC prediction.
For future investigation, omitting OM will allow a better and broader use of PTFs for investigating hydrologic questions. If data on the OM are available, PTFs should be used that use this property as a predictor (e.g., the PTF by Vereecken et al., 1989). That applies especially to soils with a high OM. Application of the functions derived in this study to soils with particle size distributions or bulk densities outside those shown in Table 1 should be treated with reasonable care.
|
ACKNOWLEDGMENTS
|
|---|
The work on this study was part of a project financed by the Deutsche Bundesstiftung Umwelt DBU (DBU file no. 21467). Some of the data sets used were inferred from the IGBP-DIS soil data set for pedotransfer function development (ISRIC, Tempel et al., 1996) and the UNSODA database (Nemes et al., 1999a). Thanks to Dr. Udo Müller, Geological Survey of Lower Saxony, Germany, and the Saxon State Institute for Agriculture, Germany, for providing additional data. We thank Prof. Klaus Bohne, University of Rostock, for his helpful comments on the manuscript.
|
NOTES
|
|---|
Abbreviations: MD, mean deviation; OM, organic matter content; PTF, pedotransfer function; r, Pearson correlation coefficient; RMSD, root mean squared deviation; WRC, water retention characteristic.
Received for publication March 3, 2006.
|
REFERENCES
|
|---|
- Arya, L.M., and J.F. Paris. 1981. A physico-empirical approach to predict the soil water moisture characteristic from particle size distribution and bulk density data. Soil Sci. Soc. Am. J. 45:1023–1030.[ISI]
- Campbell, G.S., and S. Shiozawa. 1989. Prediction of hydraulic properties of soils using particle-size distribution and bulk density data. p. 317–328. In M.Th. van Genuchten et al. (ed.) Proc. Int. Worksh. on Indirect Methods for Estimating the Hydraulic Properties of Unsaturated Soils, Riverside, CA. 11–13 Oct. 1989. Univ. of California, Riverside.
- Canarache, A. 1993. Physical-technological soil maps—A possible product of soil survey for direct use in agriculture. Soil Technol. 6:3–15.
- Cornelis, W.M., J. Ronsyn, M. van Meirvenne, and R. Hartmann. 2001. Soil Sci. Soc. Am. J. 65:638–648.
- Gijsman, A.J., S.S. Jagtap, and J.W. Jones. 2002. Wading through a swamp of complete confusion: How to choose a method for estimating soil water retention parameters for crop models. Eur. J. Agron. 18:75–105.[CrossRef]
- Gupta, S.C., and W.E. Larson. 1979. Estimating soil water retention characteristics from particle size distribution, organic matter content and bulk density. Water Resour. Res. 15:1633–1635.[CrossRef]
- Hall, D.G.M., M.J. Reeve, A.J. Thomasson, and V.F. Wright. 1977. Water retention, porosity and density of field soils. Soil Surv. Tech. Monogr. 9. Rothamsted Exp. Stn., Harpenden, UK.
- Heuscher, S.A., C.C. Brandt, and P.M. Jardine. 2005. Using soil physical and chemical properties to estimate bulk density. Soil Sci. Soc. Am. J. 69:51–56.[Abstract/Free Full Text]
- Hillel, D. 1998. Environmental soil physics. Academic Press, San Diego.
- Hudson, B.D. 1994. Soil organic matter and available water capacity. J. Soil Water Conserv. 49:189–193.
- Kay, B.D. 1997. Soil structure and organic carbon: A review. p. 169–197. In R. Lal et al. (ed.). Soil processes and the carbon cycle. CRC Press, Boca Raton, FL.
- Manrique, L.A., and C.A. Jones. 1991. Bulk density of soils in relation to soil physical and chemical properties. Soil Sci. Soc. Am. J. 55:476–481.[ISI]
- Minasny, B., A.B. McBratney, and K.L. Bristow. 1999. Comparison of different approaches to the development of pedotransfer functions for water-retention curves. Geoderma 93:225–253.[CrossRef][ISI]
- Nemes, A. 2003. Multi-scale hydraulic pedotransfer functions for Hungarian soils. Ph.D. diss. Wageningen Agric. Univ., Wageningen, the Netherlands.
- Nemes, A., M.G. Schaap, and F.J. Leij. 1999a. The UNSODA unsaturated soil hydraulic database. Version 2.0. U.S. Salinity Lab., Riverside, CA.
- Nemes, A., M.G. Schaap, and J.H.M. Wösten. 2003. Functional evaluation of pedotransfer functions derived from different scales of data collection. Soil Sci. Soc. Am. J. 67:1093–1102.[Abstract/Free Full Text]
- Nemes, A., J.H.M. Wösten, A. Lilly, and J.H. Oude Voshaar. 1999b. Evaluation of different procedures to interpolate the cumulative particle-size distribution to achieve compatibility within a soil database. Geoderma 90:187–202.[CrossRef][ISI]
- Oosterveld, M., and C. Chang. 1980. Empirical relationship between laboratory determinations of soil texture and moisture retention. Can. Agric. Eng. 22:149–151.
- Pachepsky, Y.A., R.A. Scherbakov, D. Varallyai, and K. Rajkai. 1982. Statistical analysis of correlations between soil water retention capacity and other physical properties. (In Russian.) Pochvovedenie 2:56–66.
- Pachepsky, Y.A., D. Timlin, and G. Varallyay. 1996. Artificial neural networks to estimate soil water retention from easily measurable data. Soil Sci. Soc. Am. J. 60:727–733.[Abstract/Free Full Text]
- Prevost, M. 2004. Predicting soil properties from organic matter content following mechanical site preparation of forest soils. Soil Sci. Soc. Am. J. 68:943–949.[Abstract/Free Full Text]
- Rajkai, K., S. Kabos, M.Th. van Genuchten, and P.-E. Jansson. 1996. Estimation of water-retention characteristics from the bulk density and particle-size distribution of Swedish soils. Soil Sci. 161:832–845.[CrossRef]
- Rawls, W.J. 1983. Estimating soil bulk density from particle size analysis and organic matter content. Soil Sci. 135:123–125.
- Rawls, W.J., and D.L. Brakensiek. 1982. Estimating soil water retention from soil properties. J. Irrig. Drainage Div. Am. Soc. Civ. Eng. 108:166–171.
- Rawls, W.J., and D.L. Brakensiek. 1989. Estimation of soil water retention and hydraulic properties. p. 275–300. In H.J. Morel-Seytoux (ed.) Unsaturated flow in hydrologic modeling; theory and practice. Proc. NATO Adv. Res. Worksh. Hydrology. NATO Sci. Ser. C. Kluwer Acad. Publ., Dordrecht, the Netherlands.
- Rawls, W.J., Y.A. Pachepsky, J.C. Ritchie, T.M. Sobecki, and H. Bloodworth. 2003. Effect of soil organic carbon on soil water retention. Geoderma 116:61–67.[CrossRef][ISI]
- Saxton, K.E., W.J. Rawls, J.S. Romberger, and R.I. Papendick. 1986. Estimating generalized soil-water characteristics from texture. Soil Sci. Soc. Am. J. 50:1031–1036.[Abstract/Free Full Text]
- Schaap, M.G., and W. Bouten. 1996. Modeling water retention curves of sandy soils using neural networks. Water Resour. Res. 32:3033–3040.
- Schaap, M.G., and F.J. Leij. 1998. Database-related accuracy and uncertainty of pedotransfer functions. Soil Sci. 163:765–779.[CrossRef]
- Schaap, M.G., F.L. Leij, and M.Th. van Genuchten. 1998. Neural network analysis for hierarchical prediction of soil hydraulic properties. Soil Sci. Soc. Am. J. 62:847–855.[Abstract/Free Full Text]
- Scheinost, A.C., W. Sinowski, and K. Auerswald. 1997. Regionalization of soil water retention curves in a highly variable soilscape: I. Developing a new pedotransfer function. Geoderma 78:129–143.
- Six, J., H. Bossuyt, S. Degryze, and K. Denef. 2004. A history of research on the link between (micro)aggregates, soil biota, and soil organic matter dynamics. Soil Tillage Res. 79:7–31.[CrossRef]
- Tempel, P., N.H. Batjes, and V.W.P. van Engelen. 1996. IGBP-DIS soil data set for pedotransfer function development. Int. Soil Reference and Inf. Centre, Wageningen, the Netherlands.
- Tietje, O., and M. Tapkenhinrichs. 1993. Evaluation of pedo-transfer functions. Soil Sci. Soc. Am. J. 57:1088–1095.[ISI]
- Timlin, D.J., Y.A. Pachepsky, B. Acock, and F. Whisler. 1996. Indirect estimation of soil hydraulic properties to predict soybean yield using GLYCIM. Agric. Syst. 52:331–353.[CrossRef]
- van Genuchten, M.Th. 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci. Soc. Am. J. 44:892–898.[ISI]
- Vereecken, H., J. Feyen, J. Maes, and P. Darius. 1989. Estimating the soil moisture retention characteristic from texture, bulk density and carbon content. Soil Sci. 148:389–403.[CrossRef]
- Williams, J., P. Ross, and K. Bristow. 1992. Prediction of the Campbell water retention function from texture, structure, and organic matter. p. 427–442. In M.Th. van Genuchten et al (ed.) Proc. Int. Workshop on Indirect Methods for Estimating the Hydraulic Properties of Unsaturated Soils, Riverside, CA. 11–13 Oct. 1989. Univ. of California, Riverside.
- Wösten, J.H.M., P.A. Fi, and M.J.W. Jansen. 1995. Comparison of class and continuous pedotransfer functions to generate soil hydraulic characteristics. Geoderma 66:227–237.[CrossRef][ISI]
- Wösten, J.H.M., A. Lilly, A. Nemes, and C. Le Bas. 1999. Development and use of a database of hydraulic properties of European soils. Geoderma 90:169–185.[CrossRef][ISI]
- Wösten, J.H.M., Y.A. Pachepsky, and W.J. Rawls. 2001. Pedotransfer functions: Bridging the gap between available basic soil data and missing soil hydraulic functions. J. Hydrol. 251:123–150.[CrossRef]
TOP
ABSTRACT
INTRODUCTION
