Agronomy Journal Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published in Agron J 91:984-990 (1999)
© 1999 American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Web of Science (6)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Agricola
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Related Collections
Right arrow Remote Sensing
Right arrow Other Models
Agronomy Journal 91:984-990 (1999)
© 1999 American Society of Agronomy

SOIL AND WATER

Errors Associated with the Use of Soil Survey Data for Estimating Plant-Available Water at a Regional Scale

Marie-Claude Fortina and David E. Moonb

a Pacific Agric. and Agri-food Res. Ctr. (Agassiz), Agric. Canada Res. Branch, Soil Sci. Dep., Univ. of British Columbia, 139-2357 Main Mall, Vancouver, British Columbia, Canada V6T 1Z4
b CDT Technologies, Richmond, British Columbia, Canada

fortinmc{at}interchg.ubc.ca


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 Materials and methods
 Results and discussion
 Conclusions
 REFERENCES
 
Agricultural models generally provide estimation procedures for soil properties that regularly are missing in data sets. In regional model applications, the inputs to these procedures are often derived from soil survey information. This study was conducted to determine two types of errors associated with the use of soil survey data for estimating plant-available water (PAW) for the Peace River region of British Columbia: the error associated with the use of an estimation procedure and the error associated with the use of soil survey data rather than measured data as inputs for the procedure. Two PAW estimation procedures (one used in CERES–Maize and in EPIC, and a recent update) were evaluated against laboratory-measured water-holding capacity. The original procedure did not perform adequately, with a prediction error of 0.10 compared with 0.04 for the updated procedure. Prediction error for procedure inputs derived from soil survey data were 8 to 18% of the value of the measured mean for particle size and as much as 51% for organic C. The updated procedure was relatively insensitive to input prediction errors. Prediction errors for horizon thickness were 38 mm for the Ap and 95 mm for the main B horizons, the single largest source of error in this study. Prediction errors for total PAW were 25 and 33% of the mean for the Ap and main B horizons, respectively. Tests for unbiasedness for total PAW failed. Field measurements are needed to validate the best of the two estimation procedures and to supplement the present horizon thickness values found in soil survey. These field measurements represent a significant investment of time and money, but are essential to optimize the allocation of resources for a modeling project at the regional scale.

Abbreviations: PAW, plant-available water • WHC, water-holding capacity


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
Materials and methods
 Results and discussion
 Conclusions
 REFERENCES
 
AGRICULTURAL MODELS

are increasingly being applied to large areas rather than only to fields or plots with the utilization of geographic information systems. Depending on complexity, agricultural models almost always require several soil properties as input data. These soil properties are not usually measured for large areas because data collection time and costs are prohibitive. Alternative data sources such as soil survey reports are often used because they are widely available and because they contain some of the data types needed for model input files.

Using soil survey data for such quantitative purposes goes far beyond the original purpose of most soil surveys. Soil surveys have, for the most part, been of a qualitative nature. Depending on the scale and intended purpose, surveys also vary in accuracy, detail, and complexity (Rogowski, 1996) and they rarely provide information on the level of error associated with the data (Mays, 1996). Consequently, results from modeling exercises that include soil survey data are hard to evaluate relative to their degree of accuracy or precision. This is a serious drawback as the reliability of a model is dependent on the reliability of the data (Bouma et al., 1996).

In addition to model inputs for which values can be found in soil surveys, there are other frequent model inputs, often soil physical properties, of which quantitative data are not commonly available. For example, soil water-holding capacity (WHC) and plant-available water (PAW), which are essential to soil water balance calculations, are needed in soil and plant simulation models. Soil WHC is usually defined as the difference between water retention or volumetric water at field capacity and water retention at wilting point. Soil WHC is an approximation of PAW, which is defined as the difference between the drained upper limit and lower limit of extractable water (Ritchie, 1981). Sometimes the two terms have been used interchangeably. Model users usually can access algorithms that will calculate such input variables from more readily available soil properties. For example, one of the most commonly used agricultural models, EPIC (Sharpley and Williams, 1990), uses as its default estimation procedure for soil WHC a series of equations first reported in the original version of CERES–Maize (Jones and Kiniry, 1986) for estimating PAW. This procedure predicts PAW by calculating the difference in estimated water contents at the drained upper and lower limits of water availability.

Such estimation procedures add uncertainty to model results, since their results are rarely formally tested by users for each location where the model is used. Benson et al. (1992) recognized the need for testing the estimation method for evapotranspiration and WHC in EPIC for various locations across the United States. They showed that the choice of the method used to estimate evapotranspiration and soil WHC can greatly affect model outputs (in their case, water quality indicators).

In regional applications, soil survey data with unknown reliability are being used with many of these estimation procedures. Such estimations are likely to increase the level of error associated with the model's results. On the other hand, interaction among errors from various sources may moderate the level of error associated with the model's results (Mays, 1996). Consequently, it is important for each researcher to evaluate the error associated with the various sources of information being used for modeling activities. Such evaluations provide prediction errors and help quantify the advantages associated with the use of measured data (Bouma et al., 1996).

Thus, the general goal of this study was to estimate two sources of errors associated with modeling PAW for the Peace River region of British Columbia. Specific goals were (i) to assess if the use of the PAW estimation procedure mentioned above (Jones and Kiniry, 1986) and a recent update of this procedure (Ritchie et al., 2000) are appropriate for the Peace River region of British Columbia and (ii) to assess the error associated with the use of data derived from soil survey data rather than measured data for these PAW estimation procedures and for the estimation of total PAW in the Ap and main B horizons.

The Peace River region of British Columbia is a northern Canadian Prairie farming area extending from the Alberta border to the foothills of the Rockies at the latitude of southern Alaska. Approximately 100000 ha of wheat and barley are grown annually in this region. The level of error associated with the estimation of PAW is needed to provide risk assessments with modeling results, because seasonal water deficits represent one of the major limiting factors for yield in the Peace River region (Abbaspour et al., 1992).


   Materials and methods
TOP
 ABSTRACT
 INTRODUCTION
 Materials and methods
Results and discussion
 Conclusions
 REFERENCES
 
Study Area
The study area, located in northeastern British Columbia, Canada (from latitude 55°30' to 56°30' N and from longitude 120° to 122° W), is a gently rolling till plain derived from Cordilleran bedrock, interspersed with glacial lake basins and dissected by large fluvial systems. Agriculture occurs primarily in valley soils, which are silty and clayey and have weakly calcareous stone-free glaciolacustrine or simply lacustrine deposits as parent materials (Farstad et al., 1965; Lord and Green, 1986). Forty-nine soil series are described in the area. Thirteen of these 49 soils were used in this study. They consist mostly of Alfisols (Boralfs and Cryoboralfs) and Mollisols (Natralbolls and Cryoborolls). Typical A horizons are silty loams; typical B horizons are clay.

Measured Soil Properties
In 1994, a sampling scheme was established to collect soil samples for analysis of particle size distribution, total C, bulk density, and depths of the Ap and main B horizons of the main wheat and barley growing areas of the Peace River region of British Columbia. Sample sites were selected by first stratifying on agricultural climate zone, elevation within a zone, and grain growing history within a climate–elevation combination zone. Then, within these areas, 1 to 4 fields (most often 3) were chosen for soil sampling, for a total of 30 fields. These fields ranged from 9 to 63 ha (avg. 37 ha). Each field was divided into three or four equal sections, depending on the size of the field. In July and August 1994, a randomly chosen site in each section was sampled. A pit was dug. A morphological description including horizon designation, depth, color, texture, soil structure, coarse fragments, and reaction with 0.1 M HCl was recorded at each site. Duplicate grab samples and duplicate soil cores (5 cm deep, 2 cm diam.) were taken in the Ap and main B horizons for a total of 110 sampling sites for the whole region.

Soil texture was determined in duplicate from the grab soil samples of each horizon at each sampling site using the pipette method (McKeague, 1978). Subsequently, samples of sites that belonged to the same field were pooled for the Ap and main B horizons separately. Total C was determined on these pooled samples at the Alberta Land Resource Unit of the Centre for Land and Biological Research Centre using a LECO induction furnace (Bremner and Tabatabai, 1971). Total C values were used as indicators of organic C for sites showing no presence of carbonates (no reaction to HCl) in the Ap and main B horizons. The duplicate soil cores were used for determination of water retention in the two horizons at each site. The cores were saturated from the bottom for 4 d, and water retention at 0.030 and 1.500 MPa was measured using pressure plates (Klute, 1986) to determine the volume fraction of water stored at field capacity and at permanent wilting point. The difference between the two water retention values was the soil WHC, an approximation of PAW. Although PAW is a field measurement (Ritchie, 1981), it was not measured directly, because it is very difficult to measure the lower limit of water availability in regions subject to rainfall from unpredictable convective storms during the growing season. The volume fraction of available water was multiplied by the depth of the Ap and main B horizons to determine the total water held in each horizon at each site. Bulk density was measured at 0.033 MPa.

Estimated Soil Properties
The survey procedures used for describing the Peace River region are typical of semireconnaissance surveys conducted on the Prairie region of Canada and are similar to soil survey methods in the United States. The basic soil mapping unit is the polypedon. Soil series are abstract classes of similar polypedons found to occur during the course of the survey. The series, once identified, is characterized by detailed descriptions of pedons chosen to represent the central or modal concept of that series (profile description; chemical and physical data). The 49 soil series compiled in British Columbia Soil Survey Report 42 (Lord and Green, 1986), which describes the wheat and barley growing area of the Peace River region, were mapped in approximately 100 mapping units. The survey represents a compilation of work dating from the late 1940s, but detailed soil descriptions were added between 1965 and 1980.

The study area was surveyed for presentation at a scale of 1:100000. Due to the small scale used in semireconnaissance surveys, mapping units represent a combination or a complex of soil associations. Mapping units are usually composed of an association of two or three series—sometimes up to six—but usually only the dominant two were reported for a given map delineation, with proportions determined by visual field estimations. For this reason, and because samples chosen for description were selected to represent a modal concept, these data cannot be used to estimate or describe variability.

A geographic information system comprised of a software for spatial data analysis (TerraSoft, Digital Resource Systems Ltd., Nanaimo, BC, Canada) and a database management system (dBASE III, Borland International, Scotts Valley, CA) was used to locate the randomly chosen sampling sites on the digitized soil maps of British Columbia Soil Survey Report 42 (Lord and Green, 1986). Sampling sites were located in a map delineation labeled as a specific map unit which identified the soils expected to be found in this delineation. To ensure data quality, sampling sites within 20 m of the boundary of a soil polygon or for which a profile description was not available were eliminated, reducing the number of sites from 110 to 70. Soil survey reports provided typical values for horizon depth, particle size distribution, organic C, and bulk density of the native A and main B horizons (Lord and Green, 1986; Farstad et al., 1965). Bulk density reported in the soil survey report was calculated from oven-dried weight and from volume assessments in situ at an unknown field moisture.

In the cases where map units involved a complex soil (more than one soil series), the dominant soil series properties were used for comparison with measured or observed data when the soil represented >50% of the mapping unit. If the dominant soil represented <=50% of the mapping unit, a weighted average of the properties of each soil was calculated based on the proportion of each soil present in the mapping unit. This weighted average was used for comparison with observed or measured data. Bulk density was included in the profile description for only 21% of the 70 sites. Thus, comparisons between soil survey–derived and measured bulk density were done for these sites only. Subsequently, the average soil survey–derived bulk density for the Ap and the main B horizons were used for all sites as inputs to the PAW estimation procedure.

Estimation of PAW
Several procedures exist for predicting PAW, soil water storage capacity, or the shape of the moisture retention curve based on soil properties (Oosterveld and Chang, 1980; De Jong, 1982; Rawls and Brakensiek, 1985; Ritchie et al., 1986). The PAW estimation procedure from Ritchie et al. (1986) was chosen because it is frequently used, as it is incorporated in two widely known agricultural models, CERES–Maize and EPIC. The PAW estimation procedure embedded in CERES–Maize (Jones and Kiniry, 1986) also constitutes the default EPIC soil WHC estimation procedure (Williams et al., 1990). This procedure uses percent sand, silt, clay, organic C, and bulk density as input data types to calculate the lower and the drained upper limit. PAW is calculated as the difference between the two limits. A recent improvement to this estimation procedure (Ritchie et al., 2000) was also used to estimate PAW in this study. This procedure uses percent sand, clay, organic C, and bulk density in addition to coarse fragments. Unlike the earlier procedure, it adjusts predictions for coarse fragments and organic C content, and places boundaries on the limits of water availability to avoid illogical predictions.

Results from the PAW estimation using measured input data were compared with values of soil WHC measured in the laboratory to verify that the estimation procedures were adequate for the Peace River region. The WHC values were used because no measured values of PAW were available for these sites for reasons explained above.

Subsequently, one of the two PAW estimation procedures (the new one) was used with inputs derived from soil survey data. The PAW estimations calculated using soil survey–derived inputs were compared with PAW estimations calculated using measured data. Finally, total PAW calculated from measured inputs and thickness of the Ap and main B horizons was compared with total PAW calculated from soil survey-derived inputs and horizon thickness. The error associated with the use of soil survey data for estimation of total water in the Ap and main B horizons was calculated.

Prediction Error and Statistical Analyses
Several simple methods can be used to validate estimation procedures (Bennett, 1981). First, prediction errors were calculated as

(1)
where Obs is observed value, Est is estimated value, and n is the number of observations. Prediction errors were used when comparing the values measured at individual sites with those determined from the soil survey reports. Error associated with an estimation at a regional scale is usually more important than at an individual site, because a regional value is an average over all sites in the region. Unbiasedness is very important when regional estimates are of interest because it requires that the mean of all possible estimates be equal to the mean of the parameter being estimated (Steel and Torrie, 1980). Thus, prediction–realization diagrams were made in which the predicted values of a variable were plotted against the observed ones and examined in relation to the 1:1 line. Statistical analysis to examine the relationship of the data to the 1:1 line included the calculation of a regression line forced through zero and subsequently, testing if the slope of the regression is equal to one using a Student's t-test. Testing that regression coefficient is equal to one is a test of bias. Statistical analyses were executed using the GLM procedure of SAS (SAS Institute Inc., 1988).


   Results and discussion
TOP
 ABSTRACT
 INTRODUCTION
 Materials and methods
 Results and discussion
Conclusions
 REFERENCES
 
Suitability of the Estimation Procedures for the Region
When the two estimation procedures described above (Ritchie et al., 1986, 2000) were used with measured input data, the resulting PAW as a volume fraction ranged from 0.02 to 0.12 for the 1986 method and from 0.14 to 0.18 for the 2000 method. These values are in the normal range of what is usually observed for similar mineral soils in temperate regions (Ratliff et al., 1983). However, the volume fractions estimated by the 1986 method are relatively small compared with volume fractions values typically measured for textures similar to those found at the sampling sites. These typical volume fractions usually range between 0.12 and 0.15 (Ratliff et al., 1983). Thus, the appropriateness of the 1986 procedure for estimating PAW appears questionable.

When measured values of WHC (Klute, 1986) were plotted against estimated PAW values of both estimation procedure using measured inputs, the data points did not fit well along the 1:1 line (Fig. 1 and 2) . The 1986 procedure underestimated PAW for both the Ap and main B horizons. Prediction error was 0.10. The 2000 procedure predicted similar PAW values for widely different WHC values. Prediction error was 0.04. In both cases, the original data used to develop the procedures did not include samples from this region. Another reason for the size of the prediction errors could be that laboratory-measured WHC is often different from field-measured PAW, which is the physical characteristic estimated by the two procedures (Ratliff et al., 1983). Although WHC and PAW are often used interchangeably, Ratliff et al. (1983) observed that laboratory-measured WHC values were 23% larger than field-measured water availability for silt loams, the most common soil texture in our Ap horizon. We observed a 16% difference between our laboratory-measured WHC values and our estimations of PAW using the 2000 method for the Ap horizon, the measured values being larger than the estimated ones (Table 1 ; Fig. 2). On the other hand, the 1986 procedure did not yield values that correspond to the expected difference between laboratory-measured WHC and field-measured PAW values in the Ap horizon (Table 1). In the main B horizon, there was no difference between the average of our laboratory-measured WHC values and the average of our estimations of PAW using the 2000 method, but there was a 36% difference using the 1986 method (Table 1). Given that the 2000 procedure yielded average estimations closer to measured values of WHC and a smaller prediction error than the 1986 method, the rest of the prediction analyses were done with the 2000 method only. The 2000 procedure itself is not without problems if a prediction error lower than 0.04 is needed, or if the simulation must be accurate for very small or large WHC values (Fig. 2). At present, without more work to measure field PAW rather than WHC or to validate another prediction equation, this is the best prediction error that can be obtained for this region.



View larger version (26K):
[in this window]
[in a new window]
  		Fig. 1 Regression and comparison of measured water-holding capacity and estimated plant-available water (PAW) calculated using measured soil properties as inputs to the procedure from Ritchie (1986) for 70 sites. Plant-available water is expressed as volume fraction

 


View larger version (24K):
[in this window]
[in a new window]
  		Fig. 2 Regression and comparison of measured water-holding capacity and estimated plant-available water (PAW) calculated using measured soil properties as inputs to the procedure from Ritchie et al. (2000) for 70 sites. Plant-available water is expressed as volume fraction. The estimation procedure is bound at 0.146 to avoid illogical results

 

View this table:
[in this window]
[in a new window]
  		Table 1 Mean values of measured water-holding capacity (WHC) and of estimated plant-available water availability (PAW) calculated using two different estimation methods and expressed as volume fraction. The percentage difference between means is relative to the measured values.{dagger}

 
Error Associated with the Use of Soil Survey–Derived Data for the Soil Characteristics Needed as Inputs for the two PAW Estimation Procedures
Particle-size distribution, bulk density, and organic C constitute the only inputs to the two PAW estimation procedures. Measurements of particle-size distributions indicated that the texture of most Ap horizons fell into the silt loam class, while the dominant textural class of the main B horizons was clay (data not shown). Measured particle-size fractions were compared with the soil survey–derived particle-size fractions (Farstad et al., 1965; Lord and Green, 1986). The data points were distributed around the 1:1 line for all three particle-size fractions (Fig. 3) . However, two of the three estimators failed the test for bias, because the regression equations forced through zero were significantly different from the 1:1 line for silt and clay fractions, as slopes were significantly different from one (Table 2) . Prediction errors for proportions of sand, silt, and clay ranged from 8 to 18% of the value of the measured mean (Table 3) .



View larger version (24K):
[in this window]
[in a new window]
  		Fig. 3 Regression and comparison of measured and soil survey–derived particle size fractions for 70 sites

 

View this table:
[in this window]
[in a new window]
  		Table 2 Statistics for regressions of measured and soil survey-derived particle-size fractions and for comparison of regressions to the 1:1 line for two horizons at 70 sites (Peace River region, British Columbia).{dagger}

 

View this table:
[in this window]
[in a new window]
  		Table 3 Measured mean values and prediction error (Pred. Err.) associated with the use of soil survey-derived data for various soil properties of 70 sites (Peace River region, British Columbia)

 
The prediction error for organic C was 51% of the value of the mean for the Ap horizon and 33% for the main B horizon (Table 3). The errors associated with sand, silt, clay, and organic C contents probably represent in part the variability of soil texture in the region, which is not quantified directly in the soil survey reports. In addition, some of the error associated with the Ap horizon may be due to the fact that A horizon information in the soil survey report was gathered solely from native A horizons.

The average measured bulk density was measured at 0.033 MPa, as required by the estimation procedures for the bulk density input. The average bulk density for the 70 sites was 1.30 ± 0.15 Mg m-3 for the Ap horizon and 1.58 ± 0.09 Mg m-3 for the main B horizon. On the other hand, bulk density could be derived from soil survey reports for only 15 sites, or 21% of the total number of sites. In the Ap horizon, measured bulk densities for these sites was 1.26 ± 0.01 Mg m-3 and the data derived from soil surveys averaged 1.33 ± 0.02 Mg m-3, a difference of 5% relative to the measured data. In the main B horizon, measured bulk densities for these sites averaged 1.56 ± 0.08 Mg m-3 and the data derived from soil surveys averaged 1.50 ± 0.05 Mg m-3, a difference of 4% relative to the measured data. These differences are relatively small given the fact that bulk density derived from soil survey data was determined at unknown water contents and given the fact that B horizons include clays well known to shrink and swell. Approximately half of the total clay in soils developed on lacustrine parent material in this region is montmorillonite, and a third consists of illite (Van Schaik and Pawluk, 1978).

Error Associated with the Use of Soil Survey–Derived Inputs for the Estimation of PAW
Soil survey–derived values of the four soil properties that are inputs to the 2000 estimation procedure (percent sand, percent clay, organic C, and bulk density) were used alternatively with observed values of the three other soil properties to calculate PAW (Table 4) . The average of these results was generally similar to the average of estimations obtained using measured input data only. The use of soil survey–derived data did not affect the result of the procedure except for organic C in the A horizon (Table 4). When the estimation procedure was used with all inputs derived from soil survey data, Ap horizon results were similar to results obtained with only organic C derived from soil survey data. In the main B horizon, the use of soil survey input data only instead of observed input data did not change the average PAW values (Table 4). Thus, the use of soil survey data did not affect the results of the 2000 estimation procedure, despite relatively large prediction errors for sand, clay, and organic C contents (Table 3) and despite important gaps in the bulk density information. Prediction error from using derived-only inputs relative to observed-only inputs was 0.01. Therefore, the procedure is quite insensitive to variations in soil inputs. This is in accordance with the fact that the 2000 procedure could not predict PAW values beyond a narrow range of values and could not simulate high and low WHC (Fig. 2).


View this table:
[in this window]
[in a new window]
  		Table 4 Mean and standard deviation of plant-available water (PAW) for the Ap and main B horizons of 70 sites in the Peace River region, British Columbia, calculated using an estimation procedure with either observed (Obs) input data or soil survey-derived (Drv) input data and expressed as volume fraction.{dagger}

 
Error Associated with the Use of Soil Survey–Derived Horizon Thickness and Inputs for the Estimation of Total PAW within a Horizon
The total amount of PAW in a soil horizon is defined as the volume fraction of available water in a horizon multiplied by the thickness of this horizon (Marshall and Holmes, 1988). On average, the Ap horizons were not as thick as the main B horizons (Table 3). Estimating horizon thickness from soil survey reports led to a prediction error of 38 mm for the Ap horizon and 95 mm for the main B horizon, which represents 21 and 33% of the values of the measured means (Table 3). The total amount of water calculated using soil survey–derived inputs for the 2000 procedure and soil survey–derived horizon thickness was compared with the total amount of water calculated using measured inputs and measured horizon thickness (Fig. 4) . The data were not well distributed around the 1:1 line. Consequently, the regression line was significantly different from the 1:1 line (Table 5) . The prediction errors were 7.3 mm for the Ap horizon and 13.5 mm for the main B horizon, which represents 25 and 33% of the value of the means obtained using measured data (Table 5). It appears that the use of estimated horizon thickness in total water calculation for specific horizons increases prediction error to levels that are unacceptable for agricultural or environmental applications. The description of native A horizons that no longer exist in agricultural situations and the lack of quantification of spatial variability within specific boundaries make soil survey information difficult to use for modeling purposes in our region. The Peace River survey is representative of most Canadian Great Plains surveys, and since most North American soil surveys share some of the characteristics of the soil survey used in this study, modelers who plan to use surveys as a source of data must consider examining the magnitude of the various errors associated with the estimation of their key variables.



View larger version (29K):
[in this window]
[in a new window]
  		Fig. 4 Regression and comparison of total plant-available water calculated using measured inputs and horizon thickness and of total plant-available water calculated using soil survey–derived inputs and horizon thickness. Both calculations were done using the estimation procedure for plant-available water from Ritchie et al. (2000)

 

View this table:
[in this window]
[in a new window]
  		Table 5 Mean values, prediction error, and statistics for the regression of total water calculated using measured inputs and horizon thickness in a plant-available water estimation procedure with total water calculated using inputs and horizon thickness derived from soil survey with the same estimation procedure.{dagger}{ddagger}

 

   Conclusions
TOP
 ABSTRACT
 INTRODUCTION
 Materials and methods
 Results and discussion
 Conclusions
REFERENCES
 
There is a definite need for fundamental assessments of the effect of using information from semireconnaissance-type surveys at a scale of 1:100000 for modeling purposes. Survey procedures worked well for the reconnaissance-scale planning objectives for which they were designed; i.e., samples of soil series were chosen to represent a modal concept of this series and not to estimate variability. Using soil survey information collected for the latter purpose is not adequate for predictive modeling objectives unless assessments of the level of error associated with the use of these data has been assessed.

In the case of the Peace River region of British Columbia, tests for unbiasedness failed when total PAW was estimated from soil survey–derived data. The largest source of error was due to the lack of accuracy for horizon thickness. Prediction errors associated with horizon thickness were 21 and 33% of the measured mean for the Ap and main B horizons. Such error levels are not acceptable for most modeling applications of agricultural and environmental processes.

There is also a need to test estimation methods, especially those embedded in widely used models for various locations. In our region, the two PAW estimation procedures cannot be used at present without substantial additional work. The 1986 procedure did not perform adequately and the recent update (Ritchie et al., 2000) of this procedure predicts PAW in too limited a range to account for the wide variations in WHC that were measured in the laboratory. WHC and PAW can differ substantially, but there are not enough data in the literature at present to assess if this difference can at least partially explain the performance of the 2000 procedure. Because the 2000 procedure is relatively insensitive, the use of soil survey–derived data did not greatly affect results compared with using measured data, but this situation could change if the procedure is refined and becomes more sensitive to some inputs. Consequently, the up-front cost of modeling PAW in the Peace River region will need to include expenses to (i) validate the 2000 PAW estimation procedure via field measurements of limits of water availability or else validate another estimation procedure and (ii) improve the accuracy and include measurements of variability for horizon thickness for the Ap and main B horizons of all major soil series. Procedures such as the one described here represent a significant investment of time and money, but are essential to optimize the allocation of resources for a modeling project at the regional scale. This is in turn essential if agricultural science is to make full use of recent and future improvements in scientific models and in geographic technology and not make inappropriate use of a large amount of data collected by surveyors for purposes entirely different from modeling.


   ACKNOWLEDGMENTS
 
The authors thank Drs. J.T. Ritchie, J. Hall, and W.W. Mohn for contributions to the preparation of this manuscript, Dr. C. Tarnocai for help with soil classification, and L. Kenney and S. Ulansky for technical help. This study was partially funded by the Agricultural Risk Management Branch of the British Columbia Ministry of Agriculture, Fisheries and Food.

Received for publication December 30, 1997.
   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 Materials and methods
 Results and discussion
 Conclusions
 REFERENCES
 




This article has been cited by other articles:


Home page
 Soil Sci.Home page
P. Jiang, S. H. Anderson, N. R. Kitchen, K. A. Sudduth, and E. J. Sadler
Estimating Plant-Available Water Capacity for Claypan Landscapes Using Apparent Electrical Conductivity
Soil Sci. Soc. Am. J., October 29, 2007; 71(6): 1902 - 1908.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Web of Science (6)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Agricola
Right arrow Articles by Fortin, M.-C.
Right arrow Articles by Moon, D. E.
Related Collections
Right arrow Remote Sensing
Right arrow Other Models

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Crop Science Vadose Zone Journal
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome