Journal list menu

Volume 108, Issue 3 p. 957-965
Agronomy, Soils & Environmental Quality
Open Access

Validating a Digital Soil Map with Corn Yield Data for Precision Agriculture Decision Support

Christopher W. Bobryk

Corresponding Author

Christopher W. Bobryk

USDA–ARS, Cropping Systems and Water Quality Research Unit, 269 Agriculture Engineering Building, Univ. of Missouri, Columbia, MO, 65211

Corresponding author ([email protected]).

Search for more papers by this author
D. Brenton Myers

D. Brenton Myers

DuPont Pioneer, 7300 NW 62nd Ave. P.O. Box 1004, Johnston, IA, 50131

Search for more papers by this author
Newell R. Kitchen

Newell R. Kitchen

USDA–ARS, Cropping Systems and Water Quality Research Unit, 269 Agriculture Engineering Building, Univ. of Missouri, Columbia, MO, 65211

Search for more papers by this author
John F. Shanahan

John F. Shanahan

DuPont Pioneer, 7300 NW 62nd Ave. P.O. Box 1004, Johnston, IA, 50131

Search for more papers by this author
Kenneth A. Sudduth

Kenneth A. Sudduth

USDA–ARS, Cropping Systems and Water Quality Research Unit, 269 Agriculture Engineering Building, Univ. of Missouri, Columbia, MO, 65211

Search for more papers by this author
Scott T. Drummond

Scott T. Drummond

USDA–ARS, Cropping Systems and Water Quality Research Unit, 269 Agriculture Engineering Building, Univ. of Missouri, Columbia, MO, 65211

Search for more papers by this author
Bob Gunzenhauser

Bob Gunzenhauser

DuPont Pioneer, 7300 NW 62nd Ave. P.O. Box 1004, Johnston, IA, 50131

Search for more papers by this author
Nadilia N. Gomez Raboteaux

Nadilia N. Gomez Raboteaux

DuPont Pioneer, 7300 NW 62nd Ave. P.O. Box 1004, Johnston, IA, 50131

Search for more papers by this author
First published: 01 May 2016
Citations: 25

Available freely online through the author-supported open access option

Abstract

Capturing the variability in soil-landscape properties is a challenge for grain producers attempting to integrate spatial information into the decision process of precision agriculture (PA). Digital soil maps (DSMs) use traditional soil survey information and can be the basis for PA subfield delineation (e.g., management zones). However, public soil survey maps provide only general descriptions of soil-landscape features. Therefore, improved DSMs are needed that use high-resolution data that more precisely model soil-landscape characteristics. Additionally, reliable methods are needed to validate DSM products for PA. The objective of this study was to validate with corn (Zea mays L.) yield data the performance of a new DSM product, termed Environmental Response Unit (ERU), compared with the USDA Soil Survey Geographic (SSURGO) soil map. The ERU was developed by integrating SSURGO information with high-resolution elevation data. For validation, corn yield maps were collected and corrected for common data collection errors from 409 fields across Indiana, Iowa, Minnesota, and Nebraska in 2010 to 2012. Reductions in the area-weighted variance (Rv) of corn yield for ERU and SSURGO were calculated relative to the whole-field variance. The average Rv across all site-years for SSURGO and ERU was 16 and 25%, respectively, which equated to a 57% higher median yield variance reduction with ERU over SSURGO. This variance reduction technique showed the potential of ERU as an improved model better representing soil-landscape properties that impact corn yield. This research also has application potential for determining the success of a DSM for identifying management zones in PA.

  • The variance reduction metric (Rv) provided a direct comparison between DSM models.

  • The greatest reduction in corn yield variance was exhibited by the ERU DSM model.

  • Physiographic derivatives from high-precision data sets improved DSM functionality.

  • Yield data helped outline interrelationships among various soil properties.

Abbreviations

  • DSM
  • digital soil map
  • ERU
  • environmental response unit
  • G×E×M
  • genetic potential, biophysical environment, and management
  • lidar
  • light detection and ranging
  • MZ
  • management zone
  • PA
  • precision agriculture
  • WF
  • whole field
  • Managing variability in crop productivity is a complex challenge due to fundamental interactions between genetic potential (G), biophysical environments (E), and management (M) activities (G×E×M) (Ortega and Santibáñez, 2007). The components of G×E×M are multidimensional and vary across space and time. A key PA approach taken to simplify spatial complexity is the division of fields into subfield areas, often called management zones (MZs), that are managed individually with respect to important agronomic decisions (Fridgen et al., 2004; Schepers et al., 2004; Kitchen et al., 2005; Cox and Gerard, 2007; Derby et al., 2007; Thorp et al., 2008; Moshia et al., 2014). Digital soil mapping has similar goals for partitioning the spatial variability of fundamental soil-landscape factors into more uniform delineations (McBratney et al., 2003; Roberts et al., 2012). Additionally, DSM products can be used as a basis for MZs (McBratney and Minasny, 2007; Triantafilis et al., 2009; Omran 2012; Davatgar et al., 2012). Several methodologies have been introduced for developing DSMs based on soil or geophysical terrain attributes (Moore et al., 1993; Johnson et al., 2003; Kitchen et al., 2005; Mulla 2013); however, less work has been done to validate their success in PA (Bishop et al., 2001; Whelan and McBratney 2000). Therefore, methods for developing and validating DSMs across diverse agricultural crop production regions are needed.

    One common approach for creating MZs has been to use public soil maps, such as the NRCS SSURGO database (Ortega and Santibáñez, 2007; Soil Survey Staff, 2014). The resolution needed to develop MZs for PA, however, exceeds that typically found in SSURGO products (Robert 2002; Zhu et al., 2013). The challenge is that traditional soil survey methods, like SSURGO, are difficult and expensive to apply at fine resolutions because physical observations of mapped locales are required for quality control (Grunwald et al., 2011). Thus, the low-resolution, publicly available DSM products have limited value in PA (Van Alphen and Stoorvogel 2000; McBratney et al., 2005; Miller 2012; Landrum et al., 2015). Therefore, procedures to create DSM products with higher resolution and greater accuracy could improve the quality of DSM for defining MZs (McBratney et al., 2003; Scull et al., 2003).

    Two key components in the development of DSM products include: (i) the methodology used to define discrete or continuous relationships between soil observations and environmental variables (McBratney et al., 2003); and (ii) the approach used to assess the DSM output estimates (Carré et al., 2007). Continuous DSM methods develop models of numerically and spatially continuous soil properties. These models can be used to map soil properties for surface horizons (Moore et al., 1993), as profile averages (McBratney and Pringle, 1999), in layers or horizons (Ben-Dor et al., 2008), or as continuous depth functions (Malone et al., 2011; Myers et al., 2011). Discrete DSM methods predict soil taxonomic memberships as distinct nominal classes (Bui and Moran, 2001; Behrens et al., 2010; Nauman and Thompson, 2014; Odgers et al., 2014) or membership likelihood vectors derived from continuous probabilistic or fuzzy-logic procedures (Zhu et al., 2001; McBratney and Odeh, 1997). Implementation of DSM processes is commonly executed using machine-learning (Behrens et al., 2010), expert systems (Zhu et al., 2010; Duarte de Menezes et al., 2013), or geostatistical (Lark, 2000; Cressie and Kang, 2010) techniques.

    New methodological approaches to digital soil mapping continue to emerge and are routinely being used in soil mapping programs across the globe (Grunwald et al., 2012); however, agricultural producers implementing PA practices have expressed concerns about DSM quality and relevance to MZ applications (Krol 2008; MacMillan 2008). Specifically, there is concern regarding how new DSM products directly compare with more traditional soil map-unit systems (Krol, 2008), such as SSURGO. The quality of the output from a DSM procedure needs to be assessed to indicate improvements in the digital representations of soils for PA decision support. Typically, testing DSM products stems from point observations of the target variables using cross-validation approaches (Brus et al., 2011) or mean squared error scoring functions (Gauch et al., 2003). The quality of soil map-unit delineations may also be determined as a function of a classification precision metric (Bishop et al., 2001; Nauman and Thompson, 2014) that calculates the accuracy of soil information within map units compared with known class observations. Other approaches assess the appropriate number of map units that may exist across a landscape (Bishop and McBratney, 2001; Bishop et al., 2001; Fridgen et al., 2004).

    Indices have been developed to assess pedodiversity and have application potential for assessing the success of a DSM tailored for PA. Performance indices, such as the fuzziness performance index and normalized classification entropy, have been used to evaluate the appropriate number of clusters for identifying homogeneous map units in unsupervised classification approaches (Boydell and McBratney, 2002; Fridgen et al., 2004; Davatgar et al., 2012). McBratney and Minasny (2007) introduced a theory for measuring soil variation based on a modified version of Shannon's information criterion, which could be used to decipher taxonomic similarities or differences between soil map-unit classes. Bishop et al. (2001) used information criteria as a quantitative measure of soil map-unit quality, where the proportion of information calculated in the partitioned soil map was compared with the total soil information observed for the field, as measured from empirical soil samples.

    In addition to comparing directly between different DSM products, assessment procedures have been developed to specifically determine the quality of a DSM based on the ability of map-unit partitions to capture variability in landscape attributes that include soil properties, topography, and crop yields (Schepers et al., 2004; Diacono et al., 2012). Confirming the efficacy of new DSM techniques in accounting for within-field variation remains an important step toward understanding the quality of DSM products for making PA management decisions (Triantafilis et al., 2009; Omran 2012). For example, spatial maps of yield (e.g., corn) and soil properties (e.g., apparent electrical conductivity) are common outputs from PA activities and can be used to quantify improvements between DSM products (Fraisse et al., 2001; Kitchen et al., 2005).

    A common approach for assessing the quality of a DSM for PA is needed to repeatedly test how well new map-unit areas may inform and enable better management decisions at local and regional scales. This remains a current challenge for broad implementation of DSMs in PA, and to date, there are no studies assessing DSM quality using multiple years of crop yield maps from numerous fields at regional scales. The increasing use of yield monitors on crop harvesting machines provides a robust source of spatial and temporal biomass information (McBratney et al., 2003). These biomass data can be used to implement assessment procedures for understanding how DSMs are more precisely delineating yield variation at finer scales. Ultimately, such an assessment methodology has the potential to help guide future PA management decisions if it can be easily implemented at various spatiotemporal scales and found applicable using different types of grain yield data.

    In this research, a new DSM product called ERU was investigated. This new DSM was the product of a public–private collaboration between the USDA–ARS, the University of Missouri, and DuPont Pioneer (Heggenstaller and Munaro, 2015). The ERU was developed by integrating SSURGO information, high-resolution elevation data, and watershed boundaries to create a functional DSM capable of improved identification of soil-landscape crop response environments. The objective of this research was to validate, using a yield variance reduction methodology, the performance of the ERU DSM compared with SSURGO.

    MATERIALS AND METHODS

    Study Area and Data Sets

    Corn yield data from 409 fields located in northeastern Indiana, central Iowa, southwestern Minnesota, and southeastern Nebraska (Fig. 1) were used with a variance reduction methodology for validating ERU. Corn yield information was collected by producers in 2010, 2011, and 2012. Yield data were captured using combines equipped with yield monitors coupled with GPS receivers. Specific details of the data collection on each field were not recorded in the database available to this project. To standardize the data to the extent possible, the raw yield data were screened for erroneous values and outliers using Yield Editor 2.0 (Sudduth and Drummond, 2007), and the edited data sets were used for this analysis.

    Details are in the caption following the image

    Locations of 409 study fields located throughout the midwestern United States that includes Indiana, Iowa, Minnesota, and Nebraska.

    Comparison of Environmental Response Unit Digital Soil Map with SSURGO

    Both the ERU DSM and SSURGO were evaluated relative to the whole-field (WF) variance. The whole field represents the traditional scale of management utilized by many producers (i.e., non-PA), and the field boundaries were identified using GIS vector data. Soil map units were acquired from the USDA–NRCS SSURGO (Soil Survey Staff, 2014) as vector data. The SSURGO data set represented a level of field segmentation defined by taxonomic identifiers, site, physical and chemical soil properties, and interpretive landscape attributes. The ERU DSM was based on SSURGO information enhanced with higher resolution elevation maps derived from light detection and ranging (lidar) data. We used the WF as the baseline delineation against which to compare and validate the reduction in yield variance between ERU and SSURGO soil map units across all fields.

    Digital Soil Map Procedure for Developing Environmental Response Units

    The DSM procedure used to develop ERU map units for fields in this study was part of a decision-support system offered by Encirca Services of DuPont Pioneer (Heggenstaller and Munaro, 2015) and combined methods similar to those used by Bui and Moran (2001), McBratney et al. (2003), and Nauman and Thompson (2014) (Fig. 2). The initial step was to identify the appropriate watershed containing each of the 409 fields using the publically available 12-digit Hydrologic Unit Codes (NRCS, 2013). For each location, a set of seven geospatial terrain attributes were derived from lidar, SSURGO, and a stream network data layer and stored within a GIS. The covariates used in the modeling process consisted of: (i) area above the channel network; (ii) catchment; (iii) curvature; (iv) elevation; (v) slope; (vi) topographic position index; and (vii) topographic wetness index. All variables were used as input for a machine learning process using R statistical software (R Core Team, 2015). The output of the modeling process was a 5-m-resolution DSM. A GIS area-class filtering procedure was used to eliminate small (<0.2 ha) map units, and the final DSM product was a vector data set.

    Details are in the caption following the image

    General schematic of the process for creating an Environmental Response Unit digital soil map. Soil map units were matched with various high-resolution environmental covariates that served as input predictors for a machine learning algorithm.

    Calculating Yield Variance Reduction

    We adapted the weighted-yield variance reduction approach used by Fraisse et al. (2001) to produce a statistic to quantify the performance (or ability) of SSURGO and ERU DSMs to characterize the homogeneity of corn yield environments. The syntax of the equation from Fraisse et al. (2001) was slightly modified to simplify the presentation of the final variance reduction equation and present a single value that could be easily compared between map-unit types. The area-weighted yield variances (urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0003) of each map-unit segmentation S for each field in the data set were calculated as
    urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0001(1)
    where wi is the proportion of the field area contained within the ith map unit, and σi2 is the variance within i ∈ [1,n] map units due to segmentation S. The area-weighted yield variance was used to obtain an objective comparison of map-unit polygon configurations, sizes, and densities of yield values for different map-unit types.
    We then calculated the variance reduction (Rv) for each yield map to reflect the average homogeneity of yield within map units from a given field segmentation:
    urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0002(2)
    where urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0004 is either urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0005 or urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0006, and urn:x-wiley:00021962:agj2agronj20150381:equation:agj2agronj20150381-math-0007 is the WF yield variance. Increased Rv of one field segmentation approach compared with another represents a relative improvement.

    Statistical differences in Rv between SSURGO and ERU were determined using Student's t-test at a significance level of α = 0.05. The DSM process and all analyses were performed using R statistical packages. Yield data and DSM output were visualized using ArcGIS (ArcGIS 10.2, ESRI).

    RESULTS AND DISCUSSION

    Yield and Yield Variance Reduction

    Average corn yields across all the years in each study field (Indiana, Iowa, Minnesota, and Nebraska) were 9.1, 10.9, 11, and 10.4 Mg ha−1, respectively (Fig. 3). The number of fields per each state was 77, 154, 101, and 77, respectively (Table 1). Average yield by year was 10.8, 10.9, and 9.4 Mg ha−1 for 2010, 2011, and 2012. During 2012, the US agriculture sector experienced one of the most severe and extensive droughts in recent history, where reduced precipitation and high temperatures caused significant stress in corn during critical phases of development (Al-Kaisi et al., 2013). Even though there were no statistical differences among average yields across all 3 yr in this study, we observed that Indiana fields were most negatively affected in 2012, when yields averaged 6.8 Mg ha−1. The 2012 drought was attributed to natural variations in weather patterns, which caused insufficient warm-season rains for the region (Hoerling et al., 2014). A lack of precipitation in concert with high temperatures led to an excessive soil moisture deficit (Porter and Semenov, 2005), causing portions of Indiana to reach the exceptional drought condition category (Heim, 2012; Mallya et al., 2013).

    Details are in the caption following the image

    Total corn yield for each year by state and a final average. The bars indicate standard error.

    Table 1. Summary of average total area-weighted variances of corn yield between whole-field (WF), SSURGO, and Environmental Response Unit (ERU).
    Data set Average total area-weighted variance
    n WF SSURGO ERU
    (Mg ha−1)2
    All data 409 5.6 4.6 4.1
    Indiana
    2010 17 4.6 4.2 3.8
    2011 28 4.2 3.7 3.3
    2012 32 4.9 3.7 3.3
    All years 77 4.6 3.8 3.4
    Iowa
    2010 26 4.7 4.1 3.8
    2011 37 2.6 2.3 2.1
    2012 91 7.3 5.6 5.0
    All years 154 5.7 4.5 4.1
    Minnesota
    2010 33 4.6 4.2 3.8
    2011 29 4.2 3.7 3.3
    2012 39 4.9 3.7 3.3
    All years 101 4.3 3.5 3.0
    Nebraska
    2010 26 4.6 4.2 3.8
    2011 15 4.2 3.7 3.3
    2012 36 4.9 3.7 3.3
    All years 77 8.0 6.9 6.1

    The two soil-landscape classification types (SSURGO and ERU) were assessed using the Rv metric that represents a reduction of within-map-unit variability in corn yield data. For all site-years, the average Rv for SSURGO and ERU compared with WF was 0.16 and 0.25, respectively (Table 2). Both the SSURGO map units and ERU DSM captured more yield variance than the WF; however, the ERU DSM showed a greater reduction in area-weighted variance than SSURGO (Fig. 4). The largest difference between Rv values for the ERU DSM and SSURGO was observed for Minnesota at 0.13 and the least for Iowa with 0.07 (Table 2). Significantly, the median reduction in Rv from SSURGO to ERU was 57% (Fig. 5). When comparing ERU with SSURGO, all but one state–year combination showed a significant improvement in the Rv metric (Table 2).

    Table 2. The fractions of corn yield variance reduction (Rv) for the Environmental Response Unit (ERU) and SSURGO map units, along with the probability of a significant difference between the two based on results from multiple Student's t-tests.
    State Year n Rv
    ERU SSURGO p value
    All states all years 409 0.25 0.16 <0.001
    Indiana 2010 17 0.15 0.09 0.014
    2011 28 0.21 0.11 <0.001
    2012 32 0.3 0.21 0.006
    all years 77 0.24 0.15 <0.001
    Iowa 2010 26 0.19 0.12 0.006
    2011 37 0.19 0.13 0.002
    2012 91 0.3 0.22 <0.001
    all years 154 0.25 0.18 <0.001
    Minnesota 2010 33 0.22 0.12 <0.001
    2011 29 0.26 0.13 <0.001
    2012 39 0.3 0.18 <0.001
    all years 101 0.26 0.15 <0.001
    Nebraska 2010 26 0.24 0.14 0.002
    2011 15 0.21 0.14 0.136
    2012 36 0.24 0.14 <0.001
    all years 77 0.23 0.14 <0.001
    Details are in the caption following the image

    Total corn yield variance reduction for an Environmental Response Unit (ERU) digital soil map model and NRCS SSURGO database map units compared with whole-field variance.

    Details are in the caption following the image

    Change in corn yield variance reduction from NRCS SSURGO database to Environmental Response Unit (ERU) digital soil map units. The inset provides a closer illustration of the change distribution around the median variation reduction from ERU of 57%.

    In this analysis, yield variance reduction with ERU was greater than with SSURGO for 86% of the fields. The resolution at which the ERU DSM was produced allowed a more accurate delineation of soil-landscape properties affecting corn grain yield (Fig. 6). Yield variance reduction differed from year to year and state to state, but the greatest Rv was observed for both map-unit types during 2012 (Fig. 7). Both SSURGO and ERU area-weighted variances decreased with respect to WF as the variance increased; however, the decrease was larger with ERU (Fig. 8).

    Details are in the caption following the image

    A soil map-unit diagram that illustrates the Environment Response Unit (ERU) and SSURGO map-unit boundaries in relation to the 2012 corn yield map for a field in Minnesota.

    Details are in the caption following the image

    Total corn yield variance reduction values for the Environmental Response Unit (ERU) digital soil map model and NRCS SSURGO database map units compared with whole-field boundaries by year.

    Details are in the caption following the image

    Total area-weighted variances of the Environmental Response Unit (ERU) digital soil model (black) and NRCS Soil Survey Geographic (SSURGO) database (red) with linear trend lines and associated 95% confidence intervals compared with whole-field variance (represented by solid black 1:1 line).

    This study included yield maps from 3 yr that represented a gradient of drought intensity. The 2010 growing season was generally favorable for the areas within the study. In 2011, many areas of the US Midwest experienced moderate drought late in the growing season, but in 2012 an extreme drought was experienced throughout much of the Corn Belt, resulting in diminished yields throughout the region (National Agricultural Statistics Service, 2015). Both SSURGO map units and ERU DSMs demonstrated an increasing Rv from 2010 to 2012 (Fig. 6), where the ERU DSMs had the greatest Rv, complementing the trend in drought gradient.

    The incidence of poor weather conditions during 2012 was an important factor in the observed reduction in variance produced by the ERU DSMs. Suboptimal weather conditions are important for drawing out key soil-landscape and hydrological characteristics that drive G×E×M interactions for grain production (Fullmer et al., 2014). The increased crop stress from droughty growing conditions resulted in altered patterns of corn development, which generated greater variations in observed yield (Fig. 7). The changes in the environment in turn allowed the ERU DSMs to better account for dynamics in the soil-landscape conditions that drive variability within crops.

    Value of Yield Maps

    Crop yield data were an appropriate data set to test the effectiveness of ERU DSMs for minimizing variance because yield directly and indirectly accounted for the spatial variability of soil, nutrient, landscape, and other factors that contributed to the soil-landscape component of G×E×M. Beyond soil and landscape features, variations in yield map data have also been attributed to natural (E) disturbances (e.g., pests, weeds, and diseases), as well as anthropogenic (M) factors (e.g., subfield management variability, historic management practices, field edge effects), and issues with harvesting operations (e.g., lack of good yield monitor calibration) (Shatar and McBratney, 1999; Lapen et al., 2001). Even though common errors associated with combine yield mapping were removed from this data set using proven procedures (Sudduth and Drummond, 2007), significant yield variations associated with differing yield measurement methods undoubtedly remained. Measurement-induced yield variation is magnified when fields are harvested with multiple combines and operators or have sensing systems that “drift” and do not receive regular load calibration. Digital soil maps, such as ERU, may have greater utility when combined with agronomic models that account for soil interactions with other E or M factors. While yield maps were used here to show that the ERU DSM was better than SSURGO at accounting for yield variation, soil is only one component within the E term of the G×E×M model.

    Assessing Digital Soil Maps for Application in Precision Agriculture

    The identification of soil classes across multiple fields is a difficult process and dependent on interrelationships among various physical and chemical properties (Machado et al., 2000). The quality of management decisions strongly depends on accurate information provided by soil maps, which offer insight into how crops respond to soil and landscape conditions (Bui and Moran, 2001). High-resolution DSM approaches, coupled with an assessment procedure as presented in this study, are an effective tool set for producers to gain confidence in PA approaches to management. An important outcome of this research is the knowledge gained from assessing the improved function of a DSM that was based on physiographic derivatives from high-precision data sets. This also indicated that topographic concepts used to develop traditional soil survey maps can be translated to improve DSMs by integrating newly available data sets such as real-time kinematic global navigation satellite system surveys and lidar.

    There is an increasing global demand for quantitative soil information, and DSM processes are likely to fulfill this need. Parallel to improving current soil information products, there is a need to improve approaches for assessing DSM output to better inform PA management decisions for producers (Cook et al., 2008). It remains a critical goal in DSM research that continental and global environmental covariates be incorporated to help support and guide society toward sustainable management practices (Cook et al., 2008; Grunwald et al., 2011). Spatially explicit DSMs that utilize environmental covariates with finer spatial resolutions while matching other environmental variables (e.g., hydrology, elevation, land-use change, etc.) have a greater probability for capturing the inherent variability of soil and landscape properties.

    Categorical but high-resolution DSMs, such as ERU, are promising advances toward applying PA decision models at larger landscape scales by improving the computational and on-the-ground efficiency of implementation (Zhu et al., 2013). For instance, process-based soil and landscape models are currently being utilized in large-scale commercial projects to provide management information for variable-rate seeding prescriptions (Shanahan and Kitchen, 2014), N applications (Heggenstaller and Munaro, 2015), and in the development of hyper-resolution global land surface models (Wood et al., 2011). Functional soil mapping is also gaining attention where developing high-resolution DSM specifically for crop yield management is helping to facilitate improved classification of soil types and properties (Zhu et al., 2013). Ultimately, continued model development, calibration, and implementation rely on computational simulations. Better soil information will make these efforts more effective and improve their utility for producers in the future (Finke, 2006; Cook et al., 2008).

    Future Directions in Digital Soil Map Assessment

    Investigating the accuracy of DSMs using crop yield variation as a metric of improvement continues to be an important research direction at multiple scales (McBratney et al., 2003; Johnson et al., 2005). Yield maps are an important resource for assessing DSM models and should be leveraged as an efficient source of information. The ability of DSMs to integrate higher resolution information creates the possibility of capturing G×E×M interactions across the field and with time, which makes the DSM approach highly useful in PA management. The area-weighted variance reduction performance index was one approach that provided a direct comparison between the ERU DSM and SSURGO map units. Comparing the area-weighted approach with other methods that analyze spatial structure, information content, or beyond may provide additional insight into the improvement of future DSM methods.

    Finally, while the tested ERU DSM showed a greater variance reduction than SSURGO, we presume that additional yield variation within yield-map data exists in relation to other soil and landscape features not currently captured with ERU. Within-field yield variation for any given year also originates from recent and historical management practices, genetics when not held constant within a field, and in the data collection process of yield mapping. These features aside, we suggest that yield map data are essential for future high-resolution DSM development and testing. We further suggest that DSM developments focus on improved methodologies for characterizing soil resources across landscapes, perhaps eventually moving past classification outcomes (e.g., MZ) to models that represent the soil as a continuum. We hypothesize that such developments will provide better explanations for yield variation within fields.

    CONCLUSION

    This study used corn yield maps in combination with a variance reduction algorithm to validate a high-resolution ERU DSM. The validation index (Rv) was calculated on corn yield data for both the ERU DSM and SSURGO map units, each compared with the WF variance. The greatest reduction in corn yield variance was exhibited by the ERU DSM, and as such, that model provided a better classification framework than SSURGO for capturing within-field soil variation. The variance reduction metric was a practical method for comparing measures of variability between soil map-unit types (SSURGO and ERU) and a baseline (WF) because yield data are readily available from producers and can be used to test various types of DSMs. Specifically, we concluded that the ERU DSM of this investigation classified similar soil and landscape characteristics in a way that will allow improved PA management applications by using yield to outline the interrelationships among various physical and chemical properties.