6.3   Quantitative Soil-Landscape Modeling to Define Landform Management Segments

Principal Investigators

James A. Thompson, Assistant Professor, Agronomy,

John H. Grove, Associate Professor, Agronomy

Eugenia M. Pena-Yewtukhiw, Graduate Assistant, Agronomy

Cooperators

George Hupman and Philip Lyvers, Loretto, Kentucky (Eastern Pennyroyal)
Milton and Furman Cook, Princeton, Kentucky (Western Pennyroyal)
Third location plus additional validation sites to be identified.

Introduction

Precision agriculture is the collective practice of operating a farming enterprise in consideration of the unique circumstances of soils, climate, crops, and pests within a field. The crux of precision agriculture is in knowing and accounting for the variability of site characteristics within a field. Commonly, our ability to quantify this variability is limited by the high cost, in both time and money, of compiling and analyzing data on soil and crop variability.

The objective of this research is to develop methods for assessing soil variability that minimize soil sampling, yet accurately depict within-field variability at resolutions necessary for precision agricultural management. These methods are based on a soil-landscape modeling approach and will provide a framework for assessing soil variability in a consistent manner among soils of similar environments (e.g., within a single physiographic region).

Objectives

Specific objectives of this sub-project are:

1)      To develop quantitative soil-landscape models that predict the variability in static soil properties related to crop yield (A-horizon thickness, organic C content, and clay content) for multiple fields in the Crider-Pembroke soil association in the Pennyroyal physiographic region;

2)      To examine the similarity of these quantitative models, and therefore the similarity of soil-landscape relationships, across the Pennyroyal physiographic region; and

3)      To use these results to develop sampling protocols (e.g., stratified, directed, and based on knowledge of within-field topographic variability) that reduce the number of soil samples and analyses needed to characterize the within-field soil variability.

Soil Variability and Crop Response

The most common method of assessing soil variability for precision agriculture is through grid sampling. The most common sampling intensity is a 330-ft grid, which represents one sample for every 2.5 acres. This sampling intensity can be inadequate because much soil variability occurs over distances much shorter than 330 ft (James and Wells, 1990; Pocknee et al., 1996). Another concern over grid sampling is that composite samples are clustered at the center of each grid cell, and only represent a small area in the center of each cell (Pocknee et al., 1996). Furthermore, (i) these methods normally require large numbers of samples, (ii) results are not transferable to other similar soil landscapes, and (iii) the results do not account for the primary causes of the soil variability—processes of soil formation (Moore et al., 1993).

Common soil properties used to create management zones and determine application rates include soil test P, soil test K, and soil pH. These soil properties are often directly related to crop response in the upcoming growing season. However, they are dynamic soil properties, and they can be variable over relatively short periods of time. There are, though, soil properties that are more stable over time and often have a profound influence on crop response, e.g., organic matter content and clay content. For example, soil organic matter (SOM) is a reservoir of many essential plant nutrients (N, P, S). The physical and chemical properties of SOM also (i) improve soil aggregate stability, increase infiltration, and improve aeration, (ii) increase moisture retention and available water capacity, and (iii) increase nutrient retention. The clay content of the topsoil is commonly an indicator of past erosion that has brought subsoil materials closer to the surface and incorporated them into the plow layer. In highly weathered soils, the subsoil materials tend to have a higher clay content, lower organic matter content, lower available water holding capacity, lower available nutrients, and higher bulk density. Frye et al. (1982) studied the effects of past erosion on the productivity of two Kentucky soils, and found that corn grain yields were up to 24% lower on eroded soils as compared to uneroded soils.

Soils-Landscape Modeling & Regional Variability

Soil-landscape modeling techniques (McSweeney et al., 1994) have developed as a quantitative method for predicting patterns of soil variability using observed patterns in environmental variables known to influence soil property variability. Soil-landscape modeling using terrain information has specifically been used to empirically model the spatial distribution of specific soil properties, including A-horizon thickness, organic matter content, and sand and silt content, and solum depth (Moore et al., 1993, Bell et al., 1995, Gessler et al., 1995).

A common conclusion among all of these studies is that the models that were developed may not be valid for landscapes removed from the original study site (Moore et al., 1993, Gessler et al., 1995, Thompson et al., 1997). In other landscapes, model coefficients, model variables, and/or model structure may change. However, this lack of transportability has never been tested by developing and validating models for fields from similar landscapes.

Research Plan

The key to this project is to focus effort on multiple landscapes in a single physiographic region, as opposed to examining single landscapes in multiple physiographic regions. Our approach is necessary to develop reliable predictions of soil variability across large areas.

The proposed research will be conducted in two phases over a three-year period. Both phases will be similar in that they will include both an initial sampling and modeling component, followed by a resampling and validation component. The first year of the study will include the sampling and modeling component of the first phase. The second year of the study will include both the resampling and validation component of the first phase and the sampling and modeling component of the second phase. The final year will include the resampling and validation component of the second phase, as well as dissemination of project results.

The first phase of this field study will be initiated at three locations in the Pennyroyal physiographic region of central and southwest Kentucky. Initial sites are located in Caldwell and Marion Counties (with a third study site yet to be determined). All sites have been or will be selected because they are dominated by soils of the Crider-Pembroke association.

DEM Generation and Terrain Analysis - A field survey DEM with approximate 10-m horizontal resolution and 0.1-m vertical precision will be created from a ground survey using a kinematic global positioning system. The raw data will be processed to a regular 10-m grid using ANUDEM (Hutchinson, 1995). Terrain attributes (slope gradient, slope aspect, slope curvature, upslope contributing area, etc.) will be calculated from each of the DEM using the Terrain Analysis Programs for the Environmental Sciences (TAPES) program (Moore, 1992), or similar algorithms.

Soil Sampling & Analysis - At each site, we will use the topographic variability quantified by the DEM to direct soil sampling using a stratified random sampling design. In each landscape, a total of 30 to 50 discrete soil samples will be extracted to a depth of 1 m (the approximate lower depth of the rooting zone) using a hydraulic probe and registered using a global positioning system (GPS).

Cores will be returned to the laboratory for morphological description and both physical, and chemical analysis. The thickness, color, and presence of redoximorphic features of individual horizons will be described. For each diagnostic soil horizon we will determine soil organic carbon (SOC) content by dry, clay content by the pipette method, water holding capacity, soil pH, and extractions for bioavailable nutrients.

Soil-Landscape Modeling and Landscape Prediction - The primary soil properties that will be used for modeling purposes will be A-horizon thickness, A horizon SOC content, and A horizon clay content. The relationships between these variables and other soil properties, such as available water holding capacity, pH, and bioavailable nutrients, will be examined using simple correlation and regression techniques.

The values for all terrain attributes will be extracted from the DEM for all soil sample locations. For each study site, we will develop empirical models of the distribution of A-horizon thickness, SOC content, and clay content using a split-sample method, with 75% of data used for model training and 25% used for model validation. Forward stepwise linear regression will be used to identify variables related to clay or SOC. We will validate the models using simple regression analysis, comparing the observed A-horizon thickness, SOC content, and clay content values with those predicted from the application of the individual linear models to the terrain attributes in the validation data set (Thompson et al., 1997).

Model Comparison & Transferability - To examine the effects of landscape segment size, and to test the ability of the quantitative landscape models to predict the spatial distribution of static soil properties across the Pennyroyal region, we will examine the transferability of the soil-landscape models in two ways. First, we will apply each of the models at the other study sites and compare predicted with actual values of A-horizon thickness, SOC content, and clay content using simple regression analysis. Second, we will (i) identify a second set of three fields from different parts of the Pennyroyal physiographic region (each with different landscape segment sizes), (ii) generate a DEM using the previously discussed methods, (iii) identify a stratified random sample of 10-15 samples, and (iv) collect soil core data to be used to validate the previously developed models.

The second phase of the proposed research will begin in the second year of the project and will parallel the work conducted in the first phase as described above. Three new study sites will be identified, surveyed, and intensively sampled. From these data, three new soil-landscape models will be developed, tested, and validated. Validation will occur on site, at the other two intensively sampled sites, and again at three new field sites with less intensive sampling.

Expected Benefits

The results of this work will elucidate whether soil-landscape relationships and quantitative soil-landscape models are consistent and transferable across a physiographic region. Until now, there has been no work done to examine to what extent quantitative soil-landscape models can be applied away from the study area where the model was developed.

From a practical standpoint, the results of this work will provide algorithms and procedures that can be used across all Crider-Pembroke landscapes for predicting the variation of static soil properties related to crop response, such as A-horizon thickness, SOC content, and clay content. By developing algorithms and procedures that are transferable throughout the Pennyroyal physiographic region, we will provide producers with a means to quickly and easily (requiring the collection of a DEM and a small number of soil samples) assess the nature of soil variability in their fields and divide the field into meaningful management zones. Successful application of this approach in the Pennyroyal physiographic region will provide a prototype for future work in other physiographic regions in Kentucky.

Deliverables

This research will produce such varied output as refereed scientific journal articles, extension publications, and conference presentations. However, for the successful transfer of this information to producers invested in precision agricultural production, an ideal venue will be workshops that provide ample time and opportunity for interaction with producers and training related to DEM generation and soil sampling design.

Facilities

The Department of Agronomy supports a soil and water analysis laboratory with the analytical instrumentation and supplies necessary for routine analysis of the soil samples. The College of Agriculture also supports a soil analysis laboratory for routine soil nutrient analysis. Research facilities managed by Dr. Thompson and Dr. Grove include all field equipment necessary for soil sampling, including a hydraulic probe for extracting soil cores.

Budget Justification

Personnel will be critical to the success of this project. Personnel funded through this research (graduate student, technician, or part-time post-doctoral assistant, depending on the quality of available candidates) will participate in all aspects of this project, including data collection, sample analysis, model development, and documentation. The dual-frequency real-time kinematic global positioning system is a key instrument for timely and accurate collection of digital topographic data. However, a dedicated system is not needed for this project. As it will only be needed for one to two weeks each year, it is only necessary to rent a system (at a cost significantly less than the purchase price). Initial publications should be in press by the third year of this project. Outlets for this work may include refereed journals such as Soil Science Society of America Journal, Precision Agriculture, or Geoderma.