6

The majority of field crops in Kentucky and the Southern United States are grown on fields that have a high degree of variability with respect to chemical and physical properties. The first step in managing this variability is to measure it. But there are limitations with some of the common approaches for measuring this variability such as grid sampling. We have found that there is substantial variability between grid points and that it is in many cases it may not be possible to accurately predict between grid points especially at the industry stand 1-ha grids (Mueller et al., 2000). Additionally, spatial variability of soil fertility often does not impact the spatial variability of grain yield substantially (Everett and Pierce, 1996; Blackmore et al. 1999). Soil water supplying capacity is a major factor that affects yield within many Kentucky fields. We have found that soil electrical conductivity sensor is related to some of the factors that affect plant available water including topsoil depth, depth to bedrock, and depth to fragipans on some Kentucky fields (Hartsock et al., 2000). A soil permeability sensor developed by Dr. Tim Stombaugh has been shown to relate to soil penetrability (cone index). High cone index values relate to the resistance a root experiences as it penetrates soil. Landscape position is important because it relates to the redistribution of soil materials in landscapes and the dynamics of horizontal movement of water. High-accuracy global positioning systems (GPS) can be used to measure elevation in a moving vehicle with an absolute accuracy of +/- 4 cm in the vertical dimension. Using available software programs, we can develop digital elevation models (DEM’s). Terrain attributes such as slope aspect, curvature, and land forms can be developed from the raw elevation data. The conductivity, permeability, and topographic information can be used to predict which areas in a landscape are more likely to suffer from water stress. This is useful management information because, crop responses to other crop inputs (e.g. nitrogen fertilizer) are less likely when soil water is limits crop yield (law of the limiting) and therefore may be used as a basis for variable rate nitrogen decisions. Additionally, by planting seeds at lower densities, soil moisture is conserved. Therefore the ground based sensors.

Another approach to making nitrogen management decisions is to use remote sensing to detect nutrient stress during the growing season. A number of researchers have found a relationship between remote sensing and nitrogen stress or crop response to N (Blackmer et al. (1996a; Blackmer et al., 1996b; Schepers (1998). One limitation for using remote sensing for nitrogen management is with the difficulties in obtaining timely images because periodic satellite coverage may not coincide with cloud free weather the narrow opportunity for management. For this reason, Dr. Stombaugh is developing a commercially affordable remote control platform for remote sensed visible and NIR imagery.

The use of ground and remote sensors to aid in production and management decisions has great potential to improve the economic viability of grain production in Kentucky while minimizing nitrogen losses to the environment. Therefore, our goal is to develop methods for collecting and calibrating sensor data and utilizing this information to make variable rate nitrogen decisions. Our investigation has two objectives:

As part of the phase I funding, we related soil electrical conductivity with depth of fragipans and depth to bedrock (Fig. 1), and depth to an increase in clay (Fig. 2). The depth to fragipan relationship (r² = 0.80) and the depth to clay relationship at the Trigg county location (r² = 0.74) were particularly strong. In the Hardin, Trig, and Shelby 4 fields, the relationships with depth to clay had similar shapes (Fig. 2). The differences in the intercepts and slopes can be explained by differences in clay content.

A study initiated by Dr. Stombaugh attempted to develop a continuously measuring permeability sensor (Clement, 2000). The apparatus (Fig. 3) consisted of a mass flow controller, which maintained a constant flow rate of air from an air supply to a specially designed tine. An orifice in the side of the tine allowed the air to pass into the soil as the tine was drawn through the soil. The resistance pressure gave an indication of the soil permeability. Results from tests of this initial design indicated that pressure could be correlated to compaction within a given soil type (fig. 4), and that the sensor output could delineate areas with different permeability levels.

Fig. 4. Performance of permeability sensor compared to cone index for a field test in a silt-loam soil.

We have been collecting elevation data with a survey grade GPS system that are accurate to +/-4cm in the vertical dimension (Fig 5). We are creating terrain attributes using ANUDEM and TAPES terrain analysis packages to calculate terrain attributes such as shown in Fig. 6.

We will hire a post-doc for three years who will be conduct most of the work for both objectives. Dr. Mueller’s research analysis will also be involved with this project. We will use the sensors listed in Table 1 for this project as detailed in this proposal. The procedures are listed according to objective and our specific research questions.

Remote Sensing
RC Aircraft Platform	Several digital imaging systems will be mounted on a remote control (RC) aircraft platform currently being developed by Dr. Tim Stombaugh.

Ground Based Sensors
Electrical Conductivity	Veris ® 3100 Soil Electrical Conductivity Sensor
Soil permeability sensor	Sensor developed by Dr. Tim Stombaugh, second prototype
Survey Grade GPS	Trimble Single and Dual Frequency Systems
Terrain Modeling	TAPES and Demon terrain modeling Software

Objective 1. To develop methodologies to collect and calibrate remote and ground based sensor data for precision agriculture applications.

Which spectral bands are related to soil and crop properties of agronomic importance in Kentucky agricultural fields? We are currently relating bulk soil EC, ground penetrating radar, and terrain attributes to various soil projects as part of the Phase I special USDA funding. Dr. Stombaugh’s remote control aircraft system will be used to photograph research plots using visible and infrared cameras on corn, soybeans, wheat, and bare ground fields. Leaf chlorophyll measurements will be collected for the Davis county field. Leaf tissue samples will be collected from each plot which will be analyzed for total Kjeldahl nitrogen. Soil analysis (soil texture, organic matter) will be performed by UK Division of Regulatory Services. We will use ANOVA and multivariate analyses to related the remote sensed imagery to the crop and soil treatments. The data will also be analyzed geostatistically to determine the range over which these variables vary spatially.

Should sensor data be interpolated and which interpolation procedures are best? We will ground based sensor data (Table 1) at a variety of scales (i.e. 20, 40, 60, 80, and 100-ft transect data) and interpolate with a variety of methods (e.g. kriging and inverse distance). The goal is to determine the optimum sampling intensity and interpation methods for the sensor measurements. In addition to the prediction data sets, validation data sets will be collected to evaluate the prediction maps and terrain modeling, and with terrain modeling programs. We will compare interpolations with quantitative measures of map error (mean square error, precision, and bias) and visually by observing plots of predicted vs. measured.

Can sensors be used to accurately and cost effectively map crop and soil properties of agronomic importance? At what scale and intensity should sensors be calibrated (field, farm, or regional)? Sensors will be more economical to use if a calibration is valid for a large region and if a small number of calibration points are necessary. We will calibrate a conductivity sensor and soil permeability sensors for several fields in Kentucky, for one or more farms, and at a number of points with in a geographic region (e.g. pennyrile). Calibration involves taking sensor measurements along transects as well as measuring specific soil properties (e.g. soil depth, depth to bedrock, depth of topsoil). We will also make the same measurements at certain checkpoints at each field, farm, and geographic region. We will develop a calibration relationships using regression at each scale using a different number of points (e.g. 20, 40, 60, 80, or 100 points). We will use the calibrations to predict soil properties at each checkpoint. Next, we will evaluate the prediction using plots of predicted versus measured and measures of map error. We will compare predictions created using the different calibration scales (field, farm, regional) and the various numbers of calibration points (e.g. 20, 40, 60, 80, or 100 points). We will work with Carl Dillon to evaluate the costs associated with using sensors at various scales and calibration. Then we will relate these costs to the quality of the maps we produce.

Objective 2. To develop methodologies for utilizing sensors for variable rate nitrogen management.

What are the key factors that govern crop response to nitrogen fertilizer rates in Kentucky agricultural fields? We will develop nitrogen response studies in as many locations as possible in Kentucky. The basic design of this experiment presented in Fig. 7. We will replicate this experiment throughout each field. Dr. Barnhisel will measure grain yield in each plot. For a subset or all of the

experimental units, soil properties (soil depth, depth to bedrock, depth to fragipans, soil profile texture and a sample for standard soil fertility analysis) will be measured. We will develop nitrogen response curves for each experiment. Then we will relate the parameters of the response models to the measured soil parameters.

Fig. 7. Experimental design and layout of nitrogen and experiment which will be replicated multiple times throughout each field.

Are sensors related to factors that govern crop response to fertilizer nitrogen? For the above experiments, take sensor measurements (Table 1 ) at each of the experimental units. We will relate the sensor data to the crop response parameters. We will use these multivariate techniques or neural network analysis to develop a predictive model to predict crop response using sensor data. Some of the response data will not be included in the model development. We will use it to evaluate the predictive capability of the model. We will work with Dr. Dillon to develop economic models to predict the cost and benefits of using sensors for variable rate management.

We expect to 1) develop a better understanding of crop response to variable rate nitrogen, 2) develop statistical models to predict this crop response in agricultural fields based on ground based and remote sensors, 3) methods and protocols to collect, calibrate, and utilize remote and ground based sensed data. It is important to note that the sensor component of this research will also be useful making better soil use decisions such as deciding whether or not to enroll marginal land into CRP.

We will produce one or more extension and journal publications for the following topics 1) tools for predicting crop responses to nitrogen based on sensors and 2) methods and protocols for collecting, calibrating, and utilizing remote and ground based sensors data. We also use this research to develop teaching material for the precision agriculture course and extension presentations.