6.2   Real-Time Soil Sampler and Analyzer Using Mobile Near-Infrared Reflectance Spectroscopy

Investigators

Michael D. Montross, Biosystems and Agricultural Engineering, montross@bae.uky.edu

Timothy S. Stombaugh, Biosystems and Agricultural Engineering, tstomb@bae.uky.edu

Scott A. Shearer, Biosystems and Agricultural Engineering, shearer@bae.uky.edu

Thomas G. Mueller, Agronomy, mueller@pop.uky.edu

Samuel G. McNeill, Biosystems and Agricultural Engineering, smcneill@uky.edu

Introduction

            Soil sampling for fertility requirements and variable rate application requires a tremendous amount of labor, time, and money. Each sample collected needs to be submitted for analysis at a cost between of 5 to 15 $/each. Depending on the grid size, grid soil sampling generates a tremendous amount of data. In addition, the grid size (or sample frequency) cannot be adjusted while sampling. If samples could be analyzed real-time, than grid sizes could be adjusted to produce a smart sampling scheme. For grid soil sampling to become better utilized the cost and labor must be decreased.

            A number of sensing technologies are available that could produce acceptable accuracies at a low-cost and near real-time. Near-infrared reflectance spectroscopy (NIRS) is one tool that is frequently used for the chemical analysis of numerous products. NIRS works by shining light at a given wavelength on a material in question, and measuring the intensity of the reflectance. The reflectance at certain wavelengths is correlated to specific chemical components. Developing correlations for materials requires known values of important properties from a laboratory. The reflectance is measured from each sample and statistical procedures are used to develop correlations between reflectance and the data obtained from soil analysis laboratories.

            Traditionally, NIRS instruments were very fragile and were confined to use in a laboratory.  Recently, more rugged mobile instruments have been developed.  A mobile platform could be designed such that a NIRS instrument could be used to measure soil chemical and physical properties in the field.  This project will explore the use of a mobile NIRS instrument for real-time analysis of Kentucky soils.

Objectives

The goal of this project is to develop a methodology for using NIRS to quantify the chemical and physical properties of soil. This goal will be accomplished through the following specific objectives.

1.      Develop a real-time soil sample analyzer calibration information relating NIRS data to soil physical and chemical properties of Kentucky

2.      Determine the accuracy of a NIR calibration for soil moisture, total C, total N, organic matter, pH, buffer pH, cation-exchange capacity, particle size, calcium, potassium, manganese, phosphorus, and zinc,

3.      Develop a mobile real-time soil sampling strategies.

Background

            Numerous researchers have investigated the real-time measurement of soil proprieties for precision agriculture using optical or near-infrared instruments. Chang et al. (2001) developed a NIRS analysis tool for soils found in eastern Washington, eastern Colorado, Texas panhandle, and southeastern Minnesota. Five laboratories performed the chemical analyses and good correlations (R² > 0.7) were found for many quantities including total C, total N, moisture, cation-exchange capacity, particle size, calcium, and manganese. It should be noted that the results were based on very different soil types and the results were analyzed at different soil testing laboratories. There can be a large difference in results from different laboratories due to different people, procedures, and equipment.

            Ehsani et al. (1999) used NIRS to measure soil mineral-N content. They determined that site specific calibration was necessary to accurately estimate N content. Sudduth and Hummel (1993a, 1993b) used a portable NIRS instruments to measure soil organic matter, moisture and CEC from soils in Illinois. However, the accuracy of the predictions decreased as the geographic range of samples increased (Sudduth and Hummel, 1996). Morra et al. (1991) used reflectance to accurately predict soil C and N content.

Methods

            The first task is to develop calibration information relating NIRS data to soil physical and chemical properties.  A Perstop NIRSystems 6500 laboratory grade scanning monochromator (Foss NIRSystems, Silver Spring, MD), which measures reflectance between the wavelengths of 400 and 2500 nm at 2 nm increments, is available to develop baseline calibrations for soil analyses. This instrument and associated calibration software will be used to develop a predictive model for moisture content, total C, total N, pH, buffer pH, CEC, particle size, calcium, potassium, manganese, phosphorous, and zinc. From this analysis, a reduced set of wavelengths required to analyze soil components will be identified. There are about 4,000 soil samples from various locations in Kentucky available for this project that have already been analyzed by a laboratory for particle size, buffer pH, Mehlich calcium, potassium, manganese, phosphorous, zinc, and organic matter. Approximately 3,000 samples will be analyzed and a calibration developed. The remaining 1,000 samples will be used to validate the calibration. Additional soil samples will be collected and analyzed to determine moisture, nitrogen, and cation-exchange capacity. The samples are from similar parent material, have been dried, and ground and are expected to produce very accurate predictions. However, field samples will have varying moisture contents and have not been ground. Calibrations will be developed with varying moisture contents and particles sizes to minimize the influence of those variables in the field.

            The accuracy of the prediction equations will probably be a function of the soil parent material. Increased accuracy would be achieved by having different prediction equations based on the parent soil material. It is expected that a small percentage of the soil samples will need to be analyzed using wet chemistry analysis to provide a small correction for each soil type. However, the samples that need to be chemically analyzed will be determined by the instrumentation and a small correction applied to the samples collected and analyzed in the field. It is believed that the correction factor will only be determined once for each soil type and that it will remain valid during following years.

            The next task is to develop the equipment for real-time field soil analysis.  Two methods are available for constructing a mobile NIR sensor. If a small number of wavelengths are required for measuring soil properties, photo-diodes tuned to the appropriate wavelengths can be purchased. The advantage of this system would be the low cost. This may be a suitable alternative where accuracy is not as critical, for example with sensors mounted on chemical application equipment. However, given the results obtained by other researchers, we believe that a tuned diode construction would be impractical for analyzing soil cores.

            The second option is to use a mobile scanning NIRS machine. A rugged, portable Zeiss Corona 45 (Carl Zeiss, Inc., Jena, Germany) photo-diode array instrument is available. The Corona scans between 400 and 1900 nm in 6 nm increments. This unit is operates on 12V power and is intended to be used in mobile applications. We will develop a mobile platform mounted on an ATV that an operator can take into a field, pull a soil core, insert it into the instrument, and record soil analysis along with GPS information. A few representative samples will be kept to send to a laboratory for verification analyses.

            The next logical step would be the development of an automated field sampling system. Several investigators have been developing an autonomous vehicle. This vehicle could transport the sensing equipment to predetermined locations in the field. A hydraulically driven auger system will extract a soil core. A series of mechanical devices will grind the soil sample and present it to the NIRS sensor through a duct with a sapphire window.

            A longer term goal of the project will be to develop an intelligent sampling scheme.  Since the analysis results are available immediately, the automated system could fist sample a field on a course grid.  These data can be scrutinized to determine where more intense sampling is needed. The autonomous system would then move to those areas in the field and take more readings.  This process could continue iteratively until an accurate representation of field variability is determined.

Expected Benefits

            Equipment will be developed that will allow for real time soil sample analysis. The system developed will be mounted on the back of a six-wheeler and will allow for real-time soil sample analysis. Currently the mobile NIRS instrument is expensive (~$30,000). However, the costs of the instruments have come down significantly as the technology matures and the cost of running samples through the machine are negligible. Future work with the equipment will demonstrate the feasibility of real-time soil sample analysis that would be useful with grid soil sampling. Algorithms and procedures to vary soil sampling frequency could be developed to produce optimal results. In addition, the equipment could be utilized on an autonomous tractor that would automatically produce optimal intensive soil sample results.

 Deliverables

            The project will result in a number of referred journal articles and an extension publication. The articles will demonstrate the accuracy of the calibration and the utilization for real-time soil sampling in precision agriculture.