Use of In-situ Shallow Subsurface Spectroscopy for Measuring Soil Organic Carbon

USE OF IN-SITU SHALLOW SUBSURFACE SPECTROSCOPY FOR MEASURING SOIL ORGANIC CARBON

Sponsoring Institution

National Institute of Food and Agriculture

Project Status

COMPLETE

Funding Source

SMALL BUSINESS GRANT

Reporting Frequency

Annual

Accession No.

1028373

Grant No.

2022-33530-36901

Cumulative Award Amt.

$175,000.00

Proposal No.

2022-00833

Multistate No.

(N/A)

Project Start Date

Jul 1, 2022

Project End Date

Feb 28, 2023

Grant Year

2022

Program Code

[8.4]- Air, Water and Soils

Project Director
Barrett, L.

Recipient Organization
S4 MOBILE LABORATORIES LLC
526 S MAIN ST STE 813C
AKRON,OH 443114401

Performing Department
(N/A)

Non Technical Summary
Healthy soils are essential to human well-being and to overall environmental quality. Key to soil health is maintenance of the soil's organic carbon content. Recently, carbon credit markets have emerged a means of incentivizing practices that increase the organic carbon content of soils. The resulting carbon sequestration in soils will also contribute to stabilization of atmospheric CO2 levels. However, such markets are not feasible without cost-effective and efficient means for verifying the soil organic carbon content. The outcome of this proposed project is a prototype unit, the Subterra Green Model P, that enables land managers to rapidly and accurately map the organic carbon content of their soils in three dimensions to depth of approximately one meter. The unit employs a visible/near-infrared spectroscopic probe that is pushed into the soil at intervals and is small and maneuverable enough to be operated by one person in many different vegetative conditions. In the proposed work plan, we will significantly modify the design of our existing prototype hardware by incorporating new spectrometers and a GPS unit and new chemometric analysis and mapping algorithms. To assure the required accuracy and precision, the modified prototype will be tested at four Ohio sites with varying soil texture, organic matter content, and land cover conditions, and then it will be iteratively improved based on the test results.

Animal Health Component

30%

Research Effort Categories

Basic

Applied

30%

Developmental

70%

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
101	0110	2000	100%

Knowledge Area
101 - Appraisal of Soil Resources;

Subject Of Investigation
0110 - Soil;

Field Of Science
2000 - Chemistry;

Keywords

Goals / Objectives
Major Goals:Our technology, the Subterra Green, is a mobile field unit armed with a visible and near infrared spectroscopic probe and a load cell for measuring probe insertion force. As the unit moves across a target area, the probe is pushed into the soil at intervals, measuring light reflected from a column of soil immediately adjacent to the probe. These data, and the location and force required to insert the probe, are converted into a 3D map of subsurface volumetric soil organic carbon, as well as point measurements of gravimetric soil organic carbon content and soil bulk density. These maps can be used to promote soil health and for soil carbon credit accounting.Previously S4 Mobile Laboratories has developed and successfully completed the design of a sister product, the Subterra Grey Model P, which is aimed at being able to detect clandestine human burials in the forensic market. The goal of the current project is to build on our previous forensic research, modifying the design as necessary in order to re-direct its use towards mapping and measuring soil organic carbon, while applying the Subterra technology to conduct proof-of-concept studies to measure carbon stock.Objectives:Incorporation of new spectrometer instruments to cover the entire wavelength range of the visible/near infrared spectrum.Incorporation of an appropriate GPS (GNSS) instrument.Development, testing, and iterative improvement of the machine learning algorithms for volumetric soil organic carbon, gravimetric soil organic carbon, and bulk density, including approaches for dealing with soil moisture effects (i.e., methods to achieve per-sample accuracy).Improvement of per-site accuracy, which will include identification of appropriate sampling designs for producing 3D maps of volumetric soil organic carbon, gravimetric soil organic carbon, and bulk density, and determination of the required number of calibration samples.

Project Methods
Efforts:Prototype Design. The Subterra Green Model P prototype in the current project starts with our prototype for the forensics market, and will retain the design of the probe, light source, insertion mechanism, load cell, and overall structure from the Subterra Grey Model P. The control software and user interface will also be retained. Significant changes to the Subterra Grey Model P's design will include: (1) Replacement of the spectrometer (restricted to 1550 to 1950 nm wavelength) with two or more new spectrometers to sense the entire 400 nm to 2500 nm VNIR range. (2) Incorporation of a GPS (GNSS) receiver to record location to sub-meter accuracy. (3). Alteration of the control software to accommodate the new spectrometers and GPS receiver.Field Testing. Field testing of the prototype in at least four separate field testing sites will be completed, with activities building iteratively. Each site will include the following activities, which are described in more detail in the following paragraphs: (1) Field data collection (spectral and insertion force) using the Subterra Green prototype with simultaneous soil sample collection for calibration and comparison purposes. (2) Samples are analyzed for soil organic carbon and bulk density. (3) Chemometric and machine learning analysis to develop an appropriate algorithmic model for predicting soil organic carbon (gravimetric and volumetric) and bulk density. (3) 3D mapping of soil organic carbon (gravimetric and volumetric) and bulk density. (4) Device analysis software is updated.Field data collection: The prototype unit will be used to map soil organic carbon and bulk density at each sample site. In these initial cycles, we will over-sample by placing probe insertions at a density of at least 30 points per hectare, resulting in 300 probes in a 10 hectare site. At each insertion point, spectra are recorded every 2.5 cm to a depth of 90 cm, and the force required to insert the probe is also measured. The location of each insertion point is established using the GPS unit, and stored with the spectra.Soil sample collection: The machine learning models developed from the spectral and insertion force data require calibration with soil sample test data. Therefore, at one out of every six probe insertion locations physical soil core samples will be obtained for soil organic carbon and bulk density testing, resulting in 50 cores for a 10 hectare site. Each core will be taken to a depth of 90 cm using a hand auger immediately adjacent to the probe insertion point. The core will be subdivided into 10 cm increments, from which five increments will be chosen at random and sent to a commercial soil testing laboratory for soil organic carbon analysis. The other increments will be retained for future analysis if necessary. Intact known volume samples will also be obtained at 15, 30, and 60 cm depths for bulk density and water content testing, which will be determined in house.Chemometric analysis: The soil organic carbon and bulk density measurements obtained from the soil samples are matched to the spectra and insertion force data from the adjacent probe insertion at the same depth and analyzed using chemometric techniques. In our proof-of-concept studies, we have been most successful predicting soil organic carbon using partial least squares regression, so we will begin with this method, but we will also test the efficacy of a variety of machine learning techniques, including support vector machines, random forest regressors, and artificial neural networks. We will control for moisture effects using standard techniques, as well as baseline removal of water effects by fitting Gaussian curves. Furthermore, we will directly model volumetric soil organic carbon as well as gravimetric soil organic carbon and bulk density. Once an appropriate chemometric model has been established, it will be applied to the entire probe insertion dataset, resulting in known values of soil organic carbon and bulk density at 2.5 cm depth increments to a depth of 90 cm at each probe insertion location. We expect per-sample soil organic carbon accuracy levels to be better than the RMSE < 0.5 g kg-1 and R2 > 0.80 reported in the literature for field-based spectroscopic SOC measurements.Mapping and integrated volumetric soil organic carbon determination: Interpolation of values between probe insertion points using kriging, natural neighbors, and/or inverse distance weighted algorithms will be used to map the soil organic carbon and bulk density content in three dimensions, and then used to estimate total volumetric organic carbon stock at the site. As part of this analysis, we will evaluate the effect of sampling design and density on the resulting analysis in order to establish optimal sampling design by selectively pruning our over-sampled dataset. Our ultimate goal is to develop an importance sampling strategy that will yield the highest possible volume-integrated soil organic carbon accuracy most quickly.Prototype update: The entire prototype design will be evaluated continuously based on operator feedback and chemometric accuracy metrics. The prototype will be updated as necessary, including changes to hardware, software, and sampling design that would improve accuracy or usability. In particular, the chemometric and mapping algorithms will be examined in order to understand and improve them.Final design update:Software that applies the latest chemometric analysis algorithms and 3D mapping capability will be incorporated into prototype control software to produce real time prediction of volumetric soil organic carbon. The real-time estimates are a first-order approximation that will be updated based on a limited number of calibration samples. Real-time estimation adds value because with such an estimate the importance sampling strategy can be optimized on-the-fly while measurements are being taken.Evaluation:Metrics for success:The metric for success of the protype design update will be the completion of a prototype model that includes the necessary modifications.Success of the field testing will be measured based on (1) increasing the proportions of positive informal feedback from field operators describing the usability of the prototype in the field; (2) increasing trends in R2 and RMSE, and per-sample soil organic carbon accuracy levels to be better than RMSE < 0.5 g kg-1 and R2 > 0.80; and (3) successful completion of per-site determination of volumetric soil organic carbon stock and 3D soil organic carbon and bulk density maps at each site. These metrics will be determined separately for each of the field testing sites.

Progress 07/01/22 to 02/28/23

Outputs
Target Audience:Two market segments will benefit from the data provided by the Subterra Green. The first market is precision agriculture, and the second is the carbon credit market. In agriculture, the end users are farmers, cooperatives, agricultural service providers, and soil researchers that are working on long-term changes to farming practices in order to enhance soil health. High SOC content is an indicator of healthy soils with higher yields and improved profits for farms. The carbon credit market requires an inexpensive, accurate and tamper-proof solution. Both market segments expressed frustration with the high cost and labor requirements of conventional soil sampling methods which require extraction, transportation, handling, and measurement of soil samples. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?A Ph.D. student was involved in data collection and analysis.Data collected will be used in her dissertation research. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? The purpose of this Phase I project was to modify our prior forensics-oriented prototype to produce a new prototype unit, the Subterra Green, that can rapidly and accurately map the organic carbon content of soils in three dimensions to a depth of 90 cm.The Subterra Green is a mobile field unit with a visible and near infraredspectroscopic probe and a load cell for measuring probe insertion force.As the unit moves across a target area, the probe is pushed into the soil at intervals, measuring light reflected from a column of soil immediately adjacent to the probe. These data, and the location and force required to insert the probe, are converted into a 3D mapof subsurface volumetric soil organic carbon stock, as well as point measurements of soil organic carbon concentration and soil bulk density. The maps can be used to promote soil health and for soil carbon credit accounting. Objective 1:We replaced the spectrometer on the prior forensics-oriented prototype with two new spectrometers,Hamamatsu C13053ma andHamamatsu C15511-01 FTIR Engine,that in tandem can sense nearly the entire visible and near infrared spectrum. Objective 2:We selected a GPS/GNSS instrument with submeter accuracy, the Juniper Geode GNS1, and incorporated it into our prototype. Objective 3:We field tested the Subterra Green prototype at six separate one-hectare sites in northern Ohio. Using the Subterra Green, at each site we collected spectral data in a grid pattern with 10 m spacing, and at a subset of these points also collected soil samples. The soil samples were analyzed for soil organic carbon content and bulk density. We then created models that use the Subterra Green data to predict organic carbon and bulk density. Our approach was to divide the probe insertion points into three sets, training, validation, and test. The training set is used to develop a range of alternative models, the validation set is used determine which model performs best, and the test set is used to evaluate the chosen model's performance on data not used in developing the model. We developed models for each of the six sites separately, as well as in combination. For soil organic carbon concentration, model performance exceeded our RPIQ goal of 2.0 at all of the sites, and exceed the coefficient of determination goal of 0.8 at the majority of the sites. Because at four of the sites we were unable to obtain sufficient bulk density samples due to failure of our soil sampling equipment (not the Subterra Green), we were only able to model bulk density at two sites. At those two sites the model performance for bulk density was lower than our model performance goals. Objective 4: Using the per-sample models of soil organic carbon concentration from Objective 3 and the smaller number of measured bulk density samples, we calculated the amount of soil organic carbon present in the soil at each probe insertion point to the depth the probe was inserted, usually 90 cm, and expressed the results as the carbon stock (Mg carbon per hectare to the depth of probe insertion) at each probe insertion location. Because we have data for every 10 cm depth interval at each point, we visualized the results by mapping carbon stock in 10 cm depth slices, and also as a 3D map. For each site we also created a map showing the sum of the carbon stock from the surface to the maximum probe insertion depth. With the hardware and software modifications made in the Phase I project, the Subterra Green prototype performed well for the purpose of measuring and mapping soil organic carbon at the study sites. We experienced no significant failures or data loss with the prototype under conditions that ranged from cold (just above freezing) to hot (in excess of 90 degrees F). The prototype is designed with two batteries that can be replaced without powering down, but in practice, we found that a single fully-charged battery was sufficient for a full day's data collection. The Phase I research has demonstrated that the Subterra Green can be used to estimate and map soil organic carbon concentration and stock with satisfactory accuracy at the study sites and across sites. Model metrics were consistently good both for single sites and for a combination of sites. We were less successful in modeling with our limited bulk density dataset, but by improving the quantity and quality of bulk density training data, it may be possible to improve future bulk density models. The Phase I project serves as a solid start to our broader goal of achieving commercial acceptance of the Subterra Green in the carbon credit market. Nevertheless, before it can be applied commercially, it will be necessary to demonstrate that the method consistently achieves accuracy and precision sufficient for measuring expected levels of change in soil organic carbon under a wide variety of different soil types and environmental conditions. To accomplish this, it will be essential to extend the scope of the training data to encompass a far wider range of site types, and to develop models that are applicable to broad geographical regions and to define and track the parameters (e.g., soil types, topography, land use practices) that influence the generalizability of model performance. We will address such questions in our Phase II proposal. A cost-effective method of measuring soil organic carbon stock is significant because voluntary carbon credit markets are growing rapidly, as are markets that reward sequestration of carbon in the soil. However, a major hurdle to the establishment of a functioning carbon credit market is the lack of inexpensive and accurate protocols for measurement of soil organic carbon stock. Because the Subterra Green completely eliminates the time and cost associated with removing and processing soil samples, it can be used to take far more measurements than would be economical with the established methods that rely on analyzing physical soil samples in the laboratory. By taking many more data points, the precision of the carbon stock determination can also be better than would be possible with established methods. Therefore, the Subterra Green will be both more economical and more precise than the methods that are currently available. Farmers interested in obtaining carbon credits, service providers tasked with measuring baseline carbon stock for the carbon credit market, and agencies that verify sequestered carbon will all welcome the Subterra Green's ability to accurately quantify soil organic carbon stock for a reasonable cost. These Phase I results are the necessary foundation to proving the Subterra Green's capabilities towards that end.

Publications

Type: Theses/Dissertations Status: Other Year Published: 2023 Citation: PhD thesis of Lamalani Suarez, the graduate student who participated in field measurements during the course of the project will utilize data gathered from this project.
Type: Journal Articles Status: Other Year Published: 2023 Citation: S4 will submit at least one manuscript for publication in a peer-reviewed journal to establish the credibility crucial for commercial success.