This document outlines a preliminary research presentation on using mid-infrared (MIR) spectrometry to predict soil organic carbon and soil organic matter for the USDA soil survey program. It details the methodology, calibration models, and pilot project based in Kansas involving over 225 field test samples. The findings indicate high predictive accuracy for soil properties, with ongoing research and data available upon request.

1. USING MIR SPECTROMETRY TO
PREDICT SOIL ORGANIC CARBON
AND SOIL ORGANIC MATTER FOR THE
SOIL SURVEY PROGRAM IN THE USA
Fall, 2019
Rich Ferguson, National Soil Survey Center
N a t i o n a l S o i l S u r v e y C e n t e r
The findings and conclusions in this preliminary presentation have not been formally
disseminated by the US Department of Agriculture and should not be construed to
represent any agency determination or policy. This research was supported by the
intramural research program of the US Department of Agriculture, NRCS.

2. The USDA-NRCS Kellogg Soil Survey Laboratory
Mid-infrared (MIR) Spectral Library
Over 80000 soil samples scanned, representing a great
diversity of soil types. LOTS more samples to scan
Kellogg Soil
Survey
Laboratory Soil
Archive. With
over 500,000
specimens, the
largest public
soil archive in
the USA!

3. Locations of pedons in the MIR library, to date
All spectra and data are
FREE on request (not
web-based yet)

4. The Electromagnetic Spectrum
MIR – 2500-25000 nm
*
𝜹 −
𝜹 +
𝜹 +
**
MIR radiation is
partially absorbed
by soil constituents
* By Lookang. Many thanks to Fu-Kwun Hwang and author of Easy Java Simulation = Francisco Esquembre (Own work)
[CC BY-SA 3.0 (https://creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons
** Courtesy of gfycat.com
Calibration models are being developed to
predict soil organic carbon from MIR spectra

5. Plate of samples
MIR Spectrometer: Bruker Vertex 70 / HTS-XT

6. 1) Pick a property (e.g., C) and a scale to model
2) Select soil samples already analyzed by the standard method
3) Collect MIR spectra of the samples
4) Using PCA/PLSR, prepare a model that calibrates spectra
against measured values
32 21 120 12 35 14 28 34 80 23 …
…
Org C, g/kg
U
TC TOC TIC CEC Clay pH 1500 kPa W … etc.
predictions…
Current modeling approach for pilot project:

7. Target area for the pilot project based in Central Plains
= Mollisols (Chernozems, Phaeozems, Kastanozems)
N
Lincoln
Salina
671 kilometers or 417 miles

8. Fit
PredictedM e a s u r e d M e a s u r e d
RMSEE = 2.75 RPD = 5.51 RMSEP = 2.74 RPD = 5.44
R2 = 96.71 R2 = 96.61
n = 753 n = 754
Total Carbon – A Horizons
Calibration Validation

9. Fit
PredictedM e a s u r e d M e a s u r e d
Total Carbon – B Horizons
Calibration Validation
RMSEE = 1.32 RPD = 6.48 RMSEP = 1.37 RPD = 6.19
R2 = 97.62 R2 = 97.39
n = 1049 n = 1047

10. Fit
PredictedM e a s u r e d M e a s u r e d
Organic Carbon – A Horizons
Calibration Validation
RMSEE = 2.64 RPD = 5.16 RMSEP = 2.66 RPD = 5.00
R2 = 96.24 R2 = 96.01
n = 637 n = 638

11. Fit
PredictedM e a s u r e d M e a s u r e d
Organic Carbon – B Horizons
Calibration Validation
RMSEE = 1.22 RPD = 3.59 RMSEP = 1.24 RPD = 3.63
R2 = 92.23 R2 = 92.41
n = 895 n = 895

12. Fit
PredictedM e a s u r e d M e a s u r e d
Inorganic Carbon – A Horizons
Calibration Validation
RMSEE = 0.82 RPD = 12 RMSEP = 0.81 RPD = 12.2
R2 = 99.31 R2 = 99.32
n = 149 n = 149

13. Fit
PredictedM e a s u r e d M e a s u r e d
Inorganic Carbon – B Horizons
Calibration Validation
RMSEE = 1.12 RPD = 10.9 RMSEP = 1.14 RPD = 9.99
R2 = 99.16 R2 = 99.00
n = 468 n = 469

14. Field office equipment: grinding mill and spectrometer
Mill
Retsch MM200 mill
A 5-cm3 scoop of dried, sieved
(<2 mm) soil is finely ground
using the mill

15. Field office equipment: grinding mill and spectrometer
Spectrometer
Bruker Alpha spectrometer
Sample is loaded into 6-mm
diameter hole and lightly
pressed (using a press rod) to
make a pellet
Spectra are processed through
models to obtain predicted values
for soil properties

16. Pilot project based in Kansas, USA
225 field test samples (not used to calibrate model) from > 40
scattered sites ( ) were collected, prepared, scanned and
processed; samples were also analyzed at KSSL
N

17. Field Test Samples, A & B Horizons
Predicted SOC vs Measured SOC, g / kg
RMSEP Count RPD R2
2.2 222 4.7 0.98
All All All All
3.0 67 3.8 0.97
A A A A
1.6 155 3.2 0.95
B B B B
M-distance outliers: 3

18. Field Test Samples, A & B Horizons
(Predicted TC – Predicted IC) vs Measured SOC, g / kg
RMSEP Count RPD R2
2.1 216 5.0 0.98
All All All All
2.5 63 4.4 0.98
A A A A
1.9 153 2.8 0.93
B B B B
M-distance outliers: 9

19. Field Test Samples, A Horizons
(Predicted SOC * 1.724) vs Measured SOM, g / kg
RMSEP Count RPD R2
7.5 67 2.8 0.94
A A A A

20. Work continues. Thanks for listening!
Joe Anderson Maxine Levin Suzy Riedel
Scarlett Bailey Zamir Libohova Cathy Seybold
Jonathan Comstock David Lindbo Sonam Sherpa
Michelle Etmund Steve Monteith Phil Schoeneberger
Rich Ferguson Rick Nesser Pam Thomas
Jon Hempel Brian Nester John Warner
David Hoover Kathy Newman Skye Wills
Patty Jones Chad Remley David Wolfe
Doug WysockiThe findings and conclusions in this preliminary presentation have
not been formally disseminated by the US Department of Agriculture
and should not be construed to represent any agency determination
or policy. This research was supported by the intramural research
program of the US Department of Agriculture, NRCS.