2.2.2.4 Dry Matter by Near Infrared Reflectance Spectroscopy
Reference:
Windham, W.R., F.E. Barton II and J.A. Robertson. 1988. Moisture
analysis of forage by near infrared reflectance spectroscopy:
Preliminary collaborative study and comparison between Karl
Fischer and oven drying reference methods. Journal of the
Association of Official Analytical Chemists 71:256-262.
Martin, G.C., J.S. Shenk, and F.E. Barton II. 1989. Near Infrared
Reflectance Spectroscopy (NIRS): Analysis of Forage Quality.
United States Department of Agriculture, Agricultural
Research Service. Agricultural Handbook No. 643.
Scope:
This procedure is applicable for determining laboratory dry
matter of ground, air-dry or partially dried (90 to 95% dry
matter) forage. Samples must be ground through cyclone grinder
with 1 mm screen and be 90 to 95% dry matter.
Basic Principle:
Random portions of a sample are loaded into a NIR sample holder
and reflected light from the ground sample is measured in the
infrared region (generally 1100 to 2500 nm). Instrument is part
of a system that has been calibrated using representative samples
from population to be tested. Equations selected based on
calibration statistics, which have been validated, are used to
calculate dry matter content of feed and forage samples.
Equipment:
Near infrared reflectance spectrophotometer, wavelength range at
least 1100 to 2500 nm Sample holders with infrared transmitting
quartz window Computer with software for collecting, storing and
processing spectra.
Reagents:
None.
Safety Precautions:
- Follow manufacturer's recommendation for safe operation
of instrument.
Procedure:
- Prepare samples by the same method as the calibration
samples were prepared.
- For best results run instrument (but not lamp)
continuously. If instrument is cold, warm-up time should
be 1 hr.
- Clean sample holder with a camel hair brush or vacuum.
Additional cleaning may be done with soft tissue or
lint-free cloth. Glass should be free of finger prints
and foreign material.
- Load NIR sample cup placing one scoop of forage
(previously dried to 90-95% dry matter or greater, ground
with cyclone mill to pass 1 mm screen, thoroughly mixed)
on each third of the glass surface to ensure that
portions of different subsamples are scanned. Overfill
the sample holder and scrape off any excess.
- Press back into holder so that sample is firmly pressed
against window.
- If any abnormality appears in window, remove back and
repeat process. Consistency in sample handling,
preparation, and cell packing is crucial to success.
- Scan sample, collect spectra, and process data.
Comments:
- NIR instrument should be maintained in a dust-free
environment. Clean or change filter monthly. Clean
ceramic standard and drawer assembly monthly.
- Never touch or clean grating.
- NIR instrument should be maintained in a
stable-temperature (25 ±3oC) room with controlled
humidity (60 ±5%). Room should be free from vibration
and have stable and dedicated electrical current.
- Sample must be dried and ground by the same methods as
those samples used to develop the calibration equation.
- Spectrophotometer reads only a fraction of 1 mm depth of
material touching glass of sample container. Therefore
container must be loaded carefully with different
subsample in each quadrant to make spectral reading more
representative of entire sample.
Calculations:
Prediction is made by direct comparison to calibration. No
additional calculations.
Quality Control:
Include at least one set of duplicates in each run. These
duplicates must be two subsamples each packed in a separate
holder. Scanning the same sample twice is not a true replicated
analysis using NIR. An acceptable average standard deviation (s)
among replicated analyses for moisture or dry matter using
reference methods is about ±0.10, which results in a warning
limit (2s) of about ±0.20 and a control limit (3s) of about
±0.30.
Quality control for NIR analysis involves monitoring the
accuracy of both the instrument and the calibration equation.
Instrument diagnostics should be run and recorded weekly, after
instrument warm-up, to insure that maximum response, wavelength
accuracy and repeatability are within acceptable manufacturer's
tolerances. For example, for NIRSystems model 6250 and 6500
instruments, maximum response should be between 55000 and 58000
(adjust if below 51000), suggested wavelength error should be
less than 20 to 30 (correct problems if greater). Each
manufacturer should provide acceptable performance specifications
for their instrument, and often software is provided to monitor
instrument accuracy. However, it is the responsibility of the
operator to run the diagnostic software routinely and record the
results a minimum of once a week. In addition to meeting
specifications, instrument operation should be monitored by
scanning a check sample (that has been sealed in a sample holder)
each day and storing spectral data weekly. Daily results for each
analysis (DM, CP, ADF, NDF, and minerals) should be plotted on an
X-control chart and the chart should be examined for trends.
Standard deviation (s) for the check sample can be established
after 10 scans and should be substantially lower than acceptable
standard deviations of duplicate reference method analyses (.10,
.20, .35 and .60 for DM, CP, ADF, and NDF, respectively) because
the same sample is being scanned. Results outside of ±2s upper
and lower warning limits are evidence of problems with the
analytical system. Results outside of ±3s upper and lower
control limits are evidence of loss of control and no NIR
analyses should be done without detecting and correcting the
problem. Two consecutive analyses falling on one side of the mean
between warning limits and control limits also indicate a loss of
control.
Monitoring the calibration equation consists of two tests that
determine the existence of bias and unacceptable increases in
standard error of prediction corrected for bias [SEP(C)]. Every
20 to 25th sample predicted by the NIR should be analyzed by the
reference method used to develop the calibration equation (Note:
Any bias or increased SEP(C) can be due to inaccuracies in NIR or
differences in reference method analytical procedure). A
continuous time chart of observed bias and SEP(C) should be
plotted to observe trends. After nine (N) samples have been
accumulated, analyze these samples by the reference method (A
sample size of nine is a good compromise between the number of
analyses required and the statistical accuracy desired for the
confidence limits given below). Calculate the observed bias and
SEP(C) using the equations given below. Determine the confidence
limits for bias and SEP(C) based on the standard error of the
calibration equation (SEC). If the SEP(C) of the nine samples is
less than the SEP(C) limit and absolute value of the bias is less
than the bias limit, the calibration equation is acceptable. If
the SEP(C) of the nine samples is less than the SEP(C) limit but
the absolute value of the bias exceeds the bias limit, the
calibration equation may be corrected by adding the bias to the
intercept of the calibration equation or to each value (bias
adjustment) although recalibration is recommended. If both SEP(C)
and bias exceed their limits, add samples to the calibration data
set and recalibrate.
NIR Quality Control Calculations:
Di = Xi - Yi
Bias = Di / N
Confidence Limit = ±0.55 X (SEC)
SEP(C) Confidence Limit = 1.29 X (SEC)
- where Di = difference
- Xi = reference method value
- Yi = NIR value for ith sample
- N = 9 (number of samples)
- Bias = average difference between reference and NIR
values
- SEP(C) = standard error for prediction corrected for bias
- SEC = standard error of the calibration equation.
This monitoring procedure can be used with any NIR instrument
and some manufacturers have incorporated this approach into their
quality control monitoring software
|