The NIRS Forage and Feed Testing Consortium (NIRSC) is a nonprofit association of commercial laboratories, universities, government units, plant researchers and instrument manufacturers.The organization aims to work in synergy to develop innovations, as well as unity for the use of near-infrared spectroscopy (NIRS) and to collaborate with the agricultural industry to advance NIRS knowledge and performance.

The NIRSC has worked to establish standard calibrations for NIRS analysis for over 20 years. Analyzing many samples over time will highlight some samples that are different and not represented in the calibration.

These samples have been examined by individual nutrients, for example, crude protein or digestible fiber, and selected based on those terms. NIRSC has discovered that selecting by individual nutrient (constituent) may omit valuable information.

Forage samples aren’t always predictable for nutritional quality by looking at one or two constituents. Instead, forage samples have inter-related constituents.

So it is important to test for dry matter, protein and fiber constituents, for example, as a package. This develops a database with data clouds of inter-related populations. One constituent may affect another and it is important to parse out trends in constituent values as they are related to others.

Other aspects of sample selection are important in order to develop a database of broad samples that represent many conditions. First, selecting samples based only on spectral difference may actually be an artifact of instrument variation.

Because of this reason and to best utilize the expertise of our membership in knowledge of their own sample populations, we have implemented a series of sample selection steps that we label, preliminary, primary and secondary selection before developing calibration models.

A scanning cell being packed with dried and ground forage. Photo provided by Patty Laskowski-Morren. 

This preliminary selection (1) collects physical samples available at particular times of year, are unique or a lab has a particular interest in going through the selection process, and (2) collects spectra that donor labs have identified as spectral outliers, typically looking at the Global H value.

Historically, labs identified outliers with an H-statistic cutoff of 0.6 or higher for spectral sample selection. In some cases, we now adjust the H-statistic cutoff to a lower value on specific products due to the need to “catch” spectra for species, traits and other important characteristics.

The primary selection step then narrows down possibly thousands of sample spectra to hundreds or less. Donor spectra from labs may be reviewed and physical samples requested from the donating lab. Physical samples are collected from donors and scanned on the NIRSC master instrument.

The secondary selection is then completed using the master instrument spectra. The master spectra is run through selection software and spectral outliers are selected.

From this spectra, any false outliers are eliminated, narrowing the number of original samples submitted down to the best candidate samples to be added to the calibration.

Using this three-step selection process, we can more effectively collect samples of unusual growing conditions, species or varieties, and identify real outliers. Wet chemistry is then completed on those final selected spectra and put in a queue for the next calibration release from NIRSC.

Utilizing multiple platforms for analysis

The NIRSC was first organized in 1992 as a group in order to standardize instrumentation and calibration usage. This was accomplished well with the implementation of common instrumentation and software. With the introduction of more instrument manufacturers into the forage and feed industry in the mid 2000s, a new round of calibration transfer methodology began.

At this time with more powerful computing systems, it was theorized a calibration transfer could be done mathematically with a minimal number of samples scanned on the instrument to receive transferred calibrations, even with variations in optics and sensors compared to a master instrument.

Transferability was adequate, but showed some reduction in how close the data fit the regression line (R2) and some increase in error. Over time, the NIRSC evaluated this process and began to utilize a system of calibration transfer with subsequent large-batch scanning of the same samples on all instruments of different manufacturers.

The use of spectra coming from each type of instrument to develop calibrations on that platform reduces statistical error and improves the data fit to the regression line.

A historical batch of spectra from the original NIRSC master continues to be a basis for all NIRSC instrument types. These spectra are managed for each instrument type using the instrument-recommended software packages and interchangeably to some extent.

In 2016, NIRSC took the concept of batch scanning to another level by implementing a standardized instrument, scanning, and sample library system at a dedicated NIRSC instrument hub. At this hub, one instrument representing each model within consortium membership is located in the same laboratory.

This means all instruments are controlled for temperature, humidity and vibration. In addition, sample batches can be scanned on each instrument model at nearly the same time, side-by-side.

This means that sample variation introduced as samples were shipped to different labs is controlled. Now the library samples are in the same lab stored at that location, with controlled storage containers, humidity and light.

The consortium sample library was reviewed in 2016 and viable samples were extracted. It was determined that samples over 5 years old showed more variation in storage and sample integrity. Samples 5 or fewer years old were generally still in sealed containers and good representatives of recent trends in forage breeding and what has been recently grown by producers.

These samples (a batch of 2,186) were then scanned on each of the instruments in the hub by specific NIRSC product. This provided a set of spectra for each instrument tailored specifically to the measured spectra and technology in a way never done before.

Already, one instrument platform has the spectra rescan project completed. Performance of those calibration models is very promising, and work continues on the other platform models for release in 2017.

The NIRSC is continually assessing methods, calibrations, sample library and most importantly, performance. As a group of scientists, researchers, technicians, lab managers, instrument companies and individuals with multiple strengths combined into one organization, there are continued efforts for moving forward with NIRS technology in ways never done before.

The key advancement for the next calibration release for 2017 is that years of ideas and theory have been put into practice, updating the NIRSC models across multiple platforms in a way never done before.

Further expansion of these models in 2018 will include more regional species, more attention will be given on the overall performance for analysis and continued effort for new constituents will be made available to our members.

Improving forage and feed testing is the main task for NIRSC and we welcome you to join us in our efforts.