When feed or forage analyses are returned from the lab, results are not always as expected, nor as animal performance would suggest. 

Dr Dave Davies, Silage Solutions Ltd, explains why this may be and why some types of feed should only be analysed using wet chemistry.

When you receive your silage analysis back from the lab, how did they arrive at the numbers you’ll use to ration feed for your stock?

In this article I want to provide you with a greater understanding of the NIRS (Near Infrared Spectroscopy) method for analysing feeds and forages. NIRS is quick and relatively cheap, compared with the old test tube (wet chemistry) methods that were used to build our ration formulation approaches. But NIRS can be very costly if the results are not sufficiently accurate to enable the best diet formulation from the results. So, knowing where the pitfalls are helps us to use the results to best advantage.

As Prof Stephen Hawking said: ‘The greatest enemy of knowledge is not ignorance; it is the illusion of knowledge’. For NIRS, just because you have a piece of signed headed paper showing the metabolisable energy (ME), protein, sugar, lactic and volatile fatty acid (VFA) concentration of your preserved feed, might not mean it is accurate and our background knowledge will help us determine whether or not it is believable, and indeed, usable.

Friedrich Wilhelm Herschel

Background and history to NIRS

Near infrared light is part of the electromagnetic spectrum which includes visible light, X-rays and radio-waves, and man has harnessed these physical phenomena to improve life in many ways.

Near infrared light was first discovered in 1800 by Friedrich Wilhelm Herschel who was the Astronomer Royal and the first president of the Royal Astronomical Society.

 

Figure 1

The electromagnetic spectrum is indicated in figure 1, with the visible light spectrum expanded for more detail. Rapid NIRS analysis was developed in the 1960s when Karl H Norris demonstrated NIR spectral data could be measured on samples of grain, and incorporated computers to interpret data to predict its composition. Phil Williams applied the technology to the large-scale real world of commercial commodities.

Figure 2
Figure 3

The technique involves shining NIR light at the sample which excites the organic bonds between molecules, causing further NIR light to be emitted and this is reflected and measured by the spectrophotometer. The data can be captured in a graph, as shown in figure 2. Different types of bond reflect different wavelengths and intensities of light. These differences enable mathematical models to be developed to provide detailed chemical/ nutritional analysis from the spectral data collected. So, the reflected light pattern will be different for a 65 D-value silage, compared to one at 75 D-value. Small differences in spectral data can be as a result of large differences in chemical analysis. The graph, figure 3, represents the NIRS spectral data from seven different grass silages.

The method

When a NIRS forage or feed analytical method is developed, a number of things have to happen to make it robust and accurate. There needs to be a model that takes the NIRS data from your sample and predicts the nutritional quality by comparing it to a reference set for the same type of forage or silage. The reference set, preferably at least 300 examples with a range of nutritional qualities, has been analysed by wet chemistry techniques. The reference set has also been analysed by NIRS, so a model of how the NIRS relates to the wet chemistry can be built up. This is the model that it used to interpret the NIRS data from your sample into ME, protein, sugar, lactic and VFA concentration.

The method When a NIRS forage or feed analytical method is developed, a number of things have to happen to make it robust and accurate. There needs to be a model that takes the NIRS data from your sample and predicts the nutritional quality by comparing it to a reference set for the same type of forage or silage. The reference set, preferably at least 300 examples with a range of nutritional qualities, has been analysed by wet chemistry techniques. The reference set has also been analysed by NIRS, so a model of how the NIRS relates to the wet chemistry can be built up. This is the model that it used to interpret the NIRS data from your sample into ME, protein, sugar, lactic and VFA concentration.

However, there are a few things to consider:

1. Is your forage represented in the model?

If your silage lies outside the range of the reference set, then the model won’t accurately predict the quality of your forage. For example, if your silage has a dry matter (DM) of 55% but the model was only built from silages with a DM range from 25-35%, the prediction is likely to be inaccurate.

2. Mixed forages.

NIRS is not additive, so you cannot take a prediction for one forage and add it to a prediction from a different forage and use it to predict the value from amixture of the two.

3. Changing farm practice.

The models need to be updated to keep up with changes in on-farm practice, for example when we use new plant cultivars/species, or new ensiling methodologies. Getting a reference sample set is very expensive, so this is a difficult issue.

 

Mixed forages give wrong results

Mixing forages is common practice now and the example below shows the pitfalls of using NIRS on a mixed sample of grass and maize silages, when they are combined 50:50 on a fresh matter basis. These were analysed by NIRS.

Firstly, the grass and maize were analysed separately before mixing, using the relevant NIRS prediction model, so, grass with the model from the grass reference set and maize with maize reference set. The samples were then mixed and re-analysed.

Because a mixed grass and maize silage NIRS prediction model does not exist, the mixed sample was interpreted by both the grass and maize prediction models.

The results for DM, ME, crude protein, neutral detergent fibre (NDF), starch and lactic acid are shown in the table below.

 

Table 1: Results from grass/maize silage analysis using different prediction models

In addition, the table also uses the calculated value based on the pure grass and maize silage analysis of the 50:50 mix. The individual grass and maize models were poor predictors of the nutritional quality of the mix, particularly in crude protein and starch.

The only way to get a reasonable indication of the mix was to calculate it from the individual grass and maize results on a dry weight basis, but even then, we are not sure how accurate this will be when it comes to feeding out. So, the lesson here is – don’t send a sample of mixed forages to the lab for NIRS analysis! So, this helps to understand what can go wrong with NIRS. It also shows us that getting a NIRS analysis on a total mixed ration sample is a dangerous waste of time.

A TMR sample NIRS analysis would need to have a reference set behind it with the same ingredients, which is not a practical, realistic or cost-effective proposition for the lab.

Make sure your forage is represented

As supply chains and economics encourage us to grow more variety of home-grown protein and energy sources, for example to replace the use of imported soya proteins, then we need to be very careful that the forage we want analysed is backed by a NIRS model with data from a representative set.

For example, the following analysis was a wholecrop bean silage analysed by what is the standard UK wholecrop NIRS analysis, which is based predominantly on wholecrop wheat and barley, (table 2).

Table 2: Wholecrop bean silage analysed by standard wholecrop. NIRS analysis (based predominantly on wheat and barley)

Note the negative numbers which are both chemically and biologically impossible. A typical bean wholecrop is expected to analyse at between 17% and 21% for both protein and starch.

These results indicate that the NIRS prediction model used was not appropriate, so the results were meaningless and do not even provide a guide. The best thing you can do with this type of report is light the fire with it!

 

NIRS can’t handle the sugar rush

The UK NIRS reference sets include many samples from over 25 years ago. In that time, grass varieties have changed, with many being higher in sugar, and chemical additives have changed such that more sugar is preserved. If the database is not updated in line with these changes, it will underpredict the sugar content of the grass silage being analysed. In 2017, AHDB funded work to examine factors affecting dry matter losses in English silage clamps.

As part of this work, samples were analysed both by traditional wet chemistry and the standard NIRS prediction models.

Figure 4

The graph  (figure 4), shows the comparison of the two analytical results. The graph clearly shows that at concentrations above around 3.5% water soluble carbohydrates (sugar), the NIRS models were underpredicting the content of WSC.

The underprediction became worse, the higher the content of WSC there was.

If the NIRS prediction models were accurate, most of the dots on the graph would be on or close to the solid line rather than the dotted line.

This work also found issues with the accuracy of NIRS prediction for other key measures of fermentation quality like lactic acid, VFA and ammonia-nitrogen.

Where does that leave us?

Users of forage/feed analyses need to be aware of the pitfalls of NIRS analyses.

My recommendation for the major nutrients and the common crops (eg ensiled grass, maize and wholecrop cereals harvested at the usual grain fill stage), is that the standard NIRS analysis is relatively accurate, but be sceptical about the fermentation analytes. For non-standard crops such as wholecrop legumes, crimped legumes and herbal ley forages, at this point in time, ask for wet chemistry for the key components of starch, protein and maybe NDF. The analysis will take longer to do and cost more but it will be more accurate for your rationing and in the long run could save you £1000s through getting your diets right.

Quick check on the analysis

There is a quick check I use when I am suspicious about the accuracy of an analysis. Certain components of your analysis when added together should make up the majority of the whole sample on a dry matter basis. So, if your analyses are expressed on a %DM basis, they should add up to 95. If they’re given in g/kg DM, then that number becomes 950. The 5% or 50g/kg ‘missing’ is for things like pectin. Make sure all numbers are in the same units (often the lactic and VFA are in g/kg DM and the rest of the results are in %). If some numbers are in g/kg DM then divide these by 10 before using them in the sum of the components. The components to add together are: NDF, ash, crude protein, oil, sugars (for maize and wholecrop include starch), lactic, VFA. If these were all expressed as a % then I would hope the number would be somewhere between 85-95. If it is above 100 or below 75 then I begin to question the accuracy of the analysis. If this is the case, look at the date received in the lab, and compare that back to when you sent it. Samples analysed on a Monday are generally more inaccurate than later in the week, because most have spent the weekend in the post. If you cannot find a good reason, then ask for a reanalysis or send a fresh sample. Remember if it is an unusual crop, then this is the most likely reason, and you’ll need to ask for wet chemistry.

Analyses poorly, but feeds out great

I also hear a lot of farmers saying their stated intake potential is low but the silage smells great and the animals love it and are eating a lot. The intake potential on the NIRS analysis was developed as part of ‘Feed into Milk’, a project that developed the NIRS calibrations. As part of that project, a relatively small number of cows were fed a relatively small number of silages in Northern Ireland. The cows were fed a flat rate of concentrate in the parlour. This work was done at the end of the 1990s. The intake of forage was measured and the silages were scanned by NIRS and the calibration developed. There are lots of reasons why the results now could be wrong, not least that most cows are fed TMRs and the data collected represented the genetics and grass silage making techniques of the last century, in addition to the dataset being a very small number of silages with a low number of cows. Therefore, the intake figure is at best a very broad gauge and in numerous situations should be ignored, especially if your silage is over 30% DM and is not 100% ryegrass!

Hope on the horizon for home-grown proteins

As part of the Nitrogen Efficient Plants for Climate Smart Arable Cropping Systems (NCS) project which involves Kelvin Cave Ltd and myself, amongst many industry and academic partners, samples of bean and protein forages will be analysed by both wet chemistry and NIRS. These results will help to identify weaknesses in the current NIRS reference set databases, and provide much more data to ensure the interpretation of analyses can be improved in the future.

Latest news

Newcomers to maize boost milk solids

The Wilson family wanted to lift milk solids by introducing maize and say the crop has exceeded their expectations.When...

Read more

Understanding NIRS and its role in forage and feed analysis

When feed or forage analyses are returned from the lab, results are not always as expected, nor as animal...

Read more

Adding value to homegrown grain helps secure farm’s future

Lancashire farmer, Josh Rothwell (pictured), is adding value to grain on his tenanted farm, and says it’s the way...

Read more

Problems rolling dry feeds addressed by tempering grain

A treatment to ‘temper’ grains and pulses - commonly used by the feed trade - is now being used...

Read more

News Archive

News Categories