Morning, 

Sorry, but it is not my research area.

Dear Prabhani Nisansala

It is generally agreed that NIRS equations should be specific for the feedstuff or forage of interest. Thus, that will be the explanation for the situation you are describing, as you have replaced one big component of the diet (corn) for another (wheat).

This change in the ability of the NIRS to predict the nutrient composition of the new, wheat-based diet can specifically reside in two areas:

1. Particle size of the feed: Wheat products are very fine compared to corn products, that can have a more coarse particle distribution. This will directly affect the ability of the NIRS to predict accurately, as any change in particle size will affect the processes of light reflectance, for example, by changes in the percentage of diffraction. 

2. Chemical differences between the feedstuffs: This is more related to fatty acid composition in the case of fats or monosaccharide content in the case of fiber.

What to do?

1. If the change of corn for wheat is permanent, you might want to generate a new prediction equation for the feed.

2. A first step is to check the effect of particle size of your sample. Make sure you grind your samples uniformly

3. Generally, if an adequate range of nutrient contents is available, you can generate an initial new NIRS equation with some 100 samples,plus let´s say 20 more for cross validation. Other people have used as little as 20 samples to generate a new equation, but that will depend on the variability of the nutrients you want to predict.

4. The validity of this equation needs to be tested and the equation can always be improved, if you have the proper software (i.e. WINISI in the case of FOSS NIRS).

Hope you find an early solution to your problem. Kind regards, 

Thank You very much. For your valuable answer... 

Dear Prabhani Nisansala,

generally, the number of samples depends on the nutrients you want to predict and also the ingredient. So, is important generate a specific equation for each ingredient in a ration.

For nutrients which have a small variability in a certain ingredient (crude protein in ground corn, for example), you need less samples to achieve a good coefficient of determination on your equation. However, your equation will be restricted to a small range and this increase the possibility of your ingredient get out of the equation range, making it imprecise.

So, is important to have many samples to achieve a good equation but is also important to add samples with high variability to generate a equation who can be applicable in many sources of a ingredient, considering its variability.

Regards.