NIR Spectroscopy: a strategy for the value enhancement of production
Ana Soldado visits Tom Fearn

Purpose of the visit
Dr. Ana Soldado from th Regional Institute for research and Agrofood development, Asturias (Spain) visited Tom Fearn,University Collegue of london,London(UK). In the frame of the development of the R&D project (INIA2010-00128-00-00) and the Cost Action FA0802, Feed for Health, has emerged the possibility of collaboration between different research groups in order to enhance the quality control of animal feed aimed to the reduction of existing “known and unknown risks”. In the basis of this acknowledgement we try to develop a new sensor based on NIRS "on-site" technology, for inspection, control and management of raw materials, feedstuffs and forages in receiving, storing, processing, distribution sites and farms.
In concordance with the general objective of the Cost Action, the work developed in London was focused on safety and quality control of feed and animal nutrition, improving animal health and also the quality, safety and wholesomeness of human foods of animal origin.
To attempt this general objective we have processed the NIRS (Near Infrared Spectroscopy) data of different feedstuffs and forages as raw materials to evaluate the possibility of transferring chemometric models to predict nutritive parameters between at-line to on-site and on-line NIR instruments. The complexity of the wet forages and instrumentation made necessary to collaborate with the Professor Tom Fearn, who has a wide experience in applied statistics and a particular interest in NIRS and the ‘chemometric’ methodology that has been developed to deal with the high dimensional data that arises from this application.

Description of the work carried out and main results during the visit
The silage is normally fed during periods when alternative feed supply is low. And the characterization of grass silage is a key factor in the formulation of ruminant diets and dairy animal performance, because a poor quality silage can further reduce intake of paddock feed and give only a small net gain in milk output. Being necessary to establish controls at farms level to optimize the feed intake of ruminants.
Previous research works have proved the possibility of evaluating different nutritive and fermentative parameters on raw undried grass silages using Near Infrared Spectroscopy (NIRS) calibrations, but normally, using NIR laboratory instruments. The development of these NIR methodologies was not a simple task, it needs large data base, it is tedious, expensive and time consuming.
These previous considerations make necessary to value all data sets collected using at-line NIR instruments to develop a robust chemometric model able to be transferred to other on-site instruments. This methodology has been applied widely to raw and milled feedstuffs. However to attempt this research work it is not easy due to the sample characteristics. Intact (undried) grass silage is a heterogeneous and high moisture sample, and it is not possible to be homogenized without sample pretreatment.
The specific plan is focused on development and evaluation of cloning algorithms and pretreatment of spectral data for the transfer of existing results on spectral libraries and metadata obtained with laboratory at-line instruments (Foss NIRSystem 6500 device) to on-site portable instruments (Zeiss Corona device).
To attempt this work, the steps carried out have been:

1.- To develop chemometrical models with the spectra collected using at-line instrument, in the overlapping range.
We have tried different mathematical and derivative treatments comparing results when using different wavelength ranges. No much differences have been found when applying different derivative treatments, and we decided to work with the first derivative procedure, because we observed that when comparing spectra from Corona (on-site) and FossNIRSystem (at-line) NIR instruments the differences were minimized using SNVD or Savitzky -Golay and first derivative.

2.- To try different chemometrical strategies to transfer the developed equation.
According to Fearn (2001), who indicates that the best option in standardization procedures is to use real samples, in this work Prior to my visit to UCL we scanned 11 fresh grass silages in both NIRS instruments (in triplicate with Foss, and 20 times each when using Corona NIRS), and these samples were employed in the cloning procedure previous to transference of chemometrical models.
We have evaluated the possibility of transferring models avoiding matrix development, it means applying Savitzky –Golay, first derivative and a correction of the average differences to get an acceptable prediction.
Results obtained not showed significative differences between samples scanned in one and other instrument. 

3. External validation procedure
A set of samples containing 20 different grass silages was scanned in triplicate with Foss Instrument and four times each when using Corona NIRS. And the prediction results for this spectral data were acceptable for Foss NIRS (SEPDM=1,5) and unacceptable for Corona NIR (SEPDM=3) (See Figure 1)

An explanation for these results was detected when comparing validation spectra collected in both instruments, in various examples wide differences were observed when scanning the same samples in both instruments. These results were not observed when comparing spectra of cloning set, and it is possible the establishment of a relationship between results and instrument characteristics:
a) Foss NIRSystem 6500 scanning monochromator (Foss NIRSystem, Silver Spring, MD, USA) is provided with a transport module in the scanning range of 400 to 2500 nm at 2 nm interval. The spectral data were recorded in reflectance mode (log 1/R) with WINISI 2 software v.1.05 (Infrasoft International Inc., Port Matilda, PA, USA). The analysis were carried out using natural cells.
b) Corona 45 visNIR 1.7 (Carl Zeiss, Inc.) is a diode array spectrometer that look readings from 400nm to 1690nm. Subsamples of each grass silage were scanned using this instrument with a Petri dish cell, and values were recorded as log (1/R). White referencing and dark current measurements were carried out manually. All spectra were recorded using CORA software version 3.2.2 (Carl  Zeiss,Inc.).
Taking into account the differences related with optical system and window size, we thought in the possibility of minimizing these differences increasing the number of scans when using Corona Instrument. And now, I have returned to my lab and I have scanned 20 different fresh silages in triplicate with Foss NIRSystem and 20 times with Corona. We are working via e-mail with Prof. Tom Fearn to finish this research work, a publish these results.

Future collaboration with host institution
In the frame of this research work and different R&D projects we will collaborate with Prof. Tom Fearn by e-mail, returning to London or trying to get that Prof Tom Fearn come to our country Asturias to follow these transference strategies with different samples and instruments.

Projected publications/articles resulting or to result from the STSM
At this time we are writing a paper detailing these results and evaluationg the alternative of scanning more than 4 samples with Corona to minimize prediction errors. When published we will send a copy to Feed for Health Chair.

Ana Soldado STSM
Design & Programming: Zenlex