Fjölrit RALA - 15.12.2000, Side 37

35
NIR-analyses
NIR analyses of grass samples have been used at the Agricultural Research Institute in Iceland
for several years. The NIR analyses have been calibrated using Icelandic grass samples and a
regression of NIR observations on the common analysis has been calculated. One of the aims
of the project was to test wether the Icelandic calibration database could be used for material
írom Greenland and the Faroe Islands.
The samples from the experiments were analyzed with the NIR technique as well as
with standard methods. Comparisons between these methods was carried out and the results
are presented in Tables 37 and 38.
Table 37. Mean difference between protein (% of DM) analyzed with NlRmethod and Kjeldahl method
and standard deviation of the differences. Calculated for each location in 1996 and 1997.
Average difference Standard deviation
1996 1997 1996 1997
Kollafjarður -0.47 -0.12 0.83 0.93
Upemaviarsuk -0.56 -0.50 0.76 0.52
Narsarsuaq -0.42 -0.38 1.12 1.03
Korpa -0.47 -0.10 0.54 0.48
Table 38. Mean difference between digestibility (% DM) analyzed with NIR method and in vitro method
and standard deviation of the differences. Calculated for each location in 1996 and 1997.
Average difference Standard deviation
1996 1997 1996 1997
Kollafjorður 1.57 0.38 4.89 2.76
Upemaviarsuk 0.15 5.18 3.29 3.98
Narsarsuaq -4.03 3.26 5.07 4.75
Korpa -2.10 -0.67 2.51 2.44
The protein analysis with NIR fit the Kjeldahl analysis well at all locations. The
Kjeldahl analysis was on average 0.10-0.56 percent units higher than the NIR analysis. The
standard deviation was highest in Narsarsuaq and lowest at Korpa and Upemaviarsuk. A
possible explanation of the high standard deviation in Narsarsuaq was that the samples were
transported long distances before they were dried, handling that could have affected them. The
standard deviation of the differences were higher for the first harvest dates than for the others
and lower for the Poa pratensis varieties than the others.
The results for digestibility are not as good as for the crude protein. The average
difference between NIR and in vitro analysis was 0.15-5.18 and the standard deviation of the
differences was 2.44-5.07. Again the standard deviation was highest in Narsarsuaq, possibly
for the same reason as for the crude protein content, but lowest in Korpa (both years) and
Kollafjorður inl997. The standard deviation increased with later harvest dates and the NIR
digestibility was overestimated in the early harvest dates and overestimated at the late ones.
The accuracy of the NIR estimate was best for the Phleum pratense varieties but worst for
Fylking. Further studies to clariíy the differences between these two methods are necessary
%