| Author | Message | ||
     Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 65 Registered: 5-2009 |
Hi guys, David is right about making more duplicates and even triplicates or the ref analysis. It does help to reduce the error of the ref method. But, if the ref method is as it is -it will only help a little, but will be very costly. I would examine the ref method procedures and try to analyze what is in it that contributes the largest share of the error and try to fix it. But even with that - the spectra range is not in you favor, and getting values only every 2 nm is also working against you. Gabi Levin Brimrose | ||
| David W. Hopkins (dhopkins) Senior Member Username: dhopkins Post Number: 202 Registered: 10-2002 |
Hi Maris, I would agree with Gabi, that the reason for your poor success so far is because the reference method error is so high, particularly since the SD of the ref values indicates that the range is not well covered and narrow with respect to the reference method error. If there is no immediate possibility of enlarging your sample set and the range of beta-glucans, then I would recommend concentrating on obtaining better ref values. The paper by Schmidt etal that you cited indicated that they were able to obtain 0.1% on their ref method, and that kind of reproducibility would probably make it possible for you to obtain useful calibrations. Is it possible for you to tune-up your method? It would be useful to obtain at least duplicate tests, and use the average. I think 3 or 4 reps might be justified. Good luck in your studies! Best wishes, Dave | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 64 Registered: 5-2009 |
Dear Maris, It seems that you have hit the nail on the head quite well, but let us expand on it a little more: 1. The range for case 1 - 3.54 to 10.25 is 6.71 and your ref error is about say 0.3 - so if we divide the ref error by the range we get 0.044, which barley satisfies a basic rule of thumb for all secondary type calibrations that the ref error should be less than 5% of the range. If the error is 0.4, then you exceed the rule of thumb. If you add to this the limitation posed by the spectral range you operate at, 850 to 1050, where everything is crowded, and the relatively large increments - 2nm, you get into an impossible corner from which escaping is extremely difficult and could be impossible. 2. The range for case 2 is even worse, 4.26, with even the low end of the ref error nearly exceeding the rule of thumb, and with the high end of the ref error it is nearly double the rule of thumb. If we add to that the same spectral difficulties - we are in a bad spot. My only recommendation to you would be to go to an instrument that can do at least up to 2000nm, and get a better ref method. If you go up to 2000nm, it will be difficult to do transmission, but reflectance can do the job quite well, and you will be looking at peaks with stronger absorption, improving the NIR signature. Still, the most important part is improving the ref error - as I would imagine that increasing the range is not something you control. Thanks Gabi Levin Brimrose | ||
| Francesco Davini (franz) Intermediate Member Username: franz Post Number: 17 Registered: 2-2009 |
I don't know how a beta glucan spectra looks like and how and where it differentiates from other saccharides, but surely your spectra coverage is quite narrow. I'd first seek for spectra of beta glucan and similar compounds present in barley. If the main spectral differences are found elsewhere than in your range, then I see it difficult to obtain much better results. | ||
| Maris Stegenburgs (mariss) Junior Member Username: mariss Post Number: 9 Registered: 10-2010 |
Dear colleagues! I am trying to make calibration for beta glucan in barley. Instrument used is 850 - 1050 nm, 2 nm step, transmittance. For naked barley: samples are in range 3,54...10,25%, mean=5,21%, std.dev.=0,88 For hulled barley: samples are in range 2,05...6,31%, mean=4,34%, std.dev.=0,79 Reference method accuracy is 0,2 ... 0,4% between paralel measurements. So far I have found this reference: http://findarticles.com/p/articles/mi_hb5720/is_200907/ai_n35629924/?tag=content;col1 While it tells about some success using Infratec 1241, my results are far from desirable. For the moment it seems too much noise in my data (both spectral and ref) and relative narrow range. I will appreciate any comments. Thank you in advance! |