Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Adaptive Cropping Systems Laboratory » Research » Publications at this Location » Publication #258147

Title: A career in agriculture and NIR

Author
item Reeves Iii, James

Submitted to: Near Infrared Spectroscopy International Conference Proceedings
Publication Type: Other
Publication Acceptance Date: 8/30/2010
Publication Date: 9/7/2010
Citation: Reeves III, J.B. 2010. A career in agriculture and NIR. Near Infrared Spectroscopy International Conference Proceedings. WEB publication.

Interpretive Summary: Over the last two decades near infrared spectroscopy (NIRS) has moved from what animals or humans may eat to environmental aspects there of. Both from research and reviewer activities this has led to some observations which will be the thrust of this discussion. For example, in the early years reviewers would comment that “All this is known” which was often true, but not published and thus unavailable. Today, a great deal is known and published about NIRS in general, but many new users do not seem to be availing themselves of this information. Thus, the same errors and even new variations are all too often being made. Some areas with huge potential in agriculture include determination of: ground cover, invasive species, crop health, C-sequestration, nutrient requirements, illegal crops, etc. Many of these applications will require remote sensing and it is here that some huge problems remain to be solved without the need to relearn what is already known and applicable. For example, how does one get a proper reference value when a sample is an area of 5-100 m2? Soils are at least as diverse as forages and no one has yet to determine what range of soil types can be included in one calibration and assays used are at least as empirical as those used for forages. The future potential for NIRS in agriculture, analogies to past efforts and more unique problems will be the center of this discussion from the perspective of my career.

Technical Abstract: Having spent the last three decades in research and most of that dealing with the application of near-infrared (NIR) spectroscopy (NIRS) to animal agriculture has led to observations about NRS in general and career decisions in particular. For example, over the last two decades NIRS has moved from largely an emphasis on animal nutrition applications to the environmental aspects there of. At the same time, while a great deal has been learned on how to develop NIR calibrations and is readily available through the internet, many new users do not seem to be availing themselves of this knowledge and thus are making mistakes which should be easily avoided. Observations from both research and reviewer activities will be the thrust of this discussion. For example, in the early years reviewers would comment that “All this is known” which was often true, but not published and thus unavailable. Today, a great deal is known and published about NIRS in general, but many new users do not seem to be availing themselves of this information. Thus, the same errors and even new variations are all too often being made. One error found all too often when reviewing is the use of too few samples to develop calibrations, something which may be impossible to correct after the initial sample collection. Research has demonstrated that with partial least squares it is possible to develop calibrations with rsquare values (One-out X validation results greater than .9 using only random numbers for analyte values if sample numbers in the 30-40 range. Artifical neural networks (ANN) are even more prone to over-fitting and many apply neural nets using sample numbers in this same range. Consultation with several other researchers confirmed that ANN calibrations require sample numbers in the 1000 or greater range, something I have only seen once in reviewing manuscripts. From these activities it has been become obvious that there is a great deficit in the area of educating new users in the proper use of NIRS and chemometrics, but how to address this is still in question as, as has been observed by many, most universities have no interest in doing so. Finally, some areas with huge potential in agriculture include determination of: ground cover, invasive species, crop health, C-sequestration, nutrient requirements, illegal crops, etc. Many of these applications will require remote sensing and it is here that some huge problems remain to be solved without the need to relearn what is already known and applicable. For example, how does one get a proper reference value when a sample is an area of 5-100 m2? Soils are at least as diverse as forages and no one has yet to determine what range of soil types can be included in one calibration and assays used are at least as empirical as those used for forages. The future potential for NIRS in agriculture, analogies to past efforts and more unique problems will be the center of this discussion from the perspective of my career.