Source: CORNELL UNIVERSITY submitted to NRP
IMPROVING PROFITABILITY AND REDUCING PRODUCER RISKS OF ALFALFA-GRASS BICULTURES WITH LOW-COST NIRS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
OTHER GRANTS
Reporting Frequency
Annual
Accession No.
1027162
Grant No.
2021-70005-35691
Cumulative Award Amt.
$277,288.00
Proposal No.
2021-06149
Multistate No.
(N/A)
Project Start Date
Sep 1, 2021
Project End Date
Aug 31, 2025
Grant Year
2021
Program Code
[AFRP]- Alfalfa and Forage Program
Recipient Organization
CORNELL UNIVERSITY
(N/A)
ITHACA,NY 14853
Performing Department
Soil and Crop Science
Non Technical Summary
Nationally there is a vast acreage of land on dairy farms not ideally suited to pure alfalfa stands, and alfalfa acreage could be expanded considerably by increased use of bicultures. Reducing variability between formulated and delivered dairy rations has the potential to either increase intake and milk production, or conversely, to significantly reduce feed costs. Our overall hypothesis is that the hand-held Neo Spectra Scanner, with its wide near infrared (NIR) range, will provide accurate, precise and cost-effective alfalfa and alfalfa-grass composition information on farm, and that we can successfully transfer existing calibrations based on large databases from other NIR instruments to the Neo Spectra Scanner. We will identify appropriate applications of this technology, where accuracy is better than variability, and where a management decision can be positively affected, based on NIR results. We will compare hand-held instruments for accuracy and precision of estimating alfalfa-grass nutritive value. We will evaluate the potential for calibration transfer to leverage the current investment in bench-top NIRS calibrations to rapidly develop on-farm prediction models. Specifically, we will explore calibration transfer by leveraging the data we will collect, and assess transfers among hand-held instruments (Aurora and Neo Spectra) with a commercially available portable instrument for use on harvesters (HarvestLab). We will develop calibrations for on-farm estimation of alfalfa percentage in mixtures, assisting alfalfa growers with nutrient management and crop rotation issues. This project meets ASAFS Program Area Priorities 2 and 3 by reducing risks and optimizing economic returns to both alfalfa and milk producers.
Animal Health Component
40%
Research Effort Categories
Basic
30%
Applied
40%
Developmental
30%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
10216402020100%
Knowledge Area
102 - Soil, Plant, Water, Nutrient Relationships;

Subject Of Investigation
1640 - Alfalfa;

Field Of Science
2020 - Engineering;
Goals / Objectives
Our overall hypothesis is that the hand-held Neo Spectra Scanner, with its wide near infrared (NIR) range, will provide accurate, precise and cost-effective alfalfa and alfalfa-grass composition information on farm, and that we can successfully transfer existing calibrations based on large databases from other NIR instruments to the Neo Spectra Scanner. Our specific objectives are:Evaluate hardware performance for the Neo Spectra Scanner hand-held NIRS unit, compared to a laboratory NIRS instrument.Evaluate accuracy and precision of the Neo Spectra Scanner compared to the current on-farm NIRS systems HarvestLab and AuroraNir for determining dry matter and nutritive value of fresh-chopped alfalfa-grass and haylage, as well as assessing instrument-to-instrument variability in the Neo Spectra Scanner.Develop calibrations for the Harvestlab, AuroraNir and Neo Spectra Scanner units for estimating alfalfa percentage in alfalfa-grass fresh and ensiled mixtures.Determine the feasibility of transferring calibrations between portable and hand-held NIR spectrometers.Estimate potential economic losses of under/over feeding of dairy cattle with weekly ration balancing, compared to daily re-balancing.
Project Methods
We will separate the technology differences between the Neo Spectra Scanner and a laboratory NIRS instrument from the range in wavelengths used by these instruments. In order to remove as much undesirable sampling variation as possible due to moisture, particle size, and forage species, we will use a minimum of 600 dry, ground pure alfalfa samples for all scanning with this objective. These samples are currently available and have been analyzed for nutritive value parameters. Preliminary testing of the Neo Spectra Scanner by the PIs on dry, ground alfalfa samples indicated that two stationary scans optimized precision. The Neo Spectra Scanner will be compared with the laboratory instrument constrained to the Neo Spectra Scanner wavelength range. It will also be compared using the full spectral range available on the laboratory instrument. This will not only compare differences in accuracy/precision, but will also parse out the differences due to hardware technology vs. scanning wavelength range.Three Neospectra Scanners will be evaluated with both fresh chopped alfalfa and alfalfa-grass forage as well as ensiled alfalfa-grass in NY. We plan on collecting a minimum of 300 alfalfa-grass haylage samples per year, all collected from on-farm bunker silos across NY. We plan on collecting at least 300 fresh chopped samples per year from farms to combine in known mixtures for scanning and laboratory analysis.Near infrared reflectance calibrations will be evaluated using several criteria. The coefficient of determination is the proportion of variability explained by the model. The root mean square error of prediction (RMSEP), also known as the standard error of prediction (SEP), is the average difference between measured and NIR-predicted values, and is used to calculate two additional criteria. The residual prediction variation (RPD) is the standard deviation (SD) of the reference data divided by RMSEP, and is considered a better measure than RMSEP, as it relates RMSEP to the range of the reference measurements. Additionally, the range of the reference data divided by RMSEP (RER) will be used to assess the success of NIRS calibrations.We will use pure alfalfa and pure grass samples collected above, mixed in known fractions, to calibrate instruments and assess the success of these calibrations as described above.A multi-instrument calibration model will be developed with the best performing mathematical preprocessing steps. A multi-instrument calibration would include the variability introduced by different instrument designs - i.e., optical pathway, light source, and detector into a global calibration. We would then explore data processing methods that will minimize differences in the spectra introduced by the instruments. Calibration transfer will also be attempted by correcting differences in spectra acquired. In this method of calibration transfer, a transformation will be developed to map the response (spectra) of one spectrometer onto another. This type of calibration transfer involves scanning samples on the parent and child instruments that span the variability expected in production. Several algorithms will be evaluated that accomplish this task.The Cornell Net Carbohydrate and Protein System model will be used to simulate the impact on milk performance of day-to-day forage variation in DM and nutritive value from actual ration formulation. The economic benefit of daily ration balancing compared to weekly ration balancing on milk production will be compared to the cost of a hand-held NIRS instrument, to generate a return on investment.Model statistics for each calibration objective will determine the relative success of the process. The estimated economic benefit of daily ration balancing compared to the relative success of the instrument, and the cost of the instrument, will determine the economic benefit of daily ration balancing with hand-held NIR units.Results will be transferred to forage scientists and stakeholders via refereed journal articles, workshops, and outreach publications in national magazines, as well as an extension Guide for using hand-held NIRS on-farm.

Progress 09/01/23 to 08/31/24

Outputs
Target Audience:A summary of current results was presented to audiences of farmers, consultants, and cooperative extension educators. Refereed journal articles were provided for NIR researchers. Changes/Problems:Obj. 5. It has become clear that constant improvements and changes in NIR handheld technology, with instruments regularly coming on and off the market, along with pricing fluctuations, make for an unreliable economic analysis of handheld NIRS for the near future. What opportunities for training and professional development has the project provided?Rink Tacoma-Fogal, a PhD student in Animal Science at Cornell University conducted some of the studies listed under Obj. 3 for her PhD thesis. Shirley Obih, Master's Student and Jack Jewison, Master's Student worked on Obj. 4 at the University of Wisconsin. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?Obj. 2. Calibration evaluation for the 600 haylage samples scanned with four different instruments will be completed and published. Obj. 4. Calibration transfer studies will continue.

Impacts
What was accomplished under these goals? Obj. 1 was completed and published in Sensors, 2022. Obj. 2. A total of 555 haylage samples were collected from bunkers around New York State and scanned with three Neo Spectra instruments, in both a stationary mode and by sliding the instruments over the samples. Results were published in Sensors, 2023. The sliding technique resulted in more accurate prediction of forage parameters, with some variation among instruments. Results for predicting moisture content of alfalfa-grass silage were most positive, with an R-squared value of 0.97. A total of 600 haylage samples were also collected from 111 New York State dairy farms from 2020 to 2022, and vacuum packed for eventual scanning analysis. Samples were scanned with 1) Trinamix scanner, provided by John Deere Germany, 2) a NeoSpectra scanner with revolving sample tray, provided by DairyLand Labs, 3) a Neo Spectra scanner using both stationary and sliding modes, and 4) an Agrocares Series E scanner provided by Agrocares Ltd. Calibrations were developed for each of these instruments. A manuscript focusing on the Agrocares scanner has been generated and submitted to the journal Sensors. Results indicated that all of the above scanners produced satisfactory calibrations, with minimal differences in their prediction accuracy. Existing calibrations for dry matter on a SCIO Cup instrument were also evaluated with these 600 haylage samples, and published in Crop, Forage & Turfgrass Management, 2023. Ten percent of the samples were either too wet or too dry to generate a dry matter result using the mixed silage calibration provided with the instrument. For the 90% of samples that were predicted, dry matter estimates were within 3.2 percentage units of oven dry matter 80% of the time. Instrument precision was very good. General observations: 1) Lack of sample drying and preparation limits the utility of handheld near-infrared spectra to predict forage nutritive value, 2) Sensor-to-sensor differences were observed but could be remedied by including more than one sensor in the calibration set, and 3) Calibration transfer techniques (Piecewise Direct Standardization) did not improve the performance of multi-sensor prediction models. Obj. 3 was completed and published in Crop Science, 2024. A total of 534 alfalfa-grass fresh mixtures were generated by collecting, chopping and mixing alfalfa and grass in known proportions. Samples were scanned with a NeoSpectra scanner in both stationary and sliding modes. Samples were dried and ground and analyzed for nutritive parameters, with calibrations for nutritive parameters still in progress. The instrument was determined to be effective at predicting the alfalfa proportion of fresh mixtures, and the sliding mode of scanning was more successful than a stationary scan. Obj. 4. work is underway at the University of Wisconsin to evaluate the transfer of calibrations between portable spectrometers.

Publications

  • Type: Journal Articles Status: Published Year Published: 2024 Citation: Tacoma-Fogal, R., M. Boggess, J.H. Cherney, M. Digman, and D.J.R. Cherney. 2024. Predicting grass proportion in fresh alfalfa:grass mixtures using a hand-held near-infrared spectrometer. Crop Sci. DOI: 10.1002/csc2.21254.


Progress 09/01/22 to 08/31/23

Outputs
Target Audience:A summary of current results was presented to audiences of farmers, consultants, and cooperative extension educators. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Rink Tacoma-Fogal, a PhD student in Animal Science at Cornell University conducted some of the studies listed under Obj. 3 for her PhD thesis. ShirleyObih, Master's Student and Jack Jewison, Master's Student worked on Obj. 4 at the University of Wisconsin. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?Obj. 2. Calibration evaluation for the 600 haylage samples scanned with four different instruments will be completed and published. Obj. 3. Calibration evaluation of nutritive analyses for the 534 fresh alfalfa-grass samples will be completed and published. Obj. 4. Calibration transfer studies will continue. Obj. 5. Economic evaluation of daily ration balancing, using NIRS, will be initiated.

Impacts
What was accomplished under these goals? Obj. 1 was completed and published in Sensors, 2022. Obj. 2. A total of 555 haylage samples were collected from bunkers around New York State and scanned with three Neo Spectra instruments, in both a stationary mode and by sliding the instruments over the samples. Results were published in Sensors, 2023. The sliding technique resulted in more accurate prediction of forage parameters, with some variation among instruments. Results for predicting moisture content of alfalfa-grass silage were most positive, with an R-squared value of 0.97. A total of 600 haylage samples were also collected from 111 New York State dairy farms from 2020 to 2022, and vacuum packed for eventual scanning analysis. Samples were scanned with 1) Trinamix scanner, provided by John Deere-Germany, 2) a NeoSpectra scanner with revolving sample tray, provided by DairyLand Labs, 3) a Neo Spectra scanner using both stationary and sliding modes, and 4) an Agrocares Series E scanner provided by Agrocares Ltd. Calibrations are currently being developed for each of these instruments. Existing calibrations for dry matter on a SCIO Cup instrument were also evaluated with these 600 haylage samples, and published in Crop, Forage & Turfgrass Management, 2023. Ten percent of the samples were either too wet or too dry to generate a dry matter result using the mixed silage calibration provided with the instrument. For the 90% of samples that were predicted, dry matter estimates were within 3.2 percentage units for oven dry matter 80% of the time. Instrument precision was very good. General observations: 1) Lack of sample drying and preparation limits the utility of handheld near-infrared spectra to predict forage nutritive value, 2) Sensor-to-sensor differences were observed but could be remedied by including more than one sensor in the calibration set, and 3) Calibration transfer techniques (Piecewise Direct Standardization) did not improve the performance of multi-sensor prediction models. Obj. 3. A total of 534 alfalfa-grass fresh mixtures were generated by collecting, chopping, and mixing alfalfa and grass in known proportions. Samples were scanned with a NeoSpectra scanner in both stationary and sliding modes. Samples were dried and ground and analyzed for nutritive parameters, with calibrations for nutritive parameters still in progress. The instrument was determined to be effective at predicting the alfalfa proportion of fresh mixtures. Results for predicting alfalfa proportion of mixtures are currently in review in Crop Science. Obj. 4. work is underway at the University of Wisconsin to evaluate the transfer of calibrations between portable spectrometers.

Publications

  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Feng, X., J.H. Cherney, D.J.R. Cherney, and M.F. Digman. 2023. Practical considerations for using the Neospectra-Scanner handheld near-infrared reflectance spectrometer to predict the nutritive value of undried ensiled forage. Sensors, 23, 1750 https://doi.org/10.3390/s23041750.
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Cherney, J.H., D.J.R. Cherney, and M.F. Digman. 2023. Evaluation of a handheld NIRS instrument for determining haylage dry matter. Crop, Forage & Turfgrass Management, https://doi.org/10.1002/cft2.20239.


Progress 09/01/21 to 08/31/22

Outputs
Target Audience:An overview of handheld NIR opportunities was presented to audiences of farmers, consultants, and cooperative extension educators. Changes/Problems:There have been significant advances in handheld NIRs since this study was proposed. Although we invested considerable funds into purchasing some handhelds (>$50,000, University funds), most of these earlier instruments (AuroraNir, NIR4, XNir, etc.) appear to be going extinct, replaced by considerably less expensive handheld NIR units. Therefore we have shifted our group of NIR units to include the latest developments. DairyLand Labs in Wisconsin has provided us with the latest version of the Neo Spectra Scanner, which is mounted below a revolving turntable that holds the sample. AgroCares (The Netherlands) is providing us with a prototype photo-resistive handheld NIR, with a similar scanning range to the Neo Spectra Scanner (which has a MEMS semiconductor chip). This instrument will be available for sale in late 2022. John Deere-Germany is providing us with a handheld TrinamiX NIR for evaluation, that has yet another different chip made by Spectral Engines. All new instruments are scheduled to be delivered in Sept. 2022. What opportunities for training and professional development has the project provided?Work on calibration transfer issues was initiated at the University of Wisconsin. A PhD candidate in Animal Science at Cornell University is conducting the studies involved in Obj. 3. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?Obj. 2. Additional alfalfa-grass silage samples will be collected for scanning with several handheld NIR units. These samples will be added to the current set of 417 silages vacuum-packed in oxygen-limiting bags, waiting on scanning and laboratory analyses. Obj. 3. Calibrations will be developed for predicting alfalfa proportion of mixtures using the Neo Spectra Scanner, and results will be published. Separately, calibrations will be developed for forage nutritive value parameters and evaluated for effectiveness. Obj. 4. Calibration transfer studies will continue at the University of Wisconsin.

Impacts
What was accomplished under these goals? Obj. 1 was completed and published in Sensors. Forage nutritive value mixed species prediction models were developed using a dataset of near-infrared reflectance spectra from dried alfalfa and grass samples. Nutitive value parameters predicted included NDF, ADF, ADL, CP IVTD and NDFD. Based on comparisons with a laboratory NIR instrument, the handheld Neo Spectra Scanner was successful in predicting forage nutritive value parameters. Obj. 2. A total of 555 silage samples collected from bunkers around New York State were scanned with three Neo Spectra Scanners, in both a stationary mode and by sliding the instruments over the samples. Instrument-to-instrument variability was minor. A manuscript is currnently being written to describe the results of this study. Another 417 silage samples have been collected to-date from 82 farms in New York State. These samples will be used to evaluate the functionality of several handheld NIR instruments, including two recently developed instruments. Obj. 3. A total of 534 alfalfa-grass fresh mixtures were generated by collecting, chopping and mixing alfalfa and grass samples in known proportions. Alfalfa and grass samples were collected from fields over the course of the 2022 season. Sample mixtures were immediately scanned with a Neo Spectra Scanner in both a stationary and sliding mode. Samples were dried and ground, and will be analyzed for nutritive value parameters. Preliminary analysis indicated that the instrument was effective at predicting the alfalfa proportion of mixtures. Obj. 4. Work was initiated at the Univ. of Wisconsin to evaluate the feasibility of transferring calibrations between portable spectrometers.

Publications

  • Type: Journal Articles Status: Published Year Published: 2022 Citation: Digman, M.F., Cherney, J.H., and Cherney, D.J.R. 2022. The relative performance of a benchtop scanning monochromator and handheld Fourier transform near-infrared reflectance spectrometer in predicting forage nutritive value. Sensors 22, 658. https://doi.org/10.3390/s22020658.