📄 bioengineering

Impact of Cow Parity on the Accuracy of Near-Infrared Spectroscopy for Sustainable Milk Quality Monitoring during Milking

This study demonstrates that while cow parity has a comparable impact on the accuracy of near-infrared spectroscopy for predicting milk fat and somatic cell count, it significantly affects lactose measurement reliability, indicating that parity-specific calibration is essential for sustainable, real-time milk quality monitoring.

Original authors: Iweka, P., Kawamura, S., Mitani, T.

Published 2026-04-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Iweka, P., Kawamura, S., Mitani, T.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are a dairy farmer. Your cows are like a team of athletes, and the milk they produce is their "performance report." To keep your farm running smoothly, sustainably, and profitably, you need to know exactly what's in that milk while it's being harvested. Is the fat content high? Is the lactose right? Is the cow healthy (indicated by low somatic cell counts)?

For a long time, farmers had to wait until the milk was in a lab to get these answers. But this new study introduces a high-tech "smart camera" called Near-Infrared (NIR) Spectroscopy. Think of this device as a super-fast, non-invasive scanner that shines a special light through the milk stream as it flows from the cow. It reads the "fingerprint" of the milk instantly, telling you its composition in real-time.

However, the researchers asked a crucial question: Does this scanner work the same way for every cow?

Specifically, they looked at Cow Parity. In the dairy world, "parity" is just a fancy word for "how many times a cow has had a baby."

First Parity: A cow having her first calf (a rookie).
Second Parity: A cow having her second calf (a veteran).

The Experiment: The "Rookie" vs. The "Veteran"

The researchers set up a high-tech milking station at Hokkaido University. They used a scanner that takes a picture of the milk's light spectrum every 20 seconds. To make sure the scanner was telling the truth, they also took samples to a lab to get the "gold standard" measurements.

They built a "training manual" (a calibration model) for the scanner using data from:

Only the rookies (First Parity).
Only the veterans (Second Parity).
A mix of both.

Then, they tested the manuals to see if they could predict the milk quality of the other group.

The Results: The "One-Size-Fits-All" Problem

Here is what they found, explained through a simple analogy:

1. The Fat and Health Check (SCC): The "Universal Translator"
When it came to measuring Milk Fat and Somatic Cell Count (a sign of udder health), the scanner was like a universal translator. Whether you trained it on rookies or veterans, it could still understand the other group pretty well. The "fingerprint" of fat and health cells didn't change too much between a first-time mom and a second-time mom.

2. The Lactose Problem: The "Accent" Issue
This is where it got tricky. Lactose (milk sugar) is like a specific accent in a language.

When the scanner was trained only on rookies and then tried to read a veteran's milk, it got confused. It was like a teacher who only taught in a British accent trying to understand a student speaking with a heavy American accent. The predictions for lactose became very inaccurate.
Why? Because as cows have more calves, their bodies change. A first-time cow is still growing her own body, so she uses more nutrients for herself. A second-time cow has finished growing and focuses more on making milk. This biological shift changes the sugar (lactose) levels in a way that the scanner didn't expect if it hadn't seen that specific "accent" before.

3. The Solution: The "Mixed Class" Teacher
The study found that the best way to fix this was to train the scanner on both rookies and veterans together.

Analogy: Imagine a teacher who only teaches 5-year-olds. If you suddenly put a 10-year-old in the class, the teacher might struggle to explain things correctly. But if the teacher has experience with both 5-year-olds and 10-year-olds, they can explain the lesson perfectly to everyone.
By combining the data, the scanner learned the "full dialect" of dairy cows. It became robust enough to handle the differences between a first-time mom and a veteran.

Why Does This Matter for the Planet and Your Wallet?

This isn't just about getting better numbers; it's about sustainability.

No More Guessing: If the scanner is accurate, farmers don't have to guess what to feed their cows. They can give the exact amount of nutrients needed. This stops the waste of expensive feed and reduces the cow's "carbon footprint" (less methane from over-eating).
Catch Problems Early: If the scanner detects a drop in milk quality or a rise in cell count immediately, the farmer can treat the cow right away. This prevents the cow from getting sick, reduces the need for antibiotics, and stops good milk from being wasted.
Efficiency: By knowing exactly what each cow is producing, the farm runs more efficiently. Less waste means more profit for the farmer and a smaller environmental impact for the planet.

The Bottom Line

This paper tells us that while high-tech milk scanners are amazing, they aren't "plug-and-play" for every single cow. Cow age (parity) matters.

To make these smart systems truly sustainable and accurate, we can't just use a "one-size-fits-all" setting. We need to teach the machines to understand the unique differences between a first-time mom cow and a veteran cow. Once we do that, we can run dairy farms that are smarter, cleaner, and kinder to the animals and the environment.

1. Problem Statement

Accurate, real-time monitoring of milk quality is critical for sustainable dairy farming, enabling precision feeding, herd health management, and waste reduction. Near-Infrared (NIR) spectroscopy is a promising non-destructive method for this purpose. However, the accuracy of NIR calibration models is often compromised by biological variables. While factors like lactation stage and diet are well-known, the specific impact of cow parity (the number of calvings) on the reliability of NIR predictions for key milk components (fat, lactose, and somatic cell count) remains under-explored. Without accounting for parity, monitoring systems may yield inaccurate data, leading to inefficient resource allocation and suboptimal management decisions.

2. Methodology

The study was conducted at Hokkaido University using an online NIR spectroscopic sensing system integrated into a pipeline milking setup.

Experimental Setup:
- System: An NIR spectrometer captured absorbance spectra every 20 seconds across a wavelength range of 700–1050 nm during milking. The system included a flow meter, a sampling unit holding ~30 mL of milk, and halogen lamps for optical illumination.
- Subjects: Data was collected from dairy cows during early lactation over three consecutive days (morning and evening milking). The study focused on two parity groups: First Parity (A) and Second Parity (B).
- Reference Standards: To calibrate the NIR system, milk samples were analyzed using standard laboratory instruments:
  - Fat & Lactose: Measured via a MilkoScan analyzer (infrared spectroscopy).
  - Somatic Cell Count (SCC): Measured via a Fossomatic device.
Data Analysis & Modeling:
- Algorithm: Partial Least Squares (PLS) regression was used to develop calibration models.
- Validation Scenarios: Five distinct scenarios were tested to evaluate model robustness and cross-parity applicability:
  1. A → A: Calibration and validation within First Parity.
  2. B → B: Calibration and validation within Second Parity.
  3. A → B: Validating Second Parity data using a First Parity model.
  4. B → A: Validating First Parity data using a Second Parity model.
  5. A+B → A+B: Calibration and validation using a combined dataset of both parities.
- Metrics: Model performance was evaluated using the Coefficient of Determination ( $r^2$ ), Standard Error of Prediction (SEP), Bias, and Ratio of Performance to Deviation (RPD).

3. Key Results

The study revealed significant differences in model performance depending on whether parity was accounted for.

Single-Parity Models (High Accuracy):
- Models trained and tested within the same parity group (A→A and B→B) showed excellent performance.
- Fat: $r^2$ reached 0.99 for both groups with very low SEP (0.14% and 0.18%).
- SCC: $r^2$ was 0.90 (A→A) and 0.87 (B→B).
- Lactose: $r^2$ was 0.87 (A→A) and 0.78 (B→B).
Cross-Parity Models (Significant Performance Drop):
- When models were applied to a different parity group, accuracy declined sharply, particularly for lactose.
- Lactose (A→B): $r^2$ dropped to 0.43 with a high SEP (0.18).
- Lactose (B→A): $r^2$ dropped to 0.41 with a high SEP (0.28).
- Fat: While fat prediction remained relatively robust ( $r^2$ > 0.95), it still showed increased error and bias compared to single-parity models.
- SCC: Performance degraded significantly ( $r^2$ dropped to ~0.52–0.71).
Combined Dataset Models:
- Models trained on the combined dataset (A+B) provided a balanced solution. While $r^2$ for lactose was lower (0.59) than single-parity models, the SEP was minimized, and the model avoided the catastrophic errors seen in cross-parity validation. This suggests that inclusive datasets improve generalizability.

4. Key Contributions

Quantification of Parity Bias: The study provides empirical evidence that cow parity is a critical variable affecting NIR spectral interpretation, specifically altering the prediction accuracy of lactose and SCC.
Physiological Context: The authors link the spectral variations to physiological differences:
- First Parity Cows: Allocate more nutrients to growth, resulting in different fatty acid profiles and lower milk fat synthesis.
- Second Parity Cows: Have higher circulating triglycerides and different metabolic states, altering milk composition (protein, casein, corticosteroids) and NIR absorbance.
Calibration Strategy: The paper demonstrates that "one-size-fits-all" calibration models are insufficient for multi-parity herds. It advocates for either parity-specific models or, more practically, combined datasets to ensure robustness across diverse herd demographics.

5. Significance and Implications

Sustainable Dairy Management: Accurate real-time monitoring is essential for precision agriculture. By correcting for parity, farmers can:
- Optimize Feeding: Allocate nutrients precisely based on individual cow needs, reducing overfeeding and feed waste.
- Improve Animal Welfare: Early detection of deviations in SCC or composition allows for immediate intervention in mastitis or metabolic stress, reducing antibiotic use.
- Reduce Environmental Footprint: Minimizing waste and optimizing production efficiency lowers greenhouse gas emissions per liter of milk produced.
Economic Viability: Reducing milk spoilage and improving herd productivity directly enhances farm profitability.
Technological Integration: The findings provide a framework for integrating parity data into "Smart Dairy" systems, ensuring that automated sensing technologies deliver reliable, actionable data for sustainable farming practices.

In conclusion, the study establishes that ignoring cow parity in NIR calibration leads to significant prediction errors. Integrating parity-specific data or combined datasets is a prerequisite for deploying accurate, sustainable, and economically viable real-time milk quality monitoring systems.