Performance evaluation of two handheld NIR spectrometers to determine forage nutritive value using custom calibrations

Abstract Recent technological advancements have led to the release of several commercially available near‐infrared (NIR) solutions, especially with handheld portable spectrometers. We define a NIR solution as the integration of hardware, software, and standardized operating protocols for handling and scanning samples. Rather than relying solely on pre‐loaded calibrations, which may not always meet accuracy requirements or be available for a specific application, NIR users can benefit from being able to develop custom calibrations. We hypothesized that, despite inherent differences in spectra and scanning protocols, custom calibrations developed using a single software pipeline could yield comparable performance across different NIR solutions. Spectra from 317 dried–ground forage samples (grass–legume mixtures) were collected using two portable handheld spectrometers (Trinamix and NeoSpectra) and used to develop NIR models to predict forage nutritive value (crude protein and in vitro organic matter digestibility). During the scanning process, the samples were in direct contact with the Trinamix spectrometer, whereas for the NeoSpectra spectrometer, scanning was performed through a sampling container. Despite differences in the spectra, NIR‐predicted values closely aligned with the laboratory reference values, showing a strong model fit for both NIR solutions. All models achieved r 2 values ≥0.90, with bias ranging from –0.22% to –0.17% and SE of prediction (SEP) ranging from 1.8% to 2.3%. Different model parameterization was needed to optimize the performance of each NIR solution. Our findings demonstrate that flexible software can support the development of custom NIR models, paving the way for tailored, user‐defined NIR solutions.