Estimating dry matter in African stargrass (Cynodon nlemfuensis) forage with in-field hyperspectral proximal sensing

• Hyperspectral data is cost-effective for estimating dry matter in African Stargrass. • The PLSR model for estimating dry matter achieved high statistical performance. • Hyperspectral sensing enhances forage monitoring and livestock production efficiency. Hyperspectral remote sensing, applied to forage monitoring, stands as an innovative technology promising to enhance livestock production by providing real-time results at reduced costs. Currently, forage quality testing in Costa Rica is predominantly conducted using destructive and time-consuming methods such as wet chemistry or laboratory spectrometry. This study aims to estimate the dry matter content of African Star Grass ( Cynodon nlemfuensis ) at the Alfredo Volio Mata Experimental Station (EEAVM) employing field spectroscopy with a multitemporal approach and a partial least squares regression (PLSR) mathematical model. Data were collected from two experimental plots managed for mowing and grazing at the EEAVM of the University of Costa Rica from June to September 2022. During this period, two growth cycles were considered for mowing management and three cycles for grazing. Forage quality tests were conducted using wet chemistry at the EEAVM bromatology laboratory to provide a reference dataset to create the model. Additionally, auxiliary data such as forage height and biomass were collected to correlate with spectral data and identify outliers in the data set. Spectral data were acquired using an ASD FieldSpec4 spectroradiometer with 1 nm resolution and a spectral range of 350–2,500 nm, corresponding to visible and near-infrared. The ChemFlow Galaxy platform was used to develop 18 dry matter estimation models for forage applying PLSR to two pretreatment combinations and different spectral eliminations on three data sets corresponding to the ones collected on the management plots (mowing, grazing, and both combined). Results revealed that dry matter did not exhibit the expected behavior due to the unusual development of forage within the experimental plot. The optimal model for obtaining dry matter quality parameter demonstrated robust statistical performance, with r 2 cv = 0.81, RMSECV=1.41, r 2 c =0.89, RMSEC=1.16, and an RPD of 2.31, classified as a highly satisfactory model. These findings have significant implications for forage monitoring and livestock production, as they provide a more efficient and cost-effective method for estimating dry matter content at the EEAVM. This model can be scaled up to a broader scale by replicating the experiment in different regions to collect more data and generate a more robust model, while taking into consideration the use of alternative remote sensing platforms.