Epsilon Archive for Student Projects

Abstract

The quality of forage is a critical factor influencing animal health, productivity, and the nutritional value of derived products. Accurate assessment of forage quality and yield is essential for optimizing livestock nutrition, improving productivity, and promoting animal welfare. While
laboratory analyses provide precise nutritional parameters, they are often labor intensive and time consuming. Remote sensing technology offers a promising alternative by enabling rapid, large scale monitoring of forage characteristics. Widely regarded as a key tool in precision agriculture, remote sensing facilitates efficient field monitoring and data-driven decision-making. This study
investigates the potential of multispectral imaging technology to estimate both the quantity and quality of forages. Over a two-year period, data on grass-legume mixtures were collected in Northern Sweden using an Unmanned Aerial Vehicle (UAV) equipped with RGB and multispectral cameras. Captured images underwent radiometric corrections and were processed into orthoimages to extract reflectance data. To predict the key forage quality indicators Dry
Matter (DM) yield, Crude Protein (CP), Neutral Detergent Fiber (aNDFom), Organic Matter Digestibility (OMD), and Metabolizable Energy (ME), four statistical models were tested: Multiple Linear Regression (ML), Partial Least Squares (PLS and caretPLS), and Support Vector
Machine (SVM). Model performance was evaluated using Root Mean Square Error (RMSE),Relative RMSE (RRMSE), and Coefficient of Determination (R²). Among the models, PLS
demonstrated a superior performance in predicting certain parameters in the mixed grass and clover dataset. Specifically, PLS achieved an R² of 0.71 for ME and 0.75 for DM yield. ML exhibited the lowest RMSE and RRMSE values and the highest R² values across all parameters.The ML results were likely effected by overfitting as evidence of multicollinearity was observed, suggesting potential redundancy among predictor variables. CP and aNDF showed a lower reliability due to the missing data. These findings underscore the potential of forage quality prediction using UAVs while emphasizing the need for feature selection to address multicollinearity issues.