Improving Predictability of Multisensor Data with Nonlinear Statistical Methodologies

The evaluation of the forage quality nutritive value and biomass usually takes multiple harvests and is considered time consuming, labor intensive, and expensive. The use of sensors to evaluate different forage traits has been proposed as a method to alleviate this problem. However, most analytical techniques involve the use of traditional linear methods for prediction, but prediction models can still be improved with the use of nonlinear methods. Thus, the objectives of this study are twofold: (i) to evaluate the performance of different prediction methodologies in important forage agronomic traits; and (ii) to evaluate the impact of sensor variables on predictive model performance in the different methods. The nondestructive multisensor system that accommodates spectral, ultrasonic, and laser data was tested on a bermudagrass [ Cynodon dactylon (L.) Pers] experiment. Partial least square regression, ridge regression, support vector machine with radial kernel (SVM), and random forest were tested for the prediction performance. The random forest had the best performance, with 0.89 correlation for dry matter yield trait prediction, whereas SVM performed best in the remaining 15 traits, with correlation ranging from 0.72 to 0.95. Besides the prediction performance of statistical models, this study also provided some insight about the importance of variables by removing variables and reevaluating the model performance in dry matter yield. Overall, most of the traits in this study could be reliably predicted by the SVM. The multisensor system can facilitate the measurement of agriculturally important traits in a time‐, labor‐ and cost‐efficient way.