The estimation of topsoil water content is of primary interest in the framework of precision farming, but, in general, such assessment is costly and complicated by several interfering factors which do not allow an accurate prediction. Proximal sensing can provide suitable technological facilities to support researchers and technicians in this task. GPR and EMI sensors are valuable instruments as they can provide very informative covariates to be used for improving soil water content estimation. In the present work, it was explored the single (EMI or GPR) and the combined (EMI + GPR) contribution of these proximal data sources. Furthermore, geostatistical (Ordinary Kriging and Kriging with external drift) and linear mixed effects models were applied to compare their respective predictive capabilities. As a result, GPR demonstrated to be more effective in estimating topsoil water content with respect to EMI but, combining both the information, an improvement in the prediction accuracy was observed. Moreover, adding more covariates in the models (GPR outcomes or GPR + EMI outcomes) allowed filtering out the structured spatial component of soil water content. Finally, the statistical approaches proved to behave very similarly, with a slight better performance of Kriging with external drift.
Contribution of EMI and GPR proximal sensing data in soil water content assessment by using linear mixed effects models and geostatistical approaches
Barca E;
2019
Abstract
The estimation of topsoil water content is of primary interest in the framework of precision farming, but, in general, such assessment is costly and complicated by several interfering factors which do not allow an accurate prediction. Proximal sensing can provide suitable technological facilities to support researchers and technicians in this task. GPR and EMI sensors are valuable instruments as they can provide very informative covariates to be used for improving soil water content estimation. In the present work, it was explored the single (EMI or GPR) and the combined (EMI + GPR) contribution of these proximal data sources. Furthermore, geostatistical (Ordinary Kriging and Kriging with external drift) and linear mixed effects models were applied to compare their respective predictive capabilities. As a result, GPR demonstrated to be more effective in estimating topsoil water content with respect to EMI but, combining both the information, an improvement in the prediction accuracy was observed. Moreover, adding more covariates in the models (GPR outcomes or GPR + EMI outcomes) allowed filtering out the structured spatial component of soil water content. Finally, the statistical approaches proved to behave very similarly, with a slight better performance of Kriging with external drift.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.