The accuracy of a single sensor is often low because all proximal soil sensors respond to more than one soil property of interest. Sensor data fusion can potentially overcome this inability of a single sensor and can best extract useful and complementary information from multiple sensors or sources. In this study, a data fusion was performed of a Vis–NIR spectrometer and an EM38 sensor for multiple soil properties. Stepwise multiple linear regression (SMLR), partial least squares regression (PLSR) and principal components analysis combined with stepwise multiple linear regression (PCA + SMLR) methods were used in three different fields. Soil properties investigated for data fusion included soil texture (clay, silt and sand), EC, pH, total organic carbon (TOC), total nitrogen (TN) and carbon to nitrogen ratio (CN). It was found that soil property models based on fusion methods significantly improved the accuracy of predictions of soil properties measureable by both sensors, such as clay, silt, sand, EC and pH from those based on either of the individual sensors. The accuracy of predictions of TOC, TN and CN was also improved in some cases, but was not consistent in all fields. Among data fusion methods, PLSR outperformed both SMLR and PCA + SMLR methods because it proved to have a better ability to deal with the multi-collinearity among the predictor variables of both sensors. The best data fusion results were found in a clayey field and the worst in a sandy field. It is concluded that sensor data fusion can enhance the quality of soil sensing in precision agriculture once a proper set of sensors has been selected for fusion to estimate desired soil properties. More efficient statistical data analysis methods are needed to handle a large volume of data effectively from multiple sensors for sensor data fusion.
- least-squares regression