Combining frequency domain reflectometry and visible and near infrared spectroscopy for assessment of soil bulk density

•Bulk density can be assessed be a combination of vis-NIR and FDR techniques.•ANN with multi-variable input is the best tool for predicting ω and θv.•ANN with multi-variable input enhanced considerably the prediction of ω.•ANN with multi-variable input enhanced slightly the prediction of θv.

This paper introduces a new approach for the assessment of soil bulk density (BD), which relies on an existed model to predict BD as a function of a visible and near infrared spectroscopy (vis-NIRS) measured gravimetric moisture content (ω) and a frequency domain reflectometry (FDR) measured volumetric moisture content (θv). A total of 1013 soil samples collected from England and Wales, from 32 arable and grassland fields with different soil types were measured with a vis-NIR spectrophotometer (LabSpec®Pro Near Infrared Analyzer, Analytical Spectral Devices, Inc., USA) after in situ measurement with a ThetaProbe FDR (Delta-T Device Ltd.). Two calibration methods of the vis-NIRS were tested, namely, partial least squares regression (PLSR) and artificial neural network (ANN). ThetaProbe calibration was performed with traditional methods and ANN. ANN analyses were based on a single-variable input or multiple-variable input (data fusion). During ANN – data fusion analysis, vis-NIRS spectra and ThetaProbe output voltage (V) were fused in one matrix with or without laboratory measured texture fractions and organic matter content (OM). For the vis-NIRS and ThetaProbe traditional calibration, samples were divided into calibration (75%) and prediction (25%) sets, whereas for the ANN analyses these were divided into calibration (65%), test (10%) and independent validation (25%) sets. Results proved that high measurement accuracy can be obtained for ω and θv with PLSR and the best performing traditional calibration method of the ThetaProbe with R2 values of 0.91 and 0.97, and root mean square error of prediction (RMSEp) values of 0.027 g g−1 and 0.019 cm3 cm−3, respectively. However, the ANN – data fusion resulted in improved accuracy (R2 = 0.98 and RMSEp = 0.014 g g−1 and 0.015 cm3 cm−3, respectively). This data fusion approach led to the best accuracy for BD assessment when vis-NIRS spectra and ThetaProbe V only were used as input data (R2 = 0.81 and RMSEp = 0.095 g cm−3). It can be concluded that BD can be measured by combining the vis-NIRS and FDR techniques based on ANN-data fusion approach.