PREDICTION OF SOIL PROPERTIES USING GEOPHYSICAL AND UAV-ACQUIRED DATA

Information about soil properties is important for a range of agricultural applications, including assessment of irrigation or drainage needs, soil salinity, and fertilizer runoff potential. Measuring soil properties directly is very time-consuming, but some geophysical techniques can provide information on soil properties. Ground penetrating radar (GPR) can be used to measure soil water content, and electrical conductivity (EC) can be correlated to a variety of agriculturally significant parameters, including soil texture, water content, and salinity. Although useful, geophysical techniques are still time-consuming and can only be performed on selected traverses within a field. This research investigates correlations between multi-spectral data acquired with an unmanned aerial vehicle (UAV) and ground-based geophysical data. UAV data can be collected quickly over an entire field, so has higher resolution and greater coverage than ground-based geophysical data. It is also relatively inexpensive to collect and is routinely acquired over many agricultural fields. In this project, multispectral and geophysical (GPR and electromagnetic) data were acquired over a series of plots containing corn, soybeans, and alfalfa. The multispectral data were used to calculate different vegetation indices: NDVI, NDRE, GNDVI, CIG, and VARI. Geophysical data were acquired along 18 traverses, and the vegetative indices values were determined for each geophysical footprint. These indices were compared directly to the geophysical data, and although the multispectral and geophysical data sets were acquired at different times, there is still correlation between them, with a higher correlation with the EC data than for the water content data. Neural network techniques were also used to predict water content and EC values based only on the vegetative indices values. These techniques showed that prediction of EC and water content could be improved using vegetative indices values, and the VARI index was the most useful for predicting these values.