Estimation of Soil Texture Using Ground Penetrating Radar Groundwave Techniques

Accurate estimation of soil texture over large areas is important for a range of agricultural, geotechnical, and environmental applications. Soil texture data are typically collected as a limited number of point measurements, which are often insufficient to characterize heterogeneous field sites. This study investigates the potential of ground penetrating radar (GPR) groundwave data for soil texture estimation. GPR groundwave data are non-invasive and can be collected very quickly, so GPR techniques can be used to obtain the large data sets needed to characterize heterogeneous sites. This study was performed at a 3-acre, heterogeneous field site, and GPR travel time and amplitude data were collected in a grid of ~10,000 samples across the site. Thirty-four soil texture samples were also collected across the site. Using half of the soil data and quantifying the soil texture as percent sand, variogram calculations and kriging were performed to estimate the soil texture over the entire site. The error in the kriged estimates was evaluated by comparing estimated values with the remaining soil texture measurements. Then, cross-variograms and co-kriging were performed using half of the soil data and all of the GPR travel time data. The error was again estimated using the remaining soil data, and comparison of the errors from the kriged and co-kriged data sets showed that the GPR travel time data significantly improved estimation accuracy. The co-kriging analysis was repeated using GPR amplitude data, and a similar reduction in error was observed, indicating that either travel time or amplitude data may be useful to improve soil texture estimation. Lastly, the GPR travel time data were investigated to determine whether they could provide useful soil texture information without any pre-existing soil data. The GPR travel time data were converted to water content estimates using a petrophysical model, and the field site was divided into 23 smaller "blocks" of equal size. The statistical moments of water content and the mean of the kriged values for soil texture (using all soil data) were calculated for each block. Linear regression of the soil texture and water content data suggests that GPR data may be useful for estimating soil texture, even in the absence of soil measurements.