Predicting soil texture with electrical resistivity for inputs to HYDRUS 2-D hydrological predictions and chloride transport: A case study in the Piedmont physiographic region in Georgia, USA

Abstract:

Currently available soil datasets such as SSURGO are based on limited on-site information and provide information up to a maximum depth of 2 m. They can be inadequate for activities like the installation of septic systems that warrant a detailed characterization of the sub-surface. Our study aims at investigating an alternative method to determine spatially distributed soil texture and assessing the impact of its implementation on a hydrologic model. We developed a detailed soil texture map by importing soil electrical resistivity in an Artificial Neural Network (ANN) framework to estimate soil texture at a high resolution. The ERT driven soil layering model uses three easily measured inputs: soil resistivity, relative depth of investigation, and weekly antecedent rainfall. To train and test the model, we used soil samples, collected in 30-cm increments up to 510 cm, from 11 different locations across a 42 m hillslope transect. Texture was measured using a laser particle size analyzer. We used bootstrapping to train and test the ANN framework to identify the minimum set of soil samples required to generate an acceptable dataset and concluded that models trained on 6 locations had the least uncertainty. The soil texture predictions had values of R²=0.74 and RMSE=0.4. From the obtained textures, we created a structure in HYDRUS 2-D to predict hydrological and chloride transport at the hillslope scale. The resulting model was used to investigate the effects of varying domain complexity by comparing its results with a model that was developed using the traditional pedological soil horizon-based division of the sub-surface. The two model versions differ mainly in the number of layers in the saprolite and the textural classification of the soil layers. The hillslope receives a constant sub-surface flux input of solute (akin to a septic system) and feeds into a lake. Both models are calibrated using water table elevations. In this presentation, we will discuss the soil texture determination process and the key differences in calibration and predictions for the two hydrologic models.