Qin, Mingming. Prediction of soil water dynamics and saturated hydraulic conductivity with wavelets and percolation theory. Retrieved from https://doi.org/doi:10.7282/t3-d3qe-6266
DescriptionSoil water dynamics in near-surface soils subjected to cycles of wetting and drying (vadose zone) can be inferred from frequent readings of sensors installed at various depths into the soil. The knowledge provided by the analysis of such soil water content time series is important for predicting hydrological and geochemical processes in the vadose zone. However, characterizing soil water dynamics in areas lacking sensor data remains a challenging task, especially in deep soils. One central objective of this dissertation was to estimate time series of subsurface soil water content (SWC) from surface SWC and soil properties of the entire profile. The distribution of soil water in the vadose zone over time can also be inferred using numerical models, but that approach requires information on soil hydraulic properties. Saturated hydraulic conductivity, ks, is a key soil hydraulic property for estimating subsurface SWC, which in most instances must be predicted. Therefore, another major objective of this dissertation was to predict ks. In Chapter 2 a statistical model was developed using wavelet decomposition of surface measurements of SWC to estimate subsurface SWC by segregating features at different temporal scales and projecting them to the subsurface. Climate data and SWC at various depths were collected from eight sites in the Atlantic Coastal Plain of the USA. Soil water retention and hydraulic conductivity functions of each horizon were optimized by comparing measured and predicted (using the numerical model HYDRUS-1D) soil water contents. Each time series of SWC was decomposed into 50 scale (s) components using the Mexican Hat wavelet, and later reduced to five group components. Changes in the values of each group component with depth were represented with transfer coefficients that could be estimated with predictors derived from particle size distributions and optimized soil hydraulic functions. Subsurface SWC was predicted reasonably well with the proposed approach, particularly when the vertical movement of soil water was unrestricted.
Saturated hydraulic conductivity is one of the most important predictors for the statistical model developed in Chapter 2. Selected predictive models of ks using water retention parameters from two functions were investigated in Chapter 3 with water retention data and ks measured on 378 soil cores collected from four sites in the United States and multiple sites across Norway. Three ks models based on a generalized Kozeny-Carman equation and six models based on the integration of complete water retention curves were compared. An empirical model (ROSETTA3) was also included in the comparison. Results of this work show that integral-based models of ks implemented with an exponential water retention function that contains a well-defined discontinuity near saturation (BC) produced better predictions than similar models derived with a sigmoidal water retention function (vG), especially for soils with a relative amount of macropores greater than 5%. None of the selected models predicted ks well for soils with a relative amount of macropores smaller than 5%.
Considering the limitations of the models tested in Chapter 3, ks models derived from percolation theory and critical path analysis were tested in Chapter 4. Models based on percolation theory require knowledge of the critical pore diameter and percolation porosity. Water retention data and three-dimensional (3D) images obtained from each of 169 soil cores collected across Norway were used to investigate methods to estimate critical pore diameter and percolation porosity using either the BC or the vG water retention functions, or combining information from both functions. The results were compared with two existing predictive models based on critical path analysis using information from either 3D images or water retention properties. The model developed by combining information from the two above mentioned water retention functions resulted in the best predictions of ks. However, none of the models developed using water retention data estimated the critical pore diameter well. Further research is needed to improve the estimation of this parameter.