Texture-Naïve Soil Classification and Decision Tree-Based Measurement Selection for Efficient Hydrologic Modeling and Data Collection

Abstract

Soil texture classifications are well-established and well-known tools that have been adopted for defining and visualizing soils spatially through creation of soil maps. The soil texture classifications and likewise soil maps are used in applications from geology, agriculture, hydrology, and atmospheric sciences. Despite having a long history and use, soil texture classifications have gone relatively unchanged. It has just been in recent years that alternative soil texture classifications have been developed with a focus on hydrology. Soil texture classifications are often applied to situations where soil hydraulic behavior is of importance despite being based on physical soil characteristics rather than actual hydrologic response, which is not easy to determine from soil texture alone. While soil texture is rarely classified based on hydrologic response, it is even more rare that soil maps are visualized using hydrologic response despite their widespread use. To improve visualization and expand the adoption of alternative soil texture classifications, this dissertation develops and evaluates methodologies for creating hydrologic-process-based soil texture classifications based on fundamental hydrologic scenarios. We demonstrate that viewing soils spatially in context of hydrologic response improves interpretation and more accurate application of soil texture. This dissertation also improves the development of hydrologic-process-based soil texture classifications by exploring their uncertainty and optimizing the hydrologic response classification process. Hydrologic responses were grouped more effectively into a smaller number of classes having more similar behavior than when grouped based on physical soil properties. The analysis also showed that uncertainty in the simulated hydrologic responses used to create the classification is directly related to soil hydraulic parameter uncertainty corresponding to each soil texture. The dissertation concludes by shifting the focus to development of a cost- effective decision tree algorithm for the selection of hydrologic measurements. The developed cost-weighted decision tree algorithm identifies water content and pressure measurements for predicting soil hydraulic flux that maintains predictive accuracy within 10% of traditional decision trees while reducing costs by at least 50%. This makes characterization of soil texture hydrologic responses more affordable. Collectively, these developments highlight the importance of soil texture hydrologic responses across any application. Soil maps are based on hydrologic response allowing for the soil landscape to be visualized more effectively in context of specific conditions than compared to traditional soil texture classifications. Optimization of hydrologic-process-based soil texture classifications refines our conceptual understanding and allows for more appropriate application of soil texture by representing only the necessary hydrologic responses and defining the corresponding uncertainty. With these developments, the door has been opened for exploration of new and diverse applications for hydrologic-process-based soil texture-based classifications.