PREDICTIVE MAPPING OF SOIL RESTRICTIVE LAYERS USING TOPOGRAPHIC RELATIONSHIPS AND GEOPHYSICAL TOOLS

Open Access
- Author:
- Vitko, Lauren Frances
- Graduate Program:
- Soil Science
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- October 02, 2017
- Committee Members:
- Patrick Joseph Drohan, Dissertation Advisor/Co-Advisor
Patrick Joseph Drohan, Committee Chair/Co-Chair
Richard Charles Stehouwer, Committee Member
Peter Kleinman, Committee Member
Douglas Alan Miller, Outside Member
James Hamlett, Outside Member
Anthony Buda, Special Member - Keywords:
- soil-landscape modeling
digital soil mapping
ground penetrating radar
electromagnetic induction
soil depletions - Abstract:
- Subsurface restrictive layers play a crucial role in surface runoff and lateral flow processes; however, only coarse maps (scale 1: 1:12,000 - 1:63,360) are currently available through SSURGO (Soil Survey Geographic Database) portraying their depth and distribution in the landscape. High-resolution information on soil restrictive layers is urgently needed for use in hydrologic modeling and land management planning. In this dissertation, I investigate the types and distribution of soil restrictive layers and evaluate three different methods for obtaining high-resolution datasets of soil restrictive layers, including i.) the use of soil morphological features: fragic properties or fragipans (Bx horizons) and low chroma (≤ 2) depletions (LCDs) in soil cores, ii.) ground penetrating radar (GPR), and iii.) electromagnetic induction (EMI). I relate morphological indications of restriction (i.e. LCDs) to measurements of saturated hydraulic conductivity (Ksat) from borehole permeameter measurements and water table levels from shallow wells monitored over 2.5 years. I use particle size analysis to investigate the presence of a lithologic discontinuity observed during soil morphological observations. I use logistic regression and correlation to determine significant relationships between restrictive layer presence/absence and depth (determined using the three methods) and LIDAR-derived topographic variables. Topographic variables investigated include: aspect, slope, planform curvature, profile curvature, total curvature, topographic wetness index, specific contributing area, and topographic position index. The logistic regression models were evaluated based on known soil information from well-studied watersheds in the Ridge and Valley Province of Pennsylvania, wet boot mapping of saturated regions, cross-validation, a validation dataset of 112 soil cores, and comparisons to SSURGO. I found that soils containing LCDs had significantly (p = 0.003) lower Ksat (mean Ksat = 0.018 cm hr-1) than those without LCDs (mean Ksat = 2.8 cm hr-1) and that soils with LCDs had active, shallow water tables during 2.5 years of water table monitoring. I found that 87% and 27% of soil cores that were described as having fragic properties had a lithologic discontinuity at or above the Bx horizon (for FD-36 and Mattern watersheds, respectively). Particle size analysis confirmed the presence of a lithologic discontinuity; clay-free silt percent was significantly (p < 0.05) greater above the lithologic discontinuity, and rock fragment volume percent and clay-free sand percent were significantly greater (p < 0.05) below the discontinuity. Correlations between depth to Bx/LCDs and individual topographic variables (for the FD-36 and Mattern combined dataset) were very low (|r| <0.33; p < 0.05) or insignificant (p > 0.05), which suggests that a simple, linear model predicting depth to a restrictive layer is not possible based on the data collected; this may be due to different amounts and patterns of erosion and deposition between the FD-36 and Mattern watersheds. Ground penetrating radar and EMI surveys were useful for obtaining high-resolution data on soil depth and continuity of soil restrictive layers; however, these methods had limitations for mapping precise boundaries between soils containing restrictive layers and those without soil restrictive layers. Analysis of GPR radar images showed that there were strong radar reflections for fragipans, horizons with fragic properties, high clay subsoils, and BC and CB transitional horizons; however, due to similar patterns in radar reflections, it was difficult to differentiate between the types of horizons/interfaces without corresponding morphological information. Electromagnetic induction was used to predict soil restrictive layers based on the assumption that soil restrictive layers have high apparent conductivity (ECa) because they are wet, have high subsoil clay, and/or have a high bulk density. Although some variation in ECa was explained by soil moisture and soil particle size and rock fragment content (10.5 < r 2 < 54.8); part of the unexplained variation in ECa is probably due to differences in soil depth, which varies from < 0.5 m on sideslopes to > 1.5 m on footslopes. Based on soil information and the spatial extent of surface saturated regions mapped using the wet boot technique, I conclude that ECa data is useful for modeling soil restrictive layers in steeply-sloping watersheds (approximately 25°-45°) like Mattern because they have shallow, coarse-textured soils located in backslope and shoulder landscape positions (generating low ECa values) and deep, high clay/moisture soils in footslopes positions (generating high ECa values). Shallower-sloping watersheds (< 25°) like FD-36, on the other hand, have relatively deep soils (> 1 m) and moderate clay contents (20%-30%) in backslope and shoulder positions, generating relatively high ECa values where restrictive layers do not occur. The best tool of those tested for developing landscape models of soil restrictive layers is the use of LCDs. Validation testing using skill scores showed that the two best models for predicting soil restrictive layers (as mapped using LCDs) were those based on i.) low topographic position index (TPI) and ii.) a combination of low TPI and low slope. Low TPI (low lying areas) is important for describing the distribution of soil restrictive layers because restrictive layers in these watersheds are colluvium-derived. These two models had proportion correct predictions of 0.71 and 0.74, respectively; had substantially higher hit rates compared to SSURGO (hit rates were 0.63 and 0.51, respectively for the two models; and 0.34 for SSURGO); and were able to pinpoint restrictive layers in regions that were mapped as well-drained by SSURGO. The models also intersect with a majority of the surface-saturated regions (> 92%) documented using the wet boot mapping technique. The resulting logistic regression models have applications for nutrient management and hydrologic modeling. Specifically, these models can be used to improve calculations of runoff potential in Pennsylvania’s P-Index or as an input layer in hydrologic models to improve flow routing and predictions of saturated regions. Although the specific models developed in this dissertation are only applicable for the nonglaciated region of the Ridge and Valley Province of the northeastern USA, the methods developed here may be used as framework to model soil restrictive layers in other geomorphic regions.