Use of principal components for scaling up topographic models to map soil redistribution and soil organic carbon

Authors: LI, X. - University Of Maryland; McCarty, Gregory

Submitted to: Journal of Visualized Experiments
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/1/2018
Publication Date: 10/16/2018
Citation: Li, X., McCarty, G.W. 2018. Use of principal components for scaling up topographic models to map soil redistribution and soil organic carbon. Journal of Visualized Experiments. 140:e58189. https://doi.org/10.3791/58189.
DOI: https://doi.org/10.3791/58189

Interpretive Summary: Often access to private lands is limited and the need to extrapolate the findings from representative study sites to the larger setting that include private lands can be important. Moreover, data interpolation methods such as kriging cannot be scaled to areas larger than the area physically sampled. In this report, we describe a new approach for generating large scale maps of soil properties based on the use of principal component regression which develop predictive topographic models using lidar-derived topographic metrics. The large scale lidar data permit extraction of the topographic metrics at the watershed or regional scales thus ensuring that the principal components used in the predictive models reflect characteristics of the larger setting being mapped. This increases confidence that the predictive model generated from field scale data is valid at the larger scale. Moreover, use of principal components in the predictive model reduces intercorrelations between variables within the model and further reduces problems with overfitting by the reduction in number of model variables (e.g., dimension reduction). As a result, we demonstrate that the topographic models produced by principal component regression are more robust for producing soil property maps at various scales.

Technical Abstract: Landscape topography is a critical component of soil formation and plays an important role in determining soil properties on the earth surface because it significantly affects the gravity-driven soil movement induced by runoff and tillage activities. The recent large scale application of Light Detection and Ranging (LiDAR) technology holds promise for generating high spatial resolution topographic metrics that can be used in studies of soil property variability in landscapes. In this study, fifteen topographic metrics derived from LiDAR data were selected to investigate topographic impacts on redistribution of soil and spatial distribution of soil organic carbon (SOC). We explored the use of principal components (PCs) for characterizing topography metrics and the stepwise principal component regression (SPCR) to scale-up topography-based soil erosion and SOC models developed at the site scale to the watershed scale. The performance of SPCR models was compared to stepwise ordinary least square regression (SOLSR) models developed with topographic metrics. Result showed that SPCR models were more robust than SOLSR models for predicting soil redistribution rates and SOC density with scale-up. Use of PCs removes potential collinearity between individual input variables, and dimensionality reduction by principal component analysis (PCA) diminishes the risk of overfitting the prediction models. This study proposes a new approach for modeling soil redistribution across various spatial scales. For one application, access to private lands is often limited and the need to extrapolate the findings from representative study sites to the larger setting that include private lands can be important.