Name: use of principal components for scaling up topographic models to map soil redistribution and soil organic carbon
Uploaded: 2018-10-16T14:00:00
Duration: 9 min 44 s
Description: Watch this Scientific Journal Video about Use of Principal Components for Scaling Up Topographic Models to Map Soil Redistribution and Soil Organic Carbon at JoVE.com

This research was supported by the USDA Natural Resources Conservation Service in association with the Wetland Component of the National Conservation Effects Assessment Project (NRCS 67-3A75-13-177).

1. Topographic Analyses
<ol>
	<li>Digital data preprocess
	<ol>
		<li>Collect LiDAR data from the GeoTREE LiDAR mapping project website. Select &#34;boundary type&#34; and &#34;region&#34; to zoom into a specific area. Draw a polygon to download LiDAR tiles for the selected study area.</li>
		<li>Convert the raw LiDAR data to a LAS file using the geographic information system (GIS) mapping tool.</li>
		<li>Generate DEMs with a 3-m spatial resolution using inverse distance weighted interpolation.</li>
		<li>Filter the 3-m DEMs twice with a 3-kernel low pass filter to reduce noises associate with local variation.</li>
	</ol>
	</li>
	<li>Topographic metric generation
	<ol>
		<li>To generate topographic metrics, first download the latest version of the System for Automated Geoscientific Analysis (SAGA)23. Click &#34;Import Raster&#34; in the Import/Export section to import the filtered 3-m DEMs into SAGA.</li>
		<li>Click the &#34;Slope, Aspect, Curvature&#34; module of SAGA with the default settings to generate the slope and curvature-related [profile curvature (P_Cur), plan curvature (Pl_Cur), and general curvature (G_Cur)] metrics using the filtered DEMs (Figure 1).</li>
		<li>Click the &#34;Flow Accumulation (Top-Down)&#34; module of SAGA and select &#34;Deterministic infinity&#34; as the method to generate flow accumulation (FA) metric using the filtered DEMs.</li>
		<li>Click the &#34;SAGA Topographic Openness&#34; module with the default settings to generate the positive openness (POP) metric using a filtered z-axis amplified image.</li>
		<li>Click the &#34;LS- factor (Field Based)&#34; module of SAGA with the default settings to generate the upslope slope (Upsl) and slope length factor (LS_FB) metrics using the filtered DEMs.</li>
		<li>Click the &#34;Flow Path Length&#34; module of SAGA with the default settings to generate the flow path length (FPL) metric using the filtered DEMs.</li>
		<li>Click the &#34;Downslope Distance Gradient&#34; module of SAGA with the default settings to generate the downslope index (DI) metric using the filtered DEMs.</li>
		<li>Click the &#34;SAGA Wetness Index&#34; module and select &#34;absolute catchment area&#34; as the Type of Area to generate the catchment area (CA) and topographic wetness index (TWI) metrics using the filtered DEMs.</li>
		<li>Click the &#34;Stream Power Index&#34; module of SAGA and select &#34;pseudo specific catchment area&#34; as the Area Conversion to generate the stream power index (SPI) metric using the filtered DEMs.</li>
		<li>Generate maximum elevation maps with multiple radiuses. Filter the maximum elevation maps twice through a 3-kernel low pass filter. Subtract the filtered 3-m DEM from the filtered maximum elevation maps to obtain a series of relief maps. Extract a series of relief variables to a number of locations.</li>
		<li>Perform principal component analysis (PCA) on the relief variables to convert the reliefs into topographic relief components. Select principal components that explain more than 90% variance of the relief dataset as the topographic relief metrics.</li>
	</ol>
	</li>
</ol>
2. Field Data Collection
<ol>
	<li>Field sampling
	<ol>
		<li>Select a number of cropland field locations that can adequately represent the landscape characteristics of the study area and several representative small-scale cropland fields that can be intensively sampled. 
		NOTE: The soil samples collected from the two cropland fields were used for model calibration. Soil samples collected from the entire study area were used for model validation.</li>
		<li>Upload all the sample location coordinates to a code-based geographic positioning system (GPS) and physically locate them in the fields.</li>
		<li>Collect 3 samples for each sampling location from the top 30 cm soil layer using a push probe (3.2 cm in diameter). 
		&#8203;NOTE: Soil samples from 30-50 cm layers were collected at sites where sediment deposition was expected. The volume of each sample was 241 cm3.</li>
		<li>Record geographic coordinate information of sampling locations using GPS.</li>
		<li>Weigh the soil samples after drying them at 90 &#176;C for 48 h. Calculate soil density using information of total sample volumes at sampling locations and weights. Mix the three samples from the same location to get a composite soil sample.</li>
	</ol>
	</li>
	<li>Soil sample preparation
	<ol>
		<li>Sieve the composite soil samples with a 2-mm screen.</li>
		<li>Grind a 10 g subsample of the sieved soil to a very fine powder with a roller mill.</li>
	</ol>
	</li>
	<li>Soil sample analyses
	<ol>
		<li>Measure soil total carbon (C) content in roller milled samples through combustion on a CN elemental analyzer at a temperature of 1350 &#176;C. Estimate calcium carbonate C content by analyzing the remaining C after baking soil organic matter at a temperature of 420 &#176;C for 16 h in a furnace.</li>
		<li>Calculate SOC content (%) by subtracting calcium carbonate C content from total soil C content. Convert SOC content (%) to SOC density (kg m-2) using soil density.</li>
		<li>Put the bulk 2-mm sieved soil samples in Marinelli beakers and seal them. Measure 137Cs concentration of each sample through gamma-ray analysis using a spectroscopy system that receives inputs from three high purity coaxial germanium crystals (HpCN30% efficiency) into 8192-channel analyzers (see Table of Materials).</li>
		<li>Calibrate the system using an analytic mixed radionuclide standard11. Convert 137Cs concentration to 137Cs inventory using soil density.</li>
		<li>Calculate soil redistribution rate using 137Cs inventory by applying the Mass Balance Model II (MBMII) in a spreadsheet add-in program developed by Walling et al.24.</li>
	</ol>
	</li>
</ol>
3. Topography-Based Model Development
<ol>
	<li>Topographic principal component estimation
	<ol>
		<li>Extract the topographic metrics for sampling locations in the entire study area and the small-scale cropland fields.</li>
		<li>Standardize the topographic metrics of the sampling locations in the entire study area by using mean and standard deviation. Estimate the topographic metric loadings in each component based on the standardized topographic metrics using PCA with statistical software package. Collect the topographic metric loadings in each topographic principal component (TPC) and select the top TPCs that explain 90% variance of all metrics.</li>
		<li>Standardize the topographic metrics of the sampling locations in the small-scale cropland fields. Calculate the top TPCs for each location by sum of the standardized topographic metrics weighted by the corresponding loadings from the sampling locations in WCW.</li>
	</ol>
	</li>
	<li>Model calibration
	<ol>
		<li>Perform stepwise ordinary least square regression (SOLSR) to develop topography-based SOLSRf models for SOC density and soil redistribution rates based on all topographic metrics at the small-scale cropland fields. Use Akaike information criterion (AIC) and leave-one-out cross-validation to select the optimal combination of topographic metrics for the best-fitted SOLSRf models.</li>
		<li>Check the collinearity among the topographic variables using the variance inflation factor (VIF). Remove the variables with the largest VIF (VIF &#8805; 7.525), and check VIF again. Remove the variables until the VIFs of all variables are &#60; 7.5. Perform SOLSR to develop topography-based SOLSRr models for SOC density and soil redistribution rates based on topographic metrics that were removed high collinearity variables. Use the AIC and leave-one-out cross-validation to select the optimal combination for the best-fitted SOLSRr models.</li>
		<li>Perform stepwise principal component regression (SPCR) to develop topography-based SPCR models for SOC density and soil redistribution rates based on the TPCs at the small-scale cropland fields. Use the AIC and leave-one-out cross-validation to select the optimal combination of TPCs for the best-fitted SPCR models.</li>
		<li>Calculate the adjusted coefficient of determination (Radj2), Nash-Sutcliffe efficiency (NSE), and ratio of the root mean square error to the standard deviation of measured data (RSR) to assess model efficiencies.</li>
	</ol>
	</li>
	<li>Model evaluation
	<ol>
		<li>Estimate SOC density and soil redistribution rates in the entire study area by applying the estimated models.</li>
		<li>Validate the developed model by comparing prediction with measured dataset of SOC density and soil redistribution rates in the entire study area. Evaluate the model performances using Radj2, NSE, and RSR values.</li>
	</ol>
	</li>
</ol>

The authors have nothing to disclose.

The SOLSRf models had slightly better performances than the SPCR models in calibration at the field scale. However, some of the topographic metrics, such as SPI and CA (r &#62; 0.80), are closely correlated with each other. The collinearity may add uncertainties to model predictions. Because of the multicollinearity among predictors, small changes in the input variables can significantly affect the model predictions41. Therefore, the SOLSRf models tended to be unstable and showed low efficiencies in simulations of SOC density and soil redistribution rate at the watershed scale. The SPCR models substantially outperformed the SOLSRf models in prediction of SOC distribution at the watershed scale. TPCs eliminate the multicollinearity by converting the fifteen topographic metrics into mutually independent (orthogonal) components. The conversion also uncovered underlying relationships among topographic metrics. As indicated by the high loadings (&#62; 0.35) of topographic metrics to the components, the TPC1, TPC2, TPC3, TPC6, and TPC7 were associated with runoff velocity, soil water content, runoff volume, flow divergence, and flow acceleration, respectively. Spatial patterns of soil redistribution rates and SOC distribution were highly correlated with soil water content and runoff divergence in the WCW, which is consistent with the study of Fox and Papanicolaou2, which demonstrated that eroded soil from upland could be impacted by flow divergence in a low-relief agricultural watershed.
Moreover, fewer predictor variables in the SPCR models than the SOLSRf and SOLSRr models reduced the risk of over-fitting the prediction models42,43. There were more than six variables in all the SOLSR models, which may increase the difficulty of data interpretation and induce high variance in model simulations41,44,45. This may account for the lower prediction efficiencies in WCW by the SOLSR models than by the SPCR models.
Topography- based SPCR models have advantages in simulating soil redistribution and associated SOC dynamics. First, topographic information can be easily derived from DEMs. Recent increased accessibility of the high spatial resolution LiDAR data can help improve the accuracy of DEM-derived landscape topography and benefit investigations in regions with limited field observations. Second, using a set of topographic metrics and statistical analyses, the topography-based models can efficiently quantify soil redistribution and SOC distribution patterns. Third, the application of principal component can effectively reduce biases associated with multicollinearity of topographic metrics and increase the stability of the stepwise regression models when applied to multiple spatial scales.
However, the SPCA models may be limited by variables during model development. Although application of the LiDAR data increased in ecological studies, the methods to derive useful topographic information have not yet been fully explored. In this study, the TWI and LsRe showed the highest correlations with SOC density and soil redistribution rates, respectively. However, additional topographic variables that are not considered may be equally or more important in explaining soil erosion and C dynamics. Additionally, other factors such as management practices, which may cause soil erosion variability, were not included in this study. For example, when tillage was parallel to the direction of maximum slope, soil erosion may double relative to the erosion in slantwise tillage turning soil upslope46. Therefore, different tillage practices may also be a reason for the reduced prediction efficiencies of the SPCR models.
The study is based on the paper published in Catena17. Instead of a mechanistic-based analysis of topographic influences on soil movement and soil properties as performed in the Catena paper, here we focused on the methods for quantifying topographic metrics and developing topography-based models. We discussed the feasibility and advantages of using topography-based models in studies of the spatial structure of soil properties. Meanwhile, we improved our models by updating algorithms of slope length factor and flow accumulation. The scale of slope length factor measurement was limited to field&#39;s area. Additionally, the deterministic infinity algorithm was used for flow accumulation generation. Compared with the method reported in Li et al.17 which generated flow accumulation with a deterministic eight-node algorithm, the infinity algorithm adopted in this study reduces loops in the flow direction angles and proved to be a better algorithm for low relief areas47.
In conclusion, our results demonstrate the feasibility of topography-based SPCR models in simulating SOC distribution and soil redistribution patterns in agriculture fields. As a cost-effective method to estimate SOC stocks and soil redistribution rates, it is applicable to sites with limited observational data and private lands lacking public access. In future studies, the prediction models could be improved with further refinement and availability of LiDAR data and inclusion of additional topographic metrics. The large-scale soil property maps that were developed based on the models will lead to further understanding of the mechanisms underlying the topographic impacts on soil movement in agricultural landscapes and the fate of SOC at the watershed and regional scales.

Soil redistribution (erosion and deposition) exerts significant impacts on soil organic carbon (SOC) stocks and dynamics. Increasing efforts have been devoted to investigating how SOC is detached, transported, and deposited over the landscape1,2,3. Carbon (C) sequestration and SOC distribution are influenced by gravity-driven soil movement induced by water erosion4,5,6. In cultivated fields, soil translocation by tillage is another important process contributing to C redistribution7,8,9. Tillage erosion causes a considerable net downslope movement of soil particles and leads to a within-field soil variation10. Both water and tillage erosion are significantly affected by landscape topography, which determines the locations of erosional and depositional sites11. Therefore, effective soil erosion regulation and C dynamic investigation in agricultural lands calls for a better understanding of topographic controls on soil erosion and movements.
Several studies have investigated the impacts of topography on soil redistribution and associated SOC dynamics9,12,13,14,15,16,17. Van der Perk et al.12 reported that topographic factors explained 43% of variability in soil redistribution. Rezaei and Gilkes13 found higher SOC in soils on a shady aspect, due to lower temperatures and less evaporation when compared to other aspects in rangelands. Topography may have more significant impacts on soil redistribution in agricultural lands with traditional tillage treatment than those with minimum tillage, due to the interactions between landforms and tillage practices9. However, these findings were primarily derived from field observations, which present difficulties in investigating soil properties at a broader spatial scale. There is a pressing need to develop new strategies to effectively understand spatial patterns of soil properties at watershed and regional scales.
The objective of this study is to develop efficient models to simulate soil redistribution and SOC distribution. Topography-based models using topographic metrics as predictors have been developed to quantify soil erosion and deposition processes. Compared with empirical- or process-based erosion models that employed discrete field samplings to simulate soil erosion18,19, topography-based models could be developed based on topographic information derived from digital elevation models (DEMs) with high resolutions. This approach allows for continuous soil property simulations at the watershed or regional scale. In the past several decades, accuracy of topographic information has substantially improved, with increasing availability of high resolution remotely sensed data. Although previous studies have employed topography-based models to simulate soil properties12,20,21,22, most of these investigations used a single topographic metric or single category of topographic metrics (local, non-local, or combined topographic metrics), which may not have sufficiently explored topographic impacts on soil microbial activity. Therefore, to gain a better understanding of topography controls on soil erosion and C dynamics, we examined a comprehensive set of topographic metrics including local, non-local, and combined topographic metrics and developed multi-variable topography-based models to simulate soil property dynamics. Applications of these models are expected to provide scientific support for better soil erosion control and agricultural land management.
Topographic metrics are generally categorized into one of three categories: a) local topographic metrics, b) non-local topographic metrics, or c) combined topographic metrics. Local topographic metrics refer to local features of one point on the land surface. Non-local topographic metrics refer to the relative locations of selected points. Combined topographic metrics integrate local and non-local topographic metrics. A set of topographic metrics affecting soil erosion and deposition were used in this study to investigate the topographic controls on soil movement and C stocks (Table 1). Specifically, we used four local topographic metrics [slope, profile curvature (P_Cur), plan curvature (Pl_Cur), general curvature (G_Cur)], seven non-local topographic metrics [flow accumulation (FA), topographic relief, positive openness (POP), upslope slope (UpSl), flow path length (FPL), downslope index (DI), catchment area (CA)], and three combined topographic metrics [topographic wetness index (TWI), stream power index (SPI), and slope length factor (LS)].

<table border="0" cellpadding="0" cellspacing="0" style="width:1151px;" width="1150">
	<tbody>
		<tr height="21">
			<td>Light Detection and Ranging (LiDAR) data&#160;</td>
			<td style="width:233px;"></td>
			<td style="width:256px;">http://www.geotree.uni.edu/lidar/</td>
			<td style="width:355px;">Collected from the GeoTREE LiDAR mapping project&#160;</td>
		</tr>
		<tr height="21">
			<td height="21" style="height:21px;">LECO CNS 2000 elemental analyzer&#160;</td>
			<td>LECO Corp., St. Joseph, MI</td>
			<td style="width:256px;"></td>
			<td></td>
		</tr>
		<tr height="21">
			<td height="21" style="height:21px;">Canberra Genie-2000 Spectroscopy System</td>
			<td>CANBERRA Industries</td>
			<td style="width:256px;"></td>
			<td></td>
		</tr>
		<tr height="21">
			<td height="21" style="height:21px;width:307px;">Geographic positioning system</td>
			<td>Trimble&#160;</td>
			<td style="width:256px;">RTK 4700 GPS</td>
			<td></td>
		</tr>
		<tr height="21">
			<td height="21" style="height:21px;">ArcGIS</td>
			<td>ESRI, Redlands, CA</td>
			<td style="width:256px;">10.2.2</td>
			<td></td>
		</tr>
		<tr height="21">
			<td height="21" style="height:21px;width:307px;">Statistical Analysis System&#160;</td>
			<td>SAS Institute Inc</td>
			<td style="width:256px;"></td>
			<td></td>
		</tr>
		<tr height="23">
			<td height="23" style="height:23px;">System for Automated Geoscientific Analysis&#160;</td>
			<td>University of G&#246;ttingen, Germany&#160;</td>
			<td>v. 2.2.5, http://www.saga-gis.org/</td>
			<td>GNU General Public License</td>
		</tr>
	</tbody>
</table>

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

Landscape topography is a critical factor affecting soil formation and plays an important role in determining soil properties on the earth surface, as it regulates the gravity-driven soil movement induced by runoff and tillage activities. The recent application of Light Detection and Ranging (LiDAR) data holds promise for generating high spatial resolution topographic metrics that can be used to investigate soil property variability. In this study, fifteen topographic metrics derived from LiDAR data were used to investigate topographic impacts on redistribution of soil and spatial distribution of soil organic carbon (SOC). Specifically, we explored the use of topographic principal components (TPCs) for characterizing topography metrics and stepwise principal component regression (SPCR) to develop topography-based soil erosion and SOC models at site and watershed scales. Performance of SPCR models was evaluated against stepwise ordinary least square regression (SOLSR) models. Results showed that SPCR models outperformed SOLSR models in predicting soil redistribution rates and SOC density at different spatial scales. Use of TPCs 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 findings from representative study sites to larger settings that include private lands can be important.

Watch this Scientific Journal Video about Use of Principal Components for Scaling Up Topographic Models to Map Soil Redistribution and Soil Organic Carbon at JoVE.com

Use of Principal Components for Scaling Up Topographic Models to Map Soil Redistribution and Soil Organic Carbon

Landscape processes are critical components of soil formation and play important roles in determining soil properties and spatial structure in landscapes. We propose a new approach using stepwise principal component regression to predict soil redistribution and soil organic carbon across various spatial scales.

Landscape topography is a critical factor affecting soil formation and plays an important role in determining soil properties on the earth surface, as it ...

Research

JoVE Journal

Environment

Design and Operation of a Continuous 13C and 15N Labeling Chamber for Uniform or Differential, Metabolic and Structural, Plant Isotope Labeling

Design and Operation of a Continuous 13C and 15N Labeling Chamber for Uniform or Differential, Metabolic and Structural, Plant Isotope Labeling

Tracing rare stable isotopes from plant material through the ecosystem provides the most sensitive information about ecosystem processes; from CO2 fluxes ...

A Protocol for Conducting Rainfall Simulation to Study Soil Runoff

Rainfall is a driving force for the transport of environmental contaminants from agricultural soils to surficial water bodies via surface runoff. The ...

Assessment of Labile Organic Carbon in Soil Using Sequential Fumigation Incubation Procedures

Management practices and environmental changes can alter soil nutrient and carbon cycling. Soil labile organic carbon, a readily decomposable C pool, is ...

A Gusseted Thermogradient Table to Control Soil Temperatures for Evaluating Plant Growth and Monitoring Soil Processes

Thermogradient tables were first developed in the 1950s primarily to test seed germination over a range of temperatures simultaneously without using a ...

Protocols for Quantifying Transferable Pesticide Residues in Turfgrass Systems

Plant canopies in established turfgrass systems can intercept an appreciable amount of sprayed pesticides, which can be transferred through various routes ...

Nanopore DNA Sequencing for Metagenomic Soil Analysis

This article describes the steps for construction of a DNA library from soil, preparation and use of the nanopore flow cell, and analysis of the DNA ...

Measuring and Mapping Patterns of Soil Erosion and Deposition Related to Soil Carbonate Concentrations Under Agricultural Management

Spatial patterns of soil erosion and deposition can be inferred from differences in ground elevation mapped at appropriate time increments. Such changes ...

Improving Infrared Spectroscopy Characterization of Soil Organic Matter with Spectral Subtractions

Soil organic matter (SOM) underlies numerous soil processes and functions. Fourier transform infrared (FTIR) spectroscopy detects infrared-active organic ...

Single-throughput Complementary High-resolution Analytical Techniques for Characterizing Complex Natural Organic Matter Mixtures

Natural organic matter (NOM) is composed of a highly complex mixture of thousands of organic compounds which, historically, proved difficult to ...

Microplot Design and Plant and Soil Sample Preparation for 15Nitrogen Analysis

Microplot Design and Plant and Soil Sample Preparation for 15Nitrogen Analysis

Many&#160;nitrogen fertilizer studies evaluate the overall effect of a treatment on end-of-season measurements such as grain yield or cumulative N losses. ...