Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Hydrology and Remote Sensing Laboratory » Research » Publications at this Location » Publication #382069

Research Project: Improving Agroecosystem Services by Measuring, Modeling, and Assessing Conservation Practices

Location: Hydrology and Remote Sensing Laboratory

Title: Spatial extrapolation of topographic models for mapping soil organic carbon using local samples

item DU, L. - US Department Of Agriculture (USDA)
item McCarty, Gregory
item LI, X. - Fuzhou University
item RABENHORST, M. - University Of Maryland
item WANG, Q. - University Of Maryland
item LEE, S. - University Of Seoul
item HINSON, A. - US Department Of Agriculture (USDA)
item ZOU, Z. - University Of Maryland

Submitted to: Geoderma
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/5/2021
Publication Date: 6/19/2021
Citation: Du, L., McCarty, G.W., Li, X., Rabenhorst, M., Wang, Q.L., Lee, S., Hinson, A., Zou, Z. 2021. Spatial extrapolation of topographic models for mapping soil organic carbon using local samples. Geoderma. 404:115290.

Interpretive Summary: Digital mapping of soil properties is an important emerging technology for improved management of agricultural landscapes. We investigated the potential for using a regional soil carbon prediction model based solely on readily obtained high resolution elevation data to generate a local map of soil carbon on a cropland field within the region. For this application, we updated the regional model with information from a small number of samples collected on the cropland field to generate a locally calibrated map of soil carbon. We found that an accurate local map of soil carbon could be generated without intensive soil sample collection on the cropland field of interest. This approach for generating local prediction models promises to greatly improve the efficiency of soil carbon mapping in agricultural landscapes and will facilitate improved management of agricultural landscapes.

Technical Abstract: Spatial extrapolation of soil organic carbon (SOC) mapping models provides key input for hydrological, agricultural, and ecological models and is of great importance for local agricultural land management and watershed processes assessment. Terrain attributes derived from elevation have been commonly used predictor variables in SOC mapping, due to the strong topographic control on local SOC distribution. The aim of this study is to extrapolate a regional topographic model using additional local samples to map SOC density (0-30cm) and evaluate the effects of different sample sizes and sampling schemes on model extrapolation. The regional model (TRF) was developed based on 3-m light detection and ranging (lidar)-derived topographic covariates using random forest regression at a cropland site with intensive SOC measurements. This model was then applied at another cropland site by integrating a residual kriging procedure to capture the local variance using local samples. Local samples were repeatedly selected (500 times) with different number of local samples (10-225) using two different sampling schemes: stratified random sampling (SRS) and spatial coverage sampling with close-pairs (SCS+). By analyzing the distribution of relative mean error (RME) of the overall map and all individual samples over repeated selection of samples, we found a narrower distribution of RME close to zero with larger number of local samples. The absolute RME estimates of individual samples were below 20% when the local sample number was roughly >50. There was a smaller variation in RME using SCS+ than that using SRS at the overall map level. This study demonstrates the potential of mapping soil carbon at fine scales by combining lidar-derived topographic models and a small number of local samples and suggests SCS+ is a more effective sampling scheme for unvisited areas.