Document Type

Thesis - Open Access

Award Date

2024

Degree Name

Master of Science (MS)

Department / School

Agronomy, Horticulture, and Plant Science

First Advisor

Kristopher Osterloh

Abstract

Saline and sodic soils are an increasing concern across the Northern Great Plains (NGP) due to factors of climate change and land management that are drawing geologically derived salts to the land surface. Traditional laboratory assessments, such as electrical conductivity (EC) and sodium adsorption ratio (SAR), are time and resource consumptive. Portable X-ray fluorescence (PXRF) may be a viable proximal sensing alternative, as it is able to provide elemental data in minutes, in situ or ex situ, and can directly quantify salinity-associated elements like Ca, Mg, and S. PXRF paired with predictive models has proven useful for a range of soil applications, including salinity. This study assessed EC from a 1:1 soil-to-water ratio (EC1:1) and SAR predictive models developed with elemental data from PXRF scans on NGP glacial till and glaciolacustrine soils. Models with all PXRF elements and a subset of Ca, Mg, S, and Zr were developed through multiple linear regression (MLR), random forest (RF), and cubist machine learning algorithms. Each was trained with 80% of data through k-fold cross-validation. Models were then validated on the remaining 20%. Cubist models with Ca, Mg, S, and Zr from ex-situ PXRF scans have the highest prediction accuracy for both EC1:1 and SAR (R2 = 0.83 and 0.82, respectively). In summary, ex situ PXRF models are promising alternatives to lab-based EC1:1 and SAR measurement for NGP glacial soils.

Library of Congress Subject Headings

Soils, Salts in -- Great Plains.
Soils -- Great Plains -- Analysis.
Soil management -- Great Plains.
X-ray spectroscopy.
Fluorescence spectroscopy.
Salinity.
Sodic soils.
Machine learning.

Publisher

South Dakota State University

Download

Share

COinS
 

Rights Statement

In Copyright