Author ORCID Identifier
Document Type
Dissertation
Date of Award
2026
Degree Name
Doctor of Philosophy (PhD)
Department
Sustainability
First Advisor
Meghann Jarchow
Abstract
Climate-smart agricultural practices such as conservation tillage, increased crop residue cover, and cover cropping play a critical role in improving soil health, enhancing climate resilience, and reducing carbon losses from croplands. However, their spatial distribution, long-term impacts on land–atmosphere carbon fluxes, and the ability to monitor them efficiently at regional scales remain insufficiently understood. This dissertation addresses these gaps through an integrated remote sensing, machine learning, and modeling framework applied across eastern South Dakota. Firstly, I developed a satellite-based approach to map tillage practices and detect cover crop presence using multisource spectral indices and environmental drivers. A machine learning classifier trained with in situ observations achieved accuracies exceeding 80%, revealing that only 4% of corn and soybean fields adopted cover crops during the 2022–2023 period. Seasonal precipitation, growing degree days, and soil surface texture emerged as key environmental factors influencing adoption patterns. Consecutively, I evaluated how conservation practices and changing climate conditions together were impacting land carbon dynamics during the spring season. Using remote sensing data combined with Bayesian linear and mixed-effects models, the analysis quantifies the effects of tillage, cover crops, and environmental drivers on gross primary productivity (GPP) and CO₂ emissions from 2020 to 2030. Land surface temperature and soil moisture were dominant controls on both GPP and CO₂ fluxes, while cover crops modestly increased GPP and slightly reduced emissions. Projections under the RCP 4.5 scenario indicate rising CO₂ emissions and declining GPP in northern regions due to reduced soil moisture and decreasing crop suitability. The third chapter advances residue-based soil conservation monitoring by integrating ground imagery with multispectral satellite observations. A ResNet-50 deep learning model trained on field images achieved 94% segmentation accuracy and 75% classification accuracy across residue categories. Satellite-based CRC classification using PlanetScope and Sentinel-2 imagery showed that SWIR-enabled Sentinel-2 provided higher accuracy, with moderate agreement between products. These results demonstrate the feasibility of scalable, automated residue monitoring framework to help in field-level assessments. Collectively, these studies provide a comprehensive framework for detecting conservation practices, evaluating their carbon impacts under climate change, and developing scalable monitoring tools. This work contributes to emerging measurement, reporting, and verification (MRV) systems that support climate-smart agriculture and inform sustainable land management at regional scale in U.S. Midwest.
Subject Categories
Agricultural and Resource Economics
Keywords
Bayesian Modelling Conservation Agriculture Machine learning/Deep learning Remote sensing
Number of Pages
201
Publisher
University of South Dakota
Recommended Citation
Jain, khushboo, "EXAMINING THE CAUSES AND CONSEQUENCES OF DIVERSE AGRICULTURAL MANAGEMENT PRACTICES ON LAND CARBON DYNAMICS IN EASTERN SOUTH DAKOTA" (2026). Dissertations and Theses. 394.
https://red.library.usd.edu/diss-thesis/394