Skip to main content
ARS Home » Plains Area » El Reno, Oklahoma » Grazinglands Research Laboratory » Forage and Livestock Production Research » Research » Publications at this Location » Publication #351844

Research Project: Bridging Project: Integrated Forage Systems for Food and Energy Production in the Southern Great Plains

Location: Forage and Livestock Production Research

Title: Retrospective tillage differentiation using the landsat-5 TM archive with discriminant analysis

Author
item SHARMA, SONISA - Kansas State University
item DHAKAL, KUNDAN - Oklahoma State University
item Wagle, Pradeep
item KILIC, AYSE - University Of Nebraska

Submitted to: Agrosystems, Geosciences & Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/9/2019
Publication Date: 1/27/2020
Citation: Sharma, S., Dhakal, K., Wagle, P., Kilic, A. 2020. Retrospective tillage differentiation using the landsat-5 TM archive with discriminant analysis. Agrosystems, Geosciences & Environment. 3(1). https://doi.org/10.1002/agg2.20000.
DOI: https://doi.org/10.1002/agg2.20000

Interpretive Summary: Accurate discrimination of tillage practice is important to understand the impacts of crop management on water quality, soil conservation, and soil carbon sequestration. However, collecting tillage practice data on a large scale is not possible. Thus, remote sensing can be a promising monitoring tool for tillage practices over large areas. This study evaluated Quadratic Discriminate Analysis (QDA) model to classify tillage practices using Landsat-5 Thematic Mapper (TM) imagery acquired over five counties in south-central Nebraska. The performance of QDA model was also compared with that of Logistic Regression (LR) model. Based on classification accuracy and kappa value, the QDA-based TM models performed better than LR models to classify tillage practices, showing the potential of QDA models to accurately extract tillage information.

Technical Abstract: Accurate discrimination of tillage practice is important to understand the impacts of crop management on water quality, soil conservation, and soil carbon sequestration. However, collecting such information on a large scale can be cost prohibitive. As a result, remote sensing is a promising monitoring tool for rapid assessment of tillage practice over large areas. Current models to map tillage practices do not adequately determine known classes and also do not identify what class an observation belongs to, based on knowledge of the quantitative variables. Therefore, our objective was to evaluate and apply Quadratic Discriminate Analysis (QDA) model to accurately discriminate tillage practices and compare the performance of QDA model with Logistic Regression (LR) using Landsat-5 Thematic Mapper (TM) imagery acquired over five counties in south-central Nebraska for the month of June from 2006 to 2011 (except 2009). Ground truth data were obtained from the United States Department of Agriculture Natural Resources Conservation Service (USDA-NRCS) at 48 locations [20 conventional till (CT) and 28 conservation tillage (NT)]. Classification accuracies were obtained for the QDA-based TM bands and Normalized Difference Tillage Index (NDTI) models and their performances were further assessed by comparing the results obtained from LR models. Based on classification accuracy and kappa value, results showed that the QDA-based TM models performed better to classify tillage practices than the LR-based TM models. However, both the QDA and LR-based NDTI models performed similarly in discriminating tillage practices. This study indicated that QDA-based Landsat-5 TM models can be more effective to accurately extract tillage information over large areas.