|Evett, Steven - Steve|
|Chavez Eguez, Jose|
Submitted to: International Journal of Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/5/2007
Publication Date: 6/1/2008
Citation: Gowda, P., Howell, T.A., Evett, S.R., Chavez Eguez, J.L., New, L. 2008. Remote sensing of contrasting tillage practices in the Texas Panhandle. International Journal of Remote Sensing. 29(12):3477-3487.
Interpretive Summary: Collecting tillage information manually at individual fields on a regional scale for environmental modeling purposes can be time-consuming and labor intensive. Remote sensing techniques show promise in providing such spatial data in a time and cost-effective manner. In this study, we developed and evaluated a set of Landsat TM based logistic regression models for their ability to identify tillage practices and evaluated with independent data in the Texas Panhandle. Results indicate that models that have indices derived from the combination of TM band 5 with bands 4 and 6 may provide consistent and acceptably accurate results when they are applied in the same geographic region. Further evaluation of these models is needed to determine their applicability in other geographic regions.
Technical Abstract: Tillage information is crucial in environmental modeling as it has a direct impact on water holding capacity, evapotranspiration, carbon sequestration, and soil and nutrient losses due to wind and water erosion of agricultural soils. A remote sensing approach is promising for the rapid collection of tillage information on individual fields over large areas in a time and cost-effective manner. In this study, a set of Landsat Thematic Mapper (TM) based linear logistic models were developed and verified with independent tillage data. For data collection purposes, 35 and 41 commercial fields were randomly selected in Moore and Ochiltree counties, respectively, in the Texas Panhandle. Ground-truth data including geographic location, tillage, soil organic carbon and water content were measured during the 2005 planting season. The ground-truth survey was planned to coincide with Landsat 5 Satellite overpasses and two TM scenes were acquired. Using the Moore County dataset, seven logistic regression models were developed and these were evaluated with the data collected from Ochiltree County. The overall classification accuracy of the seven developed models with the Moore County dataset varied from 86-91% during the calibration with the Moore County dataset. These models were evaluated against independent Ochiltree County dataset resulted in somewhat less accurate (classification accuracy of 67-85%) but still useful results. Analysis of these results indicate that logistic regression models that have indices derived from the combination of TM band 5 with bands 4 and 6 may provide consistent and acceptably accurate results when they are applied in the same geographic region. Overall, logistic regression models were found easy to use, cost and time effective, and produced reasonably accurate tillage classification results.