Remote Sensing (Apr 2020)

Assessing Soil Cover Levels during the Non-Growing Season Using Multitemporal Satellite Imagery and Spectral Unmixing Techniques

  • Ahmed Laamrani,
  • Pamela Joosse,
  • Heather McNairn,
  • Aaron A. Berg,
  • Jennifer Hagerman,
  • Kathryn Powell,
  • Mark Berry

DOI
https://doi.org/10.3390/rs12091397
Journal volume & issue
Vol. 12, no. 9
p. 1397

Abstract

Read online

Growing cover or winter crops and retaining crop residue on agricultural lands are considered beneficial management practices to address soil health and water quality. Remote sensing is a valuable tool to assess and map crop residue cover and cover crops. The objective of this study is to evaluate the performance of linear spectral unmixing for estimating soil cover in the non-growing season (November–May) over the Canadian Lake Erie Basin using seasonal multitemporal satellite imagery. Soil cover ground measurements and multispectral Landsat-8 imagery were acquired for two areas throughout the 2015–2016 non-growing season. Vertical soil cover photos were collected from up to 40 residue and 30 cover crop fields for each area (e.g., Elgin and Essex sites) when harvest, cloud, and snow conditions permitted. Images and data were reviewed and compiled to represent a complete coverage of the basin for three time periods (post-harvest, pre-planting, and post-planting). The correlations between field measured and satellite imagery estimated soil covers (e.g., residue and green) were evaluated by coefficient of determination (R2) and root mean square error (RMSE). Overall, spectral unmixing of satellite imagery is well suited for estimating soil cover in the non-growing season. Spectral unmixing using three-endmembers (i.e., corn residue-soil-green cover; soybean residue-soil-green cover) showed higher correlations with field measured soil cover than spectral unmixing using two- or four-endmembers. For the nine non-growing season images analyzed, the residue and green cover fractions derived from linear spectral unmixing using corn residue-soil-green cover endmembers were highly correlated with the field-measured data (mean R2 of 0.70 and 0.86, respectively). The results of this study support the use of remote sensing and spectral unmixing techniques for monitoring performance metrics for government initiatives, such as the Canada-Ontario Lake Erie Action Plan, and as input for sustainability indicators that both require knowledge about non-growing season land management over a large area.

Keywords