Evaluating metrics derived from Landsat 8 OLI imagery to map crop cover

Abstract

Developing techniques are required to generate agricultural land cover maps to monitor agricultural fields. Landsat 8 Operational Land Imager (OLI) offers reflectance data over the visible to shortwave-infrared range. OLI offers several advantages, such as adequate spatial and spectral resolution, and 16 day repeat coverage, furthermore, spectral indices derived from Landsat 8 OLI possess great potential for evaluating the status of vegetation. Additionally, classification algorithms are essential for generating accurate maps. Recently, multi-Grained Cascade Forest, which is also called deep forest, was proposed, and it was shown to give highly competitive performance for classification. However, the ability of this algorithm to generate crop maps with satellite data had not yet been evaluated. In this study, the reflectance at 7 bands and 57 spectral indices calculated from Landsat 8 OLI data were evaluated for its potential for crop type identification.

Keywords:

• Crop
• deep forest
• Landsat 8
• random forests
• reflectance
• spectral indices

Acknowledgements

The authors would like to thank Tokachi Nosai for providing the field data. We would like to thank Golden (www.goldenenglishediting.com) for editing and proofreading this manuscript.