The high spectral resolution of much satellite imagery has great potential for detailed soil mapping. However, satellite imagery often has low spatial resolution (i.e. 30 m), reducing its utility except for regional inventories. Digital aerial photographs that have been orthorectified (digital orthophoto quads (DOQ)) often have spatial resolution in the range of 1-5 m and can be inexpensive to acquire. DOQs are generally used for qualitative visual interpretation of photo tone and have had limited applications in digital image analysis due to their low spectral variability (usually panchromatic black and white). Our research examines the merging of multiple remote sensing imageries (Landsat TM, DOQ and IKONOS) and digital elevation model (DEM) data to determine optimal combinations for mapping soil drainage classes on bare soil surfaces. Spatial resolutions of these data are 30 m for Landsat TM, 4 m for IKONOS, 0.6 m for the DOQ, and 10 m for the DEM. Overall classification accuracy compared to field-verified samples was 73% using a combination of the DOQ and Landsat TM. The addition of IKONOS imagery did not provide any significant accuracy improvement. Our results demonstrate that DOQ data merged with Landsat TM and a DEM significantly improve our ability to predict spatial patterns of soil drainage classes using image classification techniques when compared to the original soil survey of the study site. For reference, the published soil survey (1:15,840 scale) had an accuracy of 55% for soil drainage class when compared to our field-verified samples. While bare soil surfaces may be limited in many areas, they do provide opportunities where rapid, automated classification techniques are possible and provide useful information for understanding and documenting soil-landscape relationships and spatial variability.
Bibliographical noteFunding Information:
We thank Alfred Bleck, the landowner, for graciously allowing access to this research site. We acknowledge Dr. Marvin Bauer (University of Minnesota) for sharing the Landsat TM data used for this research project, Dr. Stacy Lee Ozesmi (University of Minnesota) for software support, and Paul Cameron (University of Minnesota) for providing hardware support. Funding for portions of this research came from the Minnesota Board of Water and Soil Resources, the Legislative Commission of Minnesota Resources—Environment and Natural Resources Trust Fund, and the University of Minnesota Research and Outreach Center.
Copyright 2017 Elsevier B.V., All rights reserved.
- Remote sensing
- Soil mapping
- Spatial modeling
- Spatial variability