ALGORITHM FOR MAPPING BARE SOIL FROM LANDSAT TM DATA IN AGRICULTURAL TERRAIN

We are interested in correlating bare soil exposures in NW Ohio, an agricultural region, to aerosol content and to reported cases of asthma. A prerequisite of this effort is to create an algorithm that will accurately distinguish bare soil from other types of ground cover, including no-till agricultural fields, which have heavy stubble. Although bare soil is a class in most studies of land cover classification, the separation of bare soil from soil with crop residue (e.g. wheat stubble), soil with relatively small percentages of growing crop, and truly bare ground has not been very satisfying so far.

In this study, an algorithm to map percentage bare soil cover from multi-temporal LANDSAT TM data is developed, based on the compositional nature of typical ground cover. Our effort has been focused on finding a new bare soil index, and testing it with ground-truth data. The index, which employs a linear combination of dark-object-subtracted spectral ratios, is thresholded to classify pixels covered by bare soil. Those bare-soil-class pixels can be easily turned into percentage bare soil cover for user-specified geographical regions within the LANDSAT TM image. The effectiveness of the new method is compared with another compositionally based method, which employs a linear model for spectral unmixing.

The new algorithm provides a handy and timely tool for the automatic identification of bare soil cover from LANDSAT TM images. Although our specific need for this algorithm is public health remote sensing, percentage bare soil cover is needed for soil erosion studies, mapping of tilled agricultural fields, and other remote sensing applications. In non-agricultural regions, the algorithm may be helpful to geologists for planning field trips.