Efficient Sampling Schemes for Assessing Crop Residue Cover Using Remotely Sensed Data.

Crop residues on the soil surface provide a protective barrier against water and wind erosion and reduce the amounts of soil, nutrients, and pesticides that reach streams and rivers. Quantification of crop residue cover is required to evaluate the effectiveness and extent of conservation tillage practices, as well as the extent of bio-fuel harvesting.  The line-point transect method, the standard technique used to quantify crop residue is impractical for monitoring crop residue cover and tillage intensity in many fields in a timely manner. Multispectral remote sensing approaches for estimating crop residue cover have had mixed success because crop residues and soils are spectrally similar.  An alternative approach is based on detecting the absorption features near 2100 nm that are associated with cellulose and lignin. 

Both remote sensing approaches require surface reference (or ground-truth) data from at least 50 fields for estimating crop residue cover. This involves a substantial investment of man-hours. Our objectives were: (1) to estimate crop residue cover using remotely sensed data over agricultural sites in central Iowa and west-central Indiana, and (2) to evaluate alternative, less labor-intensive sampling schemes for acquiring crop residue cover surface reference data. 

Crop residue cover was measured with line point transects in >50 corn and soybean fields during May 2004 and 2005 in Iowa and >40 corn and soybean field during November 2008 and 2010 in Indiana. The Hyperion imaging spectrometer on the NASA Earth Observing-1 (EO-1) spacecraft acquired images with 220 bands at ~10 nm intervals over the 400-2500 nm wavelength region with a 30-m spatial resolution over both sites.  Crop residue cover was linearly related to the cellulose absorption index (CAI). Crop type did not significantly affect the regression lines, although fields with soybean residue typically had lower residue covers than fields with corn residue.  We examined alternative methods of selecting representative low and high samples for calibration.  Classification accuracy was not significantly degraded by the alternative sampling schemes compared to the classifications using all of surface reference data. Thus, for this limited set of Hyperion images over sites in Iowa and Indiana, the alternative sampling strategies provide adequate surface reference data using fewer observations.