Page Banner

United States Department of Agriculture

Agricultural Research Service


item Reeves Iii, James
item Madari, B
item Machado, P
item Torres, E
item Guimaraes, C
item Mccarty, Gregory

Submitted to: Geoderma
Publication Type: Abstract Only
Publication Acceptance Date: 3/24/2006
Publication Date: N/A
Citation: N/A

Interpretive Summary: Summary.

Technical Abstract: One hundred and twenty soil samples from long-term tillage experiments were collected in the Brazilian savanna region and in southern Brazil, and analyzed for total- C, N and H, sand, silt and clay and several measures of soil aggregation. These samples were fractionated into 8 aggregate size classes resulting in a set of 700 samples. These samples were also analyzed for total- C, N and hydrogen. In addition, data was available on location, depth of sample, management (forest soil, conventional till and no-till) and on cropping system (rotational or successional) for both sets of samples and aggregate size for the fractionated samples. Carbon, N and H values were determined by dry combustion, and sand, silt and clay by densitometry. Fractionation into aggregate size classes was carried out by wet sieving. All samples were also analyzed in the mid- and near-infrared by diffuse reflectance of ground, non-KBr diluted samples using a Digilab FTS-7000 Fourier Transform spectrometer equipped with a Pike Auto-Diff autosampler. Spectra were collected from 10,000 to 4000 cm-1 at 4 cm-1 resolution in the near-infrared (NIR) using a liquid N2 cooled InSB detector and in the mid-infrared (MIR) from 4000 to 400 cm-1 at 4 cm-1 resolution using a DTGS detector with 64 co-added scans per spectrum in both regions. Quantitative calibrations for the various analytes were performed using partial least squares regression (PLS) using SAS and Galactic's GRAMS software. In addition, efforts to discriminate various sample classes (i.e., location, depth, tillage, aggregate size) were attempted using either PLS (SAS) or principle components (GRAMS). With the exception of location, and to a lesser degree, soil management (forest, type of tillage), efforts at discriminant analysis were not very successful using either spectral region or type of discriminant analysis even when only surface soils alone were examined. However, this may have been due in at least part to that fact that by the time samples were segregated into all the unique classes (i.e., 2 locations X 4 depths X 3 soil uses X 2 rotations X 8 aggregates = 384 different classes for the set of 700) there were too few samples in each class (only two samples per class here). Combining samples, such as all four depths, increased the sample numbers per class, but apparently added too much variation within a class. Quantitative spectroscopic analysis was carried out separately for the original set of 120 intact (ITS) samples and the set of 700 aggregate fractionated samples (AFS). Results for the set of 120 ITS for total-C and -N were excellent using either spectral range with R2 > .97, except for C by MIR ( R2 = .93). Similar results were obtained for determinations of sand and clay (R2 > .94), but silt which is determined by difference (100% - %sand -%clay) gave much lower R2 (.80 and .63 for MIR and NIR, respectively). Several measures of aggregate size and soil aggregation state appeared to be promising in the MIR with R2 of ~.8, NIR results were not as promising. The results for C were puzzling since in all previous cases, and for other sets of these samples (set of 30 ITS surface samples and set of 700 AFS samples), MIR calibrations for total C outperform NIR calibrations. Efforts with the 700 AFS samples resulted in R2 of .97 and .96 for C, and .97 and .97 for N, for MIR and NIR calibrations respectively, all excellent results. Results for determination of H were never as good as for total-C and -N, with R2 of .6 to .7 common regardless of the sample set. Finally, efforts were made to determine the values for the set of 700 AFS samples using calibrations using the set of 120 ITS from which they were fractionated. The location from which the samples were derived was easily and very accurately determined as was found for all data sets examined. Results for C and N were promis

Last Modified: 06/25/2017
Footer Content Back to Top of Page