Assessing spatial variability in soil characteristics with geographically weighted principal components analysis

Sandeep Kumar*, Rattan Lal, Christopher D. Lloyd

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

39 Citations (Scopus)

Abstract

Dimensionality reduction methods such as principal components analysis (PCA) provide a means of identifying trends in soil characteristics which may be represented by a wide range of variables. However, these characteristics may be highly spatially variable and so the results from PCA represent, in some sense, an "average" of locally distinct characteristics. One approach to account for these local differences is to introduce a geographical weighting scheme into the PCA process. In this paper, such an approach is assessed in the exploration of soil characteristics in the state of Pennsylvania, USA. Data from 878 georeferenced soil profiles which include different soil parameters (n = 12) were extracted from the National Soil Survey Center database. Where data are parts of compositions (e. g., percentages of sand, silt, and clay), analysis using raw data is not appropriate and such data were transformed using log ratios (specifically, balances). Single variables (i. e., those which are not parts of compositions) were logged. The first two principal components explain over 50% of the variance. The mapped values suggest marked spatial variation in soil characteristics, but it is not possible to assess which of these variables explain most variation in particular regions from the simple maps of raw variables. Geographically weighted PCA (GWPCA) provides additional information which is obscured by PCA, and it also provides a set of component scores and loadings at all data locations. The soil variable with the largest loading at most locations of Pennsylvania is the logged base saturation (BSln), and this supports the findings of the conventional PCA analysis. While BSln loads most highly in most of the eastern third, the middle and the south west of the state, the northwest is less spatially consistent in terms of the variables which explain most variation. For GWPC 1, the variable with the second largest loading at most locations (i. e., primarily the south and west) is CEC. B1 (the log ratio of Ca, Mg, and Na to K and EXACID), while CEC. B2 (the log ratio of Ca and Mg to Na), pHln (logged pH) and BSln dominate in other areas. The GWPCA results suggest that there is marked spatial variation in multivariate soil characteristics across Pennsylvania state and that results from standard PCA obscure this considerable variation.

Original languageEnglish
Pages (from-to)827-835
Number of pages9
JournalComputational Geosciences
Volume16
Issue number3
DOIs
Publication statusPublished - 01 Jun 2012

Keywords

  • Geographically weighted principal component analysis
  • Log ratios
  • Principal component analysis
  • Soil organic carbon

ASJC Scopus subject areas

  • Computer Science Applications
  • Computers in Earth Sciences
  • Computational Theory and Mathematics
  • Computational Mathematics

Fingerprint

Dive into the research topics of 'Assessing spatial variability in soil characteristics with geographically weighted principal components analysis'. Together they form a unique fingerprint.

Cite this