Using soil survey data for modeling phosphorus sorption capacity

Abstract

The use of series-specific soil survey data for assessing phosphorus (P) retention and release characteristics of soils has already been reported in Sekhon et al. (Environ Earth Sci 72:2345–2356, 2014) wherein P sorption capacity (PSC, Langmuir isotherm method) of pedons collected from two benchmark upland soil series (Berks and Monongahela) and two floodplain (Huntington and Lindside) soil series of West Virginia (USA) was related with its major determinants like amorphous (ammonium oxalate-extractable) and crystalline (dithionite-citrate-bicarbonate extractable) phases of Fe and Al, exchangeable Al, Mn, Na, Ca, Mg, and K, total carbon, extractable acidity, clay content and soil pH. That study was extended into this study with the objective of predicting P sorption capacity with various statistical modeling approaches. Ordinary least squares regression (OLSR) was used to regress PSC on all variables (OLSR full model) and was compared with stepwise deletion of less important variables (OLSR reduced model) and principal components (PCs) regression (PCR) approach. The PCR approach addressed the multicollinearity issue in major determinants of PSC and gave the best prediction performance without losing information on any variable. The study thus signified the use of routinely collected soil survey data and PCR approach in statistical modeling of PSC of a soil.