BUILDING AND ASSESSMENT OF MATHEMATICAL MODELS FOR PREDICTING SORPTIVITY OF IRAQI SOILS FROM SOME PHYSICAL AND CHEMICAL PROPERTIES

Ala,; S. Atee, A; B. Salem, Salloum

	BUILDING AND ASSESSMENT OF MATHEMATICAL MODELS FOR PREDICTING SORPTIVITY OF IRAQI SOILS FROM SOME PHYSICAL AND CHEMICAL PROPERTIES
The Iraqi Journal of Agricultural Science
Article 1, Volume 38, Issue 4, 2007, Pages 83-93
Authors
; a S. Atee; Salloum B. Salem
Abstract
Five multi-linear regression equations were developed for predicting the sorptivity of 70 soil samples from easily measured soil properties. The SAS analysis system was used to produce the best multi-linear associations between dependent and independent variables using five model building procedures: l)Stepwise, 2) Maximum R-squared Improvement Technique (MAXR), 3) Minimum R-squared Improvement Technique (MINR), 4)Forward Selection, and 5) Backward Elimination. Values of measured properties ranged between 40-610, 154-635, 41.5-735, 0.5-455, 20.3-120, and 4-33.3 gm.kg"' soil for the clay , silt, sand, gypsum, carbonate, and organic matter respectively. Measured sorptivity values ranged between 0.101-1.65 cm. min_/l. Pearson bivariate correlation matrix produced significant correlation coefficients between sorptivity and each independent variable. A highly significant, yet positive linear correlation existed between sorptivity and gypsum, while no clear trend was realized between sorptivity and other independent variables. The MAXR and MINR model building procedures resulted in a six parameters model (all independent variables were included in the model) for predicting sorptivity with the highest correlation coefficient R=0.901) and a significance of %53.2 for all parameters in the model. When the first and fifth methods were used, a two parameter models were obtained for predicting sorptivity( only two independent variables were included in the models) with R- value of 0.896. All parameters in the model were significant at %99.9. The Forth method produced three parameters model(three independent variables were included in the model) with R- value of 0.897. All parameters in the model were significant at %73.3. In this study the best obtainable five regression models for predicting sorptivity were assessed upon five criteria; 1) number of the parameters in the model, 2) value of correlation coefficient, 3) Mean square error, 4)Mallows' Cp statistics, and 5)F-value. Based on our assessment, Stepwise and Backward elimination model building procedures produced the best model for predicting the sorptivity with a minimum number of parameters, highly significant R-value, lower mean square than the four parameters model, lowest Cp value, and highest F-statistic value.

Statistics Article View: 9 PDF Download: 3