assessment of tree and multiple linear regressions in estimation of cation exchange capacity

introduction: estimation of cation exchange capacity (cec) with reliable soil properties can save time and cost. pedotransfer function (ptf) is a common method in estimating certain soil properties (e.g. cec) that has been wieldy used for many years. one of the common techniques that have been used to develop ptfs is multiple linear regressions. in this method, all easily obtained soil properties are linearly related to certain soil properties. in addition to multiple linear regressions method, more complex techniques such as artificial neural networks and regression tree have been used to develop ptfs. the regression tree method is a well-known method for analyzing the environmental science which determines optimal separation point of independent variables.the purposes of this study were to evaluate and compare tree and multiple linear regressions in estimating cation exchange capacity with reliable soil properties. materials and methods: for this work, 106 soil samples of unsaturated soil hydraulic database (unsoda), which contain a wide range of soil texture classes, were used. the examples were divided into 2 sets including 81 and 25 soil samples for developing and validating multiple linear regression and tree regression, respectively. for estimating cec with tree and multiple regressions, soil texture properties, organic matter, ph and bulk density were used. to develop multiple linear regressions and create the tree structure, at first, correlation between cation exchange capacity with other soil properties were evaluated; then, soil properties that had significant correlation were chosen to introduce software. as well, the suggested linear function and tree structure were compared with 2 famous pedotranser functions including bell and van-kolen and breeuwsma et al., which have been used for estimating cec.for investigating the performance of multiple linear regression and tree regression to estimate cec 1:1 lines, determination coefficient (r2), mean error (me), root mean square error) rmse), and geometric mean error (gmer) were used. statistica 8.0 software that was developed by esri was used to develop multiple linear regressions and generate tree structure. results and discussion: the results showed for developing multiple linear regression model to estimate cec among all inputs parameters (sand, silt, clay, organic matter, ph and bulk density) only just two parameters including organic (with r=0.70) and clay percentage (with r=0.59) had a significant coefficient, so organic and clay percentage appeared, and suggested multiple linear regression models based on this two parameters, with coefficient of 3.183 and 0.274, respectively, were developed. also, only organic matter and clay percentage from inputs parameter in tree were shown. in tree structure most nods were divided into 2 childs nods based on organic matter and only in the left side of tree structure in the second level clay percentage was appeared. regression tree in two data sets (validation and development) based on r2, rmse, me and gmer had a high quality for cec estimation than regression methods. proposed linear regression model had high performance than bell and van-kolen and breeuwsma et al. to estimate cec. conclusions: the main aim of this study was to investigate the efficiency of multiple linear regression model and regression tree to predict cation exchange capacity (cec) based on relationships between cec and easily measurable soil properties. for this work, 106 soil samples of unsoda data set were used. results showed that just clay percentage and organic matter that had higher correlation with cec appeared in suggested linear regression and tree structure. based on 1:1 lines, r2 ,rmse, me and gmer, tree regression model had higher performance than all linear regression models (suggested function , bell and van-kolen and breeuwsma et. al.) to estimate cation exchange capacity. as well, suggested function had more efficiency than bell and van-kolen and breeuwsma to predict cec.