Separation Effective Soil Properties on Moisture Characteristic Curve Using Decision tree

Soft computational techniques have been widely used in scientific research and engineering in recent decades. Since the measurement of hydraulic properties by direct laboratory methods is hard, time consuming and expensive, Thus, there is need to use alternative methods based on conveniently available soil properties to estimate it with less effort, time and cost. One of the new methods for estimating soil hydraulic properties, such as soil moisture characteristic curve, is non-parametric methods. This study was performed to determine the efficiency of the decision tree method in separation of effective properties in estimating soil moisture characteristic curve parameters.

To perform this study, number of 72 points were selected in the village of Marghmalek and Sharekord city. Samples were collected from depth of 0-20 cm and then were transferred to the laboratory for required measurements. Some properties such as pH, EC, saturated moisture, calcium carbonate, organic matter, clay and sand, bulk density, mean weight diameter of dry aggregate, mean weight diameter of wet aggregate, geometric mean and standard deviation of particle diameter were measured in the laboratory. Also, the moisture characteristic curves were determined at 0, 1, 3, 5, 10, 30, 50, 100, 150, 1000, 1500 kPa suctions and were fitted to the van Genuchten model. The input variables were introduced into the MATLAB software in two scenarios (first scenario: pH, EC, %clay and sand, organic matter, calcium carbonate, mean weight diameter of wet and dry aggregate, bulk density, saturated moisture and the second scenario: pH, EC, geometric mean and standard deviation of particle diameter, organic matter, calcium carbonate, mean weight diameter of wet and dry aggregate, bulk density, saturated moisture) and modeled by decision tree and error estimators of cross validation and resub stitution. Evaluation statistics of each model including R2, RMSE and %RMSE were calculated.

The results obtained from decision tree modeling showed that the most important factors affecting moisture content in PWP suction, were saturated moisture and clay. The PWP target variable has the highest correlation in the first scenario (0.88) and in the second scenario )0.91( and the least error rate among the other variables, and after that, FC has the highest correlation (0.86) in the second scenario. Target variables n had the highest error rate and α the lowest correlation in both scenarios. Generally, the second scenario performed better than the first scenario by replacing the geometric mean and standard deviation of particle diameter with the percentage of clay and sand. The sensitivity analysis showed that PWP was the most sensitive among the input parameters to pH, BD, calcium carbonate and organic matter and FC was the most sensitive to geometric standard deviation of particle and MWDwet.

In general, modeling has been successful in both scenarios. But by substituting geometric mean and standard deviation of particle diameter instead of clay and sand percentage, a better performance was obtained in estimating moisture characteristic curve variables in the second scenario.