ABSTRACT
Deriving accurate pedo-transfer functions (PTFs) for predicting difficult-to-measure soil properties such as soil cumulative infiltration (CI) is an essential issue for saving time and cost. This study aimed to develop MLR (multiple-linear-regression), ANN (artificial-neural-network) and GMDH (group-method-of-data-handling) PTFs for predicting CI at different time intervals (i.e. 1, 2, 5, 15, 30, 60 and 120 min) and mapping in GIS in calcareous soils in northwest Iran. Soil infiltration measurements with a double-ring infiltrometer were carried out at 124 points with three replications. At each point, various easily measurable soil properties were measured. The results indicated that ANN-based PTFs provided the highest E (from 0.84 to 0.97) and the lowest RMSE (root mean square error) (from 0.19 to 8.91) compared to GMDH-based PTFs (E = 0.43–0.81 and RMSE = 0.34–19.59) and MLR-based PTFs (E = 0.41–0.83 and RMSE = 0.34–16.50). In addition, the map of soil CI at different time intervals was generated based on the best-derived PTFs (i.e. ANN-based PTFs) in GIS. We concluded that ANN-based PTFs provided an accurate prediction and a high-quality map to such costly studies by using easily measurable basic soil properties as input data in calcareous soils.
Disclosure statement
No potential conflict of interest was reported by the authors.