Abstract:
The variability of soil infiltration characteristics has outstanding scale- and location-dependent relationships. The parameter estimation on the multi-scale has been the important basis for the irrigation arrangement and management in precision agriculture. In this study, a pedo-transfer function was established to determine the multiple scales variability of soil infiltration characteristics. 52 tests of double ring infiltration were also conducted in the Guanzhong Plain. Different scaling factors were then calculated to compare their scaling effects using the Kostiakov equation. In addition, the wavelet and path analysis was utilized to quantify the relationship between the scaling factors and soil properties (soil mechanical composition, bulk density, initial water content, and organic matter content) for the multiple scales. As such, the soil properties were identified to determine the significant impact on the scaling factors. Subsequently, the pedo-transfer functions were developed with the Multiple Linear Regression (MLR), Support Vector Machine (SVM), and Back Propagation Artificial Neural Network (BP-ANN) after the scaling factor estimation. The results indicated that there was the smallest error of the scaling factor between the scaled cumulative infiltration and the measured using the least square method, where the Root Mean Square Error (RMSE), Mean Bias Error (MBE), and Mean Absolute Value of Relative Error (MARE) were 1.83 cm, 0.24 cm, and 21.2%, respectively. The scaling factor FS presented a significant relationship with the soil bulk density, sand (SA), clay (CL), and Soil Organic Matter content (SOM) at the spatial scales of 1-8, 1-2.5, 1-4, and 1-3.8 km. Among them, the scaling factor by the least square method was positively correlated with the SA and SOM, where the total path coefficients were 0.78, and 0.65, respectively, whereas, the scaling factor by the least square method was negatively correlated with the CL and bulk density, where the total path coefficients were -0.74, and -0.68, respectively. The scaling factor by the least square method also presented a significant positive correlation with the soil silt content in the range of 2.5-3.5 and 5-10 km from the position of the first test site, but there was no significant relationship in the other scales and sampling ranges. There was a complex positive and negative relationship between the scaling factor by the least square method and the initial soil water content in the sampling range of 1-4 km, indicating the outstanding scale- and location-dependent. Multiple wavelet coherence analysis demonstrated that the spatial variability of scaling factor was attributed to the combination of soil bulk density, sand, clay, and organic matter content during soil infiltration in the study area. The pedo-transfer function was achieved with the highest estimation accuracy for the scaling factors using a support vector machine. The estimated value of the infiltration from the verification set was in good agreement with the measurement, where the RMSE, MBE, and MARE were 1.92 cm, 0.05 cm, and 27.6%, respectively, indicating that the SVM was feasible to establish the soil pedo-transfer functions for the scaling factor estimation. The finding can also easily access the infiltration parameters for multiple scales.