Research Article

Artificial Neural Networks for Estimating Soil Water Retention Curve Using Fitted and Measured Data

Table 3

Sensitivity analysis of developed ANNs by substituting step by step each input parameter by its mean value.

ANN 1
(−cm)Model’s
RMSE
Bulk densityParticle densityPorositySandSiltClay
RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)

00.0640.08939.4780.0651.2420.07516.6380.0640.0520.064−0.5230.0697.5083
100.0570.08446.7490.0581.2780.06513.7040.0628.5810.055−3.8190.0581.718
200.0560.07940.5500.0560.8590.06516.9520.06712.0820.056−0.7840.0595.105
300.0510.06832.4080.0510.0380.06117.8700.06425.1530.051−0.0190.0568.416
500.0490.06427.5610.050−0.0120.0548.5110.06937.8300.0500.1680.05511.216
1000.0690.07711.0040.0690.0820.0723.5070.08725.2050.0701.2900.0757.759
2000.0820.0888.4180.081−0.6900.0842.5570.10023.0780.081−0.1970.0865.539
5000.0830.0897.3730.082−0.6050.0841.6540.09615.9550.0830.3200.0886.276
10000.0780.0858.06720.077−1.0060.0791.3770.09115.8240.077−1.1780.0836.041

ANN 2
(−cm)Model’s
RMSE
Bulk densityParticle densityPorositySandSiltClay
RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)

00.0620.09044.7950.0653.4950.07418.9840.061−1.43830.0630.7350.07113.825
100.0540.07945.6880.0551.9390.06621.2400.054−0.3460.06111.08960.06621.544
200.0550.08248.9970.0561.2180.06213.7770.06416.5150.053−4.7130.0561.897
300.0550.07944.1160.0550.3190.0608.9390.06518.4400.0551.2690.0574.422
500.0600.07525.9270.0611.6250.0659.1700.06814.0380.0636.3470.06711.943
1000.0750.08613.8800.0750.5280.0772.4540.0828.9630.0806.0470.0828.753
2000.0980.1035.3560.0980.0010.1023.9090.1079.0050.1013.2090.1057.686
5000.0870.0958.4600.087−0.0270.0891.5650.09711.7810.0880.7280.0914.904
10000.0840.0863.3390.083−0.1410.0885.3580.0874.2460.0862.6330.09210.162

ANNo
Model’s
RMSE
Bulk densityParticle densityPorositySandSiltClay
RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)RMSEVar (%)

0.0800.09417.0650.080−0.0750.0855.74640.0811.2550.0800.0370.0822.6410.11846.825

Var (%) means the increasing (+) or decreasing (−) in the RMSE value of an output (matric potentials, ) due to the substitution of a specific input parameter by its mean value.