Input (pre-processing)

Neural network prediction (Rosetta)

The dialog predicts the van Genuchten parameters θr, θs, α, n and Ks from soil data with the Rosetta pedotransfer functions Schaap et al. (2001); Zhang and Schaap (2017), as the Rosetta Lite dialog of HYDRUS-1D. It opens from Neural network prediction (Rosetta)… on the page Materials & parameters and from Rosetta… next to the soil catalog of the parameter estimation (initial estimates).

Input data

Rosetta is hierarchical: the more data, the better the prediction.

Model	Input
H1 Textural class	the USDA textural class; gives the average parameters of the class in the Rosetta database
H2 SSC	sand, silt and clay [%] (sum 99–101 %)
H3 SSC + BD	and the bulk density [g/cm³] (0.5–2)
H4 SSC + BD + θ33	and the water content at 33 kPa (field capacity)
H5 SSC + BD + θ33 + θ1500	and the water content at 1500 kPa (wilting point)

The USDA texture class of the entered fractions is shown below the inputs.

Models

Version	Description
Rosetta 3 (default)	One network for the retention parameters and Ks, recalibrated with weights from the uncertainty of the fitted retention curves and 1000 bootstrap members Zhang and Schaap (2017).
Rosetta 2	Rosetta 1 with the retention parameters corrected after Schaap et al. (2004).
Rosetta 1	The models of Rosetta Lite in HYDRUS-1D Schaap et al. (2001).

All are ensembles of neural networks trained on bootstrap samples; the prediction is the ensemble mean, shown with its standard deviation. Geometric means uses 10 to the mean of the logarithms of α, n and Ks instead of the arithmetic means. The network weights are those of the Python package rosetta-soil; the unit tests reproduce its predictions.

Result

The prediction is given in cm and days, with K0 and L of the conductivity function with a matching point Schaap and Leij (2000) for information. Accept writes θr, θs, α, n and Ks, converted to the project's units, into the material (or the initial estimates); l stays.

Pedotransfer functions give approximate parameters for a soil like the soils of their database (mostly temperate agricultural soils). Measured retention data or a parameter estimation are better when available.

Literature cited in this topic

Schaap, M. G., A. Nemes, and M. Th. van Genuchten (2004). Comparison of models for indirect estimation of water retention and available water in surface soils. Vadose Zone Journal 3(4): 1455–1463. doi:10.2136/vzj2004.1455
Schaap, M. G., and F. J. Leij (2000). Improved prediction of unsaturated hydraulic conductivity with the Mualem–van Genuchten model. Soil Science Society of America Journal 64(3): 843–851. doi:10.2136/sssaj2000.643843x
Schaap, M. G., F. J. Leij, and M. Th. van Genuchten (2001). Rosetta: a computer program for estimating soil hydraulic parameters with hierarchical pedotransfer functions. Journal of Hydrology 251(3–4): 163–176. doi:10.1016/S0022-1694(01)00466-8
USDA-ARS U.S. Salinity Laboratory. rosetta-soil, version 0.3.2: Python implementation of the Rosetta pedotransfer functions (CC0 1.0). https://github.com/usda-ars-ussl/rosetta-soil
Zhang, Y., and M. G. Schaap (2017). Weighted recalibration of the Rosetta pedotransfer model with improved estimates of hydraulic parameter distributions and summary statistics (Rosetta3). Journal of Hydrology 547: 39–53. doi:10.1016/j.jhydrol.2017.01.004

All sources: References.