Premium
An estimation of the main wetting branch of the soil water retention curve based on its main drying branch using the machine learning method
Author(s) -
Lamorski Krzysztof,
Šimůnek Jiří,
Sławiński Cezary,
Lamorska Joanna
Publication year - 2017
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1002/2016wr019533
Subject(s) - topsoil , soil science , mean squared error , soil water , wetting , bulk density , mathematics , environmental science , statistics , materials science , composite material
In this paper, we estimated using the machine learning methodology the main wetting branch of the soil water retention curve based on the knowledge of the main drying branch and other, optional, basic soil characteristics (particle size distribution, bulk density, organic matter content, or soil specific surface). The support vector machine algorithm was used for the models' development. The data needed by this algorithm for model training and validation consisted of 104 different undisturbed soil core samples collected from the topsoil layer (A horizon) of different soil profiles in Poland. The main wetting and drying branches of SWRC, as well as other basic soil physical characteristics, were determined for all soil samples. Models relying on different sets of input parameters were developed and validated. The analysis showed that taking into account other input parameters (i.e., particle size distribution, bulk density, organic matter content, or soil specific surface) than information about the drying branch of the SWRC has essentially no impact on the models' estimations. Developed models are validated and compared with well‐known models that can be used for the same purpose, such as the Mualem (1977) (M77) and Kool and Parker (1987) (KP87) models. The developed models estimate the main wetting SWRC branch with estimation errors (RMSE = 0.018 m 3 /m 3 ) that are significantly lower than those for the M77 (RMSE = 0.025 m 3 /m 3 ) or KP87 (RMSE = 0. 047 m 3 /m 3 ) models.