Name: LUCAS BROSEGHINI TOTOLA
Type: MSc dissertation
Publication date: 29/05/2020
Advisor:
Name |
Role |
|---|---|
| KÁTIA VANESSA BICALHO | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| KÁTIA VANESSA BICALHO | Advisor * |
| WILIAN HIROSHI HISATUGU | Co advisor * |
Summary: ABSTRACT
The constitutive relationship between water content or degree of saturation and suction is
determined soil-water retention curve (SWRC). The understand of the SWRC is important in the
unsaturated soil mechanics study and in solving geotechnical engineering practical problems. As
the SWRC presents spatial and variability, its direct determination can be time-consuming and
costly. Thus, the pedotransfer function concept (PTF), which use easily obtainable properties to
indirectly estimate the SWRC, have gained prominence. Based on an extensive hydrophysical
database of Brazilian tropical and subtropical soils, mostly composed by Planosols, Ferralsols and
Acrisols, this study aims to evaluate the use of artificial neural networks (ANNs) to estimate the
drying limit SWRC for the range of matric potentials from 0 to 1500 kPa. Two different topologies
(point and pseudo-continuous) are proposed, and factors influencing ANNs performance are
analyzed, such as the networks structure and geometry, and the addition of input parameters in a
hierarchical structure. Physical parameters such as particle size distribution, bulk and particle
density, total porosity and organic matter content are used as input parameters for the ANNs. The
overall performance was characterized by the coefficient of determination (r²) and the root mean
square error (RMSE). The analyzes and results obtained show the importance of the appropriate
choice of input parameters, which must include properties representative of both soil texture and
structure to represent the hydraulic behavior of the soil along the entire curve, thus providing better
results. The pseudo-continuous topology overcome the point topology performance and the RMSE
values decreased from 0.048 to 0.029 cm³.cm-³ when more predictors were used. As particles of
clay size are predominant in fines for tropical soils, the use of clay mineralogy is recommended to
minimize errors in estimates in the SWRC dry end. Although experimental testing remains
essential, and considering the geographic limitation of PTFs for Brazilian soils in the literature, the
results indicate the ANNs as a potential tool for predicting the upper limit drying SWRC over a
wide suction interval, being useful for preliminary studies and projects on unsaturated soils.
Keywords: soil-water retention curve, artificial neural networks, unsaturated soils, tropical soils.
