Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks

Amir Tophel; Thi Mai Nguyen; Jeffrey P. Walker; Jayantha Kodikara

doi:10.1007/s10706-025-03293-x

Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks

Amir Tophel
, Thi Mai Nguyen
, Jeffrey P. Walker
, Jayantha Kodikara

Civil & Environmental Engineering

Research output: Contribution to journal › Article › Research › peer-review

10 Citations (Scopus)

Abstract

Moisture content in geomaterials critically impacts road construction. Optimising the Optimum Moisture Content (OMC) during compaction reduces energy usage and construction costs, but materials often require adjustments upon delivery to reach OMC. The dry back phase is essential for efficiently allowing the pavement to release trapped moisture, preventing surface issues such as moisture resurfacing, aggregate punch-in, and premature stiffness loss in pavements. This study investigated temporal soil moisture (SM) variation within the compaction layer under varying environmental conditions such as physical soil temperature (T_soil), net radiation (Rn), initial SM, and bulk density (ρb). A Long Short-Term Memory (LSTM) model was developed and compared with a traditional three-layer water balance model to predict SM variations. The LSTM model showed superior accuracy, with a maximum Root Mean Square Error of 0.20% for Volumetric Moisture Content and 0.12% for Gravimetric Moisture Content. Integrating environmental data and site-specific soil characteristics with a data-driven model significantly advances moisture management in road construction, enhancing compaction and pavement durability.

Original language	English
Article number	318
Number of pages	19
Journal	Geotechnical and Geological Engineering
Volume	43
Issue number	6
DOIs	https://doi.org/10.1007/s10706-025-03293-x
Publication status	Published - 23 Jul 2025

Keywords

Dry back
LSTM model
Machine learning
Optimum moisture content
Pavement layers
Soil moisture

Access to Document

10.1007/s10706-025-03293-xLicence: CC BY

Cite this

@article{d6ceb164773b4ed2a710f3dc719a1b98,

title = "Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks",

abstract = "Moisture content in geomaterials critically impacts road construction. Optimising the Optimum Moisture Content (OMC) during compaction reduces energy usage and construction costs, but materials often require adjustments upon delivery to reach OMC. The dry back phase is essential for efficiently allowing the pavement to release trapped moisture, preventing surface issues such as moisture resurfacing, aggregate punch-in, and premature stiffness loss in pavements. This study investigated temporal soil moisture (SM) variation within the compaction layer under varying environmental conditions such as physical soil temperature (Tsoil), net radiation (Rn), initial SM, and bulk density (ρb). A Long Short-Term Memory (LSTM) model was developed and compared with a traditional three-layer water balance model to predict SM variations. The LSTM model showed superior accuracy, with a maximum Root Mean Square Error of 0.20\% for Volumetric Moisture Content and 0.12\% for Gravimetric Moisture Content. Integrating environmental data and site-specific soil characteristics with a data-driven model significantly advances moisture management in road construction, enhancing compaction and pavement durability.",

keywords = "Dry back, LSTM model, Machine learning, Optimum moisture content, Pavement layers, Soil moisture",

author = "Amir Tophel and Nguyen, \{Thi Mai\} and Walker, \{Jeffrey P.\} and Jayantha Kodikara",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2025.",

year = "2025",

month = jul,

day = "23",

doi = "10.1007/s10706-025-03293-x",

language = "English",

volume = "43",

journal = "Geotechnical and Geological Engineering",

issn = "0960-3182",

publisher = "Springer-Verlag London Ltd.",

number = "6",

}

TY - JOUR

T1 - Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks

AU - Tophel, Amir

AU - Nguyen, Thi Mai

AU - Walker, Jeffrey P.

AU - Kodikara, Jayantha

N1 - Publisher Copyright: © The Author(s) 2025.

PY - 2025/7/23

Y1 - 2025/7/23

N2 - Moisture content in geomaterials critically impacts road construction. Optimising the Optimum Moisture Content (OMC) during compaction reduces energy usage and construction costs, but materials often require adjustments upon delivery to reach OMC. The dry back phase is essential for efficiently allowing the pavement to release trapped moisture, preventing surface issues such as moisture resurfacing, aggregate punch-in, and premature stiffness loss in pavements. This study investigated temporal soil moisture (SM) variation within the compaction layer under varying environmental conditions such as physical soil temperature (Tsoil), net radiation (Rn), initial SM, and bulk density (ρb). A Long Short-Term Memory (LSTM) model was developed and compared with a traditional three-layer water balance model to predict SM variations. The LSTM model showed superior accuracy, with a maximum Root Mean Square Error of 0.20% for Volumetric Moisture Content and 0.12% for Gravimetric Moisture Content. Integrating environmental data and site-specific soil characteristics with a data-driven model significantly advances moisture management in road construction, enhancing compaction and pavement durability.

AB - Moisture content in geomaterials critically impacts road construction. Optimising the Optimum Moisture Content (OMC) during compaction reduces energy usage and construction costs, but materials often require adjustments upon delivery to reach OMC. The dry back phase is essential for efficiently allowing the pavement to release trapped moisture, preventing surface issues such as moisture resurfacing, aggregate punch-in, and premature stiffness loss in pavements. This study investigated temporal soil moisture (SM) variation within the compaction layer under varying environmental conditions such as physical soil temperature (Tsoil), net radiation (Rn), initial SM, and bulk density (ρb). A Long Short-Term Memory (LSTM) model was developed and compared with a traditional three-layer water balance model to predict SM variations. The LSTM model showed superior accuracy, with a maximum Root Mean Square Error of 0.20% for Volumetric Moisture Content and 0.12% for Gravimetric Moisture Content. Integrating environmental data and site-specific soil characteristics with a data-driven model significantly advances moisture management in road construction, enhancing compaction and pavement durability.

KW - Dry back

KW - LSTM model

KW - Machine learning

KW - Optimum moisture content

KW - Pavement layers

KW - Soil moisture

UR - https://www.scopus.com/pages/publications/105011285830

U2 - 10.1007/s10706-025-03293-x

DO - 10.1007/s10706-025-03293-x

M3 - Article

AN - SCOPUS:105011285830

SN - 0960-3182

VL - 43

JO - Geotechnical and Geological Engineering

JF - Geotechnical and Geological Engineering

IS - 6

M1 - 318

ER -

Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks

Abstract

Keywords

Access to Document

Other files and links

SPARC: ARC Research Hub for Smart Next Generation Transport Pavements

Cite this

Soil Moisture Prediction in Pavement Layers Using LSTM Neural Networks

Abstract

Keywords

Access to Document

Other files and links

Projects

SPARC: ARC Research Hub for Smart Next Generation Transport Pavements

Cite this