Predicting soil hydraulic conductivity using stacked deep neural networks: Long-term tillage impacts on a Vertisol in the Eastern Mediterranean

Accurate prediction of soil hydraulic conductivity (Ks) is crucial for understanding water movement and improving soil management, particularly under diverse tillage systems. The objective of this study was to develop and validate a Stacked Deep Neural Network (Stacked DNN) model for predicting Ks using easily measurable soil physical and hydro physical properties under long term tillage practices. Soil samples were collected from 0–15 cm and 15–30 cm depths across conventional tillage (CT), no-tillage (NT), reduced tillage (RT), and strategic tillage (ST). The Ks values at 15 cm depth were measured using a Guelph permeameter at 5 cm and 10 cm water heads, ranged from 0.024 to 1.101 cm/h for the 0–10 cm soil depth. The average values from these measurements were used for model training and validation. Predicted Ks values for 0–15 cm and 15–30 cm depths varied significantly among tillage systems and depths, reflecting the effects of soil management on hydraulic behavior. Conventional tillage with residue burning (CT2) recorded the highest predicted Ks at 0.339 cm h −1 (0–15 cm) and 0.091 cm h −1 (15–30 cm). In contrast, no-tillage (NT) exhibited the lowest values, averaging 0.081 cm h −1 and 0.128 cm h −1 at respective depths, likely due to surface compaction. Reduced tillage (RT2) demonstrated balanced performance, with predicted values of 0.484 cm h −1 (0–15 cm) and 0.186 cm h −1 (15–30 cm), suggesting enhanced water infiltration compared to other reduced tillage methods. The Stacked DNN model achieved superior predictive accuracy, with an R² of 0.986 and an RMSE of 0.029 cm h −1 , outperforming individual machine learning models such as Random Forest (R² = 0.902) and XGBoost (R² = 0.917). Bulk density, macroporosity, and mean weight diameter were identified as critical variables influencing Ks predictions, highlighting the model’s ability to capture complex nonlinear relationships. These findings emphasize the effectiveness of ensemble machine learning approaches in modeling Ks and the significant influence of tillage practices on soil hydraulic properties. Moreover, the ability to accurately predict Ks has substantial practical implications for sustainable agriculture, as it enables the optimization of irrigation practices, informed soil management, and improved water resource conservation to support long-term crop productivity and environmental stewardship. • Stacked_DNN model accurately predicted hydraulic conductivity under varying tillage. • Stacked models outperform RF and XGBoost in predicting soil Ks values. • Hydraulic conductivity predictions showed strong agreement with field values. • No-tillage resulted in the lowest predicted Ks, indicating reduced soil permeability. • Ensemble ML models reveal key soil variables influencing Ks predictions.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.