Halil Karahan

Accurate estimation of actual evapotranspiration (ETa) is essential for sustainable water resources management and agricultural planning. This study investigated how the availability of reference evapotranspiration (ET₀) and soil moisture (SM) affects ETa prediction performance and predictor importance using Random Forest (RF), Bagging Trees (BT), Least Squares Boosting (LSBoost), Generalized Additive Models (GAM), and Multiple Linear Regression (MLR). Global solar radiation (Rs), land surface temperature (LST), normalized difference vegetation index (NDVI), and SM were evaluated under two primary scenarios, with and without ET₀. Models were developed using 80% of the data, with five-fold cross-validation conducted within the training subset, and final performance was evaluated on an independent 20% test subset. Model interpretability was assessed using SHAP, Permutation Feature Importance (PFI), Partial Dependence Plot (PDP), and temporal signed-SHAP analyses, while an additional SM-ablation experiment quantified the incremental predictive contribution of soil moisture. In Scenario I, ET₀ was the dominant predictor, and RF and BT achieved the highest independent-test performance (R² = 0.875; RMSE = 0.398–0.399 mm day⁻¹). Excluding ET₀ reduced predictive performance, with RF, LSBoost, and BT achieving R² values of approximately 0.80 in Scenario II, while the predictor-importance structure shifted primarily toward Rs, followed by SM, LST, and NDVI. Removal of SM further reduced performance in both scenarios, with substantially greater deterioration when ET₀ was unavailable: R² decreased by 0.025–0.044 in Scenario I and by 0.051–0.099 in Scenario II. Under the SM-excluded Scenario I configuration, which more closely aligned the predictor information with previous ANN- and SAFER-based studies, R² values ranged from 0.820 to 0.832, indicating that differences in ETa prediction performance cannot be attributed solely to model structure but also depend strongly on predictor information content. Overall, ET₀ provides substantial atmospheric-demand information for ETa prediction, whereas Rs assumes the dominant predictive role, and SM provides particularly important complementary information when ET₀ is unavailable. The combined performance, ablation, and explainability analyses demonstrate the importance of considering predictor composition together with model structure when developing and comparing ETa estimation approaches.

Accurate estimation of actual evapotranspiration (ETa) is essential for sustainable water management and irrigation planning. This study presents a simulation–optimization framework, integrated with the Simple Algorithm for Evapotranspiration Retrieving (SAFER) model, for daily ETa estimation and applies it to a heterogeneous agricultural region with multiple crop types. Unlike conventional SAFER implementations that rely on linear regression or trial-and-error calibration, the proposed framework enables simultaneous calibration of model coefficients via a single optimization procedure while preserving the model’s nonlinear structure. Daily ETa was estimated using remote sensing inputs, including normalized difference vegetation ındex, land surface temperature (LST), surface albedo (αs), and meteorological data. The model was evaluated under two different training–testing scenarios to assess its robustness and generalization capability. Scenario I, which used only 38 satellite overpass days for calibration, achieved a test R2 of 0.8195 and an overall R2 of 0.8152. These results were consistent with Scenario II (test R2 = 0.8249; overall R2 = 0.8181), indicating stable predictive performance across different data partitioning schemes. The optimized coefficients remained highly consistent across scenarios (a ≈ 0.315, b ≈ −0.0012), demonstrating stable parameter convergence and robustness of the optimization process. Overall, the results indicate that reliable ETa estimation can be achieved even with limited calibration data, highlighting the practical applicability of the proposed framework for water management in data-scarce regions.

Accurate estimation of actual evapotranspiration (ETa) is essential for sustainable water resources management and agricultural planning. This study investigated the influence of reference evapotranspiration (ET₀) on ETa estimation using Random Forest (RF), Bagging, Least Squares Boosting (LSBoost), Generalized Additive Models (GAM), and Multiple Linear Regression (MLR). Global solar radiation (Rs), land surface temperature (LST), NDVI, and soil moisture (SM) were used as predictors under two scenarios: with and without ET₀. Variable importance and model interpretability were evaluated using Permutation Feature Importance (PFI) and SHAP analyses. Results showed that ET₀ was the dominant predictor when included in the models, exhibiting the highest importance and contribution values across all approaches. When ET₀ was excluded, the explan-atory structure shifted primarily toward Rs, followed by SM, LST, and NDVI. Although model accuracy generally decreased without ET₀, RF and Bagging maintained stable performance (R² ≈ 0.97), whereas LSBoost, GAM, and MLR exhibited more pronounced reductions in predictive accuracy. The analyses further revealed that SM consistently enhanced model performance by reducing prediction errors and improving robustness, particularly in the absence of ET₀. Overall, the findings demonstrate that both model selection and the physical representativeness of input variables are critical for reliable ETa estimation. Moreover, ET₀ drives an atmosphere-controlled evaporative demand regime, whereas Rs becomes the primary controlling factor when ET₀ is unavailable.

Accurate prediction of TBM penetration rate (ROP) is of critical importance for the planning of tunneling operations and performance assessment. In this study, both classical multiple linear regression (MLR) and machine learning approaches—namely Random Forest, Bagged Trees, Support Vector Machine, and LSBoost—were employed to investigate the contributions of BI, UCS, DPW, α, and BTS parameters to ROP prediction. Univariate and MLR analyses exhibited limited explanatory power (R² = 0.365), confirming that ROP is governed by complex, multivariate, and nonlinear interactions. Comparative machine learning analyses revealed that LSBoost provides the most reliable predictions, achieving the highest accuracy (R² = 0.9565) and the lowest error metrics (RMSE = 0.1794; MAPE = 5.63%) for both original and normalized datasets. While Random Forest and Bagged Trees demonstrated comparable performance, SVM showed limited predictive capability on the original dataset (R² = 0.452; RMSE = 0.637; MAPE = 18.60). However, its performance improved substantially following data normalization, approaching that of LSBoost (R² = 0.936; RMSE = 0.218; MAPE = 4.87). Feature importance analyses based on PDP-driven Jacobian sensitivity and SHAP methods indicate that UCS, BI, and DPW are the dominant factors governing TBM penetration performance, while also demonstrating that model outputs remain interpretable in an interaction-aware manner. These findings highlight that machine learning-based approaches can deliver both reliable prediction and interpretability even with small and heterogeneous datasets, and suggest that future research should focus on integrating larger datasets, hybrid modeling strategies, and advanced explainability techniques.

This study evaluated the predictive performance of Random Forest, Bagged Trees, Support Vector Machines (SVM), and Least Squares Boosting (LSBoost) for estimating Tunnel Boring Machine (TBM) penetration rate (ROP). While all models achieved acceptable accuracy, LSBoost outperformed the others, showing the highest correlation (R = 0.965) and coefficient of determination (R² = 0.909), along with the lowest RMSE and MAE. Its performance remained robust after Z-score normalization, highlighting its ability to capture nonlinear parameter interactions and generalize well on limited geotechnical datasets. Random Forest and Bagged Trees showed similar performance, with Bagged Trees only slightly improved by normalization. SVM performed less effectively, indicating limited capacity to model complex TBM penetration behavior. Feature importance and SHAP analyses identified discontinuity spacing (DPW) and uniaxial compressive strength (UCS) as the primary controlling factors, while brittleness index (BI) was more influential within the SVM model. Agreement between Jacobian-based derivative analyses and SHAP results confirmed both mathematical sensitivity and engineering interpretability. Overall, TBM penetration prediction is a multivariate and inherently nonlinear problem. LSBoost provides reliable and high-accuracy predictions even under data-constrained conditions. The combination of SHAP- and PDP-based feature importance analyses enhances interpretability, supporting engineering decision-making in TBM design and operation. These findings emphasize the applicability of machine learning approaches for accurate, interpretable, and robust TBM performance prediction.

In this study, linear and nonlinear parametric models (M1–M6) were jointly evaluated alongside machine learning (ML)-based approaches to achieve reliable and interpretable prediction of the penetration rate (ROP) of tunnel boring machines (TBMs). The analyses incorporate key geomechanical and structural variables, including the brittleness index (BI), uniaxial compressive strength (UCS), mean spacing of weakness planes (DPW), the angle between the tunnel axis and weakness planes (α), and Brazilian tensile strength (BTS). The coefficients of the parametric models were optimized using the Differential Evolution (DE) algorithm. Variable effects were systematically examined through Jacobian-based elasticity analysis under both original and standardized data scenarios. The results indicate that the M6 model, which explicitly incorporates interaction terms, delivers superior predictive accuracy and a more balanced, physically meaningful representation of variable contributions compared to widely used parametric formulations reported in the literature. While the dominant influence of BI and UCS on ROP is consistently preserved across all models, the indirect contributions of variables such as DPW and BTS are more clearly revealed in M6 owing to its interaction-based structure. Model performance improves systematically with increasing complexity, with the coefficient of determination (R²) rising from 0.62 for M1 to 0.69 for M6. Relative to the linear model, M6 achieves a 9.07% reduction in RMSE and a 10.48% increase in R², while providing additional improvements of 2.47% in RMSE and 2.37% in R² compared with the closest competing model. ML-based variable importance analyses are largely consistent with the parametric findings, highlighting BI and α in tree-based models, and UCS and α in SVM and GAM frameworks. Notably, the GAM exhibits the highest predictive performance under both data scenarios. Overall, the integrated use of parametric and ML approaches establishes a robust hybrid modeling framework that enables highly accurate and engineering-interpretable prediction of TBM penetration rate.

Accurate prediction of the tunnel boring machine (TBM) rate of penetration (ROP) is essential for improving excavation efficiency, project planning, and cost management. This study proposes an integrated machine learning and explainable artificial intelligence (XAI) framework for ROP prediction using Random Forest (RF), Least Squares Boosting (LSBoost), Bagged Trees (BT), and the Generalized Additive Model (GAM). To evaluate the robustness and applicability of the proposed framework, analyses were conducted using two independent geotechnical datasets from the New York Queens and Tehran–Karaj tunnel projects. Multicollinearity was addressed through Variance Inflation Factor (VIF)-based feature reduction, and model performance was evaluated using five-fold cross-validation under both full- and reduced-feature scenarios. The results demonstrate that ensemble learning models consistently provide high predictive accuracy and stable performance across both datasets. RF and LSBoost achieved the most reliable predictions, while BT also exhibited strong predictive capability with limited sensitivity to feature reduction. Explainability analyses based on SHapley Additive exPlanations (SHAP), Permutation Feature Importance (PFI), Partial Dependence Plots (PDP), and Local Interpretable Model-Agnostic Explanations (LIME) revealed mutually consistent interpretations of model behaviour. The dominant explanatory variables differed between datasets, with the rock brittleness index (BI) being the primary predictor in the New York Queens dataset, whereas Distance Between the Planes of Weakness (DPW), together with rock strength parameters, played a more influential role in the Tehran–Karaj dataset, highlighting the influence of geological conditions on ROP prediction. A key contribution of this study is the systematic investigation of the basis dimension (K) parameter in GAM. Although VIF-based feature reduction initially resulted in substantial performance degradation, sensitivity analyses demonstrated that increasing spline flexibility markedly improved predictive performance in both datasets. These findings indicate that appropriate specification of the basis dimension is critical for achieving reliable GAM performance and that the observed deterioration under reduced-feature scenarios primarily reflects insufficient model capacity rather than an inherent limitation of the GAM methodology. Overall, the proposed framework provides a reliable and interpretable approach for TBM ROP prediction. The results demonstrate the value of integrating machine learning with explainable artificial intelligence while highlighting the importance of hyperparameter optimization and independent validation for developing robust predictive models across different geotechnical conditions.

Accurately estimating actual evapotranspiration (ETa) is essential for sustainable water management, particularly in semi-arid regions. Although the SAFER algorithm provides a practical remote sensing-based approach, its sensitivity to parameter settings and reduced performance during dry periods limit its reliability. This study develops four parametric ETa models—two linear (LM-I, LM-II) and two nonlinear (NLM-I, NLM-II)—and recalibrates SAFER coefficients via a simulation/optimization (S/O) approach. Models were evaluated using Landsat-8 data (LST, NDVI, α ₛ ) and reference evapotranspiration (ETo), and compared with machine learning methods: Random Forest (RF), Bagged Trees (BT), Support Vector Machines (SVM), and Generalized Additive Models (GAM). Results indicate that nonlinear models better capture the physical behavior of ET processes and outperform linear models across key metrics. In particular, the NLM-II model achieved R² = 0.8295 and RMSE = 0.4913 on the test set, surpassing SAFER (R² = 0.8195, RMSE ≈ 0.5713), LM-II, and the best soft computing model, BT (R² = 0.8137, RMSE = 0.5084). Its physically grounded structure ensures stable, interpretable predictions that accurately reflect water–energy interactions and seasonal dynamics. These findings demonstrate that compact, physically based nonlinear parametric models provide a robust, operationally practical solution for ETa estimation under sparse Landsat-based datasets, outperforming both linear and black-box machine learning approaches.

In this study, a Support Vector Regression (SVR) model was developed to predict the rate of penetration (ROP) during tunnel excavation, and its hyperparameters were optimized using Grid Search (GS), Random Search (RS), and Bayesian Optimization (BO). The results indicate that BO reached the optimal parameter set with only 30–50 evaluations, whereas GS and RS required approximately 1000 evaluations. In addition, BO achieved the highest predictive accuracy (R² = 0.9625) while reducing the computational time from 25.83 s (GS) to 17.31 s. Compared with the baseline SVM model, the optimized SVR demonstrated high accuracy (R² = 0.9610–0.9625), strong stability (NSE = 0.9194–0.9231), and low error levels (MAE = 0.0927–0.1099), clearly highlighting the critical role of hyperparameter optimization in improving model performance. To enhance interpretability, a feature importance analysis was conducted using four machine learning methods: Random Forest (RF), Bagged Trees (BT), Support Vector Machines (SVM), and the Generalized Additive Model (GAM). The relative contributions of BI, UCS, ALPHA, and DPW to ROP were evaluated, providing clearer insight into the model’s decision-making process and enabling more reliable engineering interpretation. Overall, integrating hyperparameter optimization with feature importance analysis significantly improves both predictive performance and model explainability. The proposed approach offers a robust, generalizable, and scientifically sound framework for TBM operations and geotechnical modeling applications.