Okan Mert Katipoğlu

Despite increasing climate variability and anthropogenic pressures in the eastern Mediterranean, integrated approaches for assessing soil erosion susceptibility remain limited. This study evaluates soil erosion susceptibility in the Ghamqa River Basin, western Syria, a climate-sensitive coastal watershed. Ten conditioning factors were selected based on regional characteristics, literature review, field observations, and expert knowledge. Thematic raster layers were standardized, weighted through expert pairwise comparisons, and integrated to delineate erosion susceptibility zones. Model performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUC–ROC). The results identified slope as the dominant conditioning factor (weight = 0.218), followed by soil texture (0.195), rainfall (0.155), and land use/land cover (0.122). Approximately 35% of the basin was classified as high to very high susceptibility, while nearly 50% fell within moderate to high susceptibility classes. The AHP-based model achieved an AUC of 0.867, indicating very good performance. These findings highlight the combined influence of topographic, climatic, soil, and land-use factors on erosion processes and identify priority areas for soil conservation. By supporting measures such as slope stabilization, vegetation restoration, and sustainable land-use planning, the proposed GIS-based AHP framework contributes to climate-resilient watershed management and strengthens decision-making for reducing land degradation in Mediterranean coastal environments.

Hydrological drought forecasting in semi-arid basins is challenging due to the combined influence of meteorological forcing, large-scale atmospheric teleconnections, and basin memory processes, which are rarely jointly analysed within a leakage-free predictive framework. This study addresses this gap by evaluating gradient-boosted trees and neural forecasting models for one-month-ahead prediction of the Standardized Runoff Index (SRI) in two sub-basins of the Wadi Sahaouat Basin, Algeria. The models include gradient-boosted regression trees (GBRT), A-N-BEATS, A-N-HiTS, and TiDE, representing distinct forecasting architectures. Predictors consist of the Standardised Precipitation Index (SPI), seven teleconnection indices (NAO, AO, EAWR, SCAND, MEI, SOI, WeMO), and their one- to three-month lags. Two scenarios are tested: Scenario 1 uses SPI and teleconnection lags only, while Scenario 2 additionally includes lagged SRI values (SRI_lag1–3) to represent hydrological memory. A train-only Variance Inflation Factor (VIF > 10) procedure is applied to remove multicollinearity without data leakage. In Basin 1, SRI lags were excluded due to strong collinearity with SPI lags (r = 0.984), resulting in identical inputs for both scenarios. In Basin 2, SRI lags were retained to assess their predictive contribution. GBRT achieved the best overall performance across both basins and scenarios, with mean RMSE, NSE, and KGE values of 0.0682, 0.9907, and 0.8945, respectively. TiDE ranked second overall, with a mean RMSE of 0.1166, followed by A-N-HiTS in third place with a mean RMSE of 0.1203 and A-N-BEATS with the weakest overall performance, with a mean RMSE of 0.2159. These results indicate that gradient-boosted trees remain highly competitive with neural models for small monthly hydrological datasets and that the value of hydrological memory is basin-dependent and varies according to its independence from concurrent meteorological forcing.

This study developed an innovative hybrid modelling approach to predict one-month-ahead water levels (LWL (T + 1)) for two major lakes in Türkiye (Eber and Eğirdir), aiming to model complex dynamics of hydrological systems and optimize sustainable water resources management. In the proposed methodological framework, Singular Spectrum Analysis (SSA) was employed to decompose hydrometeorological time series into sub-components, while Mutual Information (MI) and Partial Autocorrelation Function (PACF) techniques were utilized to determine optimal input variables. The hybrid model architecture, incorporating Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and CatBoost algorithms, was optimized using 70% of the data for training and 30% for testing, and its performance was evaluated through deterministic and probabilistic metrics. The analyses revealed that SSA-based data preprocessing significantly improved the performance of prediction models, and depending on lake characteristics, CNN-based models demonstrated superior performance for Lake Eber, while DNN architecture excelled for Lake Eğirdir. The developed hybrid approach presents an original methodology that can be integrated into decision support systems for water resources management, contributing to the development of sustainable water management strategies and adaptability to similar hydrological systems.

Getting accurate daily estimates of reference evapotranspiration (ET₀) is essential for hydrological modeling, irrigation planning, and climate research. This study evaluated the performance of four machine learning models: Adaptive Boosting (AdaBoost), Categorical Boosting (CatBoost), Deep Neural Network (DNN), and Light Gradient Boosting Machine (LGBM) for daily ET₀ prediction. Feature selection was implemented using Partial Autocorrelation (PACF) and Autocorrelation (ACF) analysis on the basis of an 8-day lag structure, while the CatBoost feature importance framework was implemented to identify the most important inputs. Moreover, the methodologies of decomposition-based preprocessing, Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD) were implemented on the lagged inputs to build the hybrid models augmented with intrinsic constituents. These augmented hybrid models were created to improve the data representation as well as predictive capability. Model performance was evaluated using key statistical metrics, including RMSE, MAE, NSE, KGE, R2, and bias-related indicators. CatBoost produced the lowest RMSE (1.57 mm/day) and MAE (1.13 mm/day), and the highest NSE (0.67), KGE (0.78), and R2 (0.67), along with the lowest AIC score (627.15), indicating strong and reliable predictive performance. Furthermore, the VMD based CatBoost model yielded even greater improvements, highlighting the added value of decomposition enhanced learning. On the contrary, DNN showed the weakest results, with the highest RMSE (2.39 mm/day) and lowest NSE (0.23). These findings suggest that CatBoost and its hybrid variants are highly effective and reliable tools for daily ET₀ prediction, particularly in data limited environments and mountainous regions.