Katipoğlu, Okan Mert; Kartal, Veysi; Akiner, Muhammed Ernur; Terzioğlu, Zeynep Özge; Kılıç, Zeyneb

Soil moisture is vital for water resource management and agricultural productivity. With the growing availability of reanalysis data and advanced machine learning methods, it is essential to explore reliable, scalable models for long-term estimation. Comparing different algorithms enables a comprehensive assessment of their strengths and limitations in modeling complex soil moisture dynamics under varying climatic and agricultural conditions. In this study, soil moisture data for the period 1950–2022 from six stations in the Konya Closed Basin of Türkiye were estimated using Echo State Networks (ESN), Adaptive Boosting (AdaBoost), Deep Neural Networks (DNN), and Light Gradient Boosting Machine (LGBM). Precipitation (PR), solar radiation (RSDS), wind speed (WS), mean temperature (Tave), maximum temperature (Tmax), minimum temperature (Tmin), and potential evapotranspiration (ET0) values calculated by the Thornthwaite equation were used as input variables of the models, and soil moisture (SM) was determined as the output variable. Model performance was evaluated using statistical metrics and visualizations derived from training, validation, and test datasets. In addition, a feature selection approach was used to identify the most effective parameters for soil moisture estimation. The analysis showed that the DNN model achieved the highest accuracy, with R2 values ranging from 0.87 to 0.97. The second most successful model was AdaBoost or ESN, while the LGBM model showed higher deviations than the others. These findings reveal that artificial intelligence-based models can effectively estimate soil moisture.

Kartal, Veysi; Katipoğlu, Okan Mert; Karakoyun, Erkan; Simsek, Oguz; Yavuz, Veysel Süleyman; Ariman, Sema

This study aims to predict groundwater drought-based meteorological drought index using machine learning instead of traditional approaches. Groundwater drought (GWD) was predicted using machine learning methodologies such as Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), Support Vector Machine (SVM), Random Forest (RF), Least Squares Boosting Tree (LSBT), Generalized Linear Regression (GLR) and k-Nearest Neighbours (KNN). In addition, monthly, seasonal, and annual drought indices such as the Standardised Precipitation-Evapotranspiration Index (SPEI), China Z Index (CZI), Standardised Precipitation Index (SPI), Z-Score Index (ZSI), Decile Index (DI), Percent of Normal Index (PNI) and Rainfall Anomaly Index (RAI) were used to analyse the drought of groundwater. These traditional drought indices were modified for the assessment of groundwater drought. Moreover, groundwater drought was predicted based on the hydro-meteorological parameters (temperature, relative humidity, wind speed, rainfall, groundwater level). The applied models’ performances were evaluated via Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Akaike Information Criterion (AIC), R-squared (R2), Mean Bias Error (MBE), Bias Factor, and Variance Account Factor (VAF). Linear SVM is generally the best model for predicting GWD, while GLR is the second-best performing model. The KNN algorithm obtained the weakest performances. Although all types of drought and wet categories were observed, normal drought occurred more than in the other drought and wet categories. This study can contribute to the assessment of groundwater drought in regions where there is no groundwater station.

Katipoğlu, Okan Mert; Ertugay, Neşe; Elshaboury, Nehal; Aktürk, Gaye; Kartal, Veysi; Pande, Chaitanya Baliram

Drought is one of the costliest natural disasters worldwide and weakens countries economically by causing negative impacts on hydropower and agricultural production. Therefore, it is necessary to create drought risk management plans by monitoring and predicting droughts. Various drought indicators have been developed to monitor droughts. This study aims to forecast Streamflow Drought Index (SDI) values with various novel metaheuristic optimization-based Artificial Neural Network (ANN) and deep learning models to predict 1-month lead-time hydrological droughts on 1, 3, and 12-month time scales in the Konya closed basin, one of the driest basins in Turkey. To achieve this goal, the ANN model was integrated with the Firefly Algorithm (FFA), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO) techniques and compared against long short-term memory (LSTM) networks. While establishing the SDI prediction model, lag values exceeding the 95% confidence intervals in the partial autocorrelation function graphs were used. Model performance was evaluated according to scatter matrix, radar, time series, bee swarm graphs, and statistical performance metrics. As a result of the analysis, the PSO-ANN hybrid model with (R2:0.468–0.931) at station 1611 and the FFA-ANN hybrid model with (R2:0.443–0.916) at station 1612 generally have the highest accuracy.