ABSTRACT Groundwater salinization is a major cause of groundwater quality degradation in semi‐arid regions, primarily driven by evaporation and water–rock interaction. To better understand these processes, this study investigates the hydrochemical controls on groundwater mineralization in the Oued El Hjal basin (central Tunisia), where the mean total dissolved solids (TDS) concentration reaches 3111 mg/L, exceeding the World Health Organization (WHO) drinking water guideline. A combined multiple linear regression (MLR) and partial least squares structural equation modelling (PLS‐SEM) approach, supported by saturation indices and Piper and Gibbs diagrams, was applied to identify the main salinization drivers. MLR identified Cl − , SO 4 2− , Na + , Ca 2+ and Mg 2+ as the principal contributors to TDS variability. PLS‐SEM confirmed strong associations with TDS, with Cl − showing the highest standardized effect ( β = 0.954), followed by Mg 2+ , Ca 2+ , Na + and SO 4 2− , indicating dominant evaporite dissolution. Saturation indices confirmed active halite and gypsum dissolution, while Piper and Gibbs diagrams classified groundwater as Ca–Mg–Cl–SO 4 type, governed by rock–water interaction and evaporative concentration. Nitrate exceeded WHO guidelines but was not significantly associated with mineralization. This integrated framework supports groundwater salinity assessment and irrigation water management in semi‐arid aquifers.
ABSTRACT The flow regime index characterizes the sensitivity of emitter discharge to pressure variation. This study developed an interpretable machine‐learning framework to predict the flow regime index. Support vector regression (SVR), backpropagation (BP) neural networks and extreme learning machine (ELM) models were compared. The kernel and activation functions were screened using fivefold cross‐validation, and particle swarm optimization (PSO) was used to optimize the model hyperparameters. Performance was evaluated using R 2 , RMSE, MAE, MBE and CI. The results indicate that PSO‐BP 1 showed the best overall performance. A representative PSO‐BP 1 model selected according to the internal‐validation results achieved R 2 , RMSE, MAE, MBE and CI values of 0.900, 0.0010, 0.00080, 2.0 × 10 −5 and 1.6 × 10 −12 , respectively. In external validation, PSO‐BP 1 and PSO‐BP 2 both achieved an RMSE of 0.0010 and an MAE of 0.00070, providing preliminary evidence of predictive performance within the investigated parameter range. SHAP analysis identified the spacing between water‐retaining components ( f ) as the most influential input, followed by the vertex angle of the water‐retaining component ( θ 2 ); the upper‐wall spacing ( c ) and flow‐splitting‐component vertex angle ( θ 1 ) contributed less. Strong interactions occurred between θ 1 and θ 2 and between c and θ 2 . The framework supports rapid emitter screening and quantifies model‐based contributions of geometric parameters.
ABSTRACT This study integrates groundwater chemistry and climatic stressors to evaluate irrigation suitability across 33 districts of Telangana, India. Groundwater monitoring data (2018–2020) were combined with rainfall and temperature datasets from the India Meteorological Department (IMD). Key indices included sodium adsorption ratio (SAR), residual sodium carbonate (RSC), total dissolved solids (TDS), the composite irrigation suitability index (CISI+) and the climate extremes index (CEI). Most districts were safe under RSC, with only three marginal cases. Sodicity emerged as the dominant hazard: SAR correlated almost perfectly with CISI+ ( r = 0.97), while TDS showed strong associations ( r = 0.72). Rainfall deficits were linked to higher maximum temperatures ( r = −0.44) and CEI values ( r = −0.32), highlighting drought‐prone districts as more vulnerable. CEI values were generally low, but Jogulamba Gadwal, Nagarkurnool and Sangareddy recorded localised extremes. Random forest achieved perfect classification (accuracy, precision, recall, F1 = 1.0), outperforming gradient boosting, support vector machine and multilayer perceptron. SHAP analysis identified Jogulamba Gadwal, Nalgonda and Rajanna Sircilla as sodicity hotspots, while rainfall contributions were minimal. Practical implications include district‐specific adaptation strategies such as gypsum application, groundwater blending, salt‐tolerant crops, rainwater harvesting and recharge conservation. These measures are essential to safeguard irrigation resilience under evolving climate pressures.
ABSTRACT Irrigation water productivity (IWP) was assessed for major field crops, vegetables and orchards under semiarid conditions in Khak‐e‐Jabbar District, Afghanistan. Field measurements, farmer interviews and agricultural records provided 382 crop‐specific observations. IWP differed significantly among crops (Kruskal–Wallis, p < 0.001). Tomato had the highest mean IWP (3.44 kg/m 3 ), followed by onion (3.23 kg/m 3 ) and plumcot (2.79 kg/m 3 ), whereas beans had the lowest value (0.49 kg/m 3 ). Vegetables showed the highest group‐level IWP, followed by orchards, cereals and legumes. A comparison with FAO (2017) reference ranges indicated that tomato, potato, maize and apple remained below their respective benchmarks. These differences show the need for crop‐specific irrigation management in semiarid areas.
ABSTRACT Accurate estimation of evapotranspiration (ET) is essential for sustainable water management, particularly in arid and semiarid regions. This study aimed to establish relationships between the normalised difference vegetation index (NDVI) derived from satellite imagery and stress‐adjusted crop coefficients ( K c_adj ) obtained using the METRIC (mapping ET at high resolution with internalised calibration) model and to develop efficient models for rapid ET estimation. The study was conducted during the 2023 and 2024 growing seasons in Ergani (Diyarbakır, Türkiye). The water stress coefficient ( K s ) was derived from METRIC‐based ET and used to obtain K c_adj values. Regression analysis was performed to quantify the relationships between NDVI and K c_adj , and NDVI‐based K c ( K c_NDVI ) values were developed. The results indicated strong relationships between NDVI and K c_adj for maize and tobacco ( R 2 = 0.90–0.93 and 0.89–0.95, respectively), whereas the resulting NDVI‐based ET estimates showed strong internal consistency with the METRIC‐informed stress‐adjusted estimates. Conversely, weaker relationships were observed for potato and onion ( R 2 = 0.52–0.58 and 0.0003–0.56, respectively), likely driven by planting–harvest timing heterogeneity, incomplete satellite coverage and the influence of stress factors. Overall, the approach is most suitable for crops showing strong and stable NDVI– K c_adj relationships under adequate temporal data coverage conditions.
