Mahmoudreza Tabatabaei; Mohammadreza Gharib Reza
Abstract
IntroductionAccurate estimation of suspended sediment in high-sediment rivers, especially in semi-arid regions such as the Golestan River watershed, poses significant challenges in water resource management and sediment control in reservoir dams. The increase in sediment concentration not only affects ...
Read More
IntroductionAccurate estimation of suspended sediment in high-sediment rivers, especially in semi-arid regions such as the Golestan River watershed, poses significant challenges in water resource management and sediment control in reservoir dams. The increase in sediment concentration not only affects water quality but also leads to considerable economic and environmental damage by reducing the lifespan of hydraulic structures and altering river morphology. In this context, the Atrak River, as one of the most important sediment sources in northeastern Iran, serves as a prominent example of these challenges. Given the limitations of traditional direct measurement methods and labor-intensive physical models, the development of high-accuracy data-driven models emerges as an efficient solution for monitoring and predicting sediment. Materials and methodsIn this study, to estimate suspended sediment in the Atrak River at the Hootan gauging station, a combination of classical and intelligent methods was employed, including: (1) sediment rating curves (using both the midpoint and linear methods), (2) neural networks (MLP and SOM), (3) deep learning models, and (4) ensemble learning methods. The modeling process was carried out in three main stages: first, using the Random Forest algorithm, the key variables affecting sediment (including flow rate, daily precipitation, and their lagged values) were identified. Then, the data were divided into homogeneous groups through clustering, allowing for balanced sampling from each cluster to create homogeneous training (70%), evaluation (15%), and testing (15%) datasets. To enhance the model’s efficiency, various strategies were employed, including optimizing objective functions and managing data skewness. After a comprehensive evaluation of the models using standard criteria, the ensemble learning model XGBoost was selected as the best model, which was ultimately used to reconstruct and complete suspended sediment data for a 40-year period (1982-2021). Results and discussionThe comparison of the results of various models in this study indicated that the ensemble learning model (XGBoost) was recognized as the selected model among other data-driven models for estimating suspended sediment in the Atrak River, having the lowest error rates (MAE of 10,652 tons per day and RMSE of 36,219 tons per day) and the highest performance index (NSE of 0.87). This model not only outperformed other machine learning methods (with the MLP neural network achieving NSE=0.80 and the deep learning model achieving NSE=0.84), but it also demonstrated significant superiority over classical methods such as the midpoint sediment rating curve (with MAE of 20,909 tons per day and RMSE of 45,632 tons per day). A detailed analysis of the results indicates that this superiority is primarily due to the ability of the XGBoost model to manage highly skewed data and identify complex nonlinear relationships between hydrological variables. In this regard, the model’s bias (PBias) decreased significantly by approximately 15.5% from -18.93 in the sediment rating curve model to -3.52 in the XGBoost model, supporting this point. On the other hand, a comparative analysis with similar studies in other high-sediment rivers shows that the combined approach used in this research (utilizing clustering before model training and optimizing objective functions) has led to improved predictions. However, it is important to note that the effectiveness of these models largely depends on the quality and completeness of the input data, and long-term hydrological changes may necessitate periodic recalibration (training) of the models. These limitations provide a foundation for future research aimed at developing models that are more adaptable to environmental changes. ConclusionThis research examined the performance of data-driven models in estimating suspended sediment in the Atrak River at the Hootan gauging station, yielding significant results. The ensemble learning model XGBoost was identified as the best option, demonstrating high accuracy and minimal error in predicting changes in suspended sediment. These findings highlight the importance of utilizing advanced machine learning techniques in analyzing hydrological data and emphasize the need for a connection between data science and water resource management. Key points in this study include the high skewness of suspended sediment data and the specific hydrological conditions of the Atrak River, which arise from gully erosion and other natural phenomena in the region. This skewness and the sharp fluctuations in sediment concentration pose major challenges for accurate modeling. However, the combined approach used in this study provided a significant improvement in prediction accuracy compared to traditional and classical methods. This indicates that employing clustering and data optimization can serve as effective tools in future research. Nonetheless, the results of this study should be interpreted with caution, as the quality of data and environmental changes may impact model performance. Therefore, this research not only aids in a better understanding of sedimentation processes but also lays the groundwork for future studies aimed at enhancing and developing prediction models that are more adaptable to changing environmental conditions. In the end, the results of this research can serve as a model for simulating suspended sediment in other hydrometric stations across the country, providing valuable insights for researchers and experts in this field.
Mehri Dinarvand; Saba Peyrov; Seyed hossein Arami; Behzad Tajari; Kohzad Heidari
Abstract
IntroductionIran is located in the arid and semi-arid belt of the world and is very far from moisture sources. Arid and desert areas, due to a lack of moisture, high temperatures, strong winds, soil erosion, and land degradation caused by human activity, have created tough conditions for plant growth ...
Read More
IntroductionIran is located in the arid and semi-arid belt of the world and is very far from moisture sources. Arid and desert areas, due to a lack of moisture, high temperatures, strong winds, soil erosion, and land degradation caused by human activity, have created tough conditions for plant growth and development, such that only a relatively limited number of plant species can survive. Native plants of such areas are considered highly valuable species due to their ability to adapt to harsh environmental conditions and play a crucial role in the region's climate, soil formation, and hydrology; therefore, their identification is of great importance. With the aim of monitoring and recording spatial data statistics of meteorological, hydrometric, erosion and sedimentation, vegetation cover, soil and groundwater climatic parameters, the Shush representative basin station was established in 2007 by the General Directorate of Natural Resources and Watershed Management of Khuzestan Province. In addition to the Moorlands, shallow depressions and old gullies are observed in this basin, some of which have been stabilized due to enclosure and reduction of livestock pressure. These stabilized depressions themselves act as natural micro-reservoirs and have provided suitable conditions for the establishment of permanent species by increasing infiltration, reducing surface runoff, and trapping plant seeds. In this study, the floristic composition, richness, and species diversity were compared in plots located in stabilized micro-watersheds (treatment) and hills (control). The results of the study showed that the indices of percent cover, number of species, and diversity in the treatment were significantly higher than in the control. These findings indicate that stabilized gullies can act as small foci of gene pool and nurse habitats and, if properly managed, can help improve the stability of vegetation cover in arid regions and consequently soil conservation. Therefore, integrated management, including targeted enclosure, control of active gullies, and patchy protection of natural micro-watersheds, can be a low-cost and eco-friendly model for restoring degraded rangelands in southwestern Iran. Materials and methodsIn this study, vegetation cover analysis was conducted in the Shoosh representative basin area using biodiversity indices. During the appropriate growing season (early February to late March), during field visits, a list of plant species in the area was taken, and typification was performed based on the presence of Shrub and perennial species. In each type, the size of the data plot was determined using the minimum area method and using snail plots and species surface curves. The location of all plots was recorded with a GPS device. 20 plots (a total of 40 plots) were constructed separately in the shallow depressions and stabilized gullies (treatment) and the slopes of the hillocks (control). In each plot, the percentage of living cover of each species was noted with the name, the percentage of litter and debris, and the percentage of bare ground. Species diversity indices were calculated, and the vegetation cover of the plots was compared using an independent t-test. Species number, vegetation cover percentage, and species diversity indices (Simpson, Shannon, and Hill diversity indices) were calculated using PAST 2.17 software. Python version 3.13 programming environment was also used to analyze species diversity, compare different indices, and create some graphs. All data were transferred to an Excel file for subsequent statistical analysis using SPSS 27 software. Results and discussionIn the enclosed and protected area of the Shush representative area, 70 plant species belonging to 29 families were identified. The Fabaceae, Astraceae, Poaceae, and Brassicaceae families are the largest families in the region in terms of number of species, with 11, 8, 8, and 6 species, respectively. Of the 70 identified species, 45 plant species, such as Astragalus ensifer, Astragalus fasiculifolius, and Astragalus obtusifolius, have soil conservation value and are among the perennial plants of the region. In the Shush representative area, there is a significant difference at the one percent level between the average number of species, percentage of cover, dominance, Simpson, and Shannon indices in the treatment plots compared to the control plots. However, although there is a difference between the value of the uniformity index of the treatment and control sections, it is not significant. These results indicate that stabilized depressions, by increasing water infiltration and reducing surface runoff, have provided conditions for the establishment of permanent species. From a soil conservation perspective, increasing living cover and the presence of permanent species in these micro-catchments reduce the bare ground surface and, as a result, reduce the erosion. Therefore, if properly managed, these depressions can help improve soil conservation and vegetation stability as nurse habitats and gene pools; however, a distinction should be made between active erosional gullies and stabilized gullies with a positive role. ConclusionsNatural watershed management phenomena with self-organization play a very important role in the restoration of degraded rangelands. These measures help improve soil structure and increase water infiltration by reducing surface water runoff and controlling soil erosion, and provide suitable conditions for seed germination and the establishment of plant species. This research in the representative area of Shoosh as a model example can be generalized to other arid regions. The presence of shallow depressions and stabilized gullies with various depths in the region, like a mother, is actually a nurse for plant species and allows them to develop in that region. However, the rate of restoration of arid regions is slow, and these areas have a very fragile system, and more time and opportunity are needed to achieve ideal conditions. In other words, despite the emergence of suitable vegetation due to the micro-natural watersheds, they are still very sensitive and vulnerable areas, and the diversity of vegetation depends on the rainy season due to the predominance of short-lived species. Therefore, an important point that should be considered is management, planning, and the use of existing conditions. In addition to preventing destruction in the area, purposeful fencing and grazing management (appropriate grazing time, prevention of excessive grazing, timely removal of livestock from the area, resting the pasture for reconstruction and restoration) should be considered in the area. Unfortunately, in many desert and arid areas, tree planting is used for restoration, an activity that is very costly and requires subsequent care. While the best method is to take inspiration from the natural process in desert and arid areas. The results of this study showed that the stabilized depressions in the Shoosh representative area, by playing the role of natural micro-reservoirs, provided conditions for the establishment of permanent species and were effective in reducing bare land and improving soil protection. These results indicate that utilizing the capacity of these micro-reservoirs, along with targeted enclosure and grazing management, can be a low-cost and eco-friendly approach to restoring degraded rangelands in arid regions.
Rahim Kazemi; Bagher Ghermezcheshmeh
Abstract
Introduction
Baseflow, as the delayed portion of river flow, mainly supplied from subsurface water and groundwater resources, plays a key role in river flow sustainability, water resources management, aquatic ecosystem protection, and drought adaptation planning. The base flow index (BFI), which is ...
Read More
Introduction
Baseflow, as the delayed portion of river flow, mainly supplied from subsurface water and groundwater resources, plays a key role in river flow sustainability, water resources management, aquatic ecosystem protection, and drought adaptation planning. The base flow index (BFI), which is the ratio of base flow volume to the total streamflow volume, is a dimensionless measure for assessing the contribution of subsurface water to stream flow. Global studies show that, on average, more than half of river flow is provided by base flow, and BFI depends more on the inherent characteristics of the basin, such as geology, soil, topography, and morphology, than on precipitation. Despite numerous studies on baseflow estimation in Iran, limited attention has been paid to investigating its temporal changes and long-term trends at different time scales. Therefore, the aim of this study is to investigate changes in BFI on monthly, seasonal, and annual time scales, as well as analyze the long-term trend and the effect of human interventions in Barun Chay river basin.
Materials and methods
The study area is Barun Chay river basin with an area of 1025 km2, located in the Aras Basin in West Azerbaijan Province (Iran). This basin has a significant diversity in terms of permeability of geological formations, such that a major part of its surface is formed by geological formations with low and very low permeability. In this study, daily stream flow data from the Qale-Jogh hydrometric station of Barun Chay river basin were used for the period of 1976 to 2017. After extracting the physiographic characteristics of the basin using a geographic information system, the base flow was separated from the total flow hydrograph using the One Parameter Recursive Digital Filter technique and the BFI extension in the Hydro Office,2015, software. Then, time series of base flow and BFI were prepared on monthly, seasonal and annual scales. Descriptive statistical analysis and trend analysis were performed using the non-parametric Mann-Kendall method. Also, the effect of dam construction upstream of the hydrometric station on BFI was evaluated by comparing the periods before and after dam construction.
Results and discussion
The results showed that the long-term average of the annual BFI over a 41-year period was 0.552, indicating a more than 50 percent contribution of subsurface flow and groundwater to the river flow. On a seasonal scale, the highest value of BFI belongs to the winter and summer seasons and the lowest to the spring season. On a monthly scale, the maximum long-term BFI was observed in January and its minimum in May. Analysis of monthly changes in BFI showed that it had an increasing trend from the October to January and a decreasing trend from January to May, which coincided with the change in the precipitation regime and the delay in the participation of subsurface flows to stream flow. From May to September, with the decrease in precipitation, contribution of delayed flows increased and BFI showed an increasing trend again. Trend analysis of the long-term BFI using the Mann-Kendall test indicates that in most months, seasons and annual scales, the trend were negative and only in June and October an increasing trend was observed. Comparison of the annual average BFI in the periods before and after dam construction showed that the long-term average of BFI increased from 0.542 to 0.562, and the effect of human interventions resulting from dam construction on the annual BFI was estimated to be about 0.02.
Conclusion
The Overall results indicate the dominant role of subsurface flows in providing Barun Chay river flow and the greater dependence of BFI on the inherent characteristics of the basin than on precipitation changes. Understanding the temporal pattern and long-term trend of base flow can provide an effective basis for sustainable water resources management, dam operation planning, and developing drought adaptation strategies in similar watersheds. After determining BFI at different times and locations, and preparing a map of sustainable flows at regional and national scales, the initial prerequisite for developing plans appropriate to the hydrological capacity of each region will be achieved.
Moslem Borji Hassan Gaviar; Mohsen Farzin; Ali Akbar Nazari Samani
Abstract
Introduction
Gully erosion is one of the most important and destructive forms of soil degradation in arid and semi-arid regions of the world, playing a significant role in reducing agricultural land productivity, increasing sediment load in watersheds, and decreasing the useful life of dam reservoirs. ...
Read More
Introduction
Gully erosion is one of the most important and destructive forms of soil degradation in arid and semi-arid regions of the world, playing a significant role in reducing agricultural land productivity, increasing sediment load in watersheds, and decreasing the useful life of dam reservoirs. Among the different stages of gully evolution, tensile cracks in the gully wall act as early indicators of instability and play a key role in initiating wall collapse, lateral gully expansion, and accelerating erosion processes. Despite the high importance of this phenomenon in geomorphology and erosion management, quantitative and multivariate investigations of the factors controlling the occurrence and location of these cracks at the gully wall scale remain limited, and few field studies have specifically addressed the development of tensile cracks. The aim of this study is to identify the factors affecting the formation of tensile cracks and to determine the critical slope threshold in the gullies of the Kaloocheh Bijar watershed.
Materials and methods
In this study, 47 gullies (including 25 gullies with longitudinal cracks and 22 gullies without cracks) were surveyed and analyzed using UAV imagery with a vertical accuracy of 2 cm. The measured variables included wall slope, bank height, crack distance from the gully edge, bottom and top channel width, channel bed slope, gully planform curvature (outer, straight, inner), and vegetation cover status. The statistical analyses included independent t-test, Mann–Whitney U test, Kruskal–Wallis test, Fisher’s exact test, Spearman’s correlation, cluster analysis, and ROC curve analysis, all of which were performed using R and SPSS software.
Results and discussion
The results showed that wall slope is the most important factor controlling the occurrence of tensile cracks, such that the mean slope in cracked gullies (86.7 degrees) was significantly higher than in non-cracked gullies (50 degrees). Furthermore, outer curvature and lack of vegetation cover, with odds ratios of 4.2 and 4.57 respectively, were identified as the most important factors increasing the likelihood of crack occurrence. The crack distance from the gully edge (mean: 2.67 m) was mainly influenced by the geometric dimensions of the gully and showed a statistically significant positive correlation only with the bottom channel width. ROC analysis determined the critical wall slope threshold as 78.4 degrees (AUC = 0.835), with model sensitivity and specificity of 80% and 86.4%, respectively. Additionally, cluster analysis revealed three distinct patterns of cracked gullies, indicating different stages of instability evolution.
Conclusions
The findings of this study have direct implications for gully erosion management. First, wall slope can be used as a rapid and field-measurable indicator for prioritizing conservation measures. Gullies with a slope exceeding the critical threshold of 78.4 degrees should be prioritized for stabilization, as the probability of crack formation and failure propagation is significantly higher in these gullies. Second, volumetric measures without attention to slope reduction or stress concentration control have limited stability effects; therefore, engineering interventions should focus on reducing the effective wall slope or dissipating flow energy in outer curves. Third, although the role of vegetation cover was statistically weaker, mechanical evidence suggests that its removal can accelerate the failure threshold; thus, biological stabilization of walls should be considered as a complementary tool in gully management.
Negin Rashidi; Vahid Moosavi
Abstract
Introduction
Groundwater level modeling has always been a major challenge in groundwater resource management due to its complex, nonlinear, and spatiotemporally dependent nature. The main objective of this study was to develop and evaluate a distributed deep learning-based modeling approach for the ...
Read More
Introduction
Groundwater level modeling has always been a major challenge in groundwater resource management due to its complex, nonlinear, and spatiotemporally dependent nature. The main objective of this study was to develop and evaluate a distributed deep learning-based modeling approach for the simultaneous prediction of groundwater levels in a network of piezometric wells within the Kouchesfahan aquifer located in the Sefidroud watershed. In this framework, each well was considered as a node in a graph network, and the model predicted groundwater levels for each well at the next monthly time step (t+1). By integrating graph neural networks with convolutional structures and long short-term memory architectures, the study aimed to identify the spatial and temporal dependencies governing aquifer behavior and improve prediction accuracy compared to conventional methods.
Materials and methods
In this study, four graph-based deep learning architectures, including Graph Neural Network (GNN), Graph Attention Network (GAT), Graph Convolutional Network (GCN), and the hybrid GCN–LSTM model, were employed. Here, GNN refers to the baseline GCN-based graph neural architecture using the standard graph convolution operator. Each piezometric well was represented as a graph node, and the adjacency matrix was constructed using inverse-distance weighting. Input data included hydroclimatic variables, geological information, pumping characteristics, and spatial distances related to hydrogeological factors. Model performance was evaluated using R², RMSE, MAE, NSE, and KGE indices. In the proposed model, GCN layers were first applied to the spatial graph of wells to extract spatial features and hydraulic dependencies. The resulting hidden representations were then transferred to the LSTM network to model temporal dependencies and groundwater level dynamics.
Results and discussion
The results indicated that while the GNN model could partially reconstruct the general trends of groundwater level changes, it lacked accuracy in representing extreme behaviors and temporal dependencies. The GAT model showed a slight improvement over GNN but remained limited in extracting deep temporal patterns. In contrast, the GCN model demonstrated better performance in identifying spatial dependencies, leading to a significant improvement in evaluation metrics. The best performance was achieved by the GCN-LSTM model, which effectively represented both spatial and temporal features simultaneously, showing the highest overlap with observed data. This model reached an R2 of 0.896, RMSE of 0.336, and MAE of 0.269, indicating its high accuracy in predicting groundwater levels.
Conclusion
Hybrid graph-sequential architectures, especially GCN-LSTM, are highly effective for modeling complex aquifer hydrodynamics. This model predicted groundwater levels with high accuracy and outperformed purely graph-based models. Therefore, it is recommended for operational groundwater prediction and sustainable water management, serving as a novel framework to support better decision-making in water resources.
Golaleh Ghaffari
Abstract
IntroductionSuspended sediment transport constitutes a critical challenge in water resources engineering, causing reservoir capacity reduction, water quality degradation, and hydraulic infrastructure damage. Conventional sediment rating curve (SRC) methods rely on simplifying assumptions that frequently ...
Read More
IntroductionSuspended sediment transport constitutes a critical challenge in water resources engineering, causing reservoir capacity reduction, water quality degradation, and hydraulic infrastructure damage. Conventional sediment rating curve (SRC) methods rely on simplifying assumptions that frequently fail to capture nonlinear and transient flow-sediment dynamics. Recent advances in data-driven approaches, particularly artificial neural networks (ANNs) and adaptive neuro-fuzzy inference systems, have demonstrated substantial potential for modeling such complex hydrological processes. This study develops and rigorously evaluates a hybrid framework that integrates empirical SRCs with a Coactive Neuro-Fuzzy Inference System (CANFIS) to enhance predictive accuracy and model robustness at the Glinak hydrometric station in the Taleghan watershed. Materials and methodsThe analysis utilized 800 paired observations of streamflow and suspended sediment discharge collected from Glinak station between 1971 and 2023. Data preprocessing involved outlier removal, homogeneity assessment using the standard normal homogeneity test, and normalization. Six empirical SRC methods were calibrated and evaluated: logarithmic-linear regression, multi-segment regression, Logged Mean Load within Discharge Classes Method (LMLWDC), modified FAO correction factor, parametric correction factor, and non-parametric correction factor. Performance evaluation employed the coefficient of determination (R²), Nash-Sutcliffe efficiency (NSE), and relative mean error (RME). In parallel, six ANN architectures—Multilayer Perceptron (MLP), General Feed-Forward (GFF), Radial Basis Function (RBF), Support Vector Machine (SVM), Self-Organizing Feature Map (SOFM), and CANFIS—were trained using a 75%-25% train-test split. Model assessment utilized normalized mean squared error (NMSE), mean absolute error (MAE), and Pearson's correlation coefficient (R). Two hybrid configurations were examined: Hybrid Model 1 used the optimal SRC output as the sole input to CANFIS, whereas Hybrid Model 2 incorporated both SRC output and streamflow as inputs. Results and discussionA detailed interrogation of the findings at the Glinak hydrometric station reveals that conventional hydrological models encounter substantive structural constraints when estimating suspended sediment load under conditions of extreme data dispersion, as evidenced by a coefficient of variation of 206.75%. Pronounced fluctuations in sediment discharge, coupled with a right-skewed and heavy-tailed statistical distribution and the identification of 12.88% outliers, collectively suppressed the Nash–Sutcliffe efficiency (NSE) at this station. From a process-based perspective, the 3,700-fold variability in the sediment-to-flow discharge ratio corroborates the occurrence of hysteresis, underscoring the dependence of sediment yield on antecedent catchment states, particularly soil moisture conditions. Against this backdrop, Hybrid Model 1 (LMLWDC –CANFIS) achieved superior performance, attaining an NSE of 0.63 and markedly outperforming standalone empirical sediment rating curve (SRC) and artificial intelligence architectures. The robustness of this configuration stems from the strategic integration of empirically derived sediment rating-curve knowledge with the adaptive learning capacity of the Coactive Neuro-Fuzzy Inference System (CANFIS). By explicitly accommodating uncertainty and nonlinearity inherent in flow-sediment dynamics, the model establishes a coherent balance between numerical prediction and hydromorphological realism, thereby enhancing predictive reliability under highly variable sediment transport regimes. ConclusionThe results unequivocally demonstrate that the synergistic integration of empirical hydrological reasoning and artificial intelligence provides a resilient and efficient modeling framework for watersheds characterized by pronounced skewness and a high prevalence of anomalous observations. In this context, the CANFIS architecture, leveraging fuzzy logic to resolve complex nonlinear interactions, substantially improves the representation of hysteretic sediment dynamics and offers clear advantages over conventional computational structures, including multilayer perceptron (MLP), general feed-forward (GFF), radial basis function (RBF), support vector machine (SVM), and self-organizing feature map (SOFM) networks. Given the strategic significance of flood events in water resources engineering, the rigorous interrogation of outlier behavior is not optional but imperative. The proposed hybrid framework effectively captures and interprets these statistical anomalies with acceptable precision, thereby reinforcing its applicability in suspended sediment transport management under extreme hydrological conditions. To advance methodological robustness and scientific depth in future investigations, the establishment of comprehensive, high-resolution databases—particularly those documenting peak flood discharges—is strongly recommended. Incorporating auxiliary predictors such as antecedent soil moisture, rainfall intensity, and vegetation cover indices would further strengthen process representation and address hysteresis phenomena. Moreover, deploying advanced machine-learning paradigms, including Long Short-Term Memory (LSTM) networks, alongside robust preprocessing techniques such as quantile regression and outlier treatment methods, has strong potential to attenuate systematic error and enhance model confidence in integrated watershed management applications. Additionally, ensemble methods combining multiple model outputs and uncertainty quantification frameworks are recommended to improve predictive reliability and decision-making support in sediment yield estimation.
Ramtin Tavoosi Rad; Mohamad Ansari ghojghar; Arash Malekian
Abstract
Introduction
Accurate runoff prediction plays a crucial role in water resource management, flood control and climate change adaptation planning. Given the nonlinear, complex and multifactorial nature of hydrological processes, the use of data-driven methods and machine learning algorithms has become ...
Read More
Introduction
Accurate runoff prediction plays a crucial role in water resource management, flood control and climate change adaptation planning. Given the nonlinear, complex and multifactorial nature of hydrological processes, the use of data-driven methods and machine learning algorithms has become an efficient approach for runoff analysis and modeling in recent years. Two individual models, XGBoost, CatBoost and a hybrid model of this two single models, (Cat-XG)Boost were evaluated to predict runoff in the Karaj watershed. Few studies have compared these two algorithms and presented an intelligent hybrid combination at the monthly level.While filling this research gap the present study can provide practical guidance for water resource planners and resource management warning systems in climate change conditions.
Materials and method
In the present study, the efficiency of two widely used individual deep learning models, including CatBoost, XGBoost and a novel hybrid model called (Cat-XG)Boost were evaluated in predicting monthly runoff in the Karaj watershed. For this purpose, monthly average discharge data from five hydrometric stations of Gachsara, Sira-Kaloan, Sira-Karaj, Neshtarud and Pol-e-Khab Dorud with maximum monthly average discharge of 17.22, 16.56, 12.11, 0.36 and 0.28, respectively, in the Karaj watershed during the long-term statistical period of 30 years (1371-1400) were used as model inputs. After performing preprocessing, normalization, outlier removal, stationarity check and input transformation processes, the models were used as inputs. The models were measured with 4 evaluation criteria NS, R, RMSE and MAE. The predictions of the studied models were recorded in 4 states. First, the average discharge of the past 3 months was considered as the model input. In the second scenario, the average discharge of the 6 months before was entered into the model. The third scenario was designed using three consecutive delay seasons (9 months) of monthly average discharge. The fourth scenario included a year of consecutive time lag.
Results and discussion
The prediction results showed that the hybrid model (Cat-XG)Boost with a significant difference provides the best performance in predicting monthly runoff of the Karaj watershed compared to the two individual models evaluated. This model recorded NS above 0.957 and correlation above 0.939 in all stations studied. In addition, it recorded significantly fewer errors than the other two models. While the individual models XGBoost and CatBoost, especially in stations with more extensive data, faced increased errors. Analysis of correlation diagrams also showed that the integrated model had the highest agreement with the actual values and identified complex runoff patterns with higher accuracy. The results obtained from the present study were also consistent with international studies. The two individual models studied provided average performance in predicting values related to extreme climate events but by combining the two individual models and introducing the hybrid model Boost(Cat-XG), the defects in the individual models were covered and also by eliminating existing errors, much more accurate predictions were recorded.
Conclusion
The results of the present study indicate that the hybrid model (Cat-XG)Boost can be used as a powerful tool for runoff prediction in mountainous and climate change-sensitive basins and play an effective role in the development of rapid warning systems, reservoir management and strategic water resource planning. The findings of the present study confirm that combined approaches in the face of complex hydrological data have a significant advantage over single models. This model can be used as a useful tool in water resource management studies, prediction of drought and seasonal floods and presentation of decision-making scenarios at the basin level.
Aboalhasan Fathabadi; Seyed Morteza Seydian; Hossein Emami
Abstract
Introduction
Any modeling and decision-making process in watershed management fundamentally depends on accurate discharge data at the basin outlet. However, direct measurement of discharge is both costly and time-consuming. Therefore, at hydrometric stations, water stage is routinely recorded, and discharge ...
Read More
Introduction
Any modeling and decision-making process in watershed management fundamentally depends on accurate discharge data at the basin outlet. However, direct measurement of discharge is both costly and time-consuming. Therefore, at hydrometric stations, water stage is routinely recorded, and discharge is estimated using the rating-curve equation corresponding to the observed stage. Nevertheless, several sources of uncertainty, including errors in measuring flow velocity, cross-sectional area, and stage height, as well as model limitations in estimating extreme flows and temporal variations in channel morphology caused by erosion, sediment deposition, and vegetation growth, result in inaccuracies in rating-curve-based discharge estimations. Such uncertainties propagate through hydrological model outputs and management decisions, potentially leading to considerable economic and environmental losses. Consequently, it is essential to quantify the uncertainty bounds of rating-curve-derived discharge estimates in order to mitigate risks associated with measurement and estimation errors. To date, both classical and Bayesian approaches have been employed for uncertainty estimation in rating curves. The Bayesian framework, in particular, offers significant advantages: in addition to incorporating observational data through the likelihood function, it allows prior hydraulic knowledge of the station to be embedded in the model through prior distributions. With recent advances in computational power and the widespread application of Markov Chain Monte Carlo (MCMC) sampling techniques, Bayesian methods have become a powerful and flexible tool for rating-curve uncertainty estimation, and various model structures have been introduced in recent years. Accordingly, the present study estimates rating-curve uncertainty using a Bayesian approach for three hydrometric stations located in Golestan Province.
Material and methods
The study area consists of three hydrometric stations: two mountainous stations, Galikesh and Lazoreh, characterized by coarse, gravel-bed channels, and one lowland station, Araz Kooseh, with a fine-grained sedimentary bed. After preliminary evaluation, stage–discharge data corresponding to the 2022–2023 hydrological year were utilized for analysis. Following the development of the rating curve for each station, uncertainty estimation was performed using the GLUE method and three Bayesian approaches: Python Package Ratingcurve (PPR), Generalized Power Law Model (GPLM), and the Norwegian Water Resources and Energy Directorate method (NVE). For the GLUE and GPLM approaches, only the single-segment model structure was applied. In contrast, for the PPR and NVE methods, one-, two-, and three-segment structures were evaluated, and the optimal structure was selected based on statistical performance criteria. Subsequently, uncertainty bounds of the rating curve were estimated accordingly. To assess and compare the performance of the applied methods, the uncertainty ranges were quantified using Mean Prediction Interval (MPI) and Prediction Interval Coverage Probability (PICP) metrics, accompanied by graphical visualization of the estimated uncertainty interval.
Result and discussion
The findings revealed that within the GPLM framework, the simplest structural configuration consistently demonstrated superior performance across all three hydrometric stations. Furthermore, at the Lazoreh and Araz Kooseh stations, the variability of uncertainty intervals exhibited a non-uniform pattern. In several cases, the uncertainty width estimated for lower stage values exceeded that of higher stages, indicating sensitivity of the model to flow regime transitions. Additionally, at stage levels for which direct discharge measurements were available, considerably narrower uncertainty bounds were obtained, highlighting the constraining effect of empirical observations on posterior uncertainty. Application of the NVE method showed that the single-segment model exhibited the best performance at the Galikesh and Lazoreh stations, while the two-segment structure provided a more adequate representation of rating-curve behavior at the Araz Kooseh station. For all three stations, however, this method yielded extremely narrow uncertainty bands at lower stage values, resulting in the exclusion of a substantial portion of observed stage–discharge pairs from the estimated uncertainty domain. This systematic deviation suggests a consistent under-representation of uncertainty in low-flow conditions by the NVE approach. In contrast, the PPR method identified the two-segment rating curve as the most appropriate structural formulation for all stations examined. Under this optimal configuration, uncertainty estimates were initially minimal at low stage values but expanded markedly with increasing stage. The observed inflation in uncertainty intervals at higher flows is likely attributable to sparse high-discharge measurements and increased hydraulic variability under such conditions. Overall comparative analysis using MPI and PICP performance indicators demonstrated that the GLUE method exhibited the most robust and reliable behavior among all models evaluated. Among Bayesian approaches, the PPR methodology outperformed both GPLM and NVE, yielding superior uncertainty characterization and predictive coverage.
Conclusion
In the present study, rating-curve uncertainty was evaluated using the GLUE method in conjunction with three Bayesian approaches—GPLM, PPR, and NVE. The findings revealed that all Bayesian frameworks yielded notably large uncertainty intervals. This outcome can plausibly be attributed to the reliance on weakly informative priors, stemming from limited hydrometric knowledge and the absence of river-specific hydraulic characteristics within the prior structure. Given the scarcity of empirical datasets and the lack of comprehensive, nationwide analyses of stage–discharge relationships in Iranian rivers, the formulation of richer and more informative prior distributions was not practically achievable in this investigation. This limitation underscores a critical research gap, indicating the necessity for extended field measurements, systematic data collection, and the development of regional hydraulic databases. Such advancements would provide the basis for constructing hydrologically informed priors and consequently enhance the reliability and precision of Bayesian uncertainty quantification in future studies.
