Innovative Assessment of Mun River Flow Components through ANN and Isotopic End-Member Mixing Analysis

: This study innovatively assesses the Mun River flow components in Thailand, integrating artificial neural networks (ANNs) and isotopic ( δ 18 O) end-member mixing analysis (IEMMA). It quantifies the contributions of the Upper Mun River (UMR) and Chi River (CR) to the overall flow, revealing a discrepancy in their estimated contributions. The ANN method predicts that the UMR and CR contribute approximately 70.5% and 29.5% respectively, while IEMMA indicates a more pronounced disparity with 84% from UMR and 16% from CR. This divergence highlights the distinct perspectives of ANN, focusing on hydrological data patterns, and IEMMA, emphasizing isotopic signatures. Despite discrepancies, both methods validate UMR as a significant contributor to the overall flow, highlighting their utility in hydrological research. The findings emphasize the complexity of river systems and advocate for an integrated approach of river flow analysis for a comprehensive understanding, crucial for effective water resource management and planning.


Introduction
The understanding of river flow components, especially in tropical regions, is vital for water resource management, environmental protection, and sustainability.Tropical rivers face specific concerns such as seasonal variability, intense precipitation events, and unique ecological systems.The integration of artificial neural networks (ANNs) and isotopic end-member mixing analysis (IEMMA) is particularly promising in these contexts.It combines advanced computational techniques with precise isotopic measurements to navigate the complexity of hydrological processes and the diversity of water sources, offering an understanding essential for addressing the challenges unique to tropical river systems [1][2][3][4][5].
The applications of hydrogen (δ 2 H) and oxygen (δ 18 O) stable isotopes as environmental tracers in hydrology have revolutionized the understanding of river flow dynamics [6][7][8].These isotopes provide critical insights into various hydrological processes [9], such as interactions within a basin [10][11][12], evaporative fractionation effects [13][14][15], and river discharge behaviors [16,17].These isotopic compositions change within the hydrological cycle, enabling the identification of water sources across different seasons through δ 2 H or δ 18 O tracers [18,19].Standards for δ 18 O and δ 2 H measurements typically adhere to the Vienna Standard Mean Ocean Water (VSMOW) scale, ensuring comparability of data across studies.The isotopic ratios are often expressed using the delta notation (δ), calculated as: where R represents the ratio of heavy to light isotopes ( 18 O/ 16 O for δ 18 O (‰) and 2 H/ 1 H for δ 2 H (‰)) in both the sample and the standard.The d−excess is defined as d-excess (‰) = δ 2 H − 8 δ 18 O [20].This parameter is used to indicate secondary processes such as evaporation that may affect water samples [9].Isotope hydrology enables the precise identification of water sources across seasons, enhancing the analysis of how different water contributions affect river systems [9,21,22].The technique of isotopic end-member mixing analysis (IEMMA) advantages these isotopic signatures to detail the contributions of various sources to river flow, offering a deeper look into the hydrological cycle and tributary-main river interactions [23][24][25][26].IEMMA has been extensively applied to trace the sources of river water, including rainfall, groundwater, and glacial melt contributions [24,27].This helps in understanding the seasonal variability of river dynamics and the impact of climatic factors [23].Previous studies have also explored hydrological processes and the impacts of human activities on river water composition [3,23,24,[28][29][30].They demonstrated how IEMMA can effectively discern the proportions of surface and groundwater contributions to river flows.IEMMA also serves in distinguishing natural water sources from anthropogenic input [30].Furthermore, the integration of isotopic data obtained through IEMMA with hydrological models enhances the understanding of riverine processes.This synergy improves the prediction of flow dynamics and water quality under changing environmental conditions [26].Furthermore, studies utilizing IEMMA have contributed to the assessment of ecosystem health, particularly in identifying areas of concern due to pollution or over-extraction of water resources [25,31].
Artificial neural networks (ANNs) have emerged as a powerful tool in the study of river dynamics, offering innovative approaches to understanding and predicting complex hydrological processes.This literature review highlights recent advancements in the application of ANNs to river studies, emphasizing their role in predicting dissolved oxygen levels, managing reservoir sediments, landslide susceptibility modeling, and river flow prediction.The prediction of dissolved oxygen in the Nyando River Basin, Kenya, was significantly improved by implementing ANNs alongside multiple linear regression models, showcasing ANNs' ability to capture the nonlinear dynamics of water quality prediction more effectively than traditional models [32].In another study, the management of reservoir sediment in the Alpine Saalach River was optimized using an ANN, demonstrating a novel approach to sediment flushing that enhances efficiency and effectiveness [33].For river flow prediction in the Dholai River Basin, a memory-based ANN provided more accurate forecasts than conventional methods, highlighting the models' potential in hydrological forecasting [34].Furthermore, the application of Monte Carlo-optimized ANN for virtual water quality monitoring in the Danube River emphasized ANN's capability to enhance monitoring efforts at locations lacking physical monitoring infrastructure [35].These studies collectively illustrate the versatility and efficiency of ANNs in addressing a range of challenges in river dynamics studies.By employing the ability of ANNs, hydrologists can better predict hydrological changes, manage water resources, and understand the environmental impacts on river systems.
In Thailand, the Mun River represents a critical water resource, facing challenges related to water quality [36], allocation [37], and the impacts of climate change [38].Understanding the contributions of upstream flows and tributaries, such as the Chi River, to the Mun River's discharge is important for addressing these water-related issues effectively.The objectives of this study are to quantify the contributions of the Upper Mun River (UMR) and Chi River (CR) to the Mun River's overall flow and to compare the results obtained from ANN and IEMMA.While field measurements can provide direct flow data, these methods were selected due to their ability to handle complex hydrological and isotopic data, offering deeper insights into the sources and dynamics of river flow.ANN models are used for their strength in pattern recognition and prediction, while IEMMA provides de-tailed isotopic analysis to trace water sources.The combination of these methods addresses the inherent uncertainties in hydrological studies, enhancing the robustness of the results.Furthermore, this study seeks to answer two research questions: (1) What are the proportions of upstream Mun River and Chi River contributions to the Mun River's discharge?and (2) How consistent are the findings obtained from ANN and IEMMA methodologies in determining these contributions?The integration of ANN and IEMMA in this study represents a novel approach, combining advanced computational techniques with precise isotopic measurements.This synergy allows for a more comprehensive analysis of river flow components, offering an understanding that surpasses traditional methods.By the strengths of both methods, this research provides innovative solutions to the challenges of water resource management in tropical river systems.

Study Area and Climate Conditions
The Chi and Mun River basins, situated in Thailand, present unique geographic, hydrologic, and meteorologic characteristics essential for water resource management and environmental studies.Both basins are surrounded by significant mountain ranges that play a crucial role in their hydrology.The Chi River Basin is bordered by the Dong Phaya Yen, Phetchabun, and Phu Phan mountain ranges, with elevations ranging from 300 to 1300 m, contributing to its complex river system [39] and covering 49,274 km 2 (Table 1).In contrast, the Mun River Basin covers 69,701 km 2 of watershed area and features the Banthat and Phanom Dong Rak mountains, with the landscape transitioning from elevated plateaus over 1350 m above sea level to lower plains towards the Mun River, indicating a varied topography from approximately 100 to 1350 m above sea level (Figure 1).The study period of 2019-2021, though seemingly short, captures significant seasonal and hydrological variations in the Mun River Basin.This period includes multiple hydrological cycles, allowing for the analysis of both wet and dry seasons (Figure 2), which are crucial for understanding the river's behavior under different climatic conditions.While longer-term data would provide more insights, the chosen period is representative of the prevailing climatic and hydrological patterns in the region.Meteorologically, a clear cyclical pattern of rainfall across the year reflects the seasonal monsoon influences in the region (Figure 2a).For the Chi River Basin, the mean (2019-2021) annual rainfall is 1309 mm/year.The mean monthly rainfall starts low in January with 6.75 mm, and gradually increases, reaching its peak in August at 317.02 mm, before declining towards the end of the year.This peak reflects the intense rainfall during the monsoon season.The lowest rainfall occurs in December, with a mean of about 0.77 mm, indicating the dry season's onset.In comparison, the Mun River Basin shows a somewhat similar seasonal pattern but with different mean monthly rainfall values.The mean Meteorologically, a clear cyclical pattern of rainfall across the year reflects the seasonal monsoon influences in the region (Figure 2a).For the Chi River Basin, the mean (2019-2021) annual rainfall is 1309 mm/year.The mean monthly rainfall starts low in January with 6.75 mm, and gradually increases, reaching its peak in August at 317.02 mm, before declining towards the end of the year.This peak reflects the intense rainfall during the monsoon season.The lowest rainfall occurs in December, with a mean of about 0.77 mm, indicating the dry season's onset.In comparison, the Mun River Basin shows a somewhat similar seasonal pattern but with different mean monthly rainfall values.The mean annual (2019-2021) rainfall is 1686 mm/year.The rainfall begins with a mean of 5.39 mm in January, peaks in September at 318.19 mm, and then decreases towards the end of the year.The peak rainfall month for the Mun River Basin slightly shifts compared to the Chi River Basin, showcasing regional variations in monsoon impacts [40].The lowest rainfall here is also in January, similar to the Chi River Basin, but with slightly less mean rainfall.
Hydrologically, the monsoon season also markedly elevates river discharge (Figure 2b).For the Chi River Basin, the average (2019-2021) discharge is 290.57m 3 /s with 765 km of river length (Table 1).The water flow starts at a moderate level in January, with an average of 56.42 m 3 /s, and slowly increases as the rainy season progresses, reaching its highest point in September at 896.05 m 3 /s.This peak shows the maximum impact of the rainy season, after which the flow decreases towards the end of the year.By December, the flow drops to an average of 166.12 m 3 /s, indicating the start of the dry season.In the Mun River Basin, the pattern is similar, but the numbers are higher average flow of 451.33 m 3 /s, attributed to a larger catchment area collecting rainwater or different characteristics.The flow begins at 70.56 m 3 /s in January, rises gradually, and hits its highest in September at 1633.61 m 3 /s.This higher peak underlines the strong effect of the rainy season on the Mun River.The flow then reduces towards the year's end, with December seeing an average flow of 252.50 m 3 /s.

Isotope Analysis
Between 2019 and 2021, an extensive sampling campaign was undertaken for stable isotopic analysis in the Chi and Mun River basins of Thailand, focusing on both river water and daily rainwater.This effort was part of a methodology to assess river flow components using isotopic end-member mixing analysis (IEMMA) and artificial neural networks (ANNs).The study involved three strategically chosen stations: the upstream Mun River (UMR) as End-Member 1, the Chi River (CR) as End-Member 2, and the Mun River (MR) as the site for mixing river water samples.
For the collection of precipitation isotopes, the methodology adhered to protocols established by the International Atomic Energy Agency (IAEA).Rainwater was collected daily across the three sites, employing collectors designed to minimize evaporation and atmospheric exchange.These collectors featured a funnel leading into a graduated cylinder within a container, complemented by a table tennis ball to seal against evaporation.Accumulation bottles aggregated the weekly precipitation, representing the integrated rainfall per week.These samples were then stored in 50 mL high-density polyethylene bottles at 25 • C to avert condensation until their isotopic ratios could be analyzed using cavity ring-down spectroscopy.
Simultaneously, river water samples were collected weekly from the three designated stations (Figure 1), totaling 468 samples over the study period.The sampling technique involved collecting water from a depth of approximately 20 cm to ensure the purity and flow of the samples, which were then stored in 125-mL high-density polyethylene bottles and frozen to prevent evaporation.The isotopic analysis of these samples also followed IAEA guidelines [41], with the isotopic ratios of both rain and river water standardized against international references and analyzed for their deuterium (δ 2 H) and oxygen-18 (δ 18 O) contents.The deuterium excess (d-excess) was defined as d-excess Both sets of samples underwent rigorous analysis using a Picarro L2140-i cavity ring-down spectrometer at the Thailand Institute of Nuclear Technology, ensuring high precision and accuracy (1σ: ±0.15‰ for δ 18 O and ±0.5‰ for δ 2 H) in measuring the isotopic compositions, thereby facilitating a detailed assessment of the hydrologic interactions between precipitation and river flows within these basins.This innovative approach combined with river discharge and rainfall data recorded concurrently.

Artificial Neural Networks
The artificial neural network (ANN) methodology, as described in [42], utilizes the ANN's ability to navigate complex, nonlinear relationships for assessing river flow contributions from the upstream Mun River (UMR) and the Chi River (CR) to the main Mun River (MR) [43].This approach employed the weekly river isotope data (δ 18 O) and flow rates (Q) from 2019 to 2021 as the input layer (Figure 3).The observed target variables in this context are the fractions of flow rate between the Chi River (CR) and upstream Mun River (UMR) relative to the main Mun River (MR), denoted as F CR -Q and F UMR -Q, respectively.
These fractions, F CR -Q and F UMR -Q, represent how much of the main Mun River's flow comes from each tributary, providing insights into water distribution and management.They are calculated as the ratio of the flow rate from each tributary to the total flow rate into the main Mun River.If Q CR and Q UMR denote the flow rates from CR and UMR, and Q MR denotes the total flow rate into MR, the fractions can be expressed mathematically as:

Artificial Neural Networks
The artificial neural network (ANN) methodology, as described in [42], utilizes the ANN's ability to navigate complex, nonlinear relationships for assessing river flow contributions from the upstream Mun River (UMR) and the Chi River (CR) to the main Mun River (MR) [43].This approach employed the weekly river isotope data (δ 18 O) and flow rates (Q) from 2019 to 2021 as the input layer (Figure 3).The observed target variables in this context are the fractions of flow rate between the Chi River (CR) and upstream Mun River (UMR) relative to the main Mun River (MR), denoted as FCR-Q and FUMR-Q, respectively.These fractions, FCR-Q and FUMR-Q, represent how much of the main Mun River's flow comes from each tributary, providing insights into water distribution and management.They are calculated as the ratio of the flow rate from each tributary to the total flow rate into the main Mun River.If QCR and QUMR denote the flow rates from CR and UMR, and QMR denotes the total flow rate into MR, the fractions can be expressed mathematically as: The ANN model used in this study incorporates several predictor variables, including river flow rates (Q) and isotopic compositions (δ 18 O).These variables are decisive for capturing the complex interactions within the river system (Figure 3).The model architecture includes input layers for these variables (6 parameters), 2 hidden layers (10 neurons each) for processing, and an output layer for predicting the contributions of the UMR and CR to the Mun River's flow.The data preparation phase, crucial for the model's success, involves dividing the dataset into an 80% training and 20% testing set, ensuring effective model training and validation [44].The ANN architecture, featuring dense layers with the rectified linear unit activation and a sigmoid output layer, is meticulously designed to predict these continuous variables, FCR-ANN and FUMR-ANN, reflecting the river flow fractions [45].The ANN model used in this study incorporates several predictor variables, including river flow rates (Q) and isotopic compositions (δ 18 O).These variables are decisive for capturing the complex interactions within the river system (Figure 3).The model architecture includes input layers for these variables (6 parameters), 2 hidden layers (10 neurons each) for processing, and an output layer for predicting the contributions of the UMR and CR to the Mun River's flow.The data preparation phase, crucial for the model's success, involves dividing the dataset into an 80% training and 20% testing set, ensuring effective model training and validation [44].The ANN architecture, featuring dense layers with the rectified linear unit activation and a sigmoid output layer, is meticulously designed to predict these continuous variables, F CR -ANN and F UMR -ANN, reflecting the river flow fractions [45].
After training with the Adam optimizer and mean squared error loss for 100 epochs, the model's accuracy and prediction reliability are assessed on the testing set using the coefficient of determination (R²) and the root mean square error (RMSE) [46].R² assesses the model's accuracy in mirroring observed data variance, while RMSE quantifies the average prediction error magnitude, offering insights into the model's precision and reliability.

Isotopic End-Member Mixing Analysis
Isotopic end-member mixing analysis (IEMMA) is a technique used in hydrology to quantify the contributions of different sources (end-members) to a mixture, such as the fractions of river flow from two tributaries to one main river.Isotopes, particularly stable isotopes of water (e.g., δ 18 O and δ 2 H), are commonly used as tracers because they can distinguish water from different sources based on their unique isotopic signatures [47].
Applying a mixing model that incorporates the isotopic signatures of the end-members and the mixed water (main river) to calculate the fractions of each tributary's contribution (upstream Mun River (UMR) and Chi River (CR)) to the main Mun River flow, the simplest form of the mixing model for two tributaries (UMR: End-member 1 and CR: End-member 2) contributing to the Mun River can be represented as: where F UMR and F CR are the fractions of the river flow contributed by End-member 1 and End-member 2, respectively.Using isotopic data, F UMR and F CR can be solved by applying the conservation of mass for isotopes (δ 18 O): where δ 18 O UMR and δ 18 O CR are the isotopic signatures of the two end-members, and δ 18 O MR is the isotopic signature of the river mixture.Furthermore, it is important to assess the uncertainties in measurements and model assumptions.Therefore, this study also employed sensitivity analysis or the use of Monte Carlo simulations to understand how uncertainties in isotopic signatures and other parameters affect the flow component estimates [48].

Isotopic Compositions of River Water
The isotopic composition of river water within the Mun River Basin, encompassing δ 18 O, δ 2 H, and d-excess, has been meticulously analyzed across various locations (Figure 1), including the upstream Mun River (UMR), Chi River (CR), and Mun River (MR).This comprehensive dataset covers both wet and dry seasons (Figure 4).Throughout the period under study, the isotopic values displayed substantial variability: δ 18 O values ranged from −9.44‰ to 3.11‰, δ 2 H varied from −67.69‰ to 2.92‰, and d-excess exhibited significant fluctuations (−22.50‰ to 12.36‰), indicating the complicated interplay among evaporation, precipitation, and water source mixing that characterizes the basin's hydrological processes.
The UMR, CR, and MR showed distinct isotopic signatures.The mean river δ 18 O in the UMR was −4.17‰, in the CR was −4.56‰, and in the MR was slightly more enriched with a mean of −3.84‰ (Figure 4a).Similarly, the mean δ 2 H values were −36.85‰ for the UMR, −38.76‰ for the CR, and −34.04‰ for the MR, indicating a slight enrichment in the MR relative to the other locations (Figure 4b).The mean river d-excess values further elucidate the differences in hydrological processes across these locations: −3.50‰ for the UMR, −2.28‰ for the CR, and −3.33‰ for the MR (Figure 4c).
These mean values underscore the spatial variability in isotopic compositions across the Mun River Basin.The differences between locations can be attributed to a variety of factors, including local evaporation rates [49], the mixing of water sources (e.g., groundwater inputs, tributary influx) [50,51], and the regional climate's influence on precipitation patterns [8].The UMR and CR tend to show more depleted isotopic values, which could reflect the impact of higher elevation sources and more direct precipitation inputs, while the MR's relative enrichment in isotopic values may indicate lower elevation sources, increased evaporation, or the mixing of different water sources as the river progresses downstream.This spatial variability provides essential clues to the hydrological connectivity and water cycle dynamics within the Mun River Basin.
Seasonal variations in the isotopic composition of river water within the Mun River Basin are marked, with distinct shifts observed between the wet and dry seasons (Figure 4a,b).These shifts indicate the significant role of direct precipitation (Figure 4d) and the potential modulation of evaporation effects across different times of the year.For instance, the mean δ 18 O value in the UMR during the wet season is notably more depleted at −4.36‰ compared to a less depleted mean of −3.97‰ during the dry season (Figure 4a).This pattern of depletion in the wet season relative to the dry season is consistent across other locations within the basin, including the CR and the MR, emphasizing the influence of seasonal precipitation on isotopic signatures [12,52].The UMR, CR, and MR showed distinct isotopic signatures.The mean river δ 18 O in the UMR was −4.17‰, in the CR was −4.56‰, and in the MR was slightly more enriched with a mean of −3.84‰ (Figure 4a).Similarly, the mean δ 2 H values were −36.85‰ for the UMR, −38.76‰ for the CR, and −34.04‰ for the MR, indicating a slight enrichment in the MR relative to the other locations (Figure 4b).The mean river d-excess values further elucidate the differences in hydrological processes across these locations: −3.50‰ for the UMR, −2.28‰ for the CR, and −3.33‰ for the MR (Figure 4c).
These mean values underscore the spatial variability in isotopic compositions across the Mun River Basin.The differences between locations can be attributed to a variety of factors, including local evaporation rates [49], the mixing of water sources (e.g., groundwater inputs, tributary influx) [50,51], and the regional climate's influence on precipitation patterns [8].The UMR and CR tend to show more depleted isotopic values, which could reflect the impact of higher elevation sources and more direct precipitation inputs, while the MR's relative enrichment in isotopic values may indicate lower elevation sources, increased evaporation, or the mixing of different water sources as the river progresses downstream.This spatial variability provides essential clues to the hydrological connectivity and water cycle dynamics within the Mun River Basin.
Seasonal variations in the isotopic composition of river water within the Mun River Basin are marked, with distinct shifts observed between the wet and dry seasons (Figure In addition to δ 18 O, the seasonal variations in d-excess further illuminate the complexities of hydrological processes at play (Figure 4c).The mean d-excess values during the wet and dry seasons reveal differences in evaporation and moisture source conditions across the basin [53][54][55].For example, the UMR exhibits a mean d-excess of −2.79‰ in the wet season, which shifts to −4.23‰ in the dry season.This change suggests that during the dry season, evaporation effects become more pronounced, or there might be a variation in the moisture sources contributing to river water [56].
Comparing seasonal variations between locations, each river demonstrates unique patterns of isotopic changes that reflect local hydrological and meteorological influences.The CR and MR also show differences in their seasonal isotopic values, with generally more depleted δ 18 O and δ 2 H values during the wet season, indicative of enhanced precipitation input and potentially reduced evaporation.However, the extent of seasonal variation in d-excess across these locations further points to the differential impact of evaporation and moisture source dynamics within the basin [39,57].
For precipitation, the isotopic values across the basin exhibit significant variability, with δ 18 O and δ 2 H reflecting seasonal influences, moisture sources, and atmospheric circulation patterns (Figure 5).The mean δ 18 O values in precipitation range from −6.53‰ to −5.85‰ across different locations (averaged −6.11‰ for all samples), demonstrating the influence of temperature, altitude, and rainfall amount on isotopic fractionation.δ 2 H values (−136.01‰ to 26.75‰, averaged −39.65‰) and d-excess in precipitation also vary, with d-excess values (−19.82‰ to 20.85‰, averaged 9.25‰) providing insights into the evapo-rative conditions at the moisture source regions and during raindrop fall [58,59].The range of d-excess values, from negative to significantly positive, features the complex interplay between local and regional hydrological processes.The rainfall d-excess values (averaged 8.48‰ for CR, 9.42‰ for UMR, and 9.86‰ for MR) suggest that on average, precipitation in the basin tends to originate from moisture sources with similar characteristics [60].River water isotopes by comparison tend to be more enriched than those in precipitation.This depletion is generally attributed to the evaporative enrichment of heavy isotopes in open water bodies, which results in higher δ 18 O and δ 2 H values in river water compared to precipitation.The mean δ 18 O values in river water for the entire period studied across the UMR, CR, and MR locations indicate this enrichment, alongside variations in d-excess that underscore the differences in evaporation rates and moisture mixing processes between the river and atmospheric water [12,39,47].
The deviations between river water and precipitation isotopic signatures across locations reveal the complex hydrological interactions within the basin (Figure 5).These deviations are particularly obvious when analyzing the slope and intercept of the regression between δ 18 O and δ 2 H of precipitation (forming the local meteoric water line or LMWL) compared to river water.The LMWL, defined by the linear relationship between δ 18 O and δ 2 H in precipitation, serves as a benchmark for understanding hydrological processes.Deviations of river water isotopes from the LMWL indicate evaporation, mixing with differ- River water isotopes by comparison tend to be more enriched than those in precipitation.This depletion is generally attributed to the evaporative enrichment of heavy isotopes in open water bodies, which results in higher δ 18 O and δ 2 H values in river water compared to precipitation.The mean δ 18 O values in river water for the entire period studied across the UMR, CR, and MR locations indicate this enrichment, alongside variations in d-excess that underscore the differences in evaporation rates and moisture mixing processes between the river and atmospheric water [12,39,47].
The deviations between river water and precipitation isotopic signatures across locations reveal the complex hydrological interactions within the basin (Figure 5).These deviations are particularly obvious when analyzing the slope and intercept of the regression between δ 18 O and δ 2 H of precipitation (forming the local meteoric water line or LMWL) compared to river water.The LMWL, defined by the linear relationship between δ 18 O and δ 2 H in precipitation, serves as a benchmark for understanding hydrological processes.Deviations of river water isotopes from the LMWL indicate evaporation, mixing with different water sources, or both [12,61].With a slope of 7.66 and an intercept of 7.21, the LMWL represents the isotopic relationship in precipitation across the Mun River Basin.These values are indicative of the general atmospheric conditions and moisture sources influencing precipitation isotopes in the region [62].
The regression line for δ 18 O vs. δ 2 H in the Mun River has a slope of 5.44 and an intercept of −14.16 (Figure 5).The shallower slope and negative intercept compared to the LMWL suggest significant evaporation effects, altering the isotopic composition of river water [63].For the Chi River, the slope is 6.23 with an intercept of −10.33.This closer alignment with the LMWL compared to the MR indicates less evaporation influence, but still reflects a deviation from the isotopic composition expected from direct precipitation [39].The UMR features a slope of 5.85 and an intercept of −11.57.These values, while differing from the LMWL, suggest a moderate evaporation effect and possibly different mixing dynamics with groundwater or tributaries [64].The lower slopes for MR, CR, and UMR compared to the LMWL indicate evaporation's role in enriching the heavy isotopes in river water.This evaporation effect is most pronounced in the MR, as evidenced by its significantly lower slope and more negative intercept.The intercept values, particularly the negative ones for river locations, hint at the mixing of river water with other water sources that have distinct isotopic signatures, such as groundwater, which may have undergone evaporation or fractionation processes before entering the river system [65].

Performance of ANN Models
The application of the artificial neural network (ANN) method for analyzing river flow components signifies an advanced approach in comprehending water dynamics within river systems.Focusing on the Main Mun River, this study examines contributions from its tributaries, the upper Mun River (UMR) and the Chi River (CR), utilizing ANN to unravel the complexities of river flow variations and their contributing factors.The UMR's contribution to the main Mun River flow, denoted as F UMR -ANN, showed significant variability (Figure 6).Statistical analysis highlighted a maximum flow contribution rate of 0.78 and a minimum of 0.51, with an average contribution of 0.70.Analysis of the Chi River's flow contributions, F CR -ANN, revealed a maximum contribution of 0.49 and a minimum of 0.22, with an average contribution of 0.30, showing the variability in its contributions to the main Mun River.The comparative analysis of trained and test results enhances understanding of the ANN model's performance and reliability in predicting river flow contributions (Figure 6b).Minor deviations between training and testing phases for both UMR (F UMR -ANN) and CR (F CR -ANN) suggest slight variances in model predictions across different datasets [66].
Incorporating aspects of hydrology and tropical meteorology into the study of river flow components substantially enhances an understanding of water dynamics in river systems.The ANN analysis offers insights into river flow contributions during the wet and dry seasons, highlighting the seasonal dynamics of river flow.During the wet season, the analysis indicates that the UMR contributes an average fraction of 0.69 to the main Mun River flow, while the CR contributes a slightly lower average fraction of 0.31.These findings illustrate the impact of increased precipitation during the wet season on river flows, where both tributaries significantly contribute to the main Mun River, albeit with slight variations in their contributions [67].Conversely, in the dry season, the UMR shows a slight increase in contribution, averaging 0.70, suggesting a consistent or slightly enhanced flow to the main Mun River despite reduced precipitation.In contrast, the CR shows a minor decrease in its contribution, averaging 0.30, indicating a more obvious sensitivity to seasonal precipitation reductions [38].minimum of 0.22, with an average contribution of 0.30, showing the variability in its contributions to the main Mun River.The comparative analysis of trained and test results enhances understanding of the ANN model's performance and reliability in predicting river flow contributions (Figure 6b).Minor deviations between training and testing phases for both UMR (FUMR-ANN) and CR (FCR-ANN) suggest slight variances in model predictions across different datasets [66].The observed seasonal variations in river flow contributions from the UMR and CR provide critical insights into the interaction between hydrological processes and tropical meteorology.The slight increase in the UMR's contribution during the dry season may be attributed to the catchment area's characteristics, potentially including its water retention capability during wet periods and consistent release throughout the year.This suggests resilience in the UMR's flow contributions, possibly mitigating the impacts of seasonal precipitation variability [2].On the other hand, the slight decrease in the CR's contribution during the dry season suggests its heightened sensitivity to precipitation patterns, reflecting the catchment's hydrological responses to tropical meteorological variations [24,67].
Comparing ANN model predictions with observed river flow data for the UMR and CR, using the coefficient of determination (R²) and the root mean square error (RMSE) [46], provides insight into the model's precision and areas of uncertainty.For F UMR -ANN versus F UMR -Q, an R 2 of 0.33 and RMSE of 0.10 were observed, while for F CR -ANN versus F CR -Q, R 2 was 0.31 with an RMSE of 0.10 (Figure 6b).These findings suggest a moderate positive correlation, indicating that as observed flow rates increase, the model's predictions also tend to increase, although not strongly.This reveals the model's partial success in capturing observed data trends and highlights areas for significant improvement.This comprehensive analysis underscores the efficacy and potential of ANN in hydrological studies, indicating both its strengths and limitations in capturing the complexities of river flow dynamics.Enhancing the model with more detailed data, such as isotopic signatures, could improve prediction accuracy and provide deeper insights into flow components.

Isotopic End-Member Mixing Analysis Results
Incorporating oxygen-18 (δ 18 O) isotopic signatures through isotopic end-member mixing analysis (IEMMA) significantly enhances the understanding of river flow dynamics within the upper Mun River (UMR) and the Chi River (CR), contributing to the main Mun River.The IEMMA results reveal a significant variability in flow contributions, with the UMR showing a maximum contribution of 0.99 and a minimum of 0.50 (Figure 7).This variability-from a significant minimum contribution (0.50) to an average of 0.70-highlights instances where the UMR predominantly sustains the main Mun River.In contrast, the CR's contributions, while essential, are noticeably lower on average (0.30), with a maximum that coincides with the UMR's minimum and a minimal figure (0.01) emphasizing its intermittent influence [40].Figure 8 presents a scatterplot of the flow fraction results obtained by IEMMA and the observed river discharge.The analysis includes the coefficient of determination (R²) to evaluate the correlation between the observed and IEMMA-predicted flow fractions.The 95% significance band is also plotted to provide a visual representation of the confidence interval for the predictions (Figure 8).The R² value for the relationship between the observed flow fraction of the upper Mun River (FUMR-Q) and the IEMMA results (FUMR-IEMMA) is 0.89 (R 2 UMR = 0.60 overall).This high R² value indicates a strong correlation, suggesting that IEMMA is effective in predicting the flow contributions from the UMR based on isotopic data.The scatterplot shows that most data points fall within the 95% significance band, reinforcing the reliability of the IEMMA predictions.The R² value for the relationship between the observed flow fraction of the Chi River (FCR-Q) and the IEMMA results (FCR-IEMMA) is 0.76 (R 2 CR = 0.57 overall).Although lower than the UMR, this still represents a strong correlation, indicating that IEMMA provides reasonably accurate predictions for the CR contributions.The scatterplot for FCR shows a wider spread compared to FUMR, which may be attributed to the more variable nature of CR contributions or potential influences of local hydrological processes not fully captured by the isotopic analysis.Despite IEMMA's insightful nature, challenges such as isotopic data resolution, endmember selection, and the assumption of isotopic homogeneity within each end-member limit the analysis's accuracy.These challenges highlight a complex matrix of factors influencing river flow dynamics, calling for refined methodologies in future research.Addressing these limitations involves enhanced spatial and temporal sampling, advanced endmember characterization, integration with hydrological models, and the application of machine learning techniques to improve isotopic analysis resolution and accuracy, offering a more nuanced understanding of hydrological processes in river systems.
The precision of IEMMA outcomes, influenced by factors like isotopic fractionation [13,68], irrigation return flows [69], and groundwater contributions [27], introduces uncertainties in interpreting isotopic data, crucial for advancing hydrological model accuracy.Isotopic fractionation, the separation of isotopes during physical or chemical processes, can significantly affect δ 18 O signatures, potentially masking true source contributions [27,50,57].Irrigation return flows, with distinct isotopic signatures, can alter a river's isotopic composition, skewing the analysis of tributary contributions [69].Groundwater contributions, with unique isotopic signatures, can significantly change a river's isotopic composition, necessitating their inclusion as end-members in the mixing analysis [57].Seasonal variations add complexity to this hydrological narrative.The wet season's dilution effects and runoff, driven by increased rainfall (Figure 4d), might equalize contributions from both tributaries, whereas the dry season's distinct isotopic signatures could emphasize the UMR's dominance due to its larger catchment area (Figure 4a).These insights illustrate the dynamic relationship between hydro-meteorological conditions and river flow contributions in tropical settings [57].
Figure 8 presents a scatterplot of the flow fraction results obtained by IEMMA and the observed river discharge.The analysis includes the coefficient of determination (R²) to evaluate the correlation between the observed and IEMMA-predicted flow fractions.The 95% significance band is also plotted to provide a visual representation of the confidence interval for the predictions (Figure 8).The R² value for the relationship between the observed flow fraction of the upper Mun River (F UMR -Q) and the IEMMA results (F UMR -IEMMA) is 0.89 (R 2 UMR = 0.60 overall).This high R² value indicates a strong correlation, suggesting that IEMMA is effective in predicting the flow contributions from the UMR based on isotopic data.The scatterplot shows that most data points fall within the 95% significance band, reinforcing the reliability of the IEMMA predictions.The R² value for the relationship between the observed flow fraction of the Chi River (F CR -Q) and the IEMMA results (F CR -IEMMA) is 0.76 (R 2 CR = 0.57 overall).Although lower than the UMR, this still represents a strong correlation, indicating that IEMMA provides reasonably accurate predictions for the CR contributions.The scatterplot for FCR shows a wider spread compared to FUMR, which may be attributed to the more variable nature of CR contributions or potential influences of local hydrological processes not fully captured by the isotopic analysis.Despite IEMMA's insightful nature, challenges such as isotopic data resolution, endmember selection, and the assumption of isotopic homogeneity within each end-member limit the analysis's accuracy.These challenges highlight a complex matrix of factors influencing river flow dynamics, calling for refined methodologies in future research.Addressing these limitations involves enhanced spatial and temporal sampling, advanced end-member characterization, integration with hydrological models, and the application of machine learning techniques to improve isotopic analysis resolution and accuracy, offering a more nuanced understanding of hydrological processes in river systems.
The precision of IEMMA outcomes, influenced by factors like isotopic fractionation [13,68], irrigation return flows [69], and groundwater contributions [27], introduces uncertainties in interpreting isotopic data, crucial for advancing hydrological model accuracy.Isotopic fractionation, the separation of isotopes during physical or chemical processes, can significantly affect δ 18 O signatures, potentially masking true source contributions [27,50,57].Irrigation return flows, with distinct isotopic signatures, can alter a river's isotopic composition, skewing the analysis of tributary contributions [69].Groundwater contributions, with unique isotopic signatures, can significantly change a river's isotopic composition, necessitating their inclusion as end-members in the mixing analysis [57].

Comparative Analysis for River Flow Components
The combination of ANN and IEMMA methods in analyzing the contributions of the UMR and Chi River CR to the Mun River provides distinct perspectives on the hydrological dynamics within the river basin.Initial observations show general agreement between ANN and IEMMA in recognizing the UMR as a significant contributor to the Mun River's flow (Figure 9).However, closer examination reveals discrepancies in the specific fractional contributions attributed to each river by the two approaches.For instance, on an observed date, the ANN model predicts a 70.5% contribution from the UMR and a 29.5% contribution from the CR, while IEMMA suggests a greater disparity with 84% from the UMR and only 16% from the CR (Figure 9).These variances illustrate the different perspectives through which ANN and IEMMA view the river's flow dynamics: ANN through hydrological data patterns and IEMMA through isotopic signatures reflecting water mixing and source contributions [70,71].
This comparative analysis indicates that the discrepancies between ANN and IEMMA results in estimating river flow fractions stem from differences in data sources, methodologies, and sensitivities to environmental factors.ANN controls historical hydrological data to predict flow contributions through statistical patterns [5,33], whereas IEMMA uses hydrochemical isotopic signatures to understand water source contributions based on physical-chemical interactions [2,3,24].Variabilities in moisture sources, evaporation effects, land use changes, and assumptions about isotopic homogeneity can alter results between the two approaches [53,54].ANN might not fully account for complex hydrogeochemical interactions, while IEMMA may overlook temporal variations in hydrological data [69].These differences highlight the complementary nature of using both methods for a holistic understanding of river flow dynamics, suggesting an integrated approach can mitigate individual limitations and provide a comprehensive analysis.

Comparative Analysis for River Flow Components
combination of ANN and IEMMA methods in analyzing the contributions of the UMR and Chi River CR to the Mun River provides distinct perspectives on the hydrological dynamics within the river basin.Initial observations show general agreement between ANN and IEMMA in recognizing the UMR as a significant contributor to the Mun River's flow (Figure 9).However, closer examination reveals discrepancies in the specific fractional contributions attributed to each river by the two approaches.For instance, on an observed date, the ANN model predicts a 70.5% contribution from the UMR and a 29.5% contribution from the CR, while IEMMA suggests a greater disparity with 84% from the UMR and only 16% from the CR (Figure 9).These variances illustrate the different perspectives through which ANN and IEMMA view the river's flow dynamics: ANN through hydrological data patterns and IEMMA through isotopic signatures reflecting water mixing and source contributions [70,71].
This comparative analysis indicates that the discrepancies between ANN and IEMMA results in estimating river flow fractions stem from differences in data sources, methodologies, and sensitivities to environmental factors.ANN controls historical hydrological data to predict flow contributions through statistical patterns [5,33], whereas IEMMA uses hydrochemical isotopic signatures to understand water source contributions based on physical-chemical interactions [2,3,24].Variabilities in moisture sources, evaporation effects, land use changes, and assumptions about isotopic homogeneity can alter results between the two approaches [53,54].ANN might not fully account for complex hydrogeochemical interactions, while IEMMA may overlook temporal variations in hydrological data [69].These differences highlight the complementary nature of using both methods for a holistic understanding of river flow dynamics, suggesting an integrated approach can mitigate individual limitations and provide a comprehensive analysis.

Conclusions
The study represents a significant step forward in hydrological research, integrating advanced computational techniques with traditional isotopic analyses to enhance our understanding of river flow dynamics.This comprehensive approach combines the predictive power of artificial neural network (ANN) models with the detailed insights provided by isotopic end-member mixing analysis (IEMMA) based on oxygen-18, offering an insight into the contributions of various water sources to the Mun River flow, particularly from the upper Mun River (UMR) and Chi River (CR).The study highlights both consistencies and discrepancies between ANN model predictions and IEMMA outcomes, featuring the complexity of river flow dynamics.While both methods identify the UMR as a major contributor to Mun River flow, they differ in the specific fractions attributed, reflecting their underlying analytical frameworks.This delineates the significant impact of seasonal variations on river flow contributions, with wet and dry seasons influencing the relative inputs from the UMR and CR.
The integration of ANN and IEMMA provides a holistic view of river flow components, surpassing traditional methods by combining quantitative data analysis with qualitative isotopic insights.This innovative approach enhances model accuracy, addresses methodological limitations, and facilitates a deeper understanding of hydrological processes.The findings have profound implications for hydrological studies and water management in the Mun River Basin and similar contexts.The successful integration of ANN and IEMMA not only enhances an understanding of the Mun River flow components but also sets a precedent for the application of similar methodologies in other river basins globally, marking a significant advancement in the pursuit of sustainable and resilient water systems.Improved insights into flow components support better decision-making regarding water allocation, pollution control, and ecosystem services, contributing to sustainable water resource management and climate change adaptation.

Conclusions
The study represents a significant step forward in hydrological research, integrating advanced computational techniques with traditional isotopic analyses to enhance our understanding of river flow dynamics.This comprehensive approach combines the predictive power of artificial neural network (ANN) models with the detailed insights provided by isotopic end-member mixing analysis (IEMMA) based on oxygen-18, offering an insight into the contributions of various water sources to the Mun River flow, particularly from the upper Mun River (UMR) and Chi River (CR).The study highlights both consistencies and discrepancies between ANN model predictions and IEMMA outcomes, featuring the complexity of river flow dynamics.While both methods identify the UMR as a major contributor to Mun River flow, they differ in the specific fractions attributed, reflecting their underlying analytical frameworks.This delineates the significant impact of seasonal variations on river flow contributions, with wet and dry seasons influencing the relative inputs from the UMR and CR.
The integration of ANN and IEMMA provides a holistic view of river flow components, surpassing traditional methods by combining quantitative data analysis with qualitative isotopic insights.This innovative approach enhances model accuracy, addresses methodological limitations, and facilitates a deeper understanding of hydrological processes.The findings have profound implications for hydrological studies and water management in the Mun River Basin and similar contexts.The successful integration of ANN and IEMMA not only enhances an understanding of the Mun River flow components but also sets a precedent for the application of similar methodologies in other river basins globally, marking a significant advancement in the pursuit of sustainable and resilient water systems.Improved insights into flow components support better decision-making regarding water allocation, pollution control, and ecosystem services, contributing to sustainable water resource management and climate change adaptation.

Geosciences 2024 , 18 Figure 1 .
Figure 1.(a) Map of Thailand.(b) Location of Chi and Mun River basins and rainfall runoff stations.(c) Location of isotope sampling sites (reds) for rainwater and river water.

Figure 1 .
Figure 1.(a) Map of Thailand.(b) Location of Chi and Mun River basins and rainfall runoff stations.(c) Location of isotope sampling sites (reds) for rainwater and river water.

Figure 1 .
Figure 1.(a) Map of Thailand.(b) Location of Chi and Mun River basins and rainfall runoff stations.(c) Location of isotope sampling sites (reds) for rainwater and river water.

Figure 3 .
Figure 3. Structure diagram of artificial neural networks (ANNs) indicating input layer, hidden layers, and output layer.

Figure 3 .
Figure 3. Structure diagram of artificial neural networks (ANNs) indicating input layer, hidden layers, and output layer.

Geosciences 2024 , 18 Figure 5 .
Figure 5. Isotopic composition of precipitation (light-blue ellipsoid with 95% confidence level) and river water including the correlation between δ 18 O and δ 2 H (δ 18 O-δ 2 H relationship) in both rainfall and each river shown as local meteoric water line (rainfall, red line), the river trend line (Mun River (MR, black line), Chi River (CR, dotted line), and upstream Mun River (UMR, dashed line)), and box plot for δ 18 O and δ 2 H.

Figure 5 .
Figure 5. Isotopic composition of precipitation (light-blue ellipsoid with 95% confidence level) and river water including the correlation between δ 18 O and δ 2 H (δ 18 O-δ 2 H relationship) in both rainfall and each river shown as local meteoric water line (rainfall, red line), the river trend line (Mun River (MR, black line), Chi River (CR, dotted line), and upstream Mun River (UMR, dashed line)), and box plot for δ 18 O and δ 2 H .

Figure 6 .
Figure 6.The fraction of upper Mun River (FUMR) and Chi River (FCR) to the main Mun River obtained from the observed river discharge (Q) and the ANN results developed by the stable isotopes (δ 18 O) and flow rate of upper Mun River and Chi River: (a) Training period (80 percent of whole study period of January 2019-December 2021) and (b) testing period (20 percent of whole study period of January 2019-December 2021).

Figure 6 .
Figure 6.The fraction of upper Mun River (F UMR ) and Chi River (F CR ) to the main Mun River obtained from the observed river discharge (Q) and the ANN results developed by the stable isotopes (δ 18 O) and flow rate of upper Mun River and Chi River: (a) Training period (80 percent of whole study period of January 2019-December 2021) and (b) testing period (20 percent of whole study period of January 2019-December 2021).

Figure 7 .
Figure 7.The fraction of upper Mun River (FUMR) and Chi River (FCR) in the main Mun River obtained from the observed river discharge (Q) and the IEMMA results.

Figure 7 .
Figure 7.The fraction of upper Mun River (F UMR ) and Chi River (F CR ) in the main Mun River obtained from the observed river discharge (Q) and the IEMMA results.

Geosciences 2024 , 18 Figure 8 .
Figure 8. Scatterplot of the flow fraction results obtained by IEMMA and observed river discharge showing the correlation between IEMMA results and observed flow rates, with the R² and the 95% significance band indicating the model's reliability.

Figure 8 .
Figure 8. Scatterplot of the flow fraction results obtained by IEMMA and observed river discharge showing the correlation between IEMMA results and observed flow rates, with the R² and the 95% significance band indicating the model's reliability.

Figure 9 .
Figure 9. Box plot of the fraction of upper Mun River (FUMR, dark grays) and Chi River (FCR, reds) to the main Mun River obtained from the ANN results, the IEMMA results, and the observed river discharge (Q).

Figure 9 .
Figure 9. Box plot of the fraction of upper Mun River (F UMR, dark grays) and Chi River (F CR, reds) to the main Mun River obtained from the ANN results, the IEMMA results, and the observed river discharge (Q).

Table 1 .
Detailed information on the catchment characteristics.