Next-Generation River Health Monitoring: Integrating AI, GIS, and eDNA for Real-Time and Biodiversity-Driven Assessment

Abstract

Freshwater ecosystems face escalating degradation, demanding real-time, scalable, and biodiversity-aware monitoring solutions. This review proposes an integrated framework combining artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) to overcome these limitations and support next-generation river health assessment. The AI-GIS-eDNA system was applied to four representative river basins—the Mississippi, Amazon, Yangtze, and Danube—demonstrating enhanced predictive accuracy (up to 94%), spatial pollution mapping precision (85–95%), and species detection sensitivity (+18–30%) compared to conventional methods. Furthermore, the framework reduces operational costs by up to 40%, highlighting its potential for cost-effective deployment in low-resource regions. Despite its strengths, challenges persist in the areas of regulatory acceptance, data standardization, and digital infrastructure. We recommend legal recognition of AI and eDNA indicators, investment in explainable AI (XAI), and global data harmonization initiatives. The integrated AI-GIS-eDNA framework offers a scalable and policy-relevant tool for adaptive freshwater governance in the Anthropocene.

1. Introduction

Freshwater ecosystems play a critical role in maintaining biodiversity, supporting economic development, and ensuring human well-being. However, these systems are increasingly vulnerable to anthropogenic pressures such as eutrophication, urban runoff, industrial discharges, land-use change, and climate variability [1,2,3,4,5,6,7]. These stressors compromise the ecological integrity of rivers and their capacity to provide essential ecosystem services, including clean water, habitat stability, and nutrient cycling [8,9,10,11,12].

Traditionally, river health assessment has relied on physicochemical indicators—such as dissolved oxygen (DO), total nitrogen (TN), total phosphorus (TP), and heavy metals—as well as biological assemblages including macroinvertebrates, fish, and diatoms [13,14,15,16,17,18,19]. Although scientifically robust, these methods are limited by their labor intensiveness, high operational costs, and restricted spatial and temporal resolution [20,21,22,23]. Monitoring typically occurs at fixed intervals and locations, which impedes timely detection of diffuse or rapidly changing pollution events [24,25,26].

Nonetheless, the EU Water Framework Directive (WFD) has endorsed the use of such biological quality elements—namely fish, benthic macroinvertebrates, and algae/macrophytes—based on their ecological relevance, ease of sampling, and cost-effectiveness [27,28,29]. These attributes have enabled wide adoption of bioindicators in water quality assessments under standardized protocols.

Although artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) have each independently contributed to advancements in environmental monitoring, their combined application in river health assessment remains largely underexplored. AI-driven models have demonstrated superior performance over traditional statistical approaches by capturing nonlinear relationships in complex environmental datasets, thereby enhancing the accuracy of water quality forecasting [30,31,32,33,34]. GIS-based spatial analysis improves source detection and supports high-resolution hydrological modeling and pollution dispersion tracking [35,36,37]. Meanwhile, eDNA metabarcoding enables non-invasive, high-sensitivity detection of aquatic biodiversity—including rare and cryptic species—especially in complex or rapidly changing ecosystems [38,39,40,41,42]. Despite these advancements, most studies have employed these technologies independently, without considering their combined or complementary potential. This study aims to address this gap by evaluating the synergistic integration of AI, GISs, and eDNA into a unified, scalable, and real-time river health monitoring framework.

Recent technological advances offer opportunities to overcome these limitations. Artificial intelligence (AI) models—such as Random Forest (RF), Support Vector Machines (SVMs), and Long Short-Term Memory (LSTM) networks—can process large-scale multivariate data to provide near real-time water quality forecasts [8,33,43,44,45,46]. Geographic information systems (GISs) integrate remote-sensing imagery with hydrological and land-use models to identify pollution sources and visualize spatial dispersion across large basins [47,48,49,50,51]. Meanwhile, environmental DNA (eDNA) metabarcoding provides a non-invasive, high-resolution, and cost-efficient tool for detecting a wide range of aquatic species—including rare and cryptic taxa—thereby expanding the scope and sensitivity of biodiversity assessment [38,39,42,52,53]. Although initial setup costs for sequencing and laboratory processing can be significant, these are often offset by reduced field labor and the ability to conduct multi-taxa assessments from a single sample [38,53]. When integrated with AI for predictive modeling and GISs for spatial mapping, eDNA-based data contribute to a synergistic framework that enhances monitoring accuracy and management responsiveness beyond what single-method approaches can offer [39,52,54].

While substantial progress has been made in the development of each of these tools individually, few studies have proposed an integrated monitoring framework that combines AI, GIS, and eDNA technologies [36,55,56,57,58]. Given the growing complexity of freshwater systems and the demand for multidimensional, real-time data, such integration holds great promise for enhancing environmental governance and ecosystem resilience [59,60,61,62,63].

Therefore, the objective of this study is to develop and evaluate an integrated monitoring framework that combines artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) to enhance the resolution, accuracy, and responsiveness of river health assessments. Specifically, this research aims to

(1)

Benchmark the performance of the AI-GIS-eDNA framework against conventional monitoring approaches;

(2)

Examine the technological and methodological synergies among AI, GIS, and eDNA tools;

(3)

Identify the institutional, regulatory, and infrastructural conditions necessary for the scalable adoption of this framework in diverse geographical contexts.

2. Literature Review

Recent decades have witnessed significant evolution in river health assessment approaches, ranging from conventional field-based monitoring to the adoption of advanced digital and molecular tools. This section reviews the key developments in three areas: (1) traditional water quality assessment methods and their inherent limitations; (2) emerging technologies such as artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) for enhanced monitoring; and (3) the remaining research gaps that underscore the need for integrated frameworks. The synthesis aims to contextualize how the integration of multiple approaches can overcome current limitations and support more scalable, accurate, and real-time environmental governance.

2.1. Traditional Water Quality Assessment Methods and Their Limitations

Conventional river health monitoring has primarily relied on physicochemical indicators (e.g., pH, DO, TN, TP, conductivity, heavy metals) and bioindicators such as macroinvertebrates, fish, and benthic diatoms [15,16,17,18,54,64,65,66,67]. While these methods remain fundamental for regulatory compliance, they are often labor-intensive, spatially constrained, and limited in their ability to detect rapid or diffuse environmental changes [10,12,26,68,69,70]. The need for continuous, predictive, and scalable assessment tools highlights the limitations of these traditional approaches.

The need for in situ sampling and laboratory-based analyses limits the spatial resolution and monitoring program frequency [21,22,23]. Moreover, traditional systems lack predictive capability and are incapable of providing early warnings or modeling future risk scenarios. This limitation hinders proactive governance due to rapid environmental changes, such as algal blooms, industrial discharges, or stormwater pulses [71,72,73,74]. These constraints underscore the need for real-time technology-enhanced solutions that enhance monitoring coverage, improve forecasting, and support timely policy responses.

2.2. Emerging Technologies in River Health Monitoring

To address the limitations of conventional monitoring methods, studies have increasingly adopted AI, GISs, and eDNA technologies to support environmental decision making. These tools offer high temporal and spatial resolutions, automation, and predictive modeling capabilities, which collectively enable dynamic and responsive ecosystem management. Figure 1 presents a conceptual illustration showing the integration of these three technologies to form a comprehensive river health assessment system.

AI technologies, particularly machine learning (ML) and deep learning (DL) algorithms, such as RF, Gradient Boosting Machine (GBM), and LSTM networks, are increasingly employed to predict pollution trends, assess ecological risks, and optimize resource allocation [45,50,75,76,77]. These models outperform traditional regression-based methods by analyzing large-scale, nonlinear, and multivariate environmental datasets in real time [55,78]. GIS platforms have significantly enhanced spatial analysis by integrating high-resolution satellite imagery (e.g., Sentinel-2, MODIS, and Landsat-8) with hydrological models to detect land-use changes, identify pollution hotspots, and track transboundary contamination [36,48,57,79,80]. These tools allow for the geospatial modeling of pollution sources and ecosystem fragmentation, facilitating evidence-based land-use planning and watershed governance [81,82]. In parallel, eDNA metabarcoding has emerged as a non-invasive biomonitoring tool that detects species by analyzing DNA fragments present in environmental samples. This approach captures a broader range of taxa than conventional sampling does, including elusive, invasive, and low-abundance organisms [40,53,58,83,84]. Although powerful, eDNA methods face challenges related to reference database completeness, PCR biases, and contamination control [41,62], which must be addressed through standardized protocols and collaborative data-sharing platforms.

2.3. Research Gaps and the Need for an Integrated AI-GIS-eDNA Framework

Despite the independent success of the AI, GIS, and eDNA technologies, their integration into a unified river health monitoring system remains rare. Most applications continue to treat these tools as stand-alone solutions, thereby missing the opportunity to harness their synergistic potential [55,57,61,79,85]. Consequently, current monitoring strategies fail to capture the full spectrum of ecological information necessary for adaptive water governance. Three major gaps have hindered the development of integrated frameworks. First, the lack of data harmonization across platforms and regions—ranging from sampling frequency and spatial scale to data formats and metadata quality—prevents effective AI model integration of GIS layers and eDNA datasets [36,62]. These inconsistencies reduce model accuracy, impair interoperability, and introduce uncertainty in policy-relevant outputs [42,78]. Second, infrastructure and computational limitations, particularly in low- and middle-income countries, constrain the ability to deploy high-performance AI-GIS-eDNA systems. The lack of cloud-computing infrastructure, real-time sensor networks, and bioinformatics pipelines presents major implementation barriers [12,55,86]. Third, the regulatory landscape has not maintained pace with technological innovations. Most environmental policies still prioritize conventional indicators, and a few regulatory agencies recognize AI-driven forecasts or eDNA data as valid forms of assessment [3,58,87,88]. Without standardized validation protocols, legal recognition, and governance frameworks, policymakers are hesitant to rely on these tools for decision making [87,89]. To address these barriers, this study proposes a unified AI-GIS-eDNA monitoring framework that combines ML analytics, geospatial tracking, and molecular biomonitoring to support high-resolution real-time river health assessments. By promoting data standardization, investing in digital infrastructure, and integrating emerging tools into regulatory frameworks, this approach can advance global water governance and enhance freshwater ecosystem resilience [90,91,92].

3. Research Methodology

This study employed a multimethod integrative framework combining AI-based water quality modeling, GIS-based spatial pollution tracking, and eDNA-based biodiversity monitoring to evaluate river health across four major global river systems: the Mississippi, Amazon, Yangtze, and Danube. These systems represent diverse ecological conditions and anthropogenic stress levels, providing a robust empirical basis for evaluating the scalability and cross-ecoregional performance of the AI-GIS-eDNA framework [93,94,95,96,97,98,99,100].

3.1. Study Area and Data Collection

Physicochemical water quality data, such as DO, TN, TP, turbidity, and metal concentrations, were acquired from major international environmental monitoring programs, including those administered by the US EPA, ICPDR, IBAMA, and the Chinese Ministry of Ecology and Environment [16,18,65]. Satellite-based datasets were sourced from the Landsat-8, Sentinel-2, and MODIS platforms, providing enhanced temporal and spatial resolution for land-use classification, surface temperature, and vegetation indices [36,48,82]. Seasonal eDNA sampling was conducted across representative sites in each river basin during the spring, summer, and autumn of 2023. At each site, a total of three eDNA sampling events were performed using sterile single-use filtration units. Water samples were filtered in the field, stored on ice, and transported to the laboratory for processing. DNA was extracted using the Qiagen DNeasy PowerWater kit (Venlo, The Netherlands), targeting barcoding regions such as mitochondrial COI, 12S rRNA, and nuclear 16S/18S rRNA, following established protocols [53,84]. The methods were adapted from best-practice guidelines outlined in Deiner et al. [38], ensuring methodological consistency with validated eDNA workflows. Sequencing was performed using the Illumina MiSeq platform (San Diego, CA, USA) and taxonomic classification was performed using the SILVA and NCBI GenBank databases via QIIME2 and BLAST alignments [40,41]. In addition, the performance of Landsat-8 and Sentinel-2 in retrieving chlorophyll-a and turbidity has been validated in large riverine environments [101]. In eDNA studies, attention has been paid to downstream transport effects and representativeness of biodiversity detection [102]. A broader review of eDNA metabarcoding protocols confirmed the robustness of COI and 12S for freshwater biodiversity assessments [103]. Figure 2 presents a visual summary of the integrated methodology, outlining the full workflow across data acquisition, preprocessing, and cross-domain integration.

3.2. Development of AI-Based Prediction Framework

Historical water quality data (1980–2023) were used to train supervised ML models, including RF, GBM, and LSTM networks. These models were selected for their proven ability to handle nonlinear, high-dimensional environmental data and generate robust predictions of key indicators, such as DO, TN, and TP [33,45,78]. The dataset was preprocessed through missing-value imputation, outlier removal (via Z-score filtering), and min-max normalization. The training and testing datasets were split at an 80:20 ratio, and a 5-fold cross-validation method was used to reduce overfitting and enhance model generalizability. Feature importance was evaluated using Shapley Additive Explanations (SHAP), a game-theoretic interpretability method that quantifies the contribution of each input variable to the model’s output, thereby enhancing transparency in identifying critical environmental predictors [55,104,105]. Model performance was quantified using standard regression evaluation metrics, including R², RMSE (Root Mean Square Error), and MAE (Mean Absolute Error). Comparative benchmarking has revealed that AI-driven predictions achieve over 90% accuracy across most target variables, outperforming traditional regression-based models in terms of prediction accuracy and response time [90,106,107]. Recent studies have underscored the effectiveness of explainable AI (XAI) approaches, such as SHAP, for environmental modeling [105]. Additional ML applications in water quality modeling, including ensemble models and regional optimization strategies, reinforce the generalizability of this framework [108,109].

3.3. GIS-Based Spatial Analysis and Hydrological Modeling

Geospatial analyses were conducted across the four major river systems to detect pollution hotspots and support evidence-based environmental governance. These analyses were performed using ArcGIS Pro and QGIS, integrating satellite-derived land-use data, hydrological base maps, and monitoring station metadata. Sentinel-2, Landsat-8, and MODIS imagery provided temporally and spatially explicit data for environmental layer construction and land-use classification [48,57,82].

Pollution hotspots and land-use gradients were modeled using spatial interpolation techniques, such as Kriging and Inverse Distance Weighting (IDW), which were validated with field-collected concentration data [110,111]. Hydrological simulations were carried out using the Soil and Water Assessment Tool (SWAT) to quantify watershed hydrodynamics, nutrient loading, and sediment fluxes under varying land-use and climate scenarios [53,80]. The SWAT model was calibrated using region-specific parameters adapted for continental and transboundary basins [112].

Topographic features, including slope, elevation, and stream order, were extracted from digital elevation models (DEMs) and used as input features in AI models to enhance spatial prediction accuracy [36,48]. GIS-based outputs were exported as raster layers and integrated into machine learning models, enabling analysis of pollution dispersion patterns in response to anthropogenic land-use changes. Furthermore, these spatial layers facilitated the visualization of biodiversity indicators derived from eDNA metabarcoding, supporting a multidimensional view of ecosystem stress.

3.4. eDNA-Based Biodiversity Assessment

Environmental DNA (eDNA) analysis revealed detailed molecular patterns of freshwater biodiversity across the four river systems. DNA fragments shed by aquatic organisms were collected from surface water samples, filtered in the field, and subsequently extracted in the laboratory. Target gene regions included mitochondrial cytochrome oxidase I (COI), 12S rRNA, and nuclear 18S rRNA, enabling the detection of diverse taxonomic groups such as invertebrates, fish, and amphibians [38,84]. Library preparation followed standardized Illumina protocols, and sequencing was conducted using paired-end 2 × 300 bp reads. Bioinformatics processing was performed via a multi-step pipeline: quality filtering using DADA2, OTU clustering at 97% similarity, and taxonomic assignment using SILVA and GenBank reference databases [41,53]. Comparative evaluations indicate that eDNA-based methods can detect a broader range of taxa—including rare, cryptic, or invasive species—compared to traditional field-based approaches [102,103]. Furthermore, studies focused on diatom communities have validated the use of eDNA in both tropical and temperate aquatic systems as a reliable alternative to microscopy [113].

Importantly, environmental metadata—such as temperature, turbidity, and pH—were integrated with high-throughput sequencing (HTS) results to elucidate pollution–biodiversity relationships. This approach aligns with the recommendations of Rimet et al. [114], who demonstrated the added interpretive value of coupling molecular data with environmental context in comparative assessments of microscopy and eDNA. Finally, the integration of eDNA-derived biodiversity profiles with AI-based pollution forecasting models and GIS-derived spatial datasets enabled the high-resolution detection of ecological perturbations. This combined framework facilitated the identification of biotic community shifts, particularly in zones affected by nutrient enrichment and industrial discharge, thereby advancing the precision and responsiveness of river health assessments.

3.5. Model Performance Evaluation and Benchmarking

The overall performance of the integrated artificial intelligence–geographic information system–environmental DNA (AI–GIS–eDNA) framework was evaluated through a three-pronged approach focusing on prediction accuracy, spatial resolution, and ecological sensitivity. First, the prediction accuracy of the artificial intelligence models was validated against field-based water quality measurements using statistical metrics such as the coefficient of determination (R-squared, R²), root mean square error (RMSE), and mean absolute error (MAE). Second, the spatial resolution of pollution distribution was assessed by comparing geographic information system (GIS)-generated pollution maps with empirical sensor data obtained from national water quality monitoring programs, including the United States Environmental Protection Agency (US EPA) and the International Commission for the Protection of the Danube River (ICPDR). Third, ecological sensitivity was evaluated by comparing species richness and community composition derived from environmental DNA (eDNA) analysis with results obtained from traditional biological assessments using macroinvertebrate and fish assemblages [107,115]. The integrated framework demonstrated notable improvements: prediction accuracy increased by up to 25%, spatial resolution improved by 30–40%, and species detection sensitivity increased by 18–30% compared to traditional methods [88,91]. These advances are consistent with the global calls for real-time, high-resolution monitoring platforms that combine AI and molecular data [116,117]. In addition, infrastructure optimization studies have shown that the AI-GIS-eDNA system can reduce monitoring costs by 30–40% [58]. These results confirm the potential of AI-GIS-eDNA integration as a transformative tool for adaptive water governance, offering early warning capabilities, enhanced spatiotemporal resolution, and comprehensive ecosystem insights that are critical for informing regulatory frameworks and transboundary water policies.

4. Research Findings and Analysis

This study demonstrates that integrating artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) provides a robust and scalable framework for river health assessment. Compared to conventional methods—such as physicochemical sampling and bioindicator surveys—which are often constrained by limited spatial resolution, high operational costs, and time delays [5,57,93,98,99,100,118], the AI-GIS-eDNA approach offers significant advantages in predictive capacity, spatial insight, and biodiversity monitoring. AI-driven models, including Random Forest and LSTM, enabled real-time predictions of water quality across diverse hydrological regimes, reducing dependence on frequent in situ measurements [8,33,43,45,55,78]. GIS analysis supported high-resolution mapping of pollution hotspots and land-use stressors, aiding transboundary environmental governance and watershed planning [47,48,58,82,119,120]. eDNA metabarcoding enhanced the detection of aquatic species, including elusive and low-abundance taxa, thereby capturing biodiversity patterns that are often missed by traditional taxonomic methods [38,39,40,102,121,122]. Application of this integrated framework across the Mississippi, Amazon, Yangtze, and Danube Rivers revealed improved predictive accuracy, broader spatial coverage, and greater ecological resolution. These findings underscore the transformative potential of AI, GISs, and eDNA as complementary tools for real-time, cost-effective, and policy-relevant freshwater monitoring. Nevertheless, to achieve full-scale implementation, efforts must be directed toward standardizing protocols, enhancing digital infrastructure, and fostering international cooperation [7,62,92,96,97,123].

A detailed comparative evaluation of traditional and AI-GIS-eDNA-based monitoring approaches—covering categories such as accuracy, species detection sensitivity, data processing efficiency, and policy responsiveness—is provided in Supplementary Table S1.

4.1. Performance Evaluation of AI Models

This study evaluated the performance of five machine learning (ML) algorithms—Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Long Short-Term Memory (LSTM)—in predicting water quality parameters across the Mississippi, Amazon, Yangtze, and Danube Rivers. Among these, RF and GBM offered the best trade-off between accuracy and computational efficiency, while LSTM and ANN showed slightly superior predictive power at higher computational costs. Overall, model accuracy ranged from 73% to 94%, depending on river system characteristics such as sediment load and data resolution (

Table 1. Performance metrics of five AI models (RF, GBM, SVM, ANN, LSTM) in predicting water quality indicators across four major rivers. Metrics include R², RMSE, MAE, computational load, and interpretability tools.

Performance Metric	Random Forest (RF)	Gradient Boosting (GBM)	Support Vector Machine (SVM)	Artificial Neural Network (ANN)	Long Short-Term Memory (LSTM)
R2 Score	0.85–0.92	0.86–0.91	0.78–0.88	0.89–0.94	0.88–0.93
RMSE (mg/L)	2.45–3.12	2.30–3.05	2.80–3.50	2.10–2.90	2.20–2.85
MAE (mg/L)	1.85–2.60	1.75–2.50	2.00–2.75	1.70–2.40	1.75–2.55
Computational Load	Moderate	Low	High	High	High
Interpretability Tools	SHAP, LIME [126,127]	SHAP, LIME [128,129]	LIME, Permutation [130,131]	SHAP, Grad-CAM [132,133]	SHAP, Attention [105,134]
Optimal Use Case	Robust under missing data; general-purpose [135]	Effective with high-dimensional datasets [136]	Small datasets with low noise; less suited for time-series [130]	Long-term predictions with large data [137]	Time-series environmental modeling [138]

SHAP (SHapley Additive exPlanations); LIME (Local Interpretable Model-agnostic Explanations); CAM (Computer-Aided Manufacturing).

). Validation using five-fold cross-validation yielded confidence intervals of ±2.5–9.2%, indicating model robustness across hydrological gradients [106,124,125].

4.2. GIS-Based Pollution Dispersion Analysis

GIS-based monitoring integrated remote-sensing imagery with hydrological and land-use layers to improve spatial analysis of pollution dispersion [48,57,82]. Across all rivers studied, GIS-based spatial modeling demonstrated high accuracy in identifying major sources of pollution, with precision values ranging from 85% to 91% (

Table 2. Spatial accuracy of GIS-based pollution source detection and corresponding policy responses in the Mississippi, Amazon, Yangtze, and Danube Rivers. Accuracy values are derived from ground-truth validation using remote sensing and hydrological overlays.

River System	GIS Accuracy (%)	Primary Pollution Source	Policy Response	Key References
Mississippi	91%	Agricultural runoff	Nutrient control zones, riparian buffer enforcement	[57,97,118,139]
Amazon	90%	Microplastic accumulation	Urban waste separation, microplastic source tracking	[51,110,140,141]
Yangtze	87%	Industrial effluents	Discharge regulation, zoning of manufacturing corridors	[48,142]
Danube	85%	Urban wastewater	Cross-border monitoring platforms, regulatory alignment	[81,102,111,120]

). These results confirm the utility of remote sensing combined with hydrological overlays in delineating pollution dispersion patterns and informing local to transboundary management strategies. In particular, GIS mapping supported applications such as nutrient zoning in the Mississippi, urban waste policy reform in the Amazon, industrial discharge tracking in the Yangtze, and cross-border monitoring in the Danube [97,110,120]. Integration with AI further enhances zoning enforcement and real-time diagnostics of pollution events [51,111]. These spatial tools are especially valuable for collaborative basin management in politically complex or ecologically sensitive watersheds.

4.3. eDNA-Based Biodiversity Monitoring

eDNA metabarcoding enables the non-invasive detection of aquatic biodiversity by sequencing trace genetic material shed into the environment [38,103,121]. Compared with traditional taxonomic surveys, eDNA provides significantly higher detection sensitivity and broader taxonomic coverage, especially in ecologically complex or under-monitored freshwater systems [39,122]. In this study, eDNA techniques applied across the four major river systems revealed an 18–30% increase in species detection compared to conventional methods, with particularly high biodiversity observed in the Amazon Basin (

Table 3. Comparative biodiversity metrics based on eDNA metabarcoding across four river systems. Includes average read counts, species richness, detection of rare taxa, and dominant taxonomic groups identified using QIIME2 pipelines and reference databases.

River System	Total eDNA Reads (Reads/Sample Avg.)	Detected Species (Taxonomic Richness)	Rare Species (%)	Dominant Taxa Identified	Key References
Mississippi	120,500	158	18%	Fish, Mollusks, Amphibians, Crustaceans	[20,75,102]
Amazon	97,800	189	24%	Fish, Amphibians, Reptiles, Crustaceans	[20,60,141]
Yangtze	105,200	141	22%	Fish, Crustaceans, Macroinvertebrates, Amphibians	[7,18,25,120]
Danube	113,900	167	21%	Fish, Mollusks, Macroinvertebrates, Amphibians	[9,17,106,144]

). These findings confirm the ability of eDNA to identify both rare and cryptic taxa, expanding the scope of ecological assessments. Challenges related to DNA degradation, incomplete reference databases, and contamination were addressed through stringent quality control measures, including replicate PCR runs and taxon-specific bioindicator protocols [84,102]. While variability in detection was noted—especially in the Danube system due to sampling inconsistencies—this underscores the need for international standardization of eDNA protocols to improve reproducibility [62,120]. When integrated with AI-based predictions and GIS-derived pollution maps, eDNA monitoring enhances early-warning capabilities for ecosystem degradation and supports targeted conservation planning [7,81,143]. Embedding eDNA methods into regulatory frameworks may significantly strengthen the scientific basis for long-term ecological resilience and policy enforcement.

4.4. Policy Implications and Future Applications of AI-GIS-eDNA Monitoring

The AI-GIS-eDNA framework offers a transformative pathway for modernizing environmental governance and ensuring the proactive management of freshwater ecosystems. AI facilitates predictive alerts and near-real-time diagnostics, GISs support spatial zoning, risk visualization, and pollutant dispersion mapping, and eDNA enables molecular-level biodiversity surveillance and invasive species detection [40,134,145]. In transboundary or shared river systems, harmonizing these technologies promotes cross-jurisdictional data interoperability and enhances cooperative decision-making processes [81,102,142]. Dashboards powered by AI analytics can reduce administrative burdens and enhance the precision of regulatory enforcement [126,146,147,148]. Nonetheless, full-scale deployment requires substantial investment in sensor networks, cloud infrastructure, standardized workflows, and interagency alignment [3,12,58,62]. Embedding such integrative frameworks in environmental legislation and monitoring institutions can pave the way for anticipatory governance, improved risk communication, and evidence-based freshwater policies [90,92,123].

5. Discussion

5.1. Advancing River Health Monitoring: From Conventional to Integrated Approaches

The integration of artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) represents a paradigm shift in river health assessment. Traditional approaches, including physicochemical measurements and bioindicator-based surveys, remain foundational but face well-documented limitations such as temporal delays, spatial inflexibility, and high operational costs—especially in large or transboundary river systems [145].

In contrast, the AI-GIS-eDNA framework addresses these limitations by enabling near-real-time diagnostics and expanding ecological resolution. AI models reduce the need for frequent field sampling through accurate prediction of key parameters such as total nitrogen and phosphorus. GISs support the spatial mapping of pollution dispersion and identification of anthropogenic stressors. eDNA offers sensitive, non-invasive biodiversity assessment, capable of detecting rare and cryptic taxa often missed by traditional methods [38,98,149].

These advantages were reflected in our findings: ML models such as RF and LSTM demonstrated high prediction accuracy across all four river systems, while GIS analysis effectively localized pollution hotspots. When coupled with eDNA-based biodiversity profiles, these tools enhanced monitoring precision and ecological interpretation. Moreover, interpretability remains a critical challenge for the adoption of AI in regulatory settings. To address this, we incorporated explainable AI (XAI) techniques—such as SHAP and LIME—which can improve transparency and facilitate policy integration by making model decisions more interpretable to stakeholders [105,129,147,148].

In summary, the synergistic integration of AI, GISs, and eDNA enables more proactive, cost-effective, and policy-relevant river health monitoring. This multidimensional approach offers a scalable path forward in an era of increasing environmental uncertainty and data complexity. Furthermore, successful implementation of this integrated framework will require robust legal and institutional support to ensure data standardization, regulatory acceptance, and ethical compliance, thereby enabling its broader adoption and long-term sustainability.

5.2. Performance and Cost-Efficiency of the AI-GIS-eDNA Framework

The results demonstrate that the integrated AI-GIS-eDNA framework consistently outperforms conventional methods in both technical performance and operational efficiency. AI-based models achieved up to 94% accuracy in predicting key water quality indicators such as total nitrogen (TN), total phosphorus (TP), and dissolved oxygen (DO), while GIS-based spatial modeling delivered 85–95% precision in identifying pollution hotspots across variable hydrological contexts [78,97,104]. eDNA metabarcoding enables the non-invasive detection of aquatic biodiversity by sequencing trace genetic material shed into the environment [38,103,121]. Compared with traditional taxonomic surveys, eDNA provides significantly higher detection sensitivity and broader taxonomic coverage, especially in ecologically complex or under-monitored freshwater systems [39,122]. In our study, eDNA monitoring further enhanced species detection sensitivity by 18–30%, offering superior biodiversity resolution in complex ecosystems compared to traditional methods [41,150]. Beyond technical performance, the framework also yielded substantial cost savings: automation reduced long-term monitoring expenditures by up to 40%, decreased field labor demands by approximately 35%, and lowered biodiversity assessment costs by around 20% [38,58]. In practical terms, each component of the integrated system serves a complementary policy function: AI supports real-time anomaly detection and regulatory forecasting; GISs inform land-use planning and pollutant zoning; and eDNA enables early biodiversity warnings and conservation planning. These combined capabilities make the framework both scientifically robust and financially viable for scalable implementation in regional and transboundary water governance.

5.3. Implementation Challenges and Policy Considerations in Resource-Limited Regions

To empirically address the third objective, we reviewed national water policies and legal instruments associated with each river system (e.g., U.S. Clean Water Act, China’s Ecological Civilization framework, ICPDR treaties), and analyzed their compatibility with AI, GIS, and eDNA applications. We also evaluated data integration capacities and infrastructural readiness by compiling institutional mandates, sequencing capabilities, and cross-agency collaboration status. These results are based on qualitative synthesis of regulatory documents and implementation. Despite the strong technical potential of the integrated AI-GIS-eDNA monitoring framework, its implementation in low-income or resource-limited regions presents notable challenges. To address these disparities, we propose a tiered implementation model tailored to varying regional capacities. This model enables the initial deployment of more affordable and accessible components—such as GIS-based mapping and community-assisted eDNA sampling—followed by gradual integration of AI-based analytics as technical infrastructure and institutional capacity mature.

Key enablers for such progression include capacity-building partnerships between technologically advanced and resource-constrained regions, alongside the promotion of open-source tools and citizen science platforms to reduce financial and logistical barriers [22,77].

From a governance perspective, successful adoption also requires favorable legal and institutional environments. Our review of the four focal regions reveals considerable heterogeneity in regulatory readiness:

• In the Mississippi River Basin, federal agencies such as the U.S. EPA and USGS have begun integrating AI-based water quality prediction models under existing regulatory mandates [120].
• The Amazon Basin, though governed by multiple bilateral treaties, has yet to formally incorporate eDNA into national biodiversity assessments, despite increasing regional interest in molecular tools for ecological diagnostics [148].
• In China’s Yangtze River, national ecological modernization initiatives have supported pilot applications of AI for environmental forecasting, with ongoing development of regulatory instruments for emerging indicators [127].
• The Danube River, managed by the International Commission for the Protection of the Danube River (ICPDR), represents a successful transboundary governance model. Recent ICPDR initiatives have included exploratory discussions on integrating novel indicators such as eDNA and AI-based classifiers [79].

These examples underscore the necessity of aligning technological innovation with region-specific institutional and policy contexts. Ensuring legitimacy and scalability will require both legal recognition and operational capacity. Accordingly, we advocate for internationally coordinated guidelines—potentially through platforms such as GEO BON or UNEP—that promote standardization and interoperability of next-generation river health metrics [132,151]. Integration with AI further enhances zoning enforcement and real-time diagnostics of pollution events [51,111]. These spatial tools are especially valuable for collaborative basin management in politically complex or ecologically sensitive watersheds.

5.4. Institutional and Regulatory Challenges

Despite its strong technical performance, several barriers hinder the institutional adoption of the AI-GIS-eDNA framework. First, most regulatory frameworks (e.g., the EU Water Framework Directive, U.S. Clean Water Act) continue to rely on traditional indicators, limiting the acceptance of AI-based predictions or eDNA-derived metrics [3,107]. Second, a lack of standardized data formats, regional inconsistencies in eDNA reference databases, and sequencing protocol variation reduce interoperability and reproducibility [87,127]. Third, the opacity of deep learning algorithms (“black box” models) poses challenges for regulatory transparency.

To address these challenges, we recommend (1) updating legal frameworks to recognize emerging indicators, (2) investing in open-access digital infrastructure and reference libraries, and (3) expanding the use of explainable AI (XAI) tools such as SHAP and LIME to build institutional trust and improve interpretability [89,105]. Furthermore, our comparative analysis across the four case study regions revealed that scalable adoption is more feasible in jurisdictions where legal frameworks explicitly accommodate non-traditional indicators, where cloud computing infrastructure and satellite monitoring capabilities are established, and where environmental data are shared across institutional boundaries. For example, the Danube River region benefits from institutional harmonization through the ICPDR, facilitating cross-border data interoperability and eDNA pilot initiatives. In contrast, the Amazon Basin exhibits infrastructural constraints, despite growing interest in molecular diagnostics.

Importantly, when integrated with AI-based predictions and GIS-derived pollution maps, eDNA monitoring substantially enhances early-warning capabilities for ecosystem degradation and supports targeted conservation planning [7,81,143]. Embedding eDNA methodologies into formal regulatory frameworks may significantly strengthen the scientific basis for long-term ecological resilience and policy enforcement. As such, institutional adoption should prioritize not only technical validation but also legal recognition and procedural integration of these next-generation monitoring tools.

Figure 3 summarizes the multilevel governance strategies needed to operationalize the framework at scale.

5.5. Toward Scalable and Equitable Adoption

While technically sound, the scalability of this framework remains a concern in low-income or data-scarce regions. Variability in AI model training datasets, satellite resolution, and eDNA reference coverage can compromise cross-regional applicability [55,134]. Furthermore, limited access to high-throughput sequencing, cloud-based computing, and sensor infrastructure hampers deployment in developing regions. Although long-term costs are reduced by automation, initial investments remain high. To overcome this, we recommend the adoption of cost-sharing models, mobile-based sequencing units, and open-source AI platforms to lower both capital and operational barriers. For example, Google Earth Engine has supported cost-effective satellite-based monitoring in sub-Saharan Africa, while Oxford Nanopore’s MinION (Oxford, UK) has enabled portable eDNA sequencing in remote regions such as the Amazon Basin. Broader support for global initiatives like GBIF and the Earth BioGenome Project will further strengthen the foundational infrastructure needed for equitable implementation [123,144].

5.6. Future Research and Implementation Priorities

To unlock the full potential of AI-GIS-eDNA systems, future research should focus on several key areas. First, development of explainable AI (XAI) methods will be critical for policy integration and stakeholder engagement. Second, advances in portable sequencing technologies and curated taxonomic databases will enhance detection precision, particularly for underrepresented taxa. Third, integration with IoT-enabled sensor networks, UAV-based remote sensing, and real-time analytics can further improve scalability and system responsiveness [147,148,152,153,154]. These directions will help transform the framework from a research prototype into a globally scalable tool for evidence-based water governance. Additional supporting studies and detailed comparative evaluations are summarized in Supplementary Table S1 [155,156,157,158,159,160].

6. Conclusions

This study introduces and evaluates an integrated monitoring framework that combines artificial intelligence (AI), geographic information systems (GISs), and environmental DNA (eDNA) to enhance the responsiveness, accuracy, and ecological relevance of river health assessment. Applied across four global river basins, the AI-GIS-eDNA framework demonstrated superior performance in predictive water quality modeling, spatial pollution detection, and biodiversity monitoring, while also reducing long-term operational costs. These findings validate the framework’s utility for real-time and data-rich environmental governance. However, widespread adoption requires targeted reforms, including the legal recognition of AI and eDNA-derived indicators, improved data standardization, and expanded digital infrastructure. With sustained investment in explainable AI, open-access tools, and cross-border collaboration, this framework can support anticipatory and equitable water governance under accelerating environmental pressures.