Abstract
Transitional water ecosystems exhibit pronounced spatio-temporal variability and increasing anthropogenic pressures, posing substantial challenges for ecological monitoring and management. Artificial Intelligence (AI), particularly Machine Learning (ML) and Deep Learning (DL), has emerged as a powerful framework for addressing the structural complexity of these systems. This systematic review synthesizes peer-reviewed studies applying ML and DL to ecological research and monitoring in transitional waters. A structured search of the Scopus® database was conducted up to 31 December 2024, and studies were screened according to predefined eligibility criteria and PRISMA 2020 guidance; methodological quality was appraised using a structured assessment framework. Ninety-six studies met the inclusion criteria. Regression was the most frequent analytical task (44.1%), followed by classification (36.2%) and clustering (19.7%), with water quality monitoring representing the dominant thematic domain. Tree-based and kernel-based ML models prevailed overall, whereas DL architectures increased markedly after 2020, particularly in remote sensing and high-dimensional applications. Despite methodological heterogeneity and variable validation practices, the evidence indicates that ML and DL approaches effectively accommodate non-linearity, data heterogeneity, and scale mismatches typical of transitional waters. Standardized validation strategies and improved model interpretability remain essential for robust ecological inference and operational implementation.
1. Introduction
1.1. Historical Development of AI
AI is commonly defined as the discipline concerned with designing computational systems that display goal-directed behaviour in complex environments [1]. The evolution of AI has opened new frontiers in aquatic ecosystem research. In recent years, ML and DL have revolutionized many scientific fields, including aquatic ecology [2].
Over the past decade, AI has advanced rapidly, driven by (i) algorithmic innovation, (ii) increased computational power, and (iii) the widespread availability of open-source neural-network software frameworks [3]. These developments have accelerated adoption across multiple domains, including medicine [4], biology [5], finance [6], manufacturing [7], engineering [8], and entertainment [9].
During the 1990s, ensemble approaches (e.g., boosting and bagging) and RF became influential methods for data analysis and prediction, complementing and, in some settings, outperforming classical statistical baselines. Contemporary ML is commonly framed in terms of supervised and unsupervised learning paradigms, as well as reinforcement learning for sequential decision-making, implemented through algorithms ranging from RF and neural networks to clustering techniques and modern deep-learning architectures [10].
Machine learning seeks to develop models that excel in making data-driven predictions, setting itself apart from traditional approaches that rely on specific assumptions about data generation processes [11]. In ecological applications, ML is typically implemented through a modelling workflow that can include (i) data transformation and feature engineering, (ii) supervised or unsupervised learning depending on label availability and study objectives, and (iii) optimization-based calibration (e.g., parameter estimation and hyperparameter tuning) under explicit validation designs. Representative examples include data transformation workflows [12,13], clustering-based ecological stratification [14], and optimization frameworks for parameter or hyperparameter selection [15,16].
Deep learning uses ANN models, composed of multiple interconnected layers, and represents a highly effective computational method [17]. ML underpins several operational AI technologies used in ecological monitoring, including automated classification and anomaly detection, and data-driven forecasting such as autonomous driving systems have significantly improved applications like image recognition, speech processing, and language translation [18]. Unlike traditional ML, which typically depends on human expertise to manually identify relevant data features—often challenging when dealing with complex or unclear features—deep neural networks automatically identify critical patterns within datasets. Furthermore, deep learning is increasingly utilized in ecology and evolutionary biology for diverse tasks, including citizen science initiatives, environmental monitoring, genetic sequencing, population genetics, and phylogenetic studies.
In the last decade, deep neural networks (DL) have seen renewed growth, primarily driven by advances in graphics processing unit (GPU) technology. These developments have significantly improved performance across various domains, including image analysis, language processing, and ecological modeling.
In transitional water ecosystems ecology and evolution, the adoption of ML and DL has spurred significant advances. AI has been applied to automated species identification from imaging and acoustic recordings [19,20,21,22,23], and to the extraction of organismal traits or behavioural proxies from high-frequency sensor, video, or bio-logging data [24,25,26,27,28]. Additionally, these approaches have been instrumental in predictive ecosystem modeling, contributing substantially to our understanding and management of these vital environments. For example, ML and DL have been used to fill missing links in ecological networks (e.g., Ref. [29]), integrate with traditional mechanistic models [30,31], approximate differential equations [32,33]. AI has also been used to build species distribution models by learning relationships between occurrence data and environmental covariates, including both correlative ML baselines and DL-based spatial predictors [34,35,36,37].
Despite their growing popularity, the complexity of ML and DL models often leads them to be perceived as “black boxes,” raising concerns about their explainability and relationship with traditional statistical models. However, recent work on XAI and emerging causal-inference approaches has begun to complement purely predictive modelling by improving transparency and supporting hypothesis generation. However, these methods do not automatically confer causal interpretability, and their ecological validity depends on study design, confounder control, and careful model diagnostics.
Integrating ML, DL, and traditional statistics has created a powerful methodological framework, transforming complex data analysis in AI. This interdisciplinary synergy continues to drive innovation in ML and data science.
1.2. Why a Review on AI Applied to Transitional Water Ecosystems?
Transitional water ecosystems are ecologically and socio-economically significant, supporting high biodiversity and providing essential ecosystem services such as storm protection, water filtration, and fisheries support. However, these ecosystems are increasingly vulnerable to climate change, anthropogenic pressures, and pollution, which threaten their biodiversity and ecological functions [38].
Integrating AI methods—particularly ML, DL, and remote sensing—has strengthened ecological research in transitional waters by enabling early detection of harmful algal blooms, habitat mapping, and quantitative assessment of environmental stressors [39]. More broadly, AI-based predictive analytics can support conservation and management by extracting actionable signals from large, heterogeneous environmental datasets and by anticipating climate- and human-driven impacts [40].
This review analyzes methodologies, challenges, and future opportunities for AI in transitional water ecosystems, contributing to scientific discourse on improving conservation and sustainable management in these environments.
Transitional water ecosystems pose a distinctive set of methodological challenges that limit the effectiveness of traditional statistical and mechanistic modelling approaches. These systems are characterized by strong spatio-temporal non-stationarity, abrupt regime shifts driven by episodic events (e.g., storms, freshwater inflows, heatwaves), and non-linear interactions among physical, chemical, and biological processes. Moreover, ecological responses are often modulated by scale mismatches between drivers (e.g., watershed-scale nutrient loading) and observations (e.g., point-based monitoring stations), as well as by highly heterogeneous and partially observed data streams.
From a data perspective, research in transitional waters must frequently contend with:
- high-dimensional predictor spaces arising from remote sensing, multi-sensor platforms, and omics-derived variables;
- missing, irregularly sampled, or noisy observations, particularly in long-term monitoring programs;
- limited labeled data, especially for biological states or rare events (e.g., hypoxia, harmful algal blooms);
- complex response surfaces that violate linearity, normality, and independence assumptions commonly required by classical statistical models.
These characteristics motivate the adoption of AI-based approaches not merely for their predictive accuracy, but because they offer algorithmic mechanisms capable of approximating complex, non-linear mappings, integrating heterogeneous data sources, and operating under weaker distributional assumptions than traditional parametric models. The relevance of AI in this context therefore lies in its alignment with the intrinsic structure of transitional-water data, rather than in generic claims of robustness or computational efficiency.
1.3. Origins of ML and DL in Relation to Statistics
ML and DL have their origins deeply rooted in the principles of statistics and AI. Since the 1950s (Figure 1), researchers have sought to develop computational models that learn from data, infer underlying patterns, and make autonomous decisions, thereby emulating aspects of human cognition.
Figure 1.
Classification of AI methodologies applied in ecological research. The diagram illustrates the main categories of supervised and unsupervised learning methodologies commonly used for ecological monitoring and management in transitional water ecosystems.
Statistics has long provided the conceptual and methodological framework for data analysis and the modeling of complex phenomena. In addition to Bayesian statistics, key concepts such as maximum likelihood estimation (MLE) and null hypothesis significance testing (NHST) were established in the early twentieth century. These traditional parametric approaches involve formulating a model for data generation and computing the likelihood of observing specific outcomes based on assumed parameters. Pioneers such as Fisher, Neyman, and Pearson advanced these methods by introducing confidence intervals (CI) and p-values for parameter estimation and hypothesis testing, approaches that, however, recent critiques have highlighted recurring methodological risks—such as data leakage, inadequate treatment of spatio-temporal dependence, limited external validity, and insufficient ecological interpretability—particularly when models are deployed beyond the calibration domain [41,42,43,44].
Initially, practical statistical methods required simplistic models to ensure tractable probability calculations. However, the advent of early computers and novel numerical techniques like Markov Chain Monte Carlo (MCMC) methods [45] enabled the adoption of more complex parametric models. This progress paralleled advancements in ecological analysis [46]. Nonetheless, the inherent complexity of natural systems [47] often necessitated basic and inflexible statistical models, as computing likelihoods for intricate models remained mathematically challenging. Consequently, conclusions drawn from conventional statistical techniques frequently depended on oversimplified model assumptions [11].
The modern paradigm of ML emerged in part due to the availability of large datasets and the exponential increase in computational power. The 1990s witnessed a surge of innovation, with algorithms such as boosting, bagging, and RF surpassing traditional statistical models in flexibility and predictive performance. These developments firmly established ML as a fundamental approach to data analysis across diverse scientific disciplines and practical applications.
DL, which originated in the 1980s, initially faced limitations due to computational constraints and insufficient training data. Early neural network models showed promise but were restricted in scope. It was not until the advent of modern deep neural network architectures and the widespread use of GPUs that DL achieved its current potential. Throughout these developments, statistics has continued to play a crucial role by providing the theoretical underpinnings for model validation, result interpretation, and uncertainty quantification. Statistical concepts such as Bayesian inference, information theory, and sampling techniques have significantly influenced the evolution and application of ML algorithms.
Today, the integration of ML, DL, and statistical methodologies has culminated in a sophisticated methodological ecosystem, revolutionizing our approach to complex problems in data analysis and AI. This dynamic interplay among complementary disciplines continues to drive innovation and progress in ML and data science.
Evolution of AI in Transitional Water Ecosystems
Over the last two decades, the adoption of Artificial Intelligence methodologies in transitional water ecosystem research has followed a progressive and application-driven trajectory, closely linked to advances in data availability, monitoring technologies, and ecological research needs.
Early applications, emerging in the early 2000s, primarily relied on relatively simple machine-learning and neural-network models to support regression-based tasks, such as the estimation of water-quality variables (e.g., light attenuation coefficients, chlorophyll-a concentration) and the development of empirical ecological indicators. These studies represented the first attempts to move beyond purely mechanistic or linear statistical approaches by exploiting data-driven relationships in complex aquatic environments.
Between approximately 2010 and 2014, the diffusion of tree-based ensemble methods and kernel-based algorithms (e.g., Random Forests and Support Vector Machines) marked a substantial methodological shift. These approaches proved particularly well suited to transitional water ecosystems, where non-linear responses, strong interactions among predictors, and heterogeneous data structures are common. During this period, AI applications expanded to include species distribution modelling, aquaculture productivity assessment, and habitat suitability analysis in lagoons and estuarine systems.
From 2015 onwards, the integration of machine-learning models with remote-sensing data significantly broadened the spatial and temporal scope of ecological analyses. Ensemble methods and supervised classifiers were increasingly applied to land-use mapping, trophic status classification, and large-scale habitat assessments in coastal and transitional waters. This phase was characterized by the consolidation of AI as an operational tool for ecosystem monitoring rather than a purely experimental methodology.
More recently (from approximately 2020 to the present), advances in deep learning have enabled the analysis of high-dimensional and weakly structured data sources, including satellite imagery, underwater video, acoustic recordings, and high-frequency sensor time series. Convolutional and recurrent neural network architectures are now widely used for automated species detection, fine-scale habitat mapping, and short-term forecasting of ecological variables such as algal blooms and water-quality dynamics. These developments have facilitated near-real-time monitoring and the handling of data volumes that would be impractical using traditional analytical approaches.
In parallel, increasing attention has been devoted to model diagnostics, validation, and interpretability, particularly in response to concerns regarding the opacity of complex AI models. Recent studies have therefore explored the use of explainable AI techniques to support ecological interpretation, hypothesis generation, and decision-making, while acknowledging that explainability does not equate to mechanistic understanding.
Overall, the evolution of AI in transitional water ecosystem research reflects a shift from exploratory, regression-focused applications toward integrated, data-intensive frameworks capable of addressing the intrinsic spatio-temporal complexity of these systems. This progression underscores that the relevance of AI in this domain lies not in generic methodological novelty, but in its capacity to align with the specific structural and ecological challenges of transitional waters.
2. ML and DL in Transitional Water Ecosystem Research
Transitional water ecosystems (e.g., coastal lagoons, estuaries, deltas, and coastal wetlands) are characterized by pronounced spatio-temporal heterogeneity and strong coupling among physical, chemical, and biological processes. Salinity gradients, hydrodynamics, episodic disturbances, and multiple anthropogenic pressures (e.g., eutrophication, contaminant inputs, habitat alteration) generate complex, often non-linear system responses. In this context, AI—operationally discussed here through ML and DL—is increasingly adopted to (i) extract information from heterogeneous data streams (in situ sensors, remote sensing products, acoustic/video monitoring, environmental DNA (eDNA), and omics), (ii) support ecological forecasting and early-warning systems (e.g., hypoxia events, harmful algal blooms), and (iii) enable scalable monitoring and decision support. This section provides a methodological framing explicitly aligned with the aims of the present review, which compares AI approaches used in transitional waters according to task type, model family, application domain, and publication trends.
2.1. ML
ML is a pivotal technology underpinning many advances in the understanding and management of transitional water ecosystems. Unlike DL, which primarily leverages deep neural networks, ML encompasses a broader array of algorithms enabling computers to learn from data and generate predictions or decisions. In practice, ML approaches are commonly organized into supervised and unsupervised learning paradigms, as well as reinforcement learning for sequential decision-making, each offering distinct strategies for modelling complex environmental systems.
A defining characteristic of many ML methods is their algorithmic focus on optimizing predictive performance through the minimization of a broad loss function, which does not necessarily correspond to a probabilistic model of the observed data [11,48]. This paradigm reduces reliance on strict distributional assumptions (typical of classical inference) and has motivated non-parametric strategies for uncertainty assessment. In this context, bootstrap procedures have become widely used to estimate confidence intervals for parameters and predictions in both statistical and ML settings [49]. Similarly, cross-validation—partitioning data into training and test subsets to quantify generalization error—has become central to ML evaluation and is now feasible at scale due to modern computational resources [50,51].
In transitional waters, ML is widely applied to support core operational and scientific goals, including the prediction of algal blooms, monitoring of fish migration, assessment of water quality, and detection of pollution [52]. Within the analytical taxonomy adopted in this review, these applications are typically expressed as regression (e.g., estimating continuous indicators such as chlorophyll-a, turbidity, dissolved oxygen, nutrients), classification (e.g., bloom/non-bloom, ecological status classes, presence–absence outcomes), and, less frequently, clustering (e.g., grouping stations or periods by similarity to detect latent regimes).
ML can rapidly analyze large ecological datasets, identify trends, and deliver timely predictions compared to purely manual or low-dimensional approaches; however, its effectiveness depends on data quality, feature definition, and appropriate algorithm tuning, particularly under the strong space–time variability typical of transitional systems [53].
From a methodological perspective, the suitability of specific ML algorithms for transitional water ecosystems is closely linked to the structural properties of the underlying data rather than to generic performance claims. For instance, tree-based ensemble methods such as RF are particularly well adapted to ecological datasets characterized by non-linear responses, strong interaction effects, and multicollinearity among predictors—conditions that frequently arise in transitional waters due to coupled physical–biogeochemical processes.
The recursive partitioning mechanism of RF enables the approximation of complex response surfaces without requiring explicit specification of functional forms or interaction terms, making it effective for modeling threshold-driven ecological dynamics (e.g., hypoxia onset, eutrophication tipping points). Furthermore, the bootstrap aggregation and random feature selection steps reduce variance and mitigate overfitting in high-dimensional predictor spaces, which are common when integrating remote sensing, in situ monitoring, and ancillary environmental data.
Similarly, kernel-based methods such as Support Vector Machines (SVM) are well suited for transitional-water applications where sample sizes are limited but the feature space is high-dimensional, as often occurs in biological monitoring or water-quality assessment based on multi-sensor observations. By maximizing the margin between classes or fitting flexible regression functions in transformed feature spaces, SVMs can capture non-linear ecological relationships while maintaining controlled model complexity.
These algorithmic properties explain the widespread adoption of ML models in transitional-water research and clarify that their relevance arises from their ability to accommodate data heterogeneity, non-linearity, and scale mismatches intrinsic to these ecosystems.
2.2. DL
DL, a subset of ML, has transformed approaches to complex problems frequently encountered in transitional water ecosystems. DL relies on ANN models composed of multiple interconnected layers that learn hierarchical representations directly from data. This capacity is especially advantageous when relevant ecological information is embedded in high-dimensional or weakly structured inputs (e.g., satellite imagery, hyperspectral products, underwater video, passive acoustics, and high-frequency sensor time series), where manual feature engineering may be limiting.
In aquatic and transitional-water contexts, DL methods have been used to detect and classify marine species using imagery and sonar [54], and more broadly to support event forecasting (e.g., toxic algal blooms) and ecosystem health assessment in data-intensive workflows. These approaches can enhance predictive accuracy and support biodiversity-informed management by extracting non-linear relationships and complex patterns that are difficult to model with simpler formulations. At the same time, DL applications remain constrained by (i) the requirement for large, well-annotated training datasets, often difficult to obtain in transitional waters due to logistical constraints and natural variability; and (ii) challenges in model interpretability, which can limit scientific inference and operational uptake when decision-making requires transparent justification [55].
2.3. Advantages and Disadvantages of AI Applications
The application of AI (particularly ML and DL) to transitional water ecosystems offers substantial potential alongside notable challenges. On the one hand, AI can process large volumes of environmental data rapidly, enabling near-real-time monitoring and the timely synthesis of information from multiple sources [56]. This capability supports the detection of latent ecological relationships that may remain undetected under traditional analytical workflows [55], improves predictive performance for ecosystem dynamics when models are appropriately designed and validated [57], and strengthens conservation and management strategies by automating biodiversity monitoring, habitat assessment, and detection of environmental stressors [58].
On the other hand, the reliability of AI outputs is strongly dependent on the availability, quality, and representativeness of training data; biased or incomplete datasets can produce unreliable predictions and misleading inferences, particularly under the non-stationary conditions typical of transitional waters [57]. Furthermore, many high-performing models—especially DL architectures—exhibit limited transparency, complicating ecological interpretation and stakeholder trust [55]. Finally, the computational demands of model development and deployment can be substantial, with implications for accessibility and sustainability [56]. These methodological constraints, together with broader ethical and ecological risks associated with rapid deployment and misapplication, motivate a cautious and explicitly ecological framing of AI use in transitional-water monitoring and decision support [59].
2.4. Supervised and Unsupervised Learning
Artificial intelligence (AI) has become increasingly important in studying and managing transitional water ecosystems. A key methodological distinction in this domain concerns supervised versus unsupervised learning approaches (Figure 1). Supervised learning relies on labeled datasets to learn a mapping between predictors and outcomes, and it is widely used for tasks such as aquatic species classification, water quality forecasting, and organism distribution modelling based on historical observations [60,61,62]. In contrast, unsupervised learning does not require labeled data and aims to identify latent structure within multivariate datasets; in transitional waters, it is particularly useful for clustering sites or communities according to shared environmental characteristics and for exploratory analysis of time series to detect ecosystem changes [63,64]. In this review, these paradigms provide an organizing lens for synthesizing the methodological choices and application domains reported in the selected literature.
2.4.1. Supervised Learning
- What can be done
Building on the supervised-learning paradigm introduced in Section 2.4, labeled ecological observations can be leveraged to support prediction and decision-making in transitional water ecosystems. Typical applications include:
- Improve Classification Accuracy in Ecological Research: Supervised learning techniques such as Decision Trees, Random Forest, and SVMs have been successfully used to classify ecological datasets, improving prediction accuracy and efficiency [65].
- Optimize Model Selection and Hyperparameter Tuning: A comparative study of supervised learning algorithms has shown that tuning hyperparameters can significantly impact model performance, making it essential for improving prediction outcomes [66].
- Apply Supervised Learning to Predict Ecological Events: Machine learning models have been applied to predict ecological phenomena, such as mass mortality events in marine ecosystems, demonstrating the practical application of supervised learning in conservation research [67].
- Enhance Model Generalization and Adaptability: Studies suggest that incorporating multi-modal data in supervised learning models can improve generalization and adaptability in diverse ecological applications.
- Leverage Ensemble Learning for Improved Performance: Combining multiple supervised learning models using ensemble techniques such as bagging and boosting can enhance ecological modeling outcomes by reducing variance and bias [68].
- Use Supervised Learning for Real-time Monitoring: AI-driven models are increasingly being integrated with remote sensing and environmental monitoring systems to enable real-time ecological assessments [69].
- How can it be done
To implement supervised learning robustly in ecological research, a structured workflow is typically required, including algorithm selection, data preparation, model optimization, and rigorous evaluation. Key steps include:
- Choosing the Right Algorithm: Different supervised learning algorithms, such as Decision Trees, Random Forest, ANN models, and SVM models, can be applied to ecological datasets. Each method has unique advantages depending on the type and structure of the data [65].
- Data Preprocessing and Feature Selection: To improve accuracy, supervised models require proper data cleaning, normalization, and feature selection. Methods such as Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE) help refine datasets before model training [70].
- Hyperparameter Tuning for Optimization: Model performance can be enhanced through hyperparameter tuning techniques like Grid Search, Random Search, or Bayesian Optimization, ensuring that models generalize well to new ecological data [70].
- Training with Labeled Data: Supervised learning requires a sufficient amount of labeled data for effective model training. Large-scale datasets, such as GPS telemetry data in ecological studies, improve classification accuracy [71].
- Evaluating Model Performance: Performance evaluation metrics, such as Accuracy, Precision, Recall, F1-score, and Area Under the Curve (AUC), help assess the reliability of supervised learning models in ecological research [70].
- Using Ensemble Learning for Robust Predictions: Combining multiple models using ensemble learning techniques, such as Bagging, Boosting (e.g., XGBoost), and Stacking, enhances prediction stability and reduces errors [70].
- Applying Supervised Learning to Real-world Ecological Problems: Supervised learning models are used to classify animal behavioral states from environmental variables, predict species distribution, and analyze biodiversity conservation strategies [70].
The following table (Table 1) summarizes the primary supervised algorithms used in these fields.
Table 1.
Supervised learning model.
In summary, supervised learning algorithms allow ecologists to effectively analyze large ecological datasets, turning complex information into actionable insights. These tools are vital for accurate ecosystem monitoring, supporting proactive conservation and sustainable management strategies in aquatic environments.
2.4.2. Unsupervised Learning
- What can be done
As outlined in Section 2.4, unsupervised learning methods support exploratory analysis when labels are unavailable or incomplete, a frequent condition in transitional water monitoring. In aquatic ecosystem research, these techniques can uncover hidden patterns and relationships, helping to identify ecological clusters and detect temporal or spatial shifts. Key applications include:
- Clustering Ecological Data: Unsupervised learning techniques such as density-based clustering (DBSCAN) and k-means clustering have been successfully used to identify ecological patterns in marine environments [72].
- Detecting Anomalies in Ecosystems: Unsupervised learning is effective in anomaly detection, identifying unusual ecological events such as extreme climate conditions and pollution spikes without the need for labeled training data [73].
- Dimensionality Reduction for Complex Ecological Models: Techniques like t-SNE and PCA help in reducing high-dimensional ecological data, making it easier to visualize and interpret complex environmental interactions [72].
- Identifying Hidden Structures in Biodiversity: Unsupervised learning methods have been applied to analyze plankton community structures and nutrient flux patterns, leading to new insights in marine biodiversity research [72].
- Feature Extraction for Environmental Monitoring: Unsupervised neural networks have been used to extract meaningful features from satellite and sensor data, enhancing environmental monitoring efforts [74].
- Improving Conservation Strategies: By clustering species distribution data, unsupervised learning has been instrumental in helping conservationists identify priority areas for protection [75].
- How can it be done
Implementing unsupervised learning effectively typically involves careful preprocessing, algorithm selection, and validation tailored to exploratory objectives. Key steps include:
- Preprocessing and Feature Selection: Before applying unsupervised learning, datasets should be preprocessed to remove noise, handle missing values, and standardize data for better clustering and pattern detection [76].
- Choosing the Right Clustering Algorithm: Different clustering methods, such as k-means, hierarchical clustering, and density-based clustering (DBSCAN), are commonly used in ecological studies to classify species or environmental regions [72].
- Using Dimensionality Reduction Techniques: PCA and t-SNE are valuable tools for reducing high-dimensional ecological data, making it easier to visualize and analyze underlying patterns [77].
- Anomaly Detection for Ecological Monitoring: Unsupervised learning models are effective in detecting anomalies in environmental data, such as pollution events or habitat degradation, by identifying deviations from normal patterns [73].
- Leveraging DL for Unsupervised Feature Extraction: Deep learning models, such as autoencoders, can automatically learn features from large ecological datasets, enhancing species identification and habitat classification [77].
- Evaluating Model Performance: Unlike supervised learning, unsupervised models require different validation techniques, such as silhouette scores and clustering validity indices, to assess the quality of the detected patterns [78].
The following table (Table 2) summarizes key unsupervised algorithms commonly employed in aquatic ecology and evolutionary biology research.
Table 2.
Evolution of Artificial Intelligence.
These unsupervised learning models serve a variety of purposes, from elucidating ecological relationships and community structures to identifying critical conservation areas. For instance, dimensionality reduction and clustering are frequently used to delineate ecological patterns and eco-provinces, highlight dominant environmental drivers of change, and support conservation prioritization by identifying ecological hotspots and areas at risk under climate variability and anthropogenic pressures [72,73,74,75]. Considerations related to the interpretability of complex models are addressed in Section 2.5.5 [79].
2.5. Evaluating and Improving AI Models
As AI becomes more prevalent in ecological research, strong evaluation frameworks are essential to maintain model reliability, accuracy, and transparency. AI models often face challenges in generalization, especially in ecology, where complex interactions and dynamic behaviors add layers of difficulty. Overcoming these obstacles requires interdisciplinary strategies that integrate AI techniques with ecological expertise [55].
2.5.1. Model Evaluation
Assessing AI models in ecological research is crucial to ensure their predictive accuracy. While standard metrics like accuracy, precision, recall, and F1-score are widely used, ecological data often demand more specialized evaluation methods. Techniques such as cross-validation, uncertainty analysis, and domain-specific error assessments help determine the robustness and reliability of AI models [80].
2.5.2. Error Diagnosis and Analysis
AI models can exhibit systematic biases or errors due to incomplete datasets, incorrect assumptions, or improper model architectures. Conducting error analysis through confusion matrices, residual analysis, and adversarial testing helps diagnose weaknesses in model predictions. Furthermore, integrating AI with physical-based ecological models has been shown to improve prediction accuracy by incorporating fundamental environmental principles [57].
2.5.3. Model Optimization and Tuning
Hyperparameter tuning is a crucial step in AI model improvement. Techniques such as grid search, Bayesian optimization, and evolutionary algorithms allow researchers to fine-tune model parameters for optimal performance. Additionally, incorporating domain knowledge into model design can lead to more ecologically meaningful results, reducing overfitting and enhancing generalization [55].
2.5.4. Feature Analysis
Optimizing AI models requires effective feature selection and engineering to improve parsimony and computational efficiency while preserving ecological relevance. Choosing ecologically meaningful variables—such as habitat distribution, environmental factors, and species interactions—ensures that models remain scientifically valid. Techniques like PCA and RFE help streamline datasets by removing redundant or irrelevant variables [81].
2.5.5. Interpretability and Explainability
A critical methodological issue in the application of AI to ecological research concerns the distinction between interpretability and explainability, two concepts that are often conflated in the literature but refer to fundamentally different properties of models.
Interpretability denotes the extent to which a model’s internal structure and parameters can be directly understood by a human observer. Interpretable models are typically characterized by a transparent mathematical formulation or a limited number of parameters with clear semantic meaning, such as linear regression models or shallow decision trees. In these cases, the relationship between predictors and outcomes can be directly inspected and reasoned about without auxiliary tools.
Explainability, by contrast, refers to the use of post hoc analytical techniques designed to approximate, summarize, or probe the behavior of complex and intrinsically opaque models. Most machine-learning models widely adopted in transitional water ecosystem research—including Random Forests, gradient-boosting methods, and deep neural networks—are not intrinsically interpretable due to their ensemble-based or multi-layered architectures. In such cases, model-agnostic explainability methods (e.g., SHAP, LIME, partial dependence analysis) can be employed to identify influential predictors, characterize local sensitivities, or generate simplified representations of model responses.
Importantly, explainability does not confer interpretability in a strict sense, nor does it imply causal understanding. XAI methods provide descriptive insights into how a trained model behaves under specific data distributions, but their outputs remain contingent on the model structure, the training data, and the assumptions embedded in the explainer itself. Consequently, explanations derived from XAI should be interpreted as diagnostic or hypothesis-generating tools rather than as mechanistic representations of ecological processes.
In the context of transitional water ecosystems, this distinction is particularly relevant because management and policy decisions often require transparent justification. While explainable AI can enhance trust and facilitate communication with stakeholders, robust ecological inference still depends on study design, data quality, and the integration of domain knowledge, rather than on explainability techniques alone.
Accordingly, throughout this review, models such as RF, gradient boosting methods, and DL architectures are discussed as non-interpretable models, whose outputs may nonetheless be explored using explainability tools for hypothesis generation and model validation purposes.
2.5.6. A/B Testing and Continuous Monitoring
Deploying AI models in ecological applications requires continuous evaluation and refinement. A/B testing—comparing different model versions on real-world ecological data—helps identify the most effective approaches. Additionally, long-term monitoring through feedback loops ensures that AI models remain adaptive to environmental changes and evolving datasets [57].
By implementing these evaluation and improvement strategies, AI models can become more reliable and transparent, ultimately contributing to more effective ecological research and conservation efforts. Future developments should focus on integrating AI with traditional ecological modeling techniques while maintaining explainability and ethical considerations.
3. Scope and Objective
This review synthesizes the international peer-reviewed literature on the application of ML and DL to transitional water ecosystems. Specifically, we analyze 96 peer-reviewed studies (including journal articles, review papers, and peer-reviewed conference proceedings indexed in Scopus) and systematically classify them according to AI task type, AI model family, application field, and year of publication. The objectives are to (i) provide an evidence-based overview of how AI is currently used for monitoring and ecological assessment in transitional waters, (ii) identify methodological and thematic trends across the literature, and (iii) highlight recurring limitations and research gaps that should be addressed to support robust ecological inference and operational implementation. The remainder of the manuscript is organized to first describe the review methodology and then present and discuss the main patterns emerging from the classified studies.
4. Materials and Methods
4.1. Review Design and Reporting Framework
This study was conceived as a systematic literature review aimed at synthesizing peer-reviewed evidence on the application of AI—with a specific focus on ML and DL—to ecological research and monitoring in transitional water ecosystems. To ensure transparency and reproducibility, the review process was structured in sequential phases (identification, screening, eligibility assessment, and inclusion) and documented using a PRISMA-style flow scheme (Figure 2). The review questions and the analytical framework (classification by AI task type, AI model type, application field, and year of publication) were defined a priori and applied consistently throughout study selection and data extraction.
Figure 2.
PRISMA 2020 flow diagram of study selection (databases/registers).
4.2. Eligibility Criteria
Studies were considered eligible if they met all of the following criteria:
- Population/system: the study addressed transitional water ecosystems (e.g., coastal lagoons, estuaries, deltas, coastal wetlands, or clearly transitional aquatic environments);
- Intervention/method: the study applied AI approaches, with a specific focus on ML and/or DL, including hybrid or ensemble strategies, to ecological or environmental datasets;
- Outcomes/aims: the study targeted ecological research and/or monitoring objectives in transitional waters (e.g., assessment, forecasting, mapping, biodiversity evaluation, pollution detection, habitat characterization, or closely related management applications);
- Document type and source: peer-reviewed publications indexed in Scopus providing sufficient methodological detail for extraction and classification, including journal articles, review papers, and peer-reviewed conference proceedings;
- Language: English-language publications;
- Accessibility: full text available.
Exclusion criteria were:
- Studies not focused on transitional waters (e.g., exclusively open-ocean, purely freshwater, or terrestrial systems without a transitional component);
- Papers not implementing AI/ML/DL methods (e.g., purely descriptive studies, conceptual perspectives without implementation, or exclusively mechanistic modelling without an AI component);
- Excluded document types: non-peer-reviewed or non-research content (e.g., editorials, commentaries, meeting abstracts) and other records lacking sufficient methodological detail for robust extraction; additionally, records outside the eligible document types above (e.g., book notes, short communications without methods) were excluded during eligibility assessment;
- Records without accessible full text and/or insufficient information to classify AI task/model/application domain.
No a priori time restriction was imposed to capture the temporal evolution of AI methods in this field.
4.3. Information Source and Search Strategy
The bibliographic search was performed in the Scopus® database (Elsevier, Amsterdam, The Netherlands), selected for its broad coverage of peer-reviewed journals in environmental sciences and interdisciplinary research. The search was conducted on 31 December 2024. Searches were performed using a structured query string combining AI-related terms with descriptors of transitional water ecosystems. The query was applied to the title, abstract, and keywords as follows:
(TITLE-ABS-KEY (machine AND learning AND coastal AND lagoons) OR TITLE-ABS-KEY (machine AND learning AND transitional AND water)).
To enhance replicability, we explicitly report the screening constraints applied after retrieval. The Scopus® query was executed without time limits and without restricting document type at the search stage. During eligibility assessment, we retained English-language, peer-reviewed publications relevant to transitional water ecosystems that implemented AI/ML/DL approaches and provided sufficient methodological detail for extraction. Eligible records included journal articles, review papers, and peer-reviewed conference proceedings indexed in Scopus. Records such as editorials, commentaries, meeting abstracts, and other items lacking adequate methodological transparency were excluded and are summarized among the full-text exclusion categories in the PRISMA-style flow scheme (Figure 2).
To reduce the risk of missing relevant studies not captured by keyword combinations, we complemented the database query with targeted manual screening of reference lists of included papers (backward citation searching) and, where appropriate, identification of closely related articles suggested by the database interface (citation-based discovery). All retrieved records were exported and managed in a structured screening workflow.
4.4. Study Selection Process (PRISMA 2020)
Study identification and selection followed the PRISMA 2020 (Figure 2) reporting framework and its standard flow diagram for new systematic reviews based on database searches. The Scopus® search (executed on 31 December 2024) retrieved 102 records. No duplicate records were detected during de-duplication (records removed before screening: n = 0). All records (n = 102) were screened on title and abstract; one record was excluded at this stage (records excluded: n = 1). The remaining 101 reports were sought for retrieval and all were successfully retrieved (reports not retrieved: n = 0). Full-text eligibility was assessed for 101 reports; five reports were excluded for predefined reasons: (i) not focused on transitional waters (n = 2); (ii) no AI/ML/DL implementation (n = 1); (iii) ineligible document type or insufficient methodological detail for extraction (n = 2). Accordingly, 96 studies were included in the qualitative synthesis and subsequent classification.
4.5. Data Extraction and Coding Scheme
A standardized extraction matrix was used to collect key attributes from each included study and ensure consistent classification across the evidence base. Extracted items included: bibliographic information (authors, year), ecosystem type and geographic context, data source (e.g., in situ monitoring, remote sensing, acoustic/video, laboratory measurements), AI task type (regression, classification, clustering, and/or detection/segmentation where applicable), AI model family (e.g., tree-based models, kernel methods, neural networks, CNN/RNN/LSTM architectures, hybrid/statistical models), application field (e.g., water quality monitoring, habitat mapping, biodiversity assessment, pollution detection), and evaluation/validation approaches (e.g., cross-validation, train/test split, performance metrics). The extracted variables were then used to populate the classification tables (Appendix A) and to support the synthesis presented in the Results.
4.6. Quality Assessment and Risk of Bias
To enhance methodological transparency and align with PRISMA 2020 recommendations, a structured quality appraisal of included studies was conducted. Given the methodological heterogeneity of AI-based ecological research, a domain-adapted quality assessment framework was developed.
Each study was evaluated according to five predefined criteria:
- clarity of study objectives and ecological context;
- transparency of dataset description and preprocessing;
- appropriateness of AI model selection relative to research question;
- robustness of validation strategy (e.g., cross-validation, external validation, spatio-temporal independence);
- reporting of performance metrics and uncertainty measures.
Each criterion was scored on a three-level scale (low, moderate, high quality). Two reviewers independently assessed methodological quality, and discrepancies were resolved through discussion.
The quality appraisal was used to identify recurrent methodological limitations but was not employed to exclude studies from the qualitative synthesis.
4.7. Synthesis Methods
The synthesis combined descriptive and exploratory analyses. First, studies were summarized through frequency counts and distributions across task types, model families, application fields, and publication years. Second, multivariate exploratory methods (including cluster analysis and non-metric multidimensional scaling) were used to examine temporal patterns and thematic similarities among studies, supporting the identification of emerging methodological trends and application domains (Table A1, Table A2, Table A3, Table A4, Table A5 and Table A6; Figure 2, Figure 3, Figure 4, Figure 5 and Figure 6).
Figure 3.
Distribution of AI tasks in ecological research. The chart shows the proportion of regression, classification, and clustering tasks used in studies analyzing transitional water ecosystems, highlighting their frequency and relevance. Only the most frequent categories are displayed; categories with very low frequencies (<2%) are omitted for clarity. Therefore, percentages do not sum to 100%.
Figure 4.
Main research topics in AI applications for transitional water ecosystems. The figure summarizes the primary ecological and environmental research themes where Artificial Intelligence methodologies (ML and DL) have been widely applied. Only the most representative categories are displayed; categories with very low frequencies (e.g., single occurrences) are omitted for clarity. Therefore, percentages do not sum to 100%.
Figure 5.
Distribution of AI model types applied in ecological research. The figure illustrates the relative frequency of different AI methodologies, including traditional ML, DL, and statistical models, used in studies of transitional water ecosystems. Only the most representative categories are displayed; categories with very low frequencies (e.g., single occurrences) are omitted for clarity. Therefore, percentages do not sum to 100%.
Figure 6.
Temporal distribution of publications on AI in ecological research. The figure shows the evolution over time of scientific articles applying ML and DL methods to transitional water ecosystems, highlighting growth trends and periods of increased research activity.
4.8. PRISMA-Style Flow Diagram
Figure 2 reports the study selection process using the official PRISMA 2020 flow diagram template for new systematic reviews (databases/registers). The diagram explicitly documents the number of records identified, removed prior to screening, screened, excluded, assessed for eligibility, and included, together with the reasons for full-text exclusions, thereby ensuring transparency and reproducibility of the selection process.
5. Results
Under this systematic review, the final evidence base comprised 96 peer-reviewed journal articles. To facilitate interpretation of the descriptive synthesis, Appendix A provides four complementary summary tables. Table A1 reports the study inventory used for coding (Paper_ID) and links each included record to the analytical categories adopted in this review. Table A2 summarises the frequency of AI task types (regression, classification, and clustering), noting that multiple tasks may co-occur within the same study; therefore, task occurrences can exceed the number of included articles. Table A3 reports the distribution of application topics addressed by the reviewed literature, highlighting the dominant thematic areas investigated in transitional-water contexts. Finally, Table A4 summarises the AI model families adopted across studies (e.g., ML, DL, and statistical/linear approaches), thereby providing a compact overview of methodological prevalence prior to the trend analyses reported in the following subsections.
5.1. Classification of Types of AI Tasks
AI tasks can be broadly categorized based on their objectives and learning methods. Among these, three primary types—Regression, Classification, and Clustering—are particularly relevant in ecological and environmental research (Table A2, Figure 2). These methods provide essential analytical tools for predicting relationships between variables, categorizing ecological data, and identifying patterns within complex datasets.
Across the reviewed corpus, regression represents the most frequently reported task (56 occurrences), followed by classification (46) and clustering (25) (Table A2). The total exceeds the number of included studies (n = 96) because a substantial fraction of papers reported more than one analytical objective, for example, coupling regression-based estimation (e.g., continuous water-quality variables) with classification (e.g., trophic status categories) or using clustering as an exploratory step to define ecological typologies prior to supervised modelling.
- Regression
Regression is a fundamental AI task that involves predicting a continuous output variable based on input features. It is widely used in ecological and environmental sciences for predicting temperature changes, pollution levels, and species population trends. Techniques such as linear regression, support vector regression, and neural networks have been applied in various fields, including climate modeling and biodiversity assessments [81]. AI-enhanced regression models improve predictive accuracy by integrating large datasets, making them particularly useful for forecasting ecological dynamics [55].
- Classification
Classification tasks involve assigning predefined labels to data points. In ecological AI applications, classification is often used for species identification, habitat categorization, and pollution detection. Traditional methods such as decision trees, RF models, and SVMs have been supplemented by DL models, which can achieve high classification accuracy in image recognition and remote sensing applications [58]. Recent advancements in hybrid AI-human classification models have further improved performance by combining automated predictions with expert validation [82].
- Clustering
Clustering is an unsupervised learning task that groups similar data points together. It is widely used in ecological modeling to identify environmental patterns, group species with similar characteristics, and analyze habitat distributions. Common clustering algorithms include k-means, hierarchical clustering, and density-based clustering (DBSCAN). In AI-driven ecological research, clustering has been instrumental in defining ecosystem boundaries and detecting emerging environmental threats [55].
- Other AI Tasks and Their Exclusion from This Study
Beyond the three primary categories explored in this study, other AI methodologies are commonly used in various scientific domains:
- Reinforcement Learning, which optimizes decision-making in dynamic environments but requires simulated scenarios beyond the scope of this research.
- Anomaly Detection, which is valuable for identifying rare ecological events, but is less applicable to structured datasets such as those used in this study.
- Generative AI, which synthesizes new data, is primarily beneficial for model training augmentation rather than direct ecological analysis [83].
This study focuses on Regression, Classification, and Clustering due to their well-established applications in ecological modeling and environmental analysis. These methodologies offer powerful tools for predicting environmental changes, classifying ecological features, and detecting patterns in complex datasets. While other AI approaches have their merits, the selected methods provide the most direct, interpretable, and actionable insights within the scope of this research.
- Findings on AI Task Type Distribution
The analysis of 96 studies revealed that Regression tasks are the most prevalent, followed by Classification and Clustering. Figure 2 illustrates the distribution of these AI task types, demonstrating their respective applications across various ecological domains.
- Regression (44.1%)—Used primarily for environmental forecasting, such as predicting pollution levels, climate trends, and species population dynamics.
- Classification (36.2%)—Applied extensively in species identification, habitat categorization, and environmental monitoring, leveraging decision trees, neural networks, and SVMs.
- Clustering (19.7%)—Used for grouping ecological data, identifying environmental patterns, and defining ecosystem boundaries through unsupervised learning techniques.
These results highlight the dominance of predictive and classification-based AI models in ecological research while showcasing the emerging role of clustering techniques in defining ecological relationships.
The findings reinforce the importance of regression-based predictions for ecological forecasting, classification for species and habitat identification, and clustering for uncovering hidden patterns in environmental datasets. Future advancements may further integrate these approaches to enhance AI-driven conservation and monitoring efforts.
5.2. Research Topics
The application of AI in ecological and environmental sciences spans a wide range of research topics. The following section categorizes the key research domains where AI techniques have been extensively applied, highlighting the primary challenges and opportunities they address (Table A3, Figure 3).
- Water Quality Monitoring
AI has been instrumental in assessing and predicting water quality parameters such as pollutant concentrations, nutrient levels, and turbidity. These applications play a crucial role in ecosystem health monitoring, sustainable water resource management, and public health protection [84]. Machine learning models, including Random Forest (RF) and ANN models, have been applied to detect anomalies and forecast changes in water quality over time [85].
- Land Use and Coastal Ecosystem Mapping
AI-driven models contribute significantly to land use classification and coastal ecosystem mapping, which are essential for tracking habitat distribution and assessing environmental changes due to anthropogenic pressures [86]. Remote sensing data, coupled with DL techniques such as CNN models, enhance the accuracy of these mappings [87]. These methodologies support conservation efforts and urban planning strategies by providing high-resolution spatial analyses.
- Marine Ecology and Environmental Monitoring
Machine learning and DL techniques are widely used to model species distribution, track biodiversity patterns, and monitor ecosystem changes over time [88]. AI-based predictive models enable the assessment of marine ecosystem resilience and inform policy decisions on marine conservation. Data-driven AI models also facilitate the real-time detection of harmful algal blooms and other environmental hazards [89].
- Climate Change and Ecosystem Responses
AI-driven climate models are employed to predict the impact of climate change on ecosystems. These models integrate large-scale climate data with ecological variables to project species adaptation, habitat shifts, and ecosystem resilience [90]. Hybrid AI models combining statistical and ML approaches offer improved forecasting accuracy for climate-related ecological changes [91].
- Other Specialized Topics
AI applications extend to niche areas such as environmental genomics, aquaculture management, and behavioral ecology [92]. AI-powered genomics helps in biodiversity conservation through genetic sequencing and evolutionary studies, whereas ML models optimize aquaculture by predicting species growth rates and monitoring fish health. Behavioral ecology studies leverage AI to analyze species interactions and movement patterns using tracking data [93].
- Findings on Research Topics
By structuring AI research topics in this manner, we provide a framework that reflects the fundamental environmental questions AI aims to address. This classification aids researchers and policymakers in navigating the literature and underscores the unique challenges and opportunities associated with each domain.
5.3. Distribution of AI Model Types
AI models used in ecological and environmental sciences exhibit significant methodological diversity, reflecting the complexity of the studied ecosystems. In this review, AI models were categorized into three primary types: ML Models, DL Models, and Statistical and Linear Models. This classification aims to highlight their applicability, strengths, and limitations in environmental monitoring and ecological research (Table A4, Figure 4).
- ML Models
ML models, including RF, SVMs, and ensemble methods, dominate AI applications in ecology due to their robustness in handling large datasets and uncovering complex patterns [94]. These models are widely applied in species distribution modeling, habitat classification, and water quality prediction. Studies have demonstrated that RF and SVM achieve high accuracy in predictive ecological models by effectively leveraging multi-dimensional datasets [95].
Importantly, the prevalence of ML models in the reviewed literature reflects not only their predictive performance but also their alignment with the typical constraints of transitional-water datasets. These datasets often combine moderate sample sizes with high-dimensional predictor spaces, irregular temporal sampling, and strong correlations among environmental variables. Under such conditions, ML approaches that rely on flexible function approximation and weak distributional assumptions offer practical advantages over classical parametric models.
In contrast, DL models are preferentially adopted when the dominant data modality is unstructured or weakly structured (e.g., imagery, acoustic signals, dense time series), whereas traditional statistical models remain relevant when data availability, interpretability, and hypothesis-driven inference are prioritized.
- DL Models
Deep learning (DL) approaches, such as CNN models and RNN models, have revolutionized AI applications in ecology due to their superior performance in analyzing unstructured and high-dimensional data, including images, spatial maps, and time-series data. These models have been instrumental in remote sensing applications, biodiversity assessment, and marine ecosystem monitoring [96]. CNN models, in particular, have enhanced the classification accuracy of land cover and ecosystem mapping by learning intricate spatial features [97].
- Statistical and Linear Models
Traditional statistical models, including linear regression, generalized linear models (GLMs), and Bayesian models, continue to be relevant in ecological research. These models provide baseline predictions, hypothesis testing frameworks, and interpretable analytical tools, making them essential for understanding the behavior of more complex AI models. Despite the rise of DL techniques, statistical models remain integral to many ecological studies, particularly in climate modeling and population dynamics [98].
- Other Specialized AI Models
While ML, DL, and statistical models dominate AI research, other model categories exist, focusing on domain-specific ecological applications, algorithmic innovations, and simple explanations [99]. XAI models are particularly crucial in ecological research, ensuring transparency, model interpretability, and trust in AI-generated predictions [95].
- AI Model Type Distribution and Trends
The reviewed literature indicates a clear prevalence of ML models, followed by DL and statistical methods, with specialized approaches emerging as niche areas [53]. This trend highlights the ongoing refinement of AI methodologies to ensure accuracy, interpretability, and ecological relevance in environmental research.
- Findings on AI Task Type Distribution
By structuring the classification of AI models in this manner, we provide a systematic framework for researchers and policymakers, enabling a clearer understanding of how different AI methodologies align with specific ecological challenges. These findings emphasize the importance of selecting appropriate AI models based on data characteristics, research objectives, and the need for interpretability in ecological decision-making.
5.4. Temporal Distribution of Publications
The evolution of research in AI applied to ecological and environmental sciences has followed a significant trajectory, reflecting both technological advancements and shifting research priorities. Analyzing the temporal distribution of publications allows us to identify key trends in AI applications and their adoption in environmental monitoring and conservation (Table A5, Figure 5).
- Trends in AI Research Over Time
The temporal analysis of AI-related publications indicates a notable increase in research output starting from 2018, with a particularly sharp rise after 2020. This surge aligns with several factors, including:
- Advancements in computational power, particularly the accessibility of GPUs and cloud computing, which have facilitated the adoption of ML and DL approaches.
- An increase in high-quality environmental datasets, made available through remote sensing satellites and sensor networks, enabling the training of more sophisticated AI models.
- A growing emphasis on climate change research, leading to increased funding and international collaboration on AI-driven ecological monitoring.
- Key Periods of Growth and Innovation
The review of publication trends highlights distinct periods of accelerated AI adoption:
- Before 2015: the use of AI in ecological research was minimal, with a strong reliance on traditional statistical models and early ML methods like Decision Trees and Regression models.
- 2015–2019: A steady rise in publications, with increased experimentation using SVMs and RF models for species distribution modeling and environmental classification.
- 2020–2024: A substantial increase in AI publications, driven by the rise of DL methodologies such as CNN models for remote sensing, as well as hybrid AI approaches integrating ML with ecological models.
The temporal distribution of AI research thus provides valuable insights into the maturation of the field, helping to contextualize both past achievements and future directions.
From an ecological data perspective, the post-2020 increase in deep-learning studies is consistent with the growing availability of high-resolution remote-sensing products and continuous sensor records, for which convolutional and sequence-based architectures are often methodologically appropriate, provided that validation accounts for spatio-temporal dependence and domain shift across sites and seasons.
- Impact of AI on Environmental Research Priorities
The increasing number of AI-driven ecological studies over the past decade suggests a shift in research priorities:
- Early AI applications (before 2015) were primarily focused on species monitoring and environmental forecasting, relying on structured datasets with limited computational requirements.
- Recent studies (2020–2024) have expanded to real-time monitoring, anomaly detection, and predictive modeling, enabled by DL and big data analytics.
- A parallel trend in XAI has emerged to enhance model interpretability, addressing concerns about the transparency and reliability of AI models in ecological decision-making.
These trends reflect both technological progress and an increasing urgency in tackling climate-related environmental challenges, positioning AI as a critical tool in ecological research and conservation strategies.
The analysis of temporal publication trends underscores the accelerated integration of AI in environmental sciences, with a shift from traditional statistical models to more sophisticated ML and DL approaches. This growth reflects improvements in data availability, computational resources, and global environmental concerns, shaping the future of AI-driven ecological research.
5.5. Temporal Trends and AI Task Types Combination
The evolution of AI task types in ecological and environmental research follows a distinct temporal pattern, influenced by technological advancements, increasing computational resources, and the availability of high-resolution datasets. By analyzing publication years alongside AI task types—Regression, Classification, and Classification, this section highlights the shifts in modeling approaches and their alignment with research priorities over time (Table A6, Figure 6).
- Historical Evolution of AI Task Types
Before 2015, AI use in environmental sciences was largely centered on regression models, which played a key role in continuous environmental monitoring, species distribution estimation, and climate projections. Methods like linear regression, support vector regression (SVR), and Gaussian processes were commonly employed for their clarity and effectiveness in handling structured ecological data.
Between 2015 and 2019, classification models gained prominence, driven by the growing availability of labeled datasets and the adoption of tree-based methods, RF models, and SVMs. These methods were particularly useful for land cover classification, biodiversity assessments, and pollution detection.
From 2020 onwards, the increasing availability of high-resolution remote sensing data and long-term environmental monitoring datasets coincided with a rapid adoption of deep learning models, primarily for supervised tasks such as classification, regression, segmentation, and time-series forecasting. While unsupervised methods, including clustering, continue to play a role in exploratory ecological analyses, deep learning architectures are most commonly employed either in supervised settings or as feature extraction mechanisms, where learned representations are subsequently analyzed using conventional clustering algorithms.
Therefore, the observed use of clustering in recent studies should not be interpreted as a direct consequence of deep learning adoption, but rather as part of hybrid analytical pipelines combining representation learning with traditional unsupervised techniques.
- Factors Influencing the Shift in AI Task Types
The increasing adoption of classification and segmentation tasks in recent years can be attributed to:
- Enhanced computational capabilities, including cloud computing and GPUs, have facilitated the training of complex models.
- The growing availability of high-resolution environmental data, especially from satellite imagery and remote sensing, has expanded analytical possibilities.
- The demand for real-time ecological monitoring has increased, necessitating models that can detect anomalies and identify environmental patterns.
- Escalating concerns over climate change and biodiversity loss have fueled the adoption of advanced AI-driven analytical approaches.
- AI Task Type Combinations Over Time
- Before 2015: Regression models were dominant, supporting species distribution and pollution forecasting.
- 2015–2019: Classification tasks emerged, used in species identification and land use mapping.
- 2020–2024: Increased use of clustering methods was observed, primarily as part of exploratory analyses or hybrid workflows combining deep-learning-based feature extraction with traditional unsupervised techniques, rather than as a direct consequence of deep learning adoption.
This temporal shift in AI applications underscores a progressive transition toward higher-dimensional, real-time, and data-driven ecological monitoring, reflecting both methodological advancements and increasing environmental research needs.
By integrating temporal trends with AI task types, this analysis provides a comprehensive timeline of AI evolution in ecological research (Figure 7). The observed patterns highlight how AI methodologies have expanded from simple predictive models (regression) to sophisticated classification and segmentation approaches, driven by technological innovation and environmental urgency. Future research is expected to further refine detection models and enhance explainability, ensuring robust and interpretable AI-driven ecological solutions.
Figure 7.
Temporal trends and combination of AI task types in ecological research. The figure illustrates how the adoption and combination of regression, classification, and clustering tasks have evolved over time in studies of transitional water ecosystems, reflecting shifts in methodological approaches and research priorities.
6. Discussions
This review provides a comprehensive overview of the application of AI—specifically ML and DL—in the study of transitional water ecosystems. The results highlight a rapid evolution in AI methodologies, evidenced by the significant temporal increase in publications and the diversification of AI tasks and model approaches.
6.1. Shift in AI Model Adoption
A notable trend in recent research is the shift from predominantly regression-based models to a more diverse use of classification and clustering techniques. This evolution is driven by:
- Advances in computational power and DL, allowing for more complex, high-dimensional analyses.
- The increasing complexity of ecological research, necessitating sophisticated analytical approaches.
- Greater adoption of clustering models, particularly in remote sensing, habitat mapping, and pollution assessment.
For example, CNN models have significantly improved habitat classification and species monitoring by enhancing spatial accuracy and detection capabilities.
6.2. Trends in AI Model Type Distribution
The distribution of AI model types further emphasizes this evolution:
- Traditional ML methods, such as RF models, SVMs, and Gradient Boosting, remain widely used due to their robustness and their perceived transparency, although most high-performing approaches lack intrinsic interpretability in ecological applications.
- The rising adoption of DL models, particularly CNNs and RNN models, suggests that researchers are increasingly leveraging high-dimensional data, such as satellite imagery and sensor-based time-series data.
This trend reflects an increasing focus on automated feature extraction and real-time monitoring, which is crucial for climate adaptation strategies and biodiversity assessments.
Although traditional ML models are sometimes preferred for their perceived transparency, most high-performing approaches used in recent ecological studies lack intrinsic interpretability, reinforcing the need to distinguish clearly between model transparency and post hoc explainability.
6.3. AI and Global Environmental Concerns
The temporal trends in AI adoption also highlight the connection between AI advancements and growing global environmental concerns:
- Climate change and habitat degradation have catalyzed innovation in AI applications, particularly in predictive modeling and ecosystem monitoring.
- The expanding volume of AI-driven ecological publications not only marks technological advancements but also underscores the increasing urgency for effective environmental monitoring and management strategies.
These trends indicate a paradigm shift in ecological research, where AI is no longer just a supplementary tool but a critical component in policy-driven conservation efforts and environmental sustainability.
6.4. Challenges and Future Directions
Despite the promising advancements, several challenges remain:
- Model Interpretability—The increasing complexity of AI models necessitates XAI techniques to ensure transparency in ecological decision-making.
- Data Quality and Availability—AI models require large-scale, high-quality datasets, which remain limited in some ecological contexts.
- Computational Constraints—While AI models provide significant advantages, their high computational costs and energy consumption pose sustainability concerns.
Future research should focus on:
- Enhancing XAI methods to improve model transparency and stakeholder trust.
- Developing standardized AI frameworks to facilitate cross-study comparability and model reproducibility.
- Integrating AI with ecological domain expertise, ensuring that AI-generated insights align with real-world conservation priorities.
The integration of diverse AI methodologies into ecological research has transformed our ability to analyze complex, large-scale environmental datasets. While these advances offer promising avenues for improved ecosystem monitoring and conservation, they also highlight the need for interdisciplinary collaboration, improved model interpretability, and better data infrastructure. Future research should prioritize transparency, scalability, and ecological relevance, ensuring that AI serves as a reliable and sustainable tool in environmental management.
7. Conclusions and Recommendations
The adoption of AI in environmental sciences has revolutionized ecological monitoring, conservation efforts, and sustainable management. This review explores the rapid advancement of AI applications in these fields, highlighting the transition from traditional statistical techniques to more sophisticated ML and DL approaches. AI now plays a crucial role in addressing environmental challenges such as biodiversity preservation, pollution detection, climate prediction, and ecosystem management [88].
7.1. Key Findings and Implications
- Advancements in AI-driven Environmental Monitoring
AI-based models, particularly CNN models and ensemble ML techniques, have improved the accuracy and efficiency of ecosystem monitoring by analyzing vast datasets from satellites, drones, and sensor networks [87].
- AI’s Role in Conservation and Sustainable Practices
AI is widely employed in species identification, habitat assessment, and predictive conservation strategies. AI models utilizing acoustic recordings, camera trap imagery, and eDNA analysis have enhanced biodiversity tracking and anti-poaching initiatives [100].
- Climate Change Mitigation and AI
AI models contribute significantly to climate risk assessment, predictive environmental modeling, and mitigation strategies by analyzing global climate patterns and forecasting extreme weather events [101].
- Challenges in AI-driven Environmental Research
- a.
- Data Availability and Bias: Many AI models depend on high-quality datasets, yet access to consistent, standardized ecological data remains a challenge [86].
- b.
- Computational Costs: Deep learning models require significant computational resources, leading to concerns over energy consumption and sustainability [84].
- c.
- Model Interpretability: The need for XAI is growing, as complex models often lack transparency, making it difficult for policymakers and conservationists to rely on AI-based decisions [102].
7.2. Recommendations for Future Research and Policy
- Enhancing AI Transparency and Explainability
The development of XAI should be prioritized to ensure that AI models in ecological and conservation research are interpretable, trustworthy, and actionable
- Improving Data Accessibility and Standardization
Policymakers and researchers should collaborate to create open-access environmental datasets, improving model accuracy and comparability across studies [36].
- Integrating AI with Traditional Ecological Knowledge (TEK)
Future AI models should incorporate indigenous and local ecological knowledge to improve conservation efforts and align predictions with real-world observations [103].
- Ensuring Ethical and Sustainable AI Use
Governments and institutions should enforce ethical AI guidelines, including considerations of fairness, accountability, and sustainability in environmental AI applications [93].
7.3. Conclusions
AI has emerged as a transformative tool in environmental research, enhancing ecological modeling, conservation efforts, and climate change mitigation. However, its effectiveness depends on ethical deployment, cross-disciplinary collaboration, and strong data infrastructure. Tackling issues such as data bias, model interpretability, and computational sustainability is essential to ensuring AI’s long-term contribution to environmental protection. Future research should emphasize integrating AI with traditional conservation strategies while advancing XAI and data-sharing frameworks to improve ecological monitoring and climate resilience.
A responsible and collaborative AI ecosystem—driven by researchers, policymakers, and conservationists—can maximize AI’s benefits while aligning with global sustainability objectives [104].
Author Contributions
Conceptualization, M.P. and A.C.; Methodology, M.P. and A.C.; Software, A.C.; Validation, M.P.; Formal Analysis, A.C.; Investigation, M.P. and A.C.; Resources, M.P.; Data Curation, A.C.; Writing—Original Draft Preparation, A.C. and M.P.; Writing—Review and Editing, All coauthors; Supervision, M.P.; Funding Acquisition, M.P. All authors have read and agreed to the published version of the manuscript.
Funding
This research was supported by the “BlueDiversity” project funded by Interreg Italy-Croatia 2021–2027 First Call to Maurizio Pinna, by the “PRO-COAST” project funded by EU HORIZON-CL6-2022-BIODIV-01 to Maurizio Pinna and by the Italian Ministry of University and Research Funding of basic Research Activities (FFABR) 2017 awarded to M. Pinna and V. Specchia. Post-doc grants of F. Zangaro and F. Marcucci were supported by the National Biodiversity Future Center (NBFC) project CN_00000033 funded under the National Recovery and Resilience Plan (NRRP, Mission 4 Component 2 Investment 1.4) of Italian Ministry of University and Research (MUR) funded by the European Union—NextGenerationEU.
Data Availability Statement
The raw data supporting the conclusions of this article will be made 531 available by the authors on request.
Acknowledgments
The authors thank the Italian Ministry of University and Research for supporting the person’s months of staff involved. Post-doc grants of F. Zangaro and F. Marcucci were supported by the National Biodiversity Future Center (NBFC) project CN_00000033 funded under the National Recovery and Resilience Plan (NRRP, Mission 4 Component 2 Investment 1.4) of Italian Ministry of University and Research (MUR) funded by the European Union—NextGenerationEU. The authors thank also the EU—Next Generation EU Mission 4 “Education and Research”—Component 2: “From research to business”—Investment 3.1: “Fund for the realisation of an integrated system of research and innovation infrastructures”—Project IR0000032—ITINERIS—Italian Integrated Environmental Research Infrastructures System—CUP B53C22002150006, for supporting the grant of T. Semeraro.
Conflicts of Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this manuscript.
Abbreviations
AI: Artificial Intelligence; ML: Machine Learning; DL: Deep Learning; RF: Random Forest; SVM: Support Vector Machine; CNN: Convolutional Neural Network; RNN: Recurrent Neural Network; LSTM: Long Short-Term Memory; XAI: Explainable AI; GLM: Generalized Linear Model; ANN: Artificial Neural Network.
Appendix A
Summary of Reviewed Studies
Overview of Topics, Model Types, and Publication Years in 96 Peer-Reviewed Articles.
The following table provides a systematic classification of 96 articles that employ AI methodologies in aquatic ecosystem studies. The classification is based on key criteria including the research topic, the types of AI models used (e.g., neural networks, tree-based models, kernel-based models), and the publication year. This categorization aims to highlight the diversity of AI applications, from regression and classification to detection and segmentation, in addressing ecological and evolutionary challenges in coastal and transitional water environments.
Table A1.
Systematic classification of 96 articles.
Table A2.
Occurrences of types of AI Tasks.
Table A3.
Occurrences of Topics.
Table A4.
Occurrences of AI Model Types.
Table A5.
Temporal distribution of publications.
Table A6.
Occurrences of temporal trend and AI task type combination.
References
- Rajaraman, V. John McCarthy—Father of Artificial Intelligence. Resonance 2014, 19, 198–207. [Google Scholar] [CrossRef] [Scilit]
- Jordan, M.I.; Mitchell, T.M. Machine learning: Trends, perspectives, and prospects. Science 2015, 349, 255–260. [Google Scholar] [CrossRef] [Scilit]
- Mitchell, T.M. Machine learning and data mining. Commun. ACM 1999, 42, 30–36. [Google Scholar] [CrossRef] [Scilit]
- Deo, R.C. Machine learning in medicine. Circulation 2015, 132, 1920–1930. [Google Scholar] [CrossRef] [Scilit]
- Geurts, P.; Irrthum, A.; Wehenkel, L. Supervised learning with decision tree-based methods in computational and systems biology. Mol. Biosyst. 2009, 5, 1593–1605. [Google Scholar] [CrossRef] [Scilit]
- Bahrammirzaee, A. A comparative survey of artificial intelligence applications in finance: Artificial neural networks, expert systems, and hybrid intelligent systems. Neural Comput. Appl. 2010, 19, 1165–1195. [Google Scholar] [CrossRef] [Scilit]
- Tao, F.; Qi, Q.L.; Liu, A.; Kusiak, A. Data-driven smart manufacturing. J. Manuf. Syst. 2018, 48, 157–169. [Google Scholar] [CrossRef] [Scilit]
- Mishra, M.; Nayak, J.; Naik, B.; Abraham, A. Deep learning in electrical utility industry: A comprehensive review of a decade of research. Eng. Appl. Artif. Intell. 2020, 96, 104000. [Google Scholar] [CrossRef] [Scilit]
- Anderson, E.F.; McLoughlin, L.; Liarokapis, F.; Peters, C.; Petridis, P.; de Freitas, S. Developing serious games for cultural heritage: A state-of-the-art review. Virtual Real. 2010, 14, 255–275. [Google Scholar] [CrossRef] [Scilit]
- Joshi, A. Machine Learning and Artificial Intelligence; Springer: Cham, Switzerland, 2020. [Google Scholar]
- Breiman, L. Statistical modeling: The two cultures (with comments and a rejoinder by the author). Stat. Sci. 2001, 16, 199–231. [Google Scholar] [CrossRef] [Scilit]
- Ghahramani, Z. Unsupervised learning. In Advanced Lectures on Machine Learning; Bousquet, O., von Luxburg, U., Rätsch, G., Eds.; Springer: Berlin/Heidelberg, Germany, 2004; pp. 72–112. [Google Scholar] [CrossRef] [Scilit]
- Hamblin, S. On the practical usage of genetic algorithms in ecology and evolution. Methods Ecol. Evol. 2013, 4, 184–194. [Google Scholar] [CrossRef] [Scilit]
- Mundry, R.; Sommer, C. Discriminant function analysis with nonindependent data: Consequences and an alternative. Anim. Behav. 2007, 74, 965–976. [Google Scholar] [CrossRef] [Scilit]
- Reich, D.; Price, A.L.; Patterson, N. Principal component analysis of genetic data. Nat. Genet. 2008, 40, 491–492. [Google Scholar] [CrossRef] [Scilit]
- Sorensen, D.; Gianola, D. Likelihood, Bayesian, and MCMC Methods in Quantitative Genetics; Springer: New York, NY, USA, 2002. [Google Scholar] [CrossRef] [Scilit]
- Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; MIT Press: Cambridge, MA, USA, 2016; Available online: https://www.deeplearningbook.org/ (accessed on 16 March 2026).
- LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Aodha, O.M.; Gibb, R.; Barlow, K.E.; Browning, E.; Firman, M.; Freeman, R.; Harder, B.; Kinsey, L.; Mead, G.R.; Newson, S.E.; et al. Bat detective—Deep learning tools for bat acoustic signal detection. PLoS Comput. Biol. 2018, 14, e1005995. [Google Scholar] [CrossRef] [Scilit]
- Gray, P.C.; Fleishman, A.B.; Klein, D.J.; McKown, M.W.; Bézy, V.S.; Lohmann, K.J.; Johnston, D.W. A convolutional neural net- work for detecting sea turtles in drone imagery. Methods Ecol. Evol. 2019, 10, 345–355. [Google Scholar] [CrossRef] [Scilit]
- Guirado, E.; Tabik, S.; Rivas, M.L.; Alcaraz-Segura, D.; Herrera, F. Automatic whale counting in satellite images with deep learning. bioRxiv 2018. [Google Scholar] [CrossRef] [Scilit]
- Lasseck, M. Audio-based bird species identification with deep convolutional neural networks. In Working Notes of CLEF 2018; CEUR Workshop Proceedings; CEUR-WS.org: Aachen, Germany, 2018; Volume 2125, Available online: https://ceur-ws.org/Vol-2125/paper_140.pdf (accessed on 16 March 2026).
- Tabak, M.A.; Norouzzadeh, M.S.; Wolfson, D.W.; Sweeney, S.J.; Vercauteren, K.C.; Snow, N.P.; Halseth, J.M.; Salvo, P.A.D.; Lewis, J.S.; White, M.D.; et al. Machine learning to classify animal species in camera trap images: Applications in ecology. Methods Ecol. Evol. 2019, 10, 585–590. [Google Scholar] [CrossRef] [Scilit]
- Dunker, S.; Motivans, E.; Rakosy, D.; Boho, D.; Mäder, P.; Hornick, T.; Knight, T.M. Pollen analysis using multispectral imaging flow cytometry and deep learning. New Phytol. 2020, 229, 593–606. [Google Scholar] [CrossRef] [Scilit]
- Graving, J.M.; Chae, D.; Naik, H.; Li, L.; Koger, B.; Costelloe, B.R.; Couzin, I.D. Fast and robust animal pose estimation. bioRxiv 2019. [Google Scholar] [CrossRef] [Scilit]
- Mathis, A.; Mamidanna, P.; Cury, K.M.; Abe, T.; Murthy, V.N.; Mathis, M.W.; Bethge, M. DeepLabCut: Markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 2018, 21, 1281–1289. [Google Scholar] [CrossRef] [Scilit]
- Ott, T.; Lautenschlager, U. GinJinn2: Object detection and segmentation for ecology and evolution. bioRxiv 2021. [Google Scholar] [CrossRef] [Scilit]
- Pereira, T.D.; Aldarondo, D.E.; Willmore, L.; Kislin, M.; Wang, S.S.-H.; Murthy, M.; Shaevitz, J.W. Fast animal pose estima- tion using deep neural networks. Nat. Methods 2019, 16, 117–125. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Desjardins-Proulx, P.; Laigle, I.; Poisot, T.; Gravel, D. Ecological interactions and the Netflix problem. PeerJ 2017, 5, e3644. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Rammer, W.; Seidl, R. A scalable model of vegetation transitions using deep neural networks. Methods Ecol. Evol. 2019, 10, 879–890. [Google Scholar] [CrossRef] [Scilit]
- Reichstein, M.; Camps-Valls, G.; Stevens, B.; Jung, M.; Denzler, J.; Carvalhais, N. Deep learning and process understanding for data-driven Earth system science. Nature 2019, 566, 195–204. [Google Scholar] [CrossRef] [Scilit]
- Chen, R.T.Q.; Rubanova, Y.; Bettencourt, J.; Duvenaud, D. Neural ordinary differential equations. arXiv 2019. [Google Scholar] [CrossRef] [Scilit]
- Rackauckas, C.; Ma, Y.; Martensen, J.; Warner, C.; Zubov, K.; Supekar, R.; Skinner, D.; Ramadhan, A.; Edelman, A. Universal differential equations for scientific machine learning. arXiv 2021. [Google Scholar] [CrossRef] [Scilit]
- Chen, D.; Xue, Y.; Gomes, C.P. End-to-end learning for the deep multivariate Probit model. arXiv 2018, arXiv:1803.08591. [Google Scholar]
- Elith, J.; Leathwick, J.R. Species distribution models: Ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 2009, 40, 677–697. [Google Scholar] [CrossRef] [Scilit]
- Jones, J.; Harris, E.; Febriansah, Y.; Adiwijaya, A.; Hikam, I. AI for Sustainable Development: Applications in Natural Resource Management, Agriculture, and Waste Management. Int. Trans. Artif. Intell. 2024, 2, 143–149. [Google Scholar] [CrossRef] [Scilit]
- Wilkinson, D.P.; Golding, N.; Guillera-Arroita, G.; Tingley, R.; McCarthy, M.A. A comparison of joint species distribution models for presence–absence data. Methods Ecol. Evol. 2019, 10, 198–211. [Google Scholar] [CrossRef] [Scilit]
- Orfanidis, S.; Papathanasiou, V.; Sabetta, L.; Pinna, M.; Gigi, V.; Gounaris, S.; Tsiagga, E.; Nakou, K.; Theodosiou, T.H. Benthic macrophyte communities as bioindicators of transitional and coastal waters: Relevant approaches and tools. Trans. Waters Bull. 2007, 1, 45–49. [Google Scholar] [CrossRef]
- Biraghi, C.; Lotfian, M.; Carrion, D.; Brovelli, M. AI in Support to Water Quality Monitoring. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2021, XLIII-B4-2021, 167–174. [Google Scholar] [CrossRef] [Scilit]
- Arias-Real, R.; Delgado-Baquerizo, M.; Sabater, S.; Gutiérrez-Cánovas, C.; Valencia, E.; Aragón, G.; Cantón, Y.; Datry, T.; Giordani, P.; Medina, N.G.; et al. Unfolding the dynamics of ecosystems undergoing alternating wet-dry transitional states. Ecol. Lett. 2024, 27, e14488. [Google Scholar] [CrossRef] [Scilit]
- Dushoff, J.; Kain, M.P.; Bolker, B.M. I can see clearly now: Reinterpreting statistical significance. Methods Ecol. Evol. 2019, 10, 756–759. [Google Scholar] [CrossRef] [Scilit]
- Gelman, A.; Loken, E. The statistical crisis in science: Data-dependent analysis—A “garden of forking paths”—Explains why many statistically significant comparisons don’t hold up. Am. Sci. 2014, 102, 460–466. [Google Scholar] [CrossRef] [Scilit]
- Hartig, F.; Barraquand, F. The evidence contained in the P-value is context dependent. Trends Ecol. Evol. 2022, 37, 569–570. [Google Scholar] [CrossRef] [Scilit]
- Muff, S.; Nilsen, E.B.; O’Hara, R.B.; Nater, C.R. Rewriting results sections in the language of evidence. Trends Ecol. Evol. 2022, 37, 203–210. [Google Scholar] [CrossRef] [Scilit]
- Metropolis, N.; Rosenbluth, A.W.; Rosenbluth, M.N.; Teller, A.H.; Teller, E. Equation of state calculations by fast computing machines. J. Chem. Phys. 1953, 21, 1087–1092. [Google Scholar] [CrossRef] [Scilit]
- Clark, J.S. Why environmental scientists are becoming Bayesians. Ecol. Lett. 2005, 8, 2–14. [Google Scholar] [CrossRef] [Scilit]
- Grimm, V.; Revilla, E.; Berger, U.; Jeltsch, F.; Mooij, W.M.; Railsback, S.F.; Thulke, H.-H.; Weiner, J.; Wiegand, T.; DeAngelis, D.L. Pattern-oriented modeling of agent-based complex systems: Lessons from ecology. Science 2005, 310, 987–991. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Shmueli, G. To explain or to predict? Stat. Sci. 2010, 25, 289–310. [Google Scholar] [CrossRef] [Scilit]
- Efron, B. Bootstrap methods: Another look at the jackknife. In Breakthroughs in Statistics: Methodology and Distribution; Kotz, S., Johnson, N.L., Eds.; Springer: Berlin/Heidelberg, Germany, 1992; pp. 569–593. [Google Scholar] [CrossRef] [Scilit]
- Stone, M. Cross-validatory choice and assessment of statistical predictions. J. R. Stat. Soc. Ser. B (Methodol.) 1974, 36, 111–133. [Google Scholar] [CrossRef] [Scilit]
- Roberts, D.R.; Bahn, V.; Ciuti, S.; Boyce, M.S.; Elith, J.; Guillera-Arroita, G.; Hauenstein, S.; Lahoz-Monfort, J.J.; Schröder, B.; Thuiller, W.; et al. Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography 2017, 40, 913–929. [Google Scholar] [CrossRef] [Scilit]
- Lee, S.; Kim, J.; Lee, G.; Hong, J.; Bae, J.H.; Lim, K.J. Prediction of Aquatic Ecosystem Health Indices through Machine Learning Models Using the WGAN-Based Data Augmentation Method. Sustainability 2021, 13, 10435. [Google Scholar] [CrossRef] [Scilit]
- Wang, Z.; Zhang, J.; Hua, P.; Cui, Y.; Lu, C.; Wang, X.; Chen, Q.; Krebs, P. Filling in missing pieces in the co-development of artificial intelligence and environmental science. Innov. Geosci. 2023, 1, 10007. [Google Scholar] [CrossRef] [Scilit]
- Marrable, D.; Barker, K.; Tippaya, S.; Wyatt, M.; Bainbridge, S.; Stowar, M.; Larke, J. Accelerating species recognition and labelling of fish from underwater video with machine-assisted deep learning. Front. Mar. Sci. 2022, 9, 944582. [Google Scholar] [CrossRef] [Scilit]
- Han, B.; Varshney, K.; LaDeau, S.; Subramaniam, A.; Weathers, K.; Zwart, J. A synergistic future for AI and ecology. Proc. Natl. Acad. Sci. USA 2023, 120, e2220283120. [Google Scholar] [CrossRef] [Scilit]
- Kim, K.; Park, J. A Survey of Applications of Artificial Intelligence Algorithms in Eco-environmental Modelling. Environ. Eng. Res. 2009, 14, 102–110. [Google Scholar] [CrossRef] [Scilit]
- Hu, Y.; Li, N.; Jiang, Y.; Bao, X.; Li, X. Research progress on coupling artificial intelligence and eco-environmental models. Ying Yong Sheng Tai Xue Bao J. Appl. Ecol. 2023, 34, 257–263. [Google Scholar] [CrossRef] [Scilit]
- Silva, M.; Silva, J.; Delabrida, S.; Bianchi, A.; Ribeiro, S.; Silva, J.; Oliveira, R. Wearable Edge AI Applications for Ecological Environments. Sensors 2021, 21, 5082. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Zorins, A. Understanding the Essence of Artificial Intelligence: Towards Ecological Safety of AI in Human Society. In Proceedings of the 21st International Conference on Computer Systems and Technologies; Association for Computing Machinery: New York, NY, USA, 2020. [Google Scholar] [CrossRef] [Scilit]
- Cutler, D.R.; Edwards, T.C.; Beard, K.H.; Cutler, A.; Hess, K.T.; Gibson, J.; Lawler, J.J. Random forests for classification in ecology. Ecology 2007, 88, 2783–2792. [Google Scholar] [CrossRef] [Scilit]
- Liaw, A.; Wiener, M. Classification and regression by randomForest. R News 2002, 2, 18–22. [Google Scholar]
- Díaz-Uriarte, R.; Alvarez de Andrés, S. Gene selection and classification of microarray data using random forest. BMC Bioinform. 2006, 7, 3. [Google Scholar] [CrossRef] [Scilit]
- Rodriguez-Galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-Olmo, M.; Rigol-Sanchez, J.P. An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS J. Photogramm. Remote Sens. 2012, 67, 93–104. [Google Scholar] [CrossRef] [Scilit]
- Davis, R.L., II; Greene, J.K.; Dou, F.; Jo, Y.-K.; Chappell, T.M. A Practical Application of Unsupervised Machine Learning for Analyzing Plant Image Data Collected Using Unmanned Aircraft Systems. Agronomy 2020, 10, 633. [Google Scholar] [CrossRef] [Scilit]
- Siraj-Ud-Doula, M.; Alam, M. Ecological data analysis based on machine learning algorithms. arXiv 2018. [Google Scholar] [CrossRef] [Scilit]
- Narula, P. Analysis of Common Supervised Learning Algorithms Through Application. Adv. Comput. Intell. Int. J. (ACII) 2023, 10, 29–48. [Google Scholar] [CrossRef] [Scilit]
- Crisci, C.; Ghattas, B.; Perera, G. A review of supervised machine learning algorithms and their applications to ecological data. Ecol. Model. 2012, 240, 113–122. [Google Scholar] [CrossRef] [Scilit]
- Reddy, Y.; Varma, N. Review on Supervised Learning Techniques. In Emerging Research in Data Engineering Systems and Computer Communications; Advances in Intelligent Systems and Computing; Springer: Berlin/Heidelberg, Germany, 2020; pp. 577–587. [Google Scholar] [CrossRef] [Scilit]
- Guo, Q.; Jin, S.; Li, M.; Yang, Q.; Xu, K.; Ju, Y.; Zhang, J.; Xuan, J.; Liu, J.; Su, Y.; et al. Application of deep learning in ecological resource research: Theories, methods, and challenges. Sci. China Earth Sci. 2020, 63, 1457–1474. [Google Scholar] [CrossRef] [Scilit]
- Bergen, S.; Huso, M.; Duerr, A.; Braham, M.; Schmuecker, S.; Miller, T.; Katzner, T. A review of supervised learning methods for classifying animal behavioural states from environmental features. Methods Ecol. Evol. 2022, 14, 189–202. [Google Scholar] [CrossRef] [Scilit]
- Qin, X.; Wu, M.; Lock, T.; Kallenbach, R. DA: Ecological and Evolutionary Inference Using Supervised Discriminant Analysis. arXiv 2020. [Google Scholar] [CrossRef] [Scilit]
- Sonnewald, M.; Dutkiewicz, S.; Hill, C.; Forget, G. Elucidating ecological complexity: Unsupervised learning determines global marine eco-provinces. Sci. Adv. 2020, 6, eaay4740. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Usama, M.; Qadir, J.; Raza, A.; Arif, H.; Yau, K.; Elkhatib, Y.; Hussain, A.; Al-Fuqaha, A. Unsupervised Machine Learning for Networking: Techniques, Applications and Research Challenges. IEEE Access 2017, 7, 65579–65615. [Google Scholar] [CrossRef] [Scilit]
- Shanmuganathan, S. Self-Organising Maps for Integrating Data Across Environmental, Social and Economic Conditions. 2013. Available online: https://www.semanticscholar.org/paper/Self-organising-maps-for-integrating-data-across-Shanmuganathan-Sallis/d37b29a2867345de76fee1e8af81e5e03f00b5ca (accessed on 16 March 2026).
- Kanevski, M. Unsupervised learning of Swiss population spatial distribution. PLoS ONE 2021, 16, e0246529. [Google Scholar] [CrossRef] [Scilit]
- Elsaadouny, M.; Barowski, J.; Rolfes, I. Unsupervised Learning Implementation for SAR Images Clustering. In Proceedings of the 2021 International Conference on Electromagnetics in Advanced Applications (ICEAA), Honolulu, HI, USA, 9–13 August 2021; p. 104. [Google Scholar] [CrossRef] [Scilit]
- Raina, R.; Madhavan, A.; Ng, A. Large-scale deep unsupervised learning using graphics processors. In Proceedings of the 26th Annual International Conference on Machine Learning; Association for Computing Machinery: New York, NY, USA, 2009; pp. 873–880. [Google Scholar] [CrossRef] [Scilit]
- Bundzel, M.M.; Sincak, P. Combining gradient and evolutionary approaches to the artificial neural networks training according to principles of support vector machines. In Proceedings of the 2006 IEEE International Joint Conference on Neural Networks (IJCNN); IEEE: New York NY, NSA, 2006; pp. 2068–2074. [Google Scholar] [CrossRef] [Scilit]
- Xu, C.; Jackson, S.A. Machine learning and complex biological data. Genome Biol. 2019, 20, 76. [Google Scholar] [CrossRef] [Scilit]
- Bodner, K.; Fortin, M.; Molnár, P. Making predictive modelling ART: Accurate, reliable, and transparent. Ecosphere 2020, 11, e03160. [Google Scholar] [CrossRef] [Scilit]
- Semeraro, T.; Titocci, J.; Liberatore, L.; Monti, F.; Cazzetta, A.; Pinna, M.; Shokri, M.; Basset, A. A Machine Learning Framework for Assessing the Sensitivity of Regional Ocean Productivity to Climate Change. Environments 2026, 13, 137. [Google Scholar] [CrossRef] [Scilit]
- Yamashita, Y.; Ito, H.; Wakabayashi, K.; Kobayashi, M.; Morishima, A. HAEM: Obtaining Higher-Quality Classification Task Results with AI Workers. In Proceedings of the 14th ACM Web Science Conference 2022; Association for Computing Machinery: New York, NY, USA, 2022. [Google Scholar] [CrossRef] [Scilit]
- Theofanos, M. AI Use Taxonomy; National Institute of Standards and Technology (NIST): Gaithersburg, MD, USA, 2024. [CrossRef] [Scilit]
- Chapman, M. Governing AI Applications To Monitoring and Managing Our Global Environmental Commons. In Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society; Association for Computing Machinery: New York, NY, USA, 2022. [Google Scholar] [CrossRef] [Scilit]
- Yang, C.; Kuo, C.; Chen, S.; Liu, H.; Cheng, F.; Hsieh, Y.; Chue, J.; Wu, C.; Huang, C. AI-EMS: An Environmental Monitoring System with AI Prediction. In Proceedings of the 2023 IEEE SENSORS, Vienna, Austria, 29 October–1 November 2023; pp. 1–4. [Google Scholar] [CrossRef] [Scilit]
- Russo, P.; Ciaccio, F. Recent advances in AI for enhanced environmental monitoring and preservation. In Proceedings of the 2023 IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea), La Valletta, Malta, 4–6 October 2023; pp. 127–132. [Google Scholar] [CrossRef] [Scilit]
- Sharma, S.; Paranjape, V. A Data-Driven Framework for Environmental Monitoring: Integrating Machine Learning and Analytics. Int. J. Innov. Res. Comput. Commun. Eng. 2023, 12, 10423–10432. [Google Scholar] [CrossRef] [Scilit]
- Nneamaka, C.C.; Biu, P.; Umoh, A.; Obaedo, B.; Adegbite, A.; Abatan, A. Reviewing the role of AI in environmental monitoring and conservation: A data-driven revolution for our planet. World J. Adv. Res. Rev. 2024, 21, 161–171. [Google Scholar] [CrossRef] [Scilit]
- Shuford, J. Interdisciplinary Perspectives: Fusing Artificial Intelligence with Environmental Science for Sustainable Solutions. J. Artif. Intell. Gen. Sci. (JAIGS) 2024, 1, 106–123. [Google Scholar] [CrossRef] [Scilit]
- Miller, T.; Cembrowska-Lech, D.; Kisiel, A.; Krzemińska, A.; Kozlovska, P.; Jawor, M.; Kołodziejczak, M.; Durlik, I. Harnessing AI for Environmental Resilience: Mitigating Heavy Metal Pollution and Advancing Sustainable Practices in Diverse Spheres. Grail Sci. 2023, 14, 151–156. [Google Scholar] [CrossRef] [Scilit]
- Panduman, Y.; Funabiki, N.; Fajrianti, E.; Fang, S.; Sukaridhoto, S. A Survey of AI Techniques in IoT Applications with Use Case Investigations in the Smart Environmental Monitoring and Analytics in Real-Time IoT Platform. Information 2024, 15, 153. [Google Scholar] [CrossRef] [Scilit]
- Arowolo, M.; Aaron, W.; Kugbiyi, A.; Eteng, U.; Iloh, D.; Aguma, C.; Olagunju, A. Integrating AI enhanced remote sensing technologies with IOT networks for precision environmental monitoring and predictive ecosystem management. World J. Adv. Res. Rev. 2024, 23, 2156–2166. [Google Scholar] [CrossRef] [Scilit]
- Qi, Y. The Impact of Artificial Intelligence on Environmental Protection. Highlights Sci. Eng. Technol. 2024, 96, 152–156. [Google Scholar] [CrossRef] [Scilit]
- Haupt, S.; Gagne, D.; Hsieh, W.; Krasnopolsky, V.; McGovern, A.; Marzban, C.; Moninger, W.; Lakshmanan, V.; Tissot, P.; Williams, J. The History and Practice of AI in the Environmental Sciences. Bull. Am. Meteorol. Soc. 2022, 103, E1351–E1370. [Google Scholar] [CrossRef] [Scilit]
- Ryo, M.; Angelov, B.; Mammola, S.; Kass, J.; Benito, B.; Hartig, F. [Final version available] Explainable Artificial Intelligence enhances the ecological interpretability of black-box species distribution models. EcoEvoRxiv 2020, 43, 1–7. [Google Scholar] [CrossRef] [Scilit]
- Rillig, M.; Mansour, I.; Hempel, S.; Bi, M.; König-Ries, B.; Kasirzadeh, A. How widespread use of generative AI for images and video can affect the environment and the science of ecology. Ecol. Lett. 2024, 27, e14397. [Google Scholar] [CrossRef] [Scilit]
- Hassan, F.; Ilo, H.; Afolabi, N.; Akintola, K. Modelling Environmental Tree Species Volume using some selected Skewed Distributions. Int. J. Adv. Stat. Probab. 2024, 11, 29–36. [Google Scholar] [CrossRef] [Scilit]
- Komori, O.; Eguchi, S.; Ikeda, S.; Okamura, H.; Ichinokawa, M.; Nakayama, S. An asymmetric logistic regression model for ecological data. Methods Ecol. Evol. 2015, 7, 249–260. [Google Scholar] [CrossRef] [Scilit]
- Hasan, H.; Ettyem, S.; Hameed, W.; Hussien, N.; Diame, H.; Al-Tameemi, A. Environmental and Financial Asset Optimization through Multi-Restoration Informed by AI Systems for Optimizing Ecosystem Functioning. In Proceedings of the 2023 Annual International Conference on Emerging Research Areas: International Conference on Intelligent Systems (AICERA/ICIS), Kanjirapally, India, 16–18 November 2023; pp. 1–6. [Google Scholar] [CrossRef] [Scilit]
- Hegde, N.; Bargavi, D. Smart Conservation: Integrating AI for Enhanced Wildlife Monitoring. Int. Res. J. Comput. Sci. 2024, 11, 62–67. [Google Scholar] [CrossRef] [Scilit]
- Lohani, N. AI-based Environmental Sustainability: Transforming Conservation Efforts. Int. J. Multidiscip. Res. 2024, 6, 16997. [Google Scholar] [CrossRef]
- Adanma, U.; Ogunbiyi, E. Artificial intelligence in environmental conservation: Evaluating cyber risks and opportunities for sustainable practices. Comput. Sci. IT Res. J. 2024, 5, 1178–1209. [Google Scholar] [CrossRef] [Scilit]
- Swami, N. Applying AI to conservation challenges. In Data Science Applied to Sustainability Analysis; Elsevier: Amsterdam, The Netherlands, 2021; pp. 17–28. [Google Scholar] [CrossRef] [Scilit]
- Akter, S. Harnessing Technology for Environmental Sustainability: Utilizing AI to Tackle Global Ecological Challenges. Artif. Intell. Gen. Sci. (JAIGS) 2024, 2, 49–57. [Google Scholar] [CrossRef] [Scilit]
- Ressom, H.; Natarajan, P.; Srirangam, S.; Musavi, M.T.; Virnstein, R.W.; Morris, L.J.; Tweedale, W. Neural network based light attenuation model for monitoring seagrass health. IEEE Int. Conf. Neural Netw. Conf. Proc. 2004, 3, 2489–2493. [Google Scholar] [CrossRef] [Scilit]
- Mancia, A.; Warr, G.W.; Almeida, J.S.; Veloso, A.; Wells, R.S.; Chapman, R.W. Transcriptome profiles: Diagnostic signature of dolphin populations. Estuar. Coasts 2010, 33, 919–929. [Google Scholar] [CrossRef] [Scilit]
- Vincenzi, S.; Zucchetta, M.; Franzoi, P.; Pellizzato, M.; Pranovi, F.; De Leo, G.A.; Torricelli, P. Application of a Random Forest algorithm to predict spatial distribution of the potential yield of Ruditapes philippinarum in the Venice lagoon, Italy. Ecol. Model. 2011, 222, 1471–1478. [Google Scholar] [CrossRef] [Scilit]
- Margiotta, S.; Marini, G.; Fay, S.; D’Onghia, F.M.; Liso, I.S.; Parise, M.; Pinna, M. Hydro-Stratigraphic Conditions and Human Activity Leading to Development of a Sinkhole Cluster in a Mediterranean Water Ecosystem. Hydrology 2021, 8, 111. [Google Scholar] [CrossRef] [Scilit]
- Wattelez, G.; Dupouy, C.; Mangeas, M.; Lefèvre, J.; Touraivane; Frouin, R. A statistical algorithm for estimating chlorophyll concentration in the New Caledonian lagoon. Remote Sens. 2016, 8, 45. [Google Scholar] [CrossRef] [Scilit]
- Tyler, A.N.; Hunter, P.D.; Spyrakos, E.; Groom, S.; Constantinescu, A.M.; Kitchen, J. Developments in Earth observation for the assessment and monitoring of inland, transitional, coastal and shelf-sea waters. Sci. Total Environ. 2016, 572, 1307–1321. [Google Scholar] [CrossRef] [Scilit]
- Bolton, M.S.; Beaudoin, A.B. Climate reconstructions based on postglacial macrofossil assemblages from four river systems in southwestern Alberta. Can. Water Resour. J. 2017, 42, 289–305. [Google Scholar] [CrossRef] [Scilit]
- Farda, N.M. Multi-temporal land use mapping of coastal wetlands area using machine learning in Google Earth Engine. IOP Conf. Ser. Earth Environ. Sci. 2017, 98, 012042. [Google Scholar] [CrossRef] [Scilit]
- Béjaoui, B.; Ottaviani, E.; Barelli, E.; Ziadi, B.; Dhib, A.; Lavoie, M.; Gianluca, C.; Turki, S.; Solidoro, C.; Aleya, L. Machine learning predictions of trophic status indicators and plankton dynamic in coastal lagoons. Ecol. Indic. 2018, 95, 765–774. [Google Scholar] [CrossRef] [Scilit]
- Alhashem, M. Supervised machine learning in predicting multiphase flow regimes in horizontal pipes. In Proceedings of the Society of Petroleum Engineers—Abu Dhabi International Petroleum Exhibition and Conference 2019 (ADIPEC 2019), Abu Dhabi, United Arab Emirates, 11–14 November 2019. [Google Scholar] [CrossRef] [Scilit]
- Ahangarha, M.; Seydi, S.T.; Shahhoseini, R. Hyperspectral change detection in wetland and water-body areas based on machine learning. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. —ISPRS Arch. 2019, XLII-4/W18, 19–24. [Google Scholar] [CrossRef] [Scilit]
- Mamun, M.; Kim, J.-J.; Alam, M.A.; An, K.-G. Prediction of algal chlorophyll-a and water clarity in monsoon-region reservoir using machine learning approaches. Water 2020, 12, 30. [Google Scholar] [CrossRef] [Scilit]
- Jimeno-Sáez, P.; Senent-Aparicio, J.; Cecilia, J.M.; Pérez-Sánchez, J. Using machine-learning algorithms for eutrophication modeling: Case study of Mar Menor lagoon (Spain). Int. J. Environ. Res. Public Health 2020, 17, 1189. [Google Scholar] [CrossRef] [Scilit]
- Fan, Y.; Lu, W.; Miao, T.; Li, J.; Lin, J. Optimum design of a seawater intrusion monitoring scheme based on the image quality assessment method. Water Resour. Manag. 2020, 34, 2485–2502. [Google Scholar] [CrossRef] [Scilit]
- Marrocco, V.; Zangaro, F.; Sicuro, A.; Pinna, M. A scaling down mapping of Pinna nobilis (Linnaeus, 1758) through the combination of scientific literature, NATURA 2000, grey literature and citizen science data. Nat. Conserv. 2019, 33, 21–31. [Google Scholar] [CrossRef] [Scilit]
- Pudaruth, S.; Nazurally, N.; Appadoo, C.; Kishnah, S.; Chady, F.; Vinayaganidhi, M.; Mohammoodally, I.; Ally, Y.A. SuperFish: A mobile application for fish species recognition using image processing techniques and deep learning. Int. J. Comput. Digit. Syst. 2021, 10, 1157–1165. [Google Scholar] [CrossRef] [Scilit]
- Oiry, S.; Barillé, L. Using Sentinel-2 satellite imagery to develop microphytobenthos-based water quality indices in estuaries. Ecol. Indic. 2021, 121, 107184. [Google Scholar] [CrossRef] [Scilit]
- Karki, S.; Bermejo, R.; Wilkes, R.; Monagail, M.M.; Daly, E.; Healy, M.; Hanafin, J.; McKinstry, A.; Mellander, P.-E.; Fenton, O.; et al. Mapping spatial distribution and biomass of intertidal Ulva blooms using machine learning and Earth observation. Front. Mar. Sci. 2021, 8, 633128. [Google Scholar] [CrossRef] [Scilit]
- Benmokhtar, S.; Robin, M.; Maanan, M.; Bazairi, H. Mapping and quantification of the dwarf eelgrass Zostera noltei using a random forest algorithm on a SPOT 7 satellite image. ISPRS Int. J. Geo-Inf. 2021, 10, 313. [Google Scholar] [CrossRef] [Scilit]
- Talukdar, S.; Eibek, K.U.; Akhter, S.; Ziaul, S.; Towfiqul Islam, A.R.M.; Mallick, J. Modeling fragmentation probability of land-use and land-cover using the bagging, random forest and random subspace in the Teesta River Basin, Bangladesh. Ecol. Indic. 2021, 126, 107612. [Google Scholar] [CrossRef] [Scilit]
- Concepcion, R., II; Dadios, E.; Bandala, A.; Caçador, I.; Fonseca, V.F.; Duarte, B. Applying limnological feature-based machine learning techniques to chemical state classification in marine transitional systems. Front. Mar. Sci. 2021, 8, 658434. [Google Scholar] [CrossRef] [Scilit]
- Bindajam, A.A.; Mallick, J.; Talukdar, S.; Islam, A.R.M.T.; Alqadhi, S. Integration of artificial intelligence–based LULC mapping and prediction for estimating ecosystem services for urban sustainability: Past to future perspective. Arab. J. Geosci. 2021, 14, 1887. [Google Scholar] [CrossRef] [Scilit]
- Gómez, D.; Salvador, P.; Sanz, J.; Casanova, J.L. A new approach to monitor water quality in the Menor sea (Spain) using satellite data and machine learning methods. Environ. Pollut. 2021, 286, 117489. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Hansen, S.S.; Ernstsen, V.B.; Andersen, M.S.; Al-Hamdani, Z.; Baran, R.; Niederwieser, M.; Steinbacher, F.; Kroon, A. Classification of boulders in coastal environments using random forest machine learning on topo-bathymetric lidar data. Remote Sens. 2021, 13, 4101. [Google Scholar] [CrossRef] [Scilit]
- Cerim, H.; Özdemir, N.; Cremona, F.; Öğlü, B. Effect of changing in weather conditions on Eastern Mediterranean coastal lagoon fishery. Reg. Stud. Mar. Sci. 2021, 48, 102006. [Google Scholar] [CrossRef] [Scilit]
- Wei, W.; Nghiem, A.; Ma, R.; Sun, Z.; Gong, X.; Zhou, A.; Prommer, H. Factors controlling iodine enrichment in a coastal plain aquifer in the North Jiangsu Yishusi Plain, China. J. Contam. Hydrol. 2021, 243, 103894. [Google Scholar] [CrossRef] [Scilit]
- Kruk, M.; Artiemjew, P.; Paturej, E. The application of game theory-based machine learning modelling to assess climate variability effects on the sensitivity of lagoon ecosystem parameters. Ecol. Inform. 2021, 66, 101462. [Google Scholar] [CrossRef] [Scilit]
- Hadid, N.B.; Goyet, C.; Chaar, H.; Maiz, N.B.; Guglielmi, V.; Shili, A. Machine learning modeling techniques for forecasting the trophic level in a restored South Mediterranean lagoon using chlorophyll-a. Wetlands 2021, 41, 111. [Google Scholar] [CrossRef] [Scilit]
- Ha, N.T.; Pham, T.D.; Tran, T.T.H. Zoning seagrass protection in Lap An Lagoon, Vietnam using a novel integrated framework for sustainable coastal management. Wetlands 2021, 41, 122. [Google Scholar] [CrossRef] [Scilit]
- Borfecchia, F.; Micheli, C.; De Cecco, L.; Sannino, G.; Struglia, M.V.; Di Sarra, A.G.; Gomez, C.; Mattiazzo, G. Satellite multi/hyper spectral HR sensors for mapping the Posidonia oceanica in south Mediterranean islands. Sustainability 2021, 13, 13715. [Google Scholar] [CrossRef] [Scilit]
- Franklin, G.L.; Torres-Freyermuth, A. On the runup parameterisation for reef-lined coasts. Ocean Model. 2022, 169, 101929. [Google Scholar] [CrossRef] [Scilit]
- Manuel, B.-M.J.; Manuel R-Á, J.; Martín, P.-N.; Manuel, B.-F.J. Soft-ANN based correlation for air-water two-phase flow pressure drop estimation in a vertical mini-channel. Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 2022, 236, 1430–1442. [Google Scholar] [CrossRef] [Scilit]
- Wattelez, G.; Dupouy, C.; Juillot, F. Unsupervised optical classification of the seabed color in shallow oligotrophic waters from Sentinel-2 images: A case study in the Voh-Koné-Pouembout Lagoon (New Caledonia). Remote Sens. 2022, 14, 836. [Google Scholar] [CrossRef] [Scilit]
- Kraft, B.; Jung, M.; Körner, M.; Koirala, S.; Reichstein, M. Towards hybrid modeling of the global hydrological cycle. Hydrol. Earth Syst. Sci. 2022, 26, 1579–1614. [Google Scholar] [CrossRef] [Scilit]
- Wen, L.; Hughes, M.G. Coastal wetland responses to sea level rise: The losers and winners based on hydro-geomorphological settings. Remote Sens. 2022, 14, 1888. [Google Scholar] [CrossRef] [Scilit]
- Mallick, J.; Alsubih, M.; Ahmed, M.; Almesfer, M.K.; Kahla, N.B. Assessing the spatiotemporal heterogeneity of terrestrial temperature as a proxy to microclimate and its relationship with urban hydro-biophysical parameters. Front. Ecol. Evol. 2022, 10, 878375. [Google Scholar] [CrossRef] [Scilit]
- Chen, H.; Yunus, A.P.; Nukapothula, S.; Avtar, R. Modelling Arctic coastal plain lake depths using machine learning and Google Earth Engine. Phys. Chem. Earth 2022, 126, 103138. [Google Scholar] [CrossRef] [Scilit]
- Bertolini, C.; Capelle, J.; Royer, E.; Milan, M.; Witbaard, R.; Bouma, T.J.; Pastres, R. Using a clustering algorithm to identify patterns of valve-gaping behaviour in mussels reared under different environmental conditions. Ecol. Inform. 2022, 69, 101659. [Google Scholar] [CrossRef] [Scilit]
- Drummond, C.; Blacka, M.; Harley, M.; Brown, W. Smart cameras for coastal monitoring. In Proceedings of the Australasian Coasts and Ports 2021, Online, 11–13 April 2022; pp. 390–396. [Google Scholar]
- Jamil Maia, A.; Cabral Nascimento, R.; Jacques Agra Bezerra da Silva, Y.; Williams Araújo do Nascimento, C.; de Sousa Mendes, W.; Germano Veras Neto, J.; Coelho de Araújo Filho, J.; Tiecher, T.; Jacques Agra Bezerra da Silva, Y. Near-infrared spectroscopy for prediction of potentially toxic elements in soil and sediments from a semiarid and coastal humid tropical transitional river basin. Microchem. J. 2022, 179, 107544. [Google Scholar] [CrossRef] [Scilit]
- Aslan, S.; Zennaro, F.; Furlan, E.; Critto, A. Recurrent neural networks for water quality assessment in complex coastal lagoon environments: A case study on the Venice Lagoon. Environ. Model. Softw. 2022, 154, 105403. [Google Scholar] [CrossRef] [Scilit]
- López-Andreu, F.J.; López-Morales, J.A.; Atenza Juárez, J.F.; Alcaraz, R.; Hernández, M.D.; Erena, M.; Domínguez-Gómez, J.A.; García Galiano, S. Monitoring system of the Mar Menor coastal lagoon (Spain) and its watershed basin using the integration of massive heterogeneous data. Sensors 2022, 22, 6507. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Koeshidayatullah, A.; Trower, E.J.; Li, X.; Mukerji, T.; Lehrmann, D.J.; Morsilli, M.; Al-Ramadan, K.; Payne, J.L. Quantitative evaluation of the roles of ocean chemistry and climate on ooid size across the Phanerozoic: Global versus local controls. Sedimentology 2022, 69, 2486–2506. [Google Scholar] [CrossRef] [Scilit]
- Dubeux, J.C.B., Jr.; Jaramillo, D.; Santos, E.R.S.; Garcia, L.; Queiroz, L.D. Invited review: Ecosystem services provided by grasslands in the Southeast United States. Appl. Anim. Sci. 2022, 38, 648–659. [Google Scholar] [CrossRef] [Scilit]
- Bažant, Z.P.; Nguyen, H.T.; Dönmez, A.A. Scaling in size, time and risk—The problem of huge extrapolations and remedy by asymptotic matching. J. Mech. Phys. Solids 2023, 170, 105094. [Google Scholar] [CrossRef] [Scilit]
- Gimenez, J.G.; Martinez-Espana, R.; Cano, J.-C.; Cecilia, J.M. Estimation of chlorophyll-a in highly anthropized environments using machine learning and remote sensing. In Proceedings of the 19th International Conference on Intelligent Environments (IE 2023), Uniciti, Mauritius, 27–30 June 2023. [Google Scholar] [CrossRef] [Scilit]
- Mastro, P.; Pepe, A. Risk analysis of coastal areas: An AI-based perspective using SAR data. In Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS 2023), Pasadena, CA, USA, 16–21 July 2023; pp. 742–745. [Google Scholar] [CrossRef] [Scilit]
- Le Guillou, A.; Niculescu, S.; Schmullius, C. Machine and deep learning methods for detection and mapping of coastal wetlands of Crozon Peninsula (Brittany, France) used metric and sub-metric spatial resolution. In Proceedings of the Earth Resources and Environmental Remote Sensing/GIS Applications XIV, Amsterdam, The Netherlands, 3–6 September 2023; Volume 12734, p. 1273406. [Google Scholar] [CrossRef] [Scilit]
- Medina-López, E.; Navarro, G.; Santos-Echeandía, J.; Bernárdez, P.; Caballero, I. Machine learning for detection of macroalgal blooms in the Mar Menor coastal lagoon using Sentinel-2. Remote Sens. 2023, 15, 1208. [Google Scholar] [CrossRef] [Scilit]
- Yang, X.; Zhang, X.; Zhang, P.; Bidegain, G.; Dong, J.; Hu, C.; Li, M.; Zhang, Z.; Guo, H. Ensemble habitat suitability modeling for predicting optimal sites for eelgrass (Zostera marina) in the tidal lagoon ecosystem: Implications for restoration and conservation. J. Environ. Manag. 2023, 330, 117108. [Google Scholar] [CrossRef] [Scilit]
- Uddin, M.G.; Nash, S.; Rahman, A.; Olbert, A.I. A sophisticated model for rating water quality. Sci. Total Environ. 2023, 868, 161614. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Jing, L.; Zeng, Q.; He, K.; Liu, P.; Fan, R.; Lu, W.; Lei, G.; Lu, C.; Wen, L. Vegetation dynamic in a large floodplain wetland: The effects of hydroclimatic regime. Remote Sens. 2023, 15, 2614. [Google Scholar] [CrossRef] [Scilit]
- Angelini, R.; Angelats, E.; Luzi, G.; Ribas, F.; Masiero, A.; Mugnai, F. A review and test of shoreline extraction techniques. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. —ISPRS Arch. 2023, XLVIII-1/W, 17–24. [Google Scholar] [CrossRef] [Scilit]
- Giang, T.L.; Bui, Q.T.; Nguyen, T.D.L.; Dang, V.B.; Truong, Q.H.; Phan, T.T.; Nguyen, H.; Ngo, V.L.; Tran, V.T.; Yasir, M.; et al. Coastal landscape classification using convolutional neural network and remote sensing data in Vietnam. J. Environ. Manag. 2023, 335, 117537. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Larson, T.E.; Loucks, R.G.; Sivil, J.E.; Hattori, K.E.; Zahm, C.K. Machine learning classification of Austin Chalk chemofacies from high-resolution x-ray fluorescence core characterization. AAPG Bull. 2023, 107, 907–927. [Google Scholar] [CrossRef] [Scilit]
- Sun, W.; Liu, W.; Wang, Y.; Zhao, R.; Huang, M.; Wang, Y.; Yang, G.; Meng, X. Research progress and prospects of hyperspectral remote sensing for global wetland from 2010 to 2022. Natl. Remote Sens. Bull. 2023, 27, 1281–1299. [Google Scholar] [CrossRef]
- Kruk, M. Prediction of environmental factors responsible for chlorophyll a-induced hypereutrophy using explainable machine learning. Ecol. Inform. 2023, 75, 102005. [Google Scholar] [CrossRef] [Scilit]
- López-Andreu, F.J.; López-Morales, J.A.; Hernández-Guillen, Z.; Carrero-Rodrigo, J.A.; Sánchez-Alcaraz, M.; Atenza-Juárez, J.F.; Erena, M. Deep learning-based time series forecasting models evaluation for the forecast of chlorophyll a and dissolved oxygen in the Mar Menor. J. Mar. Sci. Eng. 2023, 11, 1473. [Google Scholar] [CrossRef] [Scilit]
- Ren, H.H.; Cheng, Y.; Wu, F.; Gu, Z.L.; Cao, J.J.; Huang, Y.; Xue, Y.G.; Cui, L.; Zhang, Y.W.; Chow, J.C.; et al. Spatiotemporal characteristics of ozone and the formation sensitivity over the Fenwei Plain. Sci. Total Environ. 2023, 881, 163369. [Google Scholar] [CrossRef] [Scilit]
- Mogane, L.K.; Masebe, T.; Msagati, T.A.M.; Ncube, E. A comprehensive review of water quality indices for lotic and lentic ecosystems. Environ. Monit. Assess. 2023, 195, 926. [Google Scholar] [CrossRef] [Scilit]
- O’Shea, R.E.; Pahlevan, N.; Smith, B.; Boss, E.; Gurlin, D.; Alikas, K.; Kangro, K.; Kudela, R.M.; Vaičiūtė, D. A hyperspectral inversion framework for estimating absorbing inherent optical properties and biogeochemical parameters in inland and coastal waters. Remote Sens. Environ. 2023, 295, 113706. [Google Scholar] [CrossRef] [Scilit]
- Bindajam, A.A.; Mallick, J.; Talukdar, S.; Shahfahad Shohan, A.A.A.; Rahman, A. Modeling the spatiotemporal heterogeneity of land surface temperature and its relationship with land use land cover using geo-statistical techniques and machine learning algorithms. Environ. Sci. Pollut. Res. 2023, 30, 106917–106935. [Google Scholar] [CrossRef] [Scilit]
- Wang, J.; Xue, L.; Xiang, C.; Li, X.; Xie, L. Improving ecological indicators of arid zone deserts through simulation. Ecol. Model. 2023, 485, 110506. [Google Scholar] [CrossRef] [Scilit]
- Kavhu, B.; Mashimbye, Z.E.; Luvuno, L. Predicting priority management areas for land use/cover change in the transboundary Okavango basin based on machine learning. Heliyon 2023, 9, e22762. [Google Scholar] [CrossRef] [Scilit]
- Sangiorgio, F.; Pinna, M.; Basset, A. Inter- and intra-habitat variability of plant detritus decomposition in a transitional environment (Lake Alimini, Adriatic Sea). Chem. Ecol. 2004, 20, 353–366. [Google Scholar] [CrossRef] [Scilit]
- Delnevo, G.; Tumedei, G.; Ghini, V.; Prandi, C. Evaluating the use of machine learning algorithms in environmental sensing for energy saving. In Proceedings of the NETus 2023—Proceedings of the 2023 Workshop on Networked Sensing Systems for a Sustainable Society, Madrid, Spain, 6 October 2023; pp. 201–206. [Google Scholar]
- Zhao, Z.; Shen, Y.; Zhu, H.; Zhang, Q.; Zhang, Y.; Yang, X.; Liang, P.; Chen, L. Prediction model of type and band gap for photocatalytic g-GaN-based van der Waals heterojunction of density functional theory and machine learning techniques. Appl. Surf. Sci. 2023, 640, 158400. [Google Scholar] [CrossRef] [Scilit]
- Zennaro, F.; Furlan, E.; Canu, D.; Aveytua Alcazar, L.; Rosati, G.; Solidoro, C.; Aslan, S.; Critto, A. Venice lagoon chlorophyll-a evaluation under climate change conditions: A hybrid water quality machine learning and biogeochemical-based framework. Ecol. Indic. 2023, 157, 111245. [Google Scholar] [CrossRef] [Scilit]
- Almeida, B.; Cabral, P. A hybrid modelling approach for detecting seasonal variations in inland green-blue ecosystems. Remote Sens. Appl. Soc. Environ. 2024, 33, 101121. [Google Scholar] [CrossRef] [Scilit]
- Xue, J.; Yuan, C.; Ji, X.; Zhang, M. Predictive modeling of nitrogen and phosphorus concentrations in rivers using a machine learning framework: A case study in an urban-rural transitional area in Wenzhou, China. Sci. Total Environ. 2024, 910, 168521. [Google Scholar] [CrossRef] [Scilit]
- Uddin, M.G.; Nash, S.; Rahman, A.; Dabrowski, T.; Olbert, A.I. Data-driven modelling for assessing trophic status in marine ecosystems using machine learning approaches. Environ. Res. 2024, 242, 117755. [Google Scholar] [CrossRef] [Scilit]
- Delnevo, G.; Tumedei, G.; Ghini, V.; Prandi, C. Toward a digital twin: Combining sensing, machine learning, and data visualization for the effective management of a coastal lagoon environment. In Proceedings of the—IEEE Consumer Communications and Networking Conference (CCNC 2024), Las Vegas, NV, USA, 6–9 January 2024. [Google Scholar] [CrossRef] [Scilit]
- Dalby, O.; Kennedy, D.M.; McCarroll, R.J.; Young, M.; Ierodiaconou, D. Mapping surface sediment characteristics in enclosed shallow-marine environments using spatially balanced designs and the random forest algorithm. Earth Surf. Process. Landf. 2024, 49, 2884–2897. [Google Scholar] [CrossRef] [Scilit]
- Orlić, M.; Pasarić, M. How to disentangle sea-level rise and a number of other processes influencing coastal floods? Rend. Lincei 2024, 35, 371–380. [Google Scholar] [CrossRef] [Scilit]
- de Sousa, M.F., Jr.; Uliana, E.M.; Aires, R.V.U.; Rápalo, L.M.C.; da Silva, D.D.; Moreira, M.C.; Lisboa, L.; da Silva Rondon, D. Streamflow prediction based on machine learning models and rainfall estimated by remote sensing in the Brazilian Savanna and Amazon biomes transition. Model. Earth Syst. Environ. 2024, 10, 1191–1202. [Google Scholar] [CrossRef] [Scilit]
- Wang, L.; Liang, H.; Wang, S.; Sun, D.; Li, J.; Zhang, H.; Yuan, Y. Estimating four-decadal variations of seagrass distribution using satellite data and deep learning methods in a marine lagoon. Sci. Total Environ. 2024, 919, 170936. [Google Scholar] [CrossRef] [Scilit]
- Karimpouli, S.; Kwiatek, G.; Martínez-Garzón, P.; Dresen, G.; Bohnhoff, M. Unsupervised clustering of catalogue-driven features for characterizing temporal evolution of labquake stress. Geophys. J. Int. 2024, 237, 755–771. [Google Scholar] [CrossRef] [Scilit]
- Lemenkova, P. Artificial neural networks for mapping coastal lagoon of Chilika Lake, India, using Earth observation data. J. Mar. Sci. Eng. 2024, 12, 709. [Google Scholar] [CrossRef] [Scilit]
- Fanini, L.; Defeo, O.; Elliott, M.; Paragkamian, S.; Pinna, M.; Salvo, V.S. Coupling beach ecology and macroplastics litter studies: Current trends and the way ahead. Mar. Pollut. Bull. 2021, 173, 112951. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Hanus, R.; Zych, M.; Ochał, P.; Augustyn, M. Identification of the structure of liquid–gas flow in a horizontal pipeline using the gamma-ray absorption and a convolutional neural network. Appl. Sci. 2024, 14, 4854. [Google Scholar] [CrossRef] [Scilit]
- Lin, H.; Liu, X.; Xiong, J.; Liang, L.; Wu, J.; Li, B. Reservoir horizontal principle stress prediction using intelligent fusion model based on physical model constraints: A case study of Daji Block, Eastern Ordos Basin, North China. Acta Geophys. 2024, 72, 2565–2579. [Google Scholar] [CrossRef] [Scilit]
- Kang, D.; Ryan, D.; Kim, S. Flow regime classification using orientation-independent layered spectral clustering. In Proceedings of the 20th International Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH 2023), Washington, DC, USA, 20–25 August 2023; pp. 5148–5159. [Google Scholar] [CrossRef] [Scilit]
- Santiago Borrés, S.E.; Kumar, S.; Bonzongo, J.C.; Deliz Quiñones, K.Y.; Jutla, A. A multi-factor analysis linking environmental stressors with presence of per- and polyfluoroalkyl substances (PFAS) in a coastal lagoon. ACS EST Water 2024, 4, 3464–3474. [Google Scholar] [CrossRef] [Scilit]
- Cingano, P.; Vuerich, M.; Petruzzellis, F.; Orzan, L.; Trotta, G.; Casolo, V.; Asquini, E.; Bacaro, G.; Boscutti, F. Seagrasses on the move: Tracing the multi-decadal species distribution trends in lagoon meadows using Landsat imagery. Ecol. Inform. 2024, 82, 102685. [Google Scholar] [CrossRef] [Scilit]
- Maionchi, D.D.O.; Silva, J.G.D.; Balista, F.A.; Junior, W.A.M.; Paulo, S.R.D.; Paulo, I.J.C.D.; Biudes, M.S. Estimating hourly air temperature in an Amazon-Cerrado transitional forest in Brazil using machine learning regression models. Theor. Appl. Climatol. 2024, 155, 7827–7843. [Google Scholar] [CrossRef] [Scilit]
- Martins, I.R.; Chainho, P.; Goulding, T.; Samarão, J.; Rufino, M.M. Using shell shape analysis based on landmarks to trace the geographical origin of the common cockle (Cerastoderma edule). Food Control 2024, 166, 110662. [Google Scholar] [CrossRef] [Scilit]
- Wu, W.; Tan, C.; Zhang, S.; Dong, F. Joint mining of fluid knowledge and multi-sensor data for gas–water two-phase flow status monitoring and evolution analysis. Adv. Eng. Inform. 2024, 62, 102687. [Google Scholar] [CrossRef] [Scilit]
- Narciso, G.A.M.; Tamondong, A.M.; Blanco, A.C.; Nakamura, T.; Nadaoka, K. Supervised image classification model for coral bleaching detection using a bi-temporal Sentinel-2 image stack. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2024, XLVIII-4/W8-2023, 395–401. [Google Scholar] [CrossRef] [Scilit]
- Saccotelli, L.; Verri, G.; De Lorenzis, A.; Cherubini, C.; Caccioppoli, R.; Coppini, G.; Maglietta, R. Enhancing estuary salinity prediction: A machine learning and deep learning based approach. Appl. Comput. Geosci. 2024, 23, 100173. [Google Scholar] [CrossRef] [Scilit]
- Bozzeda, F.; Zangaro, F.; Colangelo, M.A.; Pinna, M. Relationships between size and abundance in beach plastics: A power-law approach. Mar. Pollut. Bull. 2021, 173, 113005. [Google Scholar] [CrossRef] [Scilit]
- Simonetti, I.; Lubello, C.; Cappietti, L. On the use of hydrodynamic modelling and random forest classifiers for the prediction of hypoxia in coastal lagoons. Sci. Total Environ. 2024, 951, 175424. [Google Scholar] [CrossRef] [Scilit]
- Chang, S.; Chen, H.; Pan, J.; Jiang, L. Performances of three representative snow depth products originated from passive microwave sensors over the Mongolian Plateau. Int. J. Digit. Earth 2024, 17, 2395871. [Google Scholar] [CrossRef] [Scilit]
- McClanahan, T.R.; Sola, E. Comparing modeled predictions of coral reef diversity along a latitudinal gradient in Mozambique. Front. Ecol. Evol. 2024, 12, 1450383. [Google Scholar] [CrossRef] [Scilit]
- Habib, M.A.; Zarillo, G.A. Construction of a real-time forecast model with deep learning techniques for coastal engineering and processes: Nested in a basin scale suite of models. J. Mar. Sci. Eng. 2024, 12, 1152. [Google Scholar] [CrossRef] [Scilit]
- Karthick, M.; Shanmugam, P.; Saravana Kumar, G. Long-term water quality assessment in coastal and inland waters: An ensemble machine-learning approach using satellite data. Mar. Pollut. Bull. 2024, 209, 117036. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Yao, T.; Ye, H.; Wang, Y.; Zhang, J.; Guo, J.; Li, J. Kolmogorov-Zurbenko filter coupled with machine learning to reveal multiple drivers of surface ozone pollution in China from 2015 to 2022. Sci. Total Environ. 2024, 949, 175093. [Google Scholar] [CrossRef] [Scilit] [PubMed]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.






