ABSTRACT Simple elevation, defined here as the unprocessed vertical height of the land surface above a reference datum without the application of derivative computations such as slope or curvature, is frequently cited as the most or among the most important predictor variables in wetland classification. However, importance is often measured using biased metrics and there have been numerous cases in which excluding simple elevation has had little to no impact on model performance. Additionally, although wetlands can form at any elevation, they are often more abundant and/or obvious at lower elevations. This can lead to biased training conditions that result in overfitting. To advance wetland mapping, it is essential to accurately quantify the importance of predictor variables and evaluate their impact on model generalization. The objectives of this study were therefore to re-evaluate the importance and highlight the potential risks associated with incorporating simple elevation into wetland classification models. Seasonal composites of C-band Sentinel-1 Synthetic Aperture Radar are classified in combination with three different simple elevation datasets to: (i) demonstrate a scenario in which, regardless of its quality, simple elevation offers limited predictive power to the model, and (ii) simulate biased training conditions to show how its inclusion can lead to overfitting. Results show that the correlation between wetland presence and elevation is spatially variable, limiting the ability of models to generalize across domains when elevation is included.

This study evaluates the performance of artificial intelligence (AI) technologies for wetland classification in the province of Alberta, Canada, using integrated remote sensing inputs, including airborne light detection and ranging (LiDAR), orthophotography, and multi-sensor satellite imagery (Sentinel-1, Sentinel-2, PlanetScope). Our primary objective was to assess whether AI-driven modelling approaches, specifically machine learning (ML) and deep learning (DL), can meet Alberta’s provincial wetland mapping standards. We hypothesized that integrating high-resolution LiDAR with multi-seasonal optical and radar data composites into advanced AI algorithms would achieve the required classification accuracy, detail, and minimum mapping unit targets. We tested several methodologies in four ecologically distinct pilot areas representing Alberta’s Boreal, Grassland, and Parkland Natural Regions. AI models included ensemble ML using Extreme Gradient Boosting (XGBoost) and Random Forest, and a DL U-Net convolutional neural network (CNN). AI models were trained on expert-labelled photoplots and validated using in situ field surveys. Our findings demonstrate that both ML and DL models met and, in several cases, exceeded the provincial mapping standards with validation overall accuracies surpassing >70% (form), >80% (class), and >90% (wetland–upland). U-Net CNN models generally produced the highest overall accuracies and most precise wetland extent delineation, but XGBoost offered finer detail and granularity for detailed mapping of rare wetland forms. Integrating LiDAR data and derivatives further enhanced model performance, improving accuracy by as much as 13%. Based on these outcomes, we provide a set of recommendations for scaling up these approaches, focusing on model selection, LiDAR imagery integration, and the continued value of field surveys to support the operational scaling of AI-driven classification approaches for wetland inventory updates across Alberta’s diverse landscapes. However, key challenges remain in scaling up this approach due to the cost of acquiring high-resolution LiDAR and satellite imagery.

Wetlands, dynamic interfaces between land and water, serve as biodiversity hotspots and vital providers of ecosystem services, including water purification, flood control and carbon sequestration. However, rapid degradation from urbanisation, agriculture and resource extraction necessitates urgent and effective conservation strategies. This literature review synthesises global studies from 1999 to 2023 to evaluate the role of remote sensing technology in wetland conservation and management. Using a systematic literature review approach, peer-reviewed articles were selected based on their application of satellite imagery, aerial photography, LiDAR, SAR, object-based image analysis (OBIA) and machine learning techniques for wetland mapping, monitoring and management. The analysis reveals that remote sensing enables accurate boundary delineation, wetland-type identification and long-term change detection, with hybrid optical-SAR-LiDAR approaches and Random Forest classifiers consistently achieving the highest accuracy. Continuous monitoring effectively tracks land-cover shifts, vegetation health and hydrological dynamics, while integration with multi-criteria decision tools (AHP-TOPSIS) supports conservation prioritisation. Results highlight significant advancements in national-scale wetland mapping and predictive modelling of future loss, yet persistent gaps remain in satellite-ground data fusion, standardised classification systems, optimal temporal sampling and policy translation. The review concludes that remote sensing has become an indispensable, cost-effective tool for evidence-based wetland conservation worldwide, including in data-scarce regions like East Africa. Recommendations include: (1) adopting combined optical-radar-LiDAR workflows as standard practice, (2) developing regionally harmonised wetland classification frameworks, (3) promoting community-ground-truthed validation, (4) establishing open-access wetland monitoring platforms for Africa, and (5) strengthening collaboration between remote sensing experts, ecologists and policymakers to ensure research directly informs restoration and protection strategies.

ABSTRACT Wetlands play a vital role in supporting hydrologic, biogeochemical, and ecological processes across landscapes, yet understanding and quantifying their functional contributions at large spatial scales remains a challenge. In response to increasing demand for regional and national assessments, a growing number of methods have emerged that estimate wetland function using widely available geospatial data. This systematic review describes and evaluates currently available approaches to quantifying wetland function at landscape scales and highlights key lessons for future assessments. We identify two primary methodological categories: classification‐based approaches, such as hydrogeomorphic (HGM) and LLWW (Landscape Position, Landform, Water Flow Path, Waterbody Type) frameworks, which assign functional scores based on mapped wetland class; and indicator‐based approaches, which derive metrics or indices directly linked to functional processes using remote sensing and other spatial data. Across both, we find that validation against field data remains limited, habitat functions are consistently the most difficult to assess, and that many assessments estimate only the potential of wetlands to perform specific functions rather than how they are actually functioning under current landscape conditions. At the same time, advances in high‐resolution remote sensing, automation, and ecological modeling are creating new opportunities for more scalable, repeatable, and functionally relevant assessments. Hybrid approaches that bridge classification and indicator methods, and that integrate land‐use and disturbance metrics, represent a promising path toward national‐scale functional assessments. Together, these findings point to a way forward for producing wetland functional assessments that are both scientifically rigorous and actionable for conservation and policy. This article is categorized under: Water and Life > Nature of Freshwater Ecosystems Water and Life > Conservation, Management, and Awareness Science of Water > Water and Environmental Change

Wetlands are essential ecosystems increasingly threatened by human activities and climate change. This study presents a method for classifying and monitoring wetland habitats in the Čiližská Radvaň protected area (Slovak Republic) using RGB drone imagery and the Natural Numerical Network (NatNet), a mathematically based supervised deep learning approach. The primary aim was to evaluate the effectiveness of NatNet in identifying target habitat types and to assess the impact of ongoing revitalisation efforts. Habitat types were classified using RGB drone imagery and ground-truth training polygons that represented the dominant vegetation communities in Čiližská Radvaň wetland. The NatNet achieved the training classification success rate exceeding 97%, allowing the creation of relevancy maps successfully identifying spatial habitat distribution. Relevancy maps verified in the field reached classification accuracy of 0.88 and F1 score of 0.90 across all habitats together. Results showed observable shifts in habitat extent and structure after one year of restoration, confirming the suitability of the method for detecting ecological changes in wetland environments.

ABSTRACT Mapping wetlands is essential for conservation, but their dynamic nature makes them difficult to detect, especially drier-end types. These challenges are amplified in urban and peri-urban areas, where wetlands are small, fragmented, and frequently under-represented in inventories. Furthermore, many mapping efforts lack the methodological transparency required for replicability and adaptation. This study addresses these gaps by presenting a novel, replicable, and cost-effective methodological framework for mapping complex wetlands. The approach is structured as a hierarchical workflow that leverages open data and computational resources. It integrates multi-temporal optical, SAR, and LiDAR-derived features within Google Earth Engine to generate a time-series of annual Land Use/Land Cover (LULC) classifications. This is followed by a post-processing workflow that produces a final filtered Wetlands map, a LULC map (mode), and a Frequency map, allowing for a nuanced understanding of wetland dynamics. We developed and evaluated this framework in the Metropolitan area of Buenos Aires, Argentina, a landscape where such wetlands are often overlooked. The framework successfully delineated multiple wetland classes (Ponds, Wet meadows, Artificial wetlands and Rivers) from non-wetland classes. The final Wetlands map was evaluated using 372 independent points with an overall accuracy of 0.91 (95% CI: [0.87, 0.94]), estimated from the population error matrix to provide unbiased accuracy and area estimates. The primary contribution is the transparent methodological framework itself. All scripts are openly shared to facilitate full replication and adaptation, providing environmental managers a robust tool for inventorying and monitoring challenging wetland systems.

Abstract Introduction Wetlands are characterized by hydrology, hydric soils, and hydrophytic vegetation, from which numerous wetland classification schemes have been derived that propose groups of “similar” wetlands. However, trends and variation in long‐term hydrology between and among these classifications remain poorly understood. Objectives Our study aimed to answer three questions: (1) How variable is wetland hydrology between and among different reference wetland types? (2) How are attributes of the ecohydrologic regime related to wetland type? and (3) Which attributes of wetland morphometry (if any) are predictive of wetland ecohydrology across wetland types? Methods We utilized 17 years of bi‐weekly hydrology data from 30 reference wetlands. Hydrologic similarity was assessed using non‐parametric Spearman correlations. Differences in ecohydrologic regimes among wetland types were tested using Kruskal–Wallis and pairwise Wilcoxon rank‐sum tests, principal components analysis, and permutational multivariate analysis of variance (PERMANOVA). To evaluate morphometric controls, we related morphometric attributes (e.g., depth, perimeter:area ratio) to ecohydrologic metrics using redundancy analysis with variance partitioning and stepwise model selection. Results Hydrologic similarity varied widely among wetland types ( r = 0.38 to 0.95). Ecohydrologic regimes showed limited internal consistency, with up to 92% of metrics differing among cypress wetlands and up to 100% of pairwise comparisons differing across all wetland types for magnitude‐based metrics. Wetland depth and perimeter: area ratio emerged as the strongest morphometric predictors of ecohydrologic regime, together explaining 20% of observed variance. Conclusions Wetlands of the same classification often exhibit divergent hydrologic regimes, but simple morphometric attributes improve predictions of ecohydrologic similarity.

Wetlands, known as “the kidney of the Earth”, serve as critical ecological carriers for global sustainable development. The fine classification of wetlands is crucial to their utilization and protection. Wetland fine-scale classification based on remote sensing imagery has long been challenged by disturbances such as clouds, fog, and shadows. Simultaneously, the confusion of spectral information among land cover types remains a primary factor affecting classification accuracy. To address these challenges, this paper proposes a fine classification model of wetlands in remote sensing images based on multi-temporal data and feature optimization (CMW-MTFO). The model is divided into three parts: (1) a multi-satellite and multi-temporal remote sensing image fusion module; (2) a feature optimization module; and (3) a feature classification network module. Multi-satellite multi-temporal image fusion compensates for information gaps caused by cloud cover, fog, and shadows, while feature optimization reduces spectral characteristics prone to confusion. Finally, fine classification is completed using the feature classification network based on deep learning. Using coastal wetlands in Liaoning Province, China, as the experimental area, this study compares the CMW-MTFO with several classical wetland classification methods, non-feature-optimized classification, and single-temporal classification. Results show that the proposed model achieves an overall classification accuracy of 98.31% for Liaoning wetlands, with a Kappa coefficient of 0.9795. Compared to the classic random forest method, classification accuracy and Kappa coefficient improved by 11.09% and 0.1286, respectively. Compared to non-feature-based classification, classification accuracy increased by 1.06% and Kappa coefficient by 1.18%. Compared to the best classification performance using single-temporal images, the proposed method achieved a 1.81% increase in classification accuracy and a 2.19% increase in Kappa value, demonstrating the effectiveness of the model approach.

Advances in machine learning for wetland classification: a comprehensive survey of methods and applications

ABSTRACT Wetland vegetation is vital for maintaining ecosystem functions, necessitating precise mapping and dynamic monitoring for conservation. Traditional object-based image analysis (OBIA) struggles with subjective scale parameter selection in heterogeneous wetlands. This study proposes a novel scale-free classification framework tailored for coastal wetlands: (1) A scale-set structure is utilized to organize multiscale coastal wetland areas generated by region merging algorithms, avoiding inefficiency and subjectivity in selecting scale parameters during the segmentation phase. (2) A multiscale sample selection and enrichment strategy is adopted to ensure that the training model can learn multidimensional features from the macro- to microlevel, thereby enhancing the model's generalization ability and classification accuracy. (3) In terms of feature extraction, spectral features, index features, textural features, and geometric features are combined and optimized, further deepening the algorithm's understanding of the spatial distribution and layout patterns of wetland vegetation. Experiments have demonstrated significant improvements in segmenting mangroves, Spartina alterniflora, mudflats, and water bodies, with the overall accuracy increasing from 72% to 96%. The framework effectively addresses scale-related challenges in coastal wetland mapping, offering practical solutions for ecological restoration and conservation.

The Amazon wetlands are the largest and most diverse freshwater ecosystem globally, characterized by various flooded vegetation and the Amazon River’s estuary. This critical ecosystem is vulnerable to land use changes, dam construction, mining, and climate change. While several studies have utilized remote sensing to map wetlands in this region, significant uncertainty remains, which limits the assessment of impacts and the conservation priorities for Amazon wetlands. This study aims to enhance wetland mapping by integrating existing maps, remote sensing data, expert knowledge, and cloud computing via Earth Engine. We developed a harmonized regional wetland classification system adaptable to individual countries, enabling us to train and build a random forest model to classify wetlands using a robust remote sensing dataset. In 2020, wetlands spanned 151.7 million hectares (Mha) or 22.0% of the study area, plus an additional 7.4 Mha in deforested zones. The four dominant wetland classes accounted for 98.5% of the total area: Forest Floodplain (89.0 Mha; 58.6%), Lowland Herbaceous Floodplain (29.6 Mha; 19.6%), Shrub Floodplain (16.7 Mha; 11.0%), and Open Water (14.1 Mha; 9.3%). The overall mapping accuracy was 82.2%. Of the total wetlands in 2020, 52.6% (i.e., 79.8 Mha) were protected in Indigenous Territories, Conservation Units, and Ramsar Sites. Threats to the mapped wetlands included 7.4 Mha of loss due to fires and deforestation, with an additional 800,000 ha lost from 2021 to 2024 due to agriculture, urban expansion, and gold mining. Notably, 21 Mha of wetlands were directly affected by both reduced precipitation and surface water in 2020. Our mapping efforts will help identify priorities for wetland protection and support informed decision-making by local governments and ancestral communities to implement conservation and management plans. As 47.4% of the mapped wetlands are unprotected and have some level of threats and pressure, there are also opportunities to expand protected areas and implement effective management and conservation practices.

Accurate and detailed mapping of coastal wetlands is essential for effective wetland resource management. However, due to periodic tidal inundation, frequent cloud cover, and spectral similarity of land cover types, reliable coastal wetland classification methods remain limited. To address these issues, we developed an integrated pixel- and object-based hierarchical classification strategy based on multi-source remote sensing data to achieve fine-grained coastal wetland classification on Google Earth Engine. With the random forest classifier, pixel-level classification was performed to classify rough wetland and non-wetland types, followed by object-based classification to differentiate artificial and natural attributes of water bodies. In this process, multi-dimensional features including water level, phenology, variation, topography, geography, and geometry were extracted from Sentinel-1/2 time-series images, topographic data and shoreline data, which can fully capture the variability and dynamics of coastal wetlands. Feature combinations were then optimized through Recursive Feature Elimination and Jeffries–Matusita analysis to ensure the model’s ability to distinguish complex wetland types while improving efficiency. The classification strategy was applied to typical coastal wetlands in central Jiangsu in 2020 and finally generated a 10 m wetland map including 7 wetland types and 3 non-wetland types, with an overall accuracy of 92.50% and a Kappa coefficient of 0.915. Comparative analysis with existing datasets confirmed the reliability of this strategy, particularly in extracting intertidal mudflats, salt marshes, and artificial wetlands. This study can provide a robust framework for fine-grained wetland mapping and support the inventory and conservation of coastal wetland resources.

Wetlands are composed of the interaction of water, soil and suitable vegetation, which has rich biological resources and strong ecological benefits. Due to increasing human disturbance and the effects of climate change, wetlands are being dramatically degraded and destroyed. However, the existing wetland products lack the ability to capture and update the dynamic changes in time and space, with less attention to the classification based on hydrological processes and vegetation types. Therefore, we developed a Decision Tree (DT)-based classification method, incorporating water frequency (WF) and vegetation frequency (VF) calibrated with field observations, to monitor wetland dynamics using Landsat-5/7/8/9 time-series images (2000–2022) and Google Earth Engine (GEE). Taking Beidagang Wetland as the study area, six classes were extracted with high overall accuracy (0.89) and Kappa coefficient (0.85) in 2022. Interannual dynamics during 2000–2022 revealed two distinct periods: terrestrial vegetation (TerV) dominance with permanent water (PW) below 10% (2000–2014), and PW exceeding 20% while temporary vegetation (TemV) decreased (2015–2022). Spatially, land cover types radiated outward from Tiane Lake, with northwestern regions primarily covered by TerV and southeastern regions by TemV and barren (B). Frequent type conversions occurred between adjacent classes, with the most significant changes in Guanqi Lake. Despite declining wetland water volumes due to rising temperatures and reduced precipitation, ecological compensation measures, including functional zoning, water replenishment, and phragmites restoration, have continuously improved the wetland environment. This study presents a promising method combining Landsat time-series images, DT and GEE for continuous land cover monitoring. Threshold optimization using local data and interpretability based on vegetation physiological characteristics demonstrate enhanced applicability for large-scale wetland classification. The generated annual maps represent the most current dataset for Beidagang Wetland, providing scientific support for wetland monitoring, protection and management.

Wetlands play a crucial role in climate regulation, pollutant filtration, and biodiversity conservation. Accurate wetland classification through high-resolution remote sensing imagery is pivotal for the scientific management, ecological monitoring, and sustainable development of these ecosystems. However, the intricate spatial details in such imagery pose significant challenges to conventional interpretation techniques, necessitating precise boundary extraction and multi-scale contextual modeling. In this study, we propose WetSegNet, an edge-guided Multi-Scale Feature Interaction network for wetland classification, which integrates a convolutional neural network (CNN) and Swin Transformer within a U-Net architecture to synergize local texture perception and global semantic comprehension. Specifically, the framework incorporates two novel components: (1) a Multi-Scale Feature Interaction (MFI) module employing cross-attention mechanisms to mitigate semantic discrepancies between encoder–decoder features, and (2) a Multi-Feature Fusion (MFF) module that hierarchically enhances boundary delineation through edge-guided spatial attention (EGA). Experimental validation on GF-2 satellite imagery of Dongting Lake wetlands demonstrates that WetSegNet achieves state-of-the-art performance, with an overall accuracy (OA) of 90.81% and a Kappa coefficient of 0.88. Notably, it achieves classification accuracies exceeding 90% for water, sedge, and reed habitats, surpassing the baseline U-Net by 3.3% in overall accuracy and 0.05 in Kappa. The proposed model effectively addresses heterogeneous wetland classification challenges, validating its capability to reconcile local–global feature representation.

Using Landsat series imagery and the deep learning model CITNet, this study conducted high-accuracy classification and spatiotemporal change analysis of wetlands on Chongming Island from 2000–2020 and explored the driving mechanisms by integrating climatic and anthropogenic factors. The results demonstrate that the total wetland area decreased by approximately 125.5 km2 over the two decades. Among natural wetlands, tidal mudflats and shallow seawater zones continuously shrank, while herbaceous marshes exhibited a “decline recovery” trajectory. Artificial wetlands expanded before 2005 but contracted significantly thereafter, mainly due to aquaculture pond reduction. Wetland transformation was dominated by wetland-to-non-wetland conversions, peaking during 2005–2010. Driving factor analysis revealed a “human pressure dominated, climate modulated” pattern: nighttime light index (NTL) and GDP demonstrated strong negative correlations with wetland extent, while minimum temperature and the Palmer Drought Severity Index (PDSI) promoted herbaceous marsh expansion and accelerated artificial wetland contraction, respectively. The findings indicate that wetland changes on Chongming Island result from the combined effects of policy, economic growth, and ecological processes. Sustainable management should focus on restricting urban expansion in ecologically sensitive zones, optimizing water resource allocation under drought conditions, and incorporating climate adaptation and invasive species control into restoration programs to maintain both the extent and ecological quality of wetlands.

Vegetation ecological feature-aware multimodal network: Fine-scale classification of the Yellow River Delta wetlands using UAV-based hyperspectral and LiDAR data

Assessing the spatio evolution of carbon sequestration and optimizing ecological restoration strategies using the InVEST model: A case study of the Yellow River Estuary, China.

(1) Background: Wetlands are ecologically significant ecosystems that support biodiversity and contribute to essential environmental functions such as water purification, carbon storage and flood regulation. However, these ecosystems face increasing pressures from land-use change and degradation, prompting the need for scalable and accurate classification methods to support conservation and policy efforts. In this research, our motivation was to test whether high-spatial-resolution PlanetScope imagery can be used with pixel-based machine learning to support the mapping and monitoring of wetlands at a national scale. (2) Methods: This study compared four machine learning classification models—Random Forest (RF), XGBoost (XGB), Histogram-Based Gradient Boosting (HGB) and a Multi-Layer Perceptron Classifier (MLPC)—to detect and map wetland areas across New Zealand. All models were trained using eight-band SuperDove satellite imagery from PlanetScope, with a spatial resolution of ~3 m, and ancillary geospatial datasets representing topography and soil drainage characteristics, each of which is available globally. (3) Results: All four machine learning models performed well in detecting wetlands from SuperDove imagery and environmental covariates, with varying strengths. The highest accuracy was achieved using all eight image bands alongside features created from supporting geospatial data. For binary wetland classification, the highest F1 scores were recorded by XGB (0.73) and RF/HGB (both 0.72) when including all covariates. MLPC also showed competitive performance (wetland F1 score of 0.71), despite its relatively lower spatial consistency. However, each model over-predicts total wetland area at a national level, an issue which was able to be reduced by increasing the classification probability threshold and spatial filtering. (4) Conclusions: The comparative analysis highlights the strengths and trade-offs of RF, XGB, HGB and MLPC models for wetland classification. While all four methods are viable, RF offers some key advantages, including ease of deployment and transferability, positioning it as a promising candidate for scalable, high-resolution wetland monitoring across diverse ecological settings. Further work is required for verification of small-scale wetlands (<~0.5 ha) and the addition of fine-spatial-scale covariates.

ABSTRACT Wetland monitoring is essential for understanding dynamics, protecting ecological environments, and supporting sustainable development. The potential of multiple spatial–temporal SAR features derived from time series InSAR for fine-scale wetland mapping is not yet fully understood. In this study, we utilized time series Sentinel-1 VH/VV images to derive multi-modal features (Amplitude, coherence, and phase) and explored the accuracy of these features in fine-scale coastal wetland classification. Furthermore, we proposed an auto-weighted ensemble machine learning (AWEML) classifier to optimize wetland monitoring accuracy. Finally, we employed SHAP to assess the influence of features with different temporal baselines on various ground objects. The results of the Yellow River Delta wetland indicate that: (1) Integrating multi-modal features improved classification accuracy under XGBoost by 13.1%, 6.55%, and 27.21% compared to single input amplitude, coherence, and phase features, respectively. (2) AWEML demonstrated superior performance, achieving OA of 94.36%, which is 2.42% and 2.81% higher than ensemble machine learning with soft-voting and XGBoost. (3) VH short temporal baseline coherence features significantly improved the extraction accuracy of salt marsh vegetation, while VV long temporal baseline coherence features performed well for identifying stable ground objects. Our method provides a new reference for coastal wetland fine mapping using SAR data. Abbreviations: SA: Spartina alterniflora; PA: Phragmites australis; TC: Tamarix chinensis; WT: water; FL: farmland; TF: tidal flat; BB: building and bare land; SS: Sueada salsa; IC: Imperata cylindrica.

Abstract. The coastal area of New Hanover County in North Carolina encompasses diverse wetland habitats influenced by unique coastal and tidal dynamics, with researchers examining the impacts of landscape changes, sea-level rise, and climate fluctuations on wetland health and biodiversity. This study integrates multispectral imagery data, LiDAR, and additional sources to enhance classification accuracy. The study also addresses binary classification for wetland and non-wetland classification and a multi-classification for different wetland classes, leveraging on the Random Forest algorithm which significantly improved the overall accuracy of wetland mapping. The Random Forest model’s performance in different scenarios was evaluated, with Scenario 1 achieving an overall accuracy of nearly 93.9%, Scenario 2 achieving an overall accuracy of 93.5%, Scenario 3 achieving an overall accuracy of 94.1%, and Scenario 4 achieving an overall accuracy of 88.2%. These results underscore the model’s effectiveness in accurately classifying coastal wetland areas under diverse remote sensing scenarios, highlighting its potential for practical applications in wetland mapping and ecological research.