The management of natural resources is increasingly critical and challenging due to complex interactions among environmental, industrial, and societal processes. Traditional approaches often fail to integrate heterogeneous data, limiting predictive and decision-support capabilities. This study presents a conceptual architecture for an Artificial Intelligence (AI)-assisted Digital Twin (DT) of the Centre-Val de Loire region, designed to unify time-dependent multi-source data. Based on the ENVRI Reference Model, it covers Science, Information, Computational, Engineering, and Technology layers, defining standardized data exchange, communication protocols, and prototype functionalities. A proof of concept FIWARE implementation supports ingestion, monitoring and analytical services for piezometric and meteorological data, exemplified through groundwater dynamics in the Beauce aquifer. It integrates daily observations from 53 piezometric stations over more than five years, managing approximately 2.8 million records in a containerized environment.
Results show that the proposed DT architecture can enhance sustainability-oriented decision making, integrating heterogeneous data and predictive analyses while enabling collaboration across scientific and technical domains. Its modular design offers a replicable template for future AI-assisted environmental DTs, scalable to larger regions. Hence, this work illustrates how DTs can improve environmental monitoring and understanding, providing a pathway toward resilient, data-driven management of natural resources.
{"title":"A conceptual architecture for AI-assisted Digital Twins in natural resource management","authors":"Félix Iglesias , Frédéric Ros , Lynh Hoang Vy Thuy , Laurence Gourcy , Jean-Sébastien Moquet , Véronique Daële , Sébastien Dupraz","doi":"10.1016/j.ecoinf.2026.103635","DOIUrl":"10.1016/j.ecoinf.2026.103635","url":null,"abstract":"<div><div>The management of natural resources is increasingly critical and challenging due to complex interactions among environmental, industrial, and societal processes. Traditional approaches often fail to integrate heterogeneous data, limiting predictive and decision-support capabilities. This study presents a conceptual architecture for an Artificial Intelligence (AI)-assisted Digital Twin (DT) of the Centre-Val de Loire region, designed to unify time-dependent multi-source data. Based on the ENVRI Reference Model, it covers Science, Information, Computational, Engineering, and Technology layers, defining standardized data exchange, communication protocols, and prototype functionalities. A proof of concept FIWARE implementation supports ingestion, monitoring and analytical services for piezometric and meteorological data, exemplified through groundwater dynamics in the Beauce aquifer. It integrates daily observations from 53 piezometric stations over more than five years, managing approximately 2.8 million records in a containerized environment.</div><div>Results show that the proposed DT architecture can enhance sustainability-oriented decision making, integrating heterogeneous data and predictive analyses while enabling collaboration across scientific and technical domains. Its modular design offers a replicable template for future AI-assisted environmental DTs, scalable to larger regions. Hence, this work illustrates how DTs can improve environmental monitoring and understanding, providing a pathway toward resilient, data-driven management of natural resources.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103635"},"PeriodicalIF":7.3,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080074","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-25DOI: 10.1016/j.ecoinf.2026.103625
Qiqi Ding , Haojun Xi , Pinjian Li , Hongzhe Fang , Yibin Yuan , Tianhong Li
Rapid urbanisation intensifies multiple pollution pressures on river ecosystems, generating heterogeneous and nonlinear spatiotemporal water quality dynamics that challenge conventional evaluation methods. Although machine-learning (ML) models are increasingly used for water-quality assessment and prediction, most applications remain evaluation-centric and rarely link their outputs to receptor-based source apportionment or explicit spatial zoning, which limits interpretability and management relevance. Here, we proposed an integrated framework that couples ML-assisted water quality index (WQI) optimisation with receptor modelling and unsupervised spatial clustering. Using monthly observations of ten water quality indicators from 19 sites in the Jinma River Basin (2018–2022), we trained an eXtreme Gradient Boosting (XGBoost) model to derive optimised weights of water quality indicators for WQI aggregation. We then applied positive matrix factorisation (PMF) to resolve latent source factors and quantify their contributions, and used self-organising maps (SOM) to cluster monitoring sites into spatially coherent zones based on both WQI status and source composition. Four dominant contributors to the basin water pollution were identified: seasonal hydrological influences (28.16%), domestic sewage (27.36%), agricultural runoff (27.30%) and industrial emissions (17.18%). Integrating the XGBoost-optimised WQI, PMF-resolved source contributions, and SOM clusters delineated three functional management zones, specifically, forested headwaters with high WQI and minimal anthropogenic influence, midstream transition reaches dominated by nutrient-enriched agricultural runoff, and urban downstream corridors affected by combined industrial and domestic inputs. This modular, code-driven workflow translates routine multi-indicator monitoring data into management outputs and can be retrained for other river basins facing complex pollution regimes.
{"title":"Linking water quality assessment to source apportionment with machine learning-assisted WQI, PMF, and SOM: A case study of the Jinma River basin","authors":"Qiqi Ding , Haojun Xi , Pinjian Li , Hongzhe Fang , Yibin Yuan , Tianhong Li","doi":"10.1016/j.ecoinf.2026.103625","DOIUrl":"10.1016/j.ecoinf.2026.103625","url":null,"abstract":"<div><div>Rapid urbanisation intensifies multiple pollution pressures on river ecosystems, generating heterogeneous and nonlinear spatiotemporal water quality dynamics that challenge conventional evaluation methods. Although machine-learning (ML) models are increasingly used for water-quality assessment and prediction, most applications remain evaluation-centric and rarely link their outputs to receptor-based source apportionment or explicit spatial zoning, which limits interpretability and management relevance. Here, we proposed an integrated framework that couples ML-assisted water quality index (WQI) optimisation with receptor modelling and unsupervised spatial clustering. Using monthly observations of ten water quality indicators from 19 sites in the Jinma River Basin (2018–2022), we trained an eXtreme Gradient Boosting (XGBoost) model to derive optimised weights of water quality indicators for WQI aggregation. We then applied positive matrix factorisation (PMF) to resolve latent source factors and quantify their contributions, and used self-organising maps (SOM) to cluster monitoring sites into spatially coherent zones based on both WQI status and source composition. Four dominant contributors to the basin water pollution were identified: seasonal hydrological influences (28.16%), domestic sewage (27.36%), agricultural runoff (27.30%) and industrial emissions (17.18%). Integrating the XGBoost-optimised WQI, PMF-resolved source contributions, and SOM clusters delineated three functional management zones, specifically, forested headwaters with high WQI and minimal anthropogenic influence, midstream transition reaches dominated by nutrient-enriched agricultural runoff, and urban downstream corridors affected by combined industrial and domestic inputs. This modular, code-driven workflow translates routine multi-indicator monitoring data into management outputs and can be retrained for other river basins facing complex pollution regimes.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103625"},"PeriodicalIF":7.3,"publicationDate":"2026-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080075","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Planning resilient plant communities for ecological restoration under climate change requires tools that integrate functional trait data with explicit climatic constraints. This study presents a multi-objective optimisation framework that identifies species assemblages balancing hydraulic safety (drought resistance) with functional diversity. We apply this approach to a Mediterranean forest system using three key traits, xylem vulnerability (P50), specific leaf area (SLA), and leaf dry matter content (LDMC), to represent species' physiological performance and resource-use strategies. Climatic filtering is included by deriving community-weighted P50 targets from the Standardised Precipitation Evapotranspiration Index (SPEI), classified into drought categories. We report two representative scenarios—Near Normal (–0.99 < SPEI<0.99) and Extra Dry (SPEI<-2.0)—thereby aligning species selection with scenario-specific drought conditions. Functional diversity is quantified using Rao's quadratic entropy, which captures trait dissimilarity across communities. Using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), the model generates Pareto fronts describing the trade-offs between hydraulic alignment and functional divergence. Across climatic scenarios, the increasing drought severity progressively constrains the solution space and promotes the selection of moderately drought-tolerant and functionally distinct species, shifting community-weighted P50 towards more negative values. In the Near Normal scenario (target P50 ≈ −2.0 MPa), the Pareto front spans P50 ≈ −4.0 to −2.0 MPa and Rao's Q ≈ 0.36–5.3. In contrast, in the Extra Dry scenario (target P50 ≈ −3.8 MPa), P50 narrows to ≈ −4.3 to −3.8 MPa while diversity remains high (Rao's Q ≈ 5.0–5.4). Kernel density estimation and pairwise overlap analyses across 500 optimisation runs demonstrate a strong convergence, particularly under extreme drought (in the Extra Dry scenario, 80.5% of solutions fall within the top 5% kernel-density region). Compositional similarity to field communities, measured using Bray-Curtis' dissimilarity, corroborates this pattern, with a lower median dissimilarity under Extra Dry than Near Normal (median BC = 0.365 vs 0.478). This framework provides a robust, flexible, and scalable method for trait-based restoration planning. By explicitly modelling trade-offs and uncertainty, it enhances the ecological relevance and reproducibility of species selection under future climate scenarios, offering practical support for data-informed restoration strategies.
{"title":"Functional trait-based multi-objective optimisation of plant communities for ecological restoration under climate change","authors":"Kristina Micalizzi, Danilo Lombardi, Giulia Bardino, Marcello Vitale","doi":"10.1016/j.ecoinf.2026.103623","DOIUrl":"10.1016/j.ecoinf.2026.103623","url":null,"abstract":"<div><div>Planning resilient plant communities for ecological restoration under climate change requires tools that integrate functional trait data with explicit climatic constraints. This study presents a multi-objective optimisation framework that identifies species assemblages balancing hydraulic safety (drought resistance) with functional diversity. We apply this approach to a Mediterranean forest system using three key traits, xylem vulnerability (P50), specific leaf area (SLA), and leaf dry matter content (LDMC), to represent species' physiological performance and resource-use strategies. Climatic filtering is included by deriving community-weighted P50 targets from the Standardised Precipitation Evapotranspiration Index (SPEI), classified into drought categories. We report two representative scenarios—Near Normal (–0.99 < SPEI<0.99) and Extra Dry (SPEI<-2.0)—thereby aligning species selection with scenario-specific drought conditions. Functional diversity is quantified using Rao's quadratic entropy, which captures trait dissimilarity across communities. Using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), the model generates Pareto fronts describing the trade-offs between hydraulic alignment and functional divergence. Across climatic scenarios, the increasing drought severity progressively constrains the solution space and promotes the selection of moderately drought-tolerant and functionally distinct species, shifting community-weighted P50 towards more negative values. In the Near Normal scenario (target P50 ≈ −2.0 MPa), the Pareto front spans P50 ≈ −4.0 to −2.0 MPa and Rao's Q ≈ 0.36–5.3. In contrast, in the Extra Dry scenario (target P50 ≈ −3.8 MPa), P50 narrows to ≈ −4.3 to −3.8 MPa while diversity remains high (Rao's Q ≈ 5.0–5.4). Kernel density estimation and pairwise overlap analyses across 500 optimisation runs demonstrate a strong convergence, particularly under extreme drought (in the Extra Dry scenario, 80.5% of solutions fall within the top 5% kernel-density region). Compositional similarity to field communities, measured using Bray-Curtis' dissimilarity, corroborates this pattern, with a lower median dissimilarity under Extra Dry than Near Normal (median BC = 0.365 vs 0.478). This framework provides a robust, flexible, and scalable method for trait-based restoration planning. By explicitly modelling trade-offs and uncertainty, it enhances the ecological relevance and reproducibility of species selection under future climate scenarios, offering practical support for data-informed restoration strategies.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103623"},"PeriodicalIF":7.3,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.ecoinf.2026.103610
Javier Lopatin , Rocío Araya-López , Iryna Dronova
Plant phenology is often used as an indicator of ecological processes and responses to changing environmental conditions. Remote sensing enables phenological monitoring across space and time, yet separating vegetation composition or environmental drivers remains challenging in heterogeneous tidal marshes. We analyzed Sentinel-2 EVI time series to derive phenological metrics, grouped pixels into phenological types via clustering, and linked these to vegetation composition and environmental variation in Suisun Marsh, California. Using phenology metrics alone, a PLS-DA classifier achieved an overall accuracy of 0.69 (per-class balanced accuracy of 0.50–0.81), demonstrating that phenology captures meaningful community patterns. However, transition zones exhibited a complex interplay among vegetation, phenology, elevation, and hydrology: mean mixing rates ranged from 1 to 45%, with class-specific error structures (sensitivity = 0–0.80), indicating limited separability where inundation and salinity covary with phenology. The variation in the timing and magnitude of greenness, alongside the differing proportions of vegetation types across phenological types, suggests that these interacting drivers jointly shape seasonal vegetation cycles. Core phenology metrics (start, peak, end of season) effectively distinguished wetland communities with similar aboveground function and aided delineation of wetland–upland transitions. Yet, despite ecological differences, several vegetation types expressed similar phenological behavior, likely due to shared hydrologic and microclimatic regimes and, potentially, spectral mixing at moderate spatial resolution. We provide a comprehensive work that combines management and vegetation classes to disentangle the complex interplay between wetland communities and remotely sensed phenology predictions.
{"title":"Remotely sensed phenology reveals environmental and management controls on coastal wetland plant communities","authors":"Javier Lopatin , Rocío Araya-López , Iryna Dronova","doi":"10.1016/j.ecoinf.2026.103610","DOIUrl":"10.1016/j.ecoinf.2026.103610","url":null,"abstract":"<div><div>Plant phenology is often used as an indicator of ecological processes and responses to changing environmental conditions. Remote sensing enables phenological monitoring across space and time, yet separating vegetation composition or environmental drivers remains challenging in heterogeneous tidal marshes. We analyzed Sentinel-2 EVI time series to derive phenological metrics, grouped pixels into phenological types via clustering, and linked these to vegetation composition and environmental variation in Suisun Marsh, California. Using phenology metrics alone, a PLS-DA classifier achieved an overall accuracy of 0.69 (per-class balanced accuracy of 0.50–0.81), demonstrating that phenology captures meaningful community patterns. However, transition zones exhibited a complex interplay among vegetation, phenology, elevation, and hydrology: mean mixing rates ranged from 1 to 45%, with class-specific error structures (sensitivity = 0–0.80), indicating limited separability where inundation and salinity covary with phenology. The variation in the timing and magnitude of greenness, alongside the differing proportions of vegetation types across phenological types, suggests that these interacting drivers jointly shape seasonal vegetation cycles. Core phenology metrics (start, peak, end of season) effectively distinguished wetland communities with similar aboveground function and aided delineation of wetland–upland transitions. Yet, despite ecological differences, several vegetation types expressed similar phenological behavior, likely due to shared hydrologic and microclimatic regimes and, potentially, spectral mixing at moderate spatial resolution. We provide a comprehensive work that combines management and vegetation classes to disentangle the complex interplay between wetland communities and remotely sensed phenology predictions.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103610"},"PeriodicalIF":7.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080647","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-22DOI: 10.1016/j.ecoinf.2026.103628
Ying Liu , Sihan Wang , Zhaohua Liu , Dongmei Lyu , Sijia Li , Bingxue Zhu , Ge Liu , Kaishan Song
Orchard plantations play a crucial role in the rural economy of southern China, making accurate orchard surveys essential for effective management and resource allocation. Owing to the distinct seasonal growth patterns of orchards, extracting phenological features from multi-temporal remote sensing data has become a primary approach for obtaining orchard information. However, the subtropical monsoon climate of southern China brings frequent cloud cover and rainfall. This poses major challenges to constructing continuous, high-resolution optical remote sensing datasets. To overcome these limitations, this study integrates high-temporal-resolution MODIS data with medium-spatial-resolution Landsat imagery to generate monthly composite images that capture key stages of orchard growth. Based on more than 9000 sampling sites across the province, phenological information was extracted from three conventional features, including spectral reflectance, vegetation indices, and texture features, to build multiple machine learning classification models for high-precision orchard mapping. The results demonstrate that the proposed multi-feature fusion framework yields a substantial accuracy gain of up to 17 percentage points compared to traditional methods. While the baseline method relying solely on single-phase spectral features achieved 72.2% accuracy, the optimal combination of spectral, texture, and phenological features using the LightGBM model reached an accuracy of 89.2% (F1-score: 88.6%).Furthermore, SHAP analysis enhanced model interpretability by revealing the key factors influencing the decision-making process. The results indicate that orchards in Hunan Province are primarily distributed in hilly regions, where large- and small-scale orchards coexist, with Huaihua and Yongzhou containing the largest orchard areas. From 1995 to 2022, the province's orchard area expanded significantly, growing from approximately 45,000 ha to nearly 140,000 ha, which represents an increase of more than 200%. This study demonstrates the effectiveness of spatiotemporal data fusion in mitigating cloud-related challenges in subtropical regions and underscores the novel role of texture features in capturing key phenological information. It provides a reliable framework for large-scale orchard mapping and supports protective land utilization strategies and sustainable agricultural development in the region.
{"title":"Orchard plantation mapping using remote sensing phenological feature fusion and interpretable ML algorithms in Hunan Province, China","authors":"Ying Liu , Sihan Wang , Zhaohua Liu , Dongmei Lyu , Sijia Li , Bingxue Zhu , Ge Liu , Kaishan Song","doi":"10.1016/j.ecoinf.2026.103628","DOIUrl":"10.1016/j.ecoinf.2026.103628","url":null,"abstract":"<div><div>Orchard plantations play a crucial role in the rural economy of southern China, making accurate orchard surveys essential for effective management and resource allocation. Owing to the distinct seasonal growth patterns of orchards, extracting phenological features from multi-temporal remote sensing data has become a primary approach for obtaining orchard information. However, the subtropical monsoon climate of southern China brings frequent cloud cover and rainfall. This poses major challenges to constructing continuous, high-resolution optical remote sensing datasets. To overcome these limitations, this study integrates high-temporal-resolution MODIS data with medium-spatial-resolution Landsat imagery to generate monthly composite images that capture key stages of orchard growth. Based on more than 9000 sampling sites across the province, phenological information was extracted from three conventional features, including spectral reflectance, vegetation indices, and texture features, to build multiple machine learning classification models for high-precision orchard mapping. The results demonstrate that the proposed multi-feature fusion framework yields a substantial accuracy gain of up to 17 percentage points compared to traditional methods. While the baseline method relying solely on single-phase spectral features achieved 72.2% accuracy, the optimal combination of spectral, texture, and phenological features using the LightGBM model reached an accuracy of 89.2% (F1-score: 88.6%).Furthermore, SHAP analysis enhanced model interpretability by revealing the key factors influencing the decision-making process. The results indicate that orchards in Hunan Province are primarily distributed in hilly regions, where large- and small-scale orchards coexist, with Huaihua and Yongzhou containing the largest orchard areas. From 1995 to 2022, the province's orchard area expanded significantly, growing from approximately 45,000 ha to nearly 140,000 ha, which represents an increase of more than 200%. This study demonstrates the effectiveness of spatiotemporal data fusion in mitigating cloud-related challenges in subtropical regions and underscores the novel role of texture features in capturing key phenological information. It provides a reliable framework for large-scale orchard mapping and supports protective land utilization strategies and sustainable agricultural development in the region.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103628"},"PeriodicalIF":7.3,"publicationDate":"2026-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080649","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-21DOI: 10.1016/j.ecoinf.2026.103620
Badiu Badams , Usman Ullah Sheikh , Norhaliza Abdul Wahab , Syed A.R. Abu Bakar , Muhammad I. Masud , Mohammed Khouj , Urooj Waheed , Zeeshan Ahmad Arfeen , Kam Meng Goh
Floating marine debris such as plastics, cans, and discarded packaging has become one of the most persistent threats to aquatic ecosystems and coastal sustainability. Detecting and tracking this debris in real time is vital for protecting biodiversity and guiding cleanup and policy actions. In this study, we introduce A2ANet, a lightweight deep learning framework that combines multi-scale atrous convolutions and Enhanced Channel Attention (ECA) to detect small, submerged, and visually ambiguous debris under challenging aquatic conditions. These mechanisms-expanding the receptive field and highlighting salient cues-reduce errors from reflections, glare, and clutter in aquatic scenes.
A2ANet was evaluated on two datasets: a newly developed six-class dataset (D_six) representing real-world river conditions, and the publicly available FloW-Img benchmark. The model achieved [email protected] of 0.841 and 0.892 on D_six and FloW-Img datasets, respectively, with inference time as low as 39 ms/image. Beyond detection performance, by enabling automated and frequent monitoring, A2ANet provides actionable insights for mapping pollution, tracking trends, and supporting ecosystem management. The framework offers a practical pathway toward intelligent aquatic observation systems aligned with Sustainable Development Goal 14: Life Below Water.
All code and datasets are openly available (see Data Availability).
{"title":"A2ANet: Real-time detection of floating marine debris using atrous convolution and channel attention","authors":"Badiu Badams , Usman Ullah Sheikh , Norhaliza Abdul Wahab , Syed A.R. Abu Bakar , Muhammad I. Masud , Mohammed Khouj , Urooj Waheed , Zeeshan Ahmad Arfeen , Kam Meng Goh","doi":"10.1016/j.ecoinf.2026.103620","DOIUrl":"10.1016/j.ecoinf.2026.103620","url":null,"abstract":"<div><div>Floating marine debris such as plastics, cans, and discarded packaging has become one of the most persistent threats to aquatic ecosystems and coastal sustainability. Detecting and tracking this debris in real time is vital for protecting biodiversity and guiding cleanup and policy actions. In this study, we introduce A2ANet, a lightweight deep learning framework that combines multi-scale atrous convolutions and Enhanced Channel Attention (ECA) to detect small, submerged, and visually ambiguous debris under challenging aquatic conditions. These mechanisms-expanding the receptive field and highlighting salient cues-reduce errors from reflections, glare, and clutter in aquatic scenes.</div><div>A2ANet was evaluated on two datasets: a newly developed six-class dataset (D_six) representing real-world river conditions, and the publicly available FloW-Img benchmark. The model achieved [email protected] of 0.841 and 0.892 on D_six and FloW-Img datasets, respectively, with inference time as low as 39 ms/image. Beyond detection performance, by enabling automated and frequent monitoring, A2ANet provides actionable insights for mapping pollution, tracking trends, and supporting ecosystem management. The framework offers a practical pathway toward intelligent aquatic observation systems aligned with Sustainable Development Goal 14: Life Below Water.</div><div>All code and datasets are openly available (see Data Availability).</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103620"},"PeriodicalIF":7.3,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025988","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-21DOI: 10.1016/j.ecoinf.2026.103626
R. Bentivogli , L. Pezzolesi , N. Caputo , B. Casarotto , S. Silvestri
Monitoring phytoplankton communities is essential for assessing ecosystem health and detecting harmful algal blooms (HABs). Hyperspectral imaging has emerged as a promising tool to discriminate among microalgal species based on their unique reflectance signatures. This study presents a laboratory spectral analysis of five phytoplankton species, including bloom-forming and toxin-producing taxa common in coastal waters. Reflectance spectra were measured at multiple cell concentrations and analyzed using two normalization approaches, second- and fourth-derivative transformations, and dimensionality reduction techniques including principal component analysis (PCA) and linear discriminant analysis (LDA).
Our results demonstrate that specific spectral features, particularly in the 470–500 nm and 620–680 nm ranges, enable species-level discrimination. PCA and LDA effectively enhanced separability by reducing spectral redundancy and emphasizing class features. We further applied linear spectral unmixing (LSU) to estimate fractional species abundances in synthetic mixtures. LSU performed well in simple mixtures but revealed limitations in complex communities, where nonlinear effects and spectral similarity reduced accuracy.
Beyond classification, LSU enables quantitative assessment of species contributions, providing a valuable complement to PCA and LDA for ecological interpretation and bloom dynamics investigation. This integrated approach lays the foundation for future development of operational tools that combine spectral unmixing and machine learning for automated HAB detection. The combined use of hyperspectral reflectance data and computational methods supports scalable, real-time monitoring of phytoplankton diversity and abundance, with strong potential for deployment in early-warning systems and coastal observatories.
{"title":"Laboratory-based hyperspectral reflectance analysis for phytoplankton species identification","authors":"R. Bentivogli , L. Pezzolesi , N. Caputo , B. Casarotto , S. Silvestri","doi":"10.1016/j.ecoinf.2026.103626","DOIUrl":"10.1016/j.ecoinf.2026.103626","url":null,"abstract":"<div><div>Monitoring phytoplankton communities is essential for assessing ecosystem health and detecting harmful algal blooms (HABs). Hyperspectral imaging has emerged as a promising tool to discriminate among microalgal species based on their unique reflectance signatures. This study presents a laboratory spectral analysis of five phytoplankton species, including bloom-forming and toxin-producing taxa common in coastal waters. Reflectance spectra were measured at multiple cell concentrations and analyzed using two normalization approaches, second- and fourth-derivative transformations, and dimensionality reduction techniques including principal component analysis (PCA) and linear discriminant analysis (LDA).</div><div>Our results demonstrate that specific spectral features, particularly in the 470–500 nm and 620–680 nm ranges, enable species-level discrimination. PCA and LDA effectively enhanced separability by reducing spectral redundancy and emphasizing class features. We further applied linear spectral unmixing (LSU) to estimate fractional species abundances in synthetic mixtures. LSU performed well in simple mixtures but revealed limitations in complex communities, where nonlinear effects and spectral similarity reduced accuracy.</div><div>Beyond classification, LSU enables quantitative assessment of species contributions, providing a valuable complement to PCA and LDA for ecological interpretation and bloom dynamics investigation. This integrated approach lays the foundation for future development of operational tools that combine spectral unmixing and machine learning for automated HAB detection. The combined use of hyperspectral reflectance data and computational methods supports scalable, real-time monitoring of phytoplankton diversity and abundance, with strong potential for deployment in early-warning systems and coastal observatories.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103626"},"PeriodicalIF":7.3,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025990","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-21DOI: 10.1016/j.ecoinf.2026.103617
Luisa S.R. Nogueira , Mariana A.S. de Carvalho , Berilo de O. Santos , Roland Yonaba , Apoorva Bamal , Md Galal Uddin , Matteo Bodini , Leonardo Goliatt
Accurate prediction of river water quality is fundamental to environmental sustainability and public health, particularly amid increasing freshwater scarcity. This study develops a robust Machine Learning (ML) framework to forecast the River Pollution Index (RPI) using a comprehensive 36-year national dataset from Taiwan’s Environmental Protection Administration, covering over 500 monitoring stations. We conducted a systematic comparison of ensemble methods (CatBoost, XGBoost, NGBoost) and non-ensemble benchmarks (SVM, ElasticNet, and 1D CNN). Hyperparameters were optimized via Bayesian optimization, and statistical significance was ensured by evaluating model stability using a suite of complementary indicators (RMSE, MAE, R, A10 index) across 30 independent experimental runs. The results demonstrated the consistent superiority of ensemble models over non-ensemble counterparts. Among them, CatBoost achieved the highest accuracy and stability (RMSE 0.85, MAE 0.61, R = 0.78), reducing prediction error by approximately 20% relative to SVM and ElasticNet. These findings highlight the capacity of ensemble learning techniques to capture complex, non-linear interactions inherent in water quality data. The study makes two principal contributions: (1) the systematic implementation, optimization, and comparison of ensemble and non-ensemble ML models for river pollution prediction on a long-term national dataset; and (2) the identification of ensemble-based methods, particularly CatBoost, as robust and data-driven tools to enhance RPI forecasting and to support informed decision-making in sustainable water resource management.
{"title":"A comparative study of ensemble and non-ensemble machine learning methods for predicting river pollution index","authors":"Luisa S.R. Nogueira , Mariana A.S. de Carvalho , Berilo de O. Santos , Roland Yonaba , Apoorva Bamal , Md Galal Uddin , Matteo Bodini , Leonardo Goliatt","doi":"10.1016/j.ecoinf.2026.103617","DOIUrl":"10.1016/j.ecoinf.2026.103617","url":null,"abstract":"<div><div>Accurate prediction of river water quality is fundamental to environmental sustainability and public health, particularly amid increasing freshwater scarcity. This study develops a robust Machine Learning (ML) framework to forecast the River Pollution Index (RPI) using a comprehensive 36-year national dataset from Taiwan’s Environmental Protection Administration, covering over 500 monitoring stations. We conducted a systematic comparison of ensemble methods (CatBoost, XGBoost, NGBoost) and non-ensemble benchmarks (SVM, ElasticNet, and 1D CNN). Hyperparameters were optimized via Bayesian optimization, and statistical significance was ensured by evaluating model stability using a suite of complementary indicators (RMSE, MAE, R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>, A10 index) across 30 independent experimental runs. The results demonstrated the consistent superiority of ensemble models over non-ensemble counterparts. Among them, CatBoost achieved the highest accuracy and stability (RMSE <span><math><mo>≈</mo></math></span> 0.85, MAE <span><math><mo>≈</mo></math></span> 0.61, R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span> = 0.78), reducing prediction error by approximately 20% relative to SVM and ElasticNet. These findings highlight the capacity of ensemble learning techniques to capture complex, non-linear interactions inherent in water quality data. The study makes two principal contributions: (1) the systematic implementation, optimization, and comparison of ensemble and non-ensemble ML models for river pollution prediction on a long-term national dataset; and (2) the identification of ensemble-based methods, particularly CatBoost, as robust and data-driven tools to enhance RPI forecasting and to support informed decision-making in sustainable water resource management.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103617"},"PeriodicalIF":7.3,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025989","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-17DOI: 10.1016/j.ecoinf.2026.103612
Adrian Pascual , Juan Guerra-Hernández , Brigite Botequim , Eduardo González-Ferreiro
The next-generation of fire behavior models must integrate 3D forest structural metrics to better explain fire spread, risk and severity. Canopy base height (CBH) and canopy bulk density (CBD) can be calibrated using lidar data collocated over field plots. Where no airborne lidar scanning data (ALS) exist, GEDI spaceborne lidar can provide 25-m predictions of CBH and CBD contingent to ALS-calibrated workflows. Our research presents GEDI footprint-level estimates of CBH and CBD for Mediterranean forests and builds upon collocated ALS-GEDI crossovers. Our accuracies are based on Random Forests classification: the R2 values for CBH and CBD were < 0.4 for all plant functional types evaluated. The classification of vertical continuity was satisfactory to inform on fire-prone conditions at the GEDI footprint level. Predictions for CBD were aggregated to produce a regional baseline maps, one at 1-km resolution. The usability of these coarse-scale aggregations of fuel estimates is limited because of resolution, presence of gaps and high heterogeneity of forest fuels within small steps. To inform about this heterogeneity and change estimates over time, we predict CBH and CBD over adjacent GEDI tracks collected 5-years apart (2019/24). These change estimates are relevant to show the high variability of the forest fuels that compromises the ability to depict change adding the issue of exact collocation between GEDI measurements that are adjacent, not collocated and therefore not repeated. We discuss methodological differences between our approach and recent studies on mapping fuel baselines and their approach to inform on dynamics.
{"title":"Supporting fire behavior modelling with canopy base height and canopy bulk density estimates using airborne and spaceborne lidar","authors":"Adrian Pascual , Juan Guerra-Hernández , Brigite Botequim , Eduardo González-Ferreiro","doi":"10.1016/j.ecoinf.2026.103612","DOIUrl":"10.1016/j.ecoinf.2026.103612","url":null,"abstract":"<div><div>The next-generation of fire behavior models must integrate 3D forest structural metrics to better explain fire spread, risk and severity. Canopy base height (CBH) and canopy bulk density (CBD) can be calibrated using lidar data collocated over field plots. Where no airborne lidar scanning data (ALS) exist, GEDI spaceborne lidar can provide 25-m predictions of CBH and CBD contingent to ALS-calibrated workflows. Our research presents GEDI footprint-level estimates of CBH and CBD for Mediterranean forests and builds upon collocated ALS-GEDI crossovers. Our accuracies are based on Random Forests classification: the R2 values for CBH and CBD were < 0.4 for all plant functional types evaluated. The classification of vertical continuity was satisfactory to inform on fire-prone conditions at the GEDI footprint level. Predictions for CBD were aggregated to produce a regional baseline maps, one at 1-km resolution. The usability of these coarse-scale aggregations of fuel estimates is limited because of resolution, presence of gaps and high heterogeneity of forest fuels within small steps. To inform about this heterogeneity and change estimates over time, we predict CBH and CBD over adjacent GEDI tracks collected 5-years apart (2019/24). These change estimates are relevant to show the high variability of the forest fuels that compromises the ability to depict change adding the issue of exact collocation between GEDI measurements that are adjacent, not collocated and therefore not repeated. We discuss methodological differences between our approach and recent studies on mapping fuel baselines and their approach to inform on dynamics.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103612"},"PeriodicalIF":7.3,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025991","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-17DOI: 10.1016/j.ecoinf.2026.103613
Georgia K. Dwyer, Galen Holt, Rebecca E. Lester
Decision making in natural resource management is increasingly challenged by the complexity, uncertainty and volume of information available. This leads to information overload and reduces decision quality. Here, we aim to reduce the degree of redundant information in a highly complex model used to manage environmental outcomes in the Murray-Darling Basin, Australia's largest river system. We identified extremely high levels of redundancy with 6–8 hydrological indicators able to explain 98–100% of the variation in the full suite of >2500. Many sets of 6–8 had similar explanatory power, providing flexibility for managers to tailor the subset used, and sets were produced for individual management targets (e.g. native fish, waterbirds) and for individual catchments. These subsets maintained the ability to distinguish between scenarios of plausible future climates and different policy settings, while traditional aggregation techniques did not (particularly for policy settings). Aggregations using key flow categories (e.g. overbank, low flows) were also largely able to distinguish among scenarios. This study is novel in its use of decision science to reduce information overload associated with a commonly used ecological model, illustrating the value of explicitly considering whether complexity in such a model is necessary. Such assessments are rarely undertaken. We illustrated how subsets of complex models provide simple, parsimonious approaches for credible, salient and legitimate decision making under uncertainty, and increase the likelihood of good evidence-based decision making in water management.
{"title":"Effective indicators to enable robust decision making in the Murray-Darling Basin","authors":"Georgia K. Dwyer, Galen Holt, Rebecca E. Lester","doi":"10.1016/j.ecoinf.2026.103613","DOIUrl":"10.1016/j.ecoinf.2026.103613","url":null,"abstract":"<div><div>Decision making in natural resource management is increasingly challenged by the complexity, uncertainty and volume of information available. This leads to information overload and reduces decision quality. Here, we aim to reduce the degree of redundant information in a highly complex model used to manage environmental outcomes in the Murray-Darling Basin, Australia's largest river system. We identified extremely high levels of redundancy with 6–8 hydrological indicators able to explain 98–100% of the variation in the full suite of >2500. Many sets of 6–8 had similar explanatory power, providing flexibility for managers to tailor the subset used, and sets were produced for individual management targets (e.g. native fish, waterbirds) and for individual catchments. These subsets maintained the ability to distinguish between scenarios of plausible future climates and different policy settings, while traditional aggregation techniques did not (particularly for policy settings). Aggregations using key flow categories (e.g. overbank, low flows) were also largely able to distinguish among scenarios. This study is novel in its use of decision science to reduce information overload associated with a commonly used ecological model, illustrating the value of explicitly considering whether complexity in such a model is necessary. Such assessments are rarely undertaken. We illustrated how subsets of complex models provide simple, parsimonious approaches for credible, salient and legitimate decision making under uncertainty, and increase the likelihood of good evidence-based decision making in water management.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103613"},"PeriodicalIF":7.3,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080650","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号