Ecological Informatics最新文献

{"title":"Copula–information gain-based identification of GPP response thresholds under multiscale agricultural drought","authors":"Tianlong She,&nbsp;Chen Xu,&nbsp;Quanwang Chen,&nbsp;Yanan Wang,&nbsp;Yuanyuan Hong,&nbsp;Yechao Sun,&nbsp;Qiang Wang","doi":"10.1016/j.ecoinf.2026.103654","DOIUrl":"10.1016/j.ecoinf.2026.103654","url":null,"abstract":"<div><div>Scientifically determining the threshold at which gross primary productivity (GPP) enters a significantly drought-affected state is a critical prerequisite for effective drought risk mitigation and agricultural management. Previous studies have predominantly relied on empirical criteria or probability-based analyses to determine this threshold, while objective, information-driven frameworks to characterize nonlinear drought–ecosystem response thresholds across multiple time scales are lacking. To address this limitation, we propose a Copula–Information Gain (Copula–IG) framework that integrates copula-based joint dependence modeling with information-theoretic discrimination, thereby enabling probabilistic and information-driven identification of GPP drought thresholds across multiple temporal scales. The results indicate pronounced spatiotemporal heterogeneity and strong scale dependence in agricultural drought responses across China. As the drought duration extends from short-term to medium- and long-term periods, the discriminative performance of the Copula–IG model consistently improves, while the GPP drought threshold gradually shifts from a spatially dispersed pattern toward greater convergence and stability. IG values were primarily concentrated within the range of 0.05–0.10, indicating enhanced separation between drought-affected and non-affected GPP states. Meanwhile, the mechanism underlying GPP drought threshold formation transitions from a rapid, evapotranspiration-dominated short-term response to a structurally constrained suppression process governed by cumulative soil moisture deficits under prolonged drought conditions. Overall, this study presents a probabilistic, quantitative, and spatially refined framework for identifying agricultural drought–ecosystem response thresholds, thereby providing valuable scientific support for high-resolution agricultural climate monitoring and ecological early-warning systems.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103654"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174103","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}

{"title":"Deep learning models used to detect fish movement over resistivity counters","authors":"Sophie A.M. Elliott ,&nbsp;Keerthan Boraiah ,&nbsp;Chun Kee Tham ,&nbsp;William R.C. Beaumont ,&nbsp;Paul Elsmere ,&nbsp;Luke Scott ,&nbsp;Adrian Fewings","doi":"10.1016/j.ecoinf.2026.103606","DOIUrl":"10.1016/j.ecoinf.2026.103606","url":null,"abstract":"<div><div>Diadromous fish are one of the most threatened groups of fish species, being subject to pressures from freshwater, estuarine and marine environments. Of these fish, Atlantic salmon is the most economically important and increasingly threatened. To assess salmonid (Atlantic salmon and sea trout) stocks, resistivity counters have been widely used. However, verification of data from the counters can be challenging due to miscounts, misidentification and biases in human verification of fish counts.</div><div>We applied deep learning models to identify diadromous fish using continuous electrical resistivity data from resistivity fish counters. Our models were tested on three rivers (Frome, Fowey and Test in the South and South-West of England) and compared with a minimum of one year's manually validated data.</div><div>We detected fish signals from background noise with an F1-score of 99%, large from small fish (≥30 cm) with a precision of 95%, and an increase of &gt;38% small and large fish waveforms. The F1-score for salmonids was 92%, and a significantly greater proportion (&gt;173%) of upstream-moving large salmonids (≥30 cm) were detected compared to manual methods.</div><div>To date, abundance estimates for resistivity counters have only been applied to salmonids because of labour-intensive waveform identification. Using deep learning methods, we quantified salmonids and other diadromous fish with varying accuracies. Our method can be applied to resistivity counters to detect diadromous fish globally, reducing human bias and improving detection accuracy.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103606"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174412","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}

{"title":"Real-time hyperspectral decision support for precision herbicide management in Maize: Optimizing herbicide efficacy and plant response","authors":"Sk Asraful Ali ,&nbsp;Ramanjit Kaur ,&nbsp;Sudhir Kumar ,&nbsp;Allimuthu Elangovan ,&nbsp;Rahul Kumar ,&nbsp;Arjun Shreepad Hegde ,&nbsp;Rashmi Sharma ,&nbsp;Yogeshwar Singh ,&nbsp;Shiv Vendra Singh","doi":"10.1016/j.ecoinf.2026.103642","DOIUrl":"10.1016/j.ecoinf.2026.103642","url":null,"abstract":"<div><div>Hyperspectral VNIR imaging (400–1000 nm) offers a swift, non-invasive method for identifying early herbicide-induced stress in maize and its accompanying weed flora. Unlike traditional visual scoring and biomass measurements, this technology captures subtle changes in pigment content, water status, and canopy structure with high precision and accuracy. The present study employed hyperspectral vegetation indices and multivariate analysis to identify spectral responses to sequential pre- and post-emergence herbicide combinations, measure the dynamics of chlorophyll, carotenoid, and anthocyanin-related pigments under herbicide stress, and differentiate treatment efficacy patterns to enhance precision weed management. Hyperspectral reflectance data collected before and after herbicide application were used to calculate indices, including NDVI, CIgreen, CIred-edge, NPQI, CRI1, CRI2, ARI1, and ARI2. This was shown by a drop in CIred-edge (26.2%) and CIgreen (8.8%), and a considerable increase in NPQI (+68.4%), CRI2 (+63.4%), and ARI1 (+824.1%) within four days of application, indicating that pigments break down quickly and weed species are sensitive. In contrast, combinations based on halosulfuron methyl showed very little spectral divergence and mostly resembled the weedy check because <em>Cyperus</em> spp., the main target of the herbicide, was not present. Principal Component Analysis showed that the first two components explained 78.5% of the total variance (PC1:56.9%; PC2:21.6%), successfully distinguished tembotrione-based combinations from other regimens, whereas hierarchical clustering categorised treatments based on their temporal spectral response patterns. These results show that hyperspectral imaging and multivariate analysis can provide an objective and early indication of herbicide efficacy. This study presents a real-time decision-support framework that improves precision herbicide management, reduces dependence on subjective evaluations, and fosters more sustainable maize production methodologies.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103642"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174424","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}

{"title":"A comparative study of ensemble and non-ensemble machine learning methods for predicting river pollution index","authors":"Luisa S.R. Nogueira ,&nbsp;Mariana A.S. de Carvalho ,&nbsp;Berilo de O. Santos ,&nbsp;Roland Yonaba ,&nbsp;Apoorva Bamal ,&nbsp;Md Galal Uddin ,&nbsp;Matteo Bodini ,&nbsp;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":"2026-03-01","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}

{"title":"Predicting wildfires triggered by human-caused ignition: A spatial framework integrating AI models","authors":"Sujung Heo,&nbsp;Sujung Ahn","doi":"10.1016/j.ecoinf.2026.103640","DOIUrl":"10.1016/j.ecoinf.2026.103640","url":null,"abstract":"<div><div>Human-caused ignitions—including agricultural residue burning, land clearing, and various open-fire practices—are now a major contributor to wildfire occurrence in South Korea, accounting for more than 40% of recent events and concentrating along rapidly expanding wildland–urban interface (WUI) zones. To improve national-scale understanding of these ignition processes, this study develops an integrated wildfire risk assessment framework that combines Random Forest (RF), Generalized Additive Models (GAM), and Geographically Weighted Regression (GWR). Using 3703 georeferenced human-caused wildfire records (2001–2025) and high-resolution climatic, land-use, vegetation, and demographic datasets, we identify key ignition drivers, quantify nonlinear environmental thresholds, and map spatial heterogeneity in human-driven wildfire susceptibility. RF achieved the strongest predictive performance (AUC = 0.829), while GAM revealed sharp increases in ignition probability below 15% relative humidity, around 15 m/s wind speeds, and at intermediate NDVI levels (0.2–0.4). GWR showed substantial regional variability in these effects, particularly within peri-urban landscapes where human activity and fuel continuity intersect. High-risk areas (probability ≥0.70) accounted for 16.96% of the national territory and were concentrated in northern Gyeonggi-do, eastern Gangwon-do, and parts of Chungcheong and Jeolla provinces. By integrating the complementary strengths of RF, GAM, and GWR, this study provides operational ignition thresholds and high-resolution risk maps that support evidence-based land-use planning, targeted burning restrictions, and climate-adaptive fire management. The framework offers a transferable approach for regions facing similar challenges associated with human-caused wildfire ignitions and rapidly evolving socio-ecological landscapes.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103640"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147421595","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}

{"title":"Solar radiation times-series forecasting in southern Brazil: A comprehensive analysis","authors":"Ricardo H.G. Furiati ,&nbsp;Filipe Sacchetto ,&nbsp;Simon Malinowski ,&nbsp;Zenilton Kleber G. do Patrocínio Jr. ,&nbsp;Felipe D. Cunha ,&nbsp;Cristiana B. Maia ,&nbsp;Silvio Jamil F. Guimarães","doi":"10.1016/j.ecoinf.2026.103601","DOIUrl":"10.1016/j.ecoinf.2026.103601","url":null,"abstract":"<div><div>To enable the study of solar behavior without installing expensive sensor equipment, machine learning time-series models can be highly useful. In this study, we forecast future values of solar radiation incident on a horizontal surface by comparing five different models: Holt-Winters, LSTM, SARIMAX, SVM, and XGBoost, using a comprehensive dataset of satellite meteorological observations from NASA spanning over 30 years. 90 points in the Brazilian southeast (in the state of Minas Gerais and its surroundings) were analyzed using two different cross-validation methods (Fixed Start and Rolling Window) and compared using the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) metrics. Our analysis revealed that the Holt-Winters model yielded the lowest error, with an MAE of 0.302 kWh/m<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>/day, followed by the LSTM (0.314), SARIMAX (0.338), SVM (0.39), and XGBoost (0.336) models. The statistical analysis of the cross-validation methods revealed that although the fixed start method yields lower error metrics, it requires substantially longer training times (due to the increased input data) and is only slightly superior to the rolling window method. The most significant divergence between the models and the actual solar radiation values was observed along the eastern border of the state. An exploratory analysis of solar behavior showed that greater data variability (standard deviation and variance) is associated with worse forecasting performance. Given the worldwide availability of the data, the methodology presented in our work can be replicated to make solar radiation predictions anywhere, facilitating new developments in sustainable renewable energy production.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103601"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173893","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}

{"title":"Remote sensing monitoring of cropland abandonment at the parcel level based on time-series fitting of cultivation probability values","authors":"Xinyu Yang ,&nbsp;Yufeng Liu ,&nbsp;Hu Li ,&nbsp;Ji Ma ,&nbsp;Zhicheng Ye","doi":"10.1016/j.ecoinf.2026.103632","DOIUrl":"10.1016/j.ecoinf.2026.103632","url":null,"abstract":"<div><div>Abandoned cropland is a critical component of agricultural land-use change, with notable implications for food production and the ecological sustainability. This study proposes an integrated monitoring framework, termed OBIA–LT, to address the temporal complexity and spatial fragmentation of abandoned cropland. Cultivation probability is constructed as a time series and analyzed through LandTrendr segmentation fitting. The use of land parcels as analysis units suppresses pixel-level noise, enabling precise identification of the timing and dynamics of abandoned cropland. The Jianghuai hilly region was selected as the study area, and multi-temporal remote sensing imagery and field samples were used for analysis. Results indicate that abandoned cropland exhibits pronounced spatial clustering, with high-density concentrations in the central and western hilly areas and a scattered distribution in the southern plains. The timing of abandonment shows seasonal patterns, occurring predominantly in spring and winter. Long-term continuous abandonment is rare, with more than half the parcels abandoned for only a single quarter, demonstrating the sensitivity of the OBIA–LT framework to short-term cultivation gaps at a monthly scale. This study confirms the effectiveness of the method in achieving high accuracy and spatiotemporal consistency and provides a valuable reference for large-scale monitoring of abandoned cropland dynamics.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103632"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173891","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}

{"title":"Functional trait-based multi-objective optimisation of plant communities for ecological restoration under climate change","authors":"Kristina Micalizzi,&nbsp;Danilo Lombardi,&nbsp;Giulia Bardino,&nbsp;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 &lt; SPEI&lt;0.99) and Extra Dry (SPEI&lt;-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":"2026-03-01","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}

{"title":"Evaluating key environmental variables in dust occurrence through artificial intelligence methods","authors":"Mohammad Mahdi Pourhanifeh,&nbsp;Hassan Khosravi,&nbsp;Tayebeh Mesbah Zadeh,&nbsp;Esmaeil Heydari Alamdarloo,&nbsp;Azam Abolhasani","doi":"10.1016/j.ecoinf.2026.103653","DOIUrl":"10.1016/j.ecoinf.2026.103653","url":null,"abstract":"<div><div>Dust storms in arid and semi-arid regions present a critical environmental challenge, arising from complex interactions between climatic patterns, landform features, and surface conditions. The Sistan and Baluchestan Province in southeastern Iran is a global dust hotspot, yet the quantitative impact of its diverse environmental drivers remains poorly understood. This study quantifies these relationships by evaluating the influence of 18 environmental variables on dust occurrence over a 20-year period (2003−2023). To achieve this, we developed a robust framework combining six machine learning algorithms (Logistic Regression, KNN, SVM, MLP, Random Forest, and XGBoost) with a rigorous hybrid feature selection strategy (VIF-RFE). While ensemble models (Random Forest and XGBoost) and KNN demonstrated superior performance (AUC ∼0.89), capturing the non-linear nature of dust generation, the Shapley Additive exPlanations (SHAP) analysis revealed a clear hierarchy of drivers. Topographic features, specifically elevation, were the dominant broad-scale predictor, followed closely by hydro-climatic factors (precipitation and NDMI) and soil chemical properties (Salinity/VSSI). In alignment with the region's hyper-arid characteristics, vegetation indices (e.g., NDVI, MSAVI) showed minimal predictive power. As over 93% of the landscape constitutes naturally barren soil or sparse cover, the vegetation falls below the critical threshold required to act as a significant wind buffer, leaving abiotic drivers (soil moisture and salinity) as the primary determinants of dust generation. These findings highlight that mitigation strategies must prioritize soil moisture preservation and salinity stabilization over isolated revegetation efforts, providing a scientific foundation for targeted management in hyper-arid regions.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103653"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147421594","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}

{"title":"A probabilistic deep learning framework for retrieving chlorophyll-a from hyperspectral imagery: Integrating channel attention and mixture density networks","authors":"Wenzhao Li ,&nbsp;Surendra Maharjan ,&nbsp;Rejoice Thomas ,&nbsp;Junde Chen ,&nbsp;Hesham Morgan ,&nbsp;Michael J. Garay ,&nbsp;Olga V. Kalashnikova ,&nbsp;Shahryar Fazli ,&nbsp;Charles Ichoku ,&nbsp;Hesham El-Askary","doi":"10.1016/j.ecoinf.2026.103677","DOIUrl":"10.1016/j.ecoinf.2026.103677","url":null,"abstract":"<div><div>Accurate monitoring of Chlorophyll-a (Chla) is critical for assessing aquatic ecosystem health, yet ecological complexity often leads to ambiguous spectral signatures in satellite data. Traditional deterministic models assume a one-to-one mapping between spectra and pigments, often failing to capture these high-dimensional analytical challenges. In this study, we propose a novel deep learning architecture, the Channel Attention-Mixture Density Network (CA-MDN), to retrieve Chla from the National Aeronautics and Space Administration (NASA) Earth Surface Mineral Dust Source Investigation (EMIT) hyperspectral mission. The CA-MDN integrates an attention mechanism to dynamically select ecologically relevant spectral bands and employs a probabilistic output layer to quantify retrieval uncertainty. Trained and tested against the global GLObal Reflectance community dataset for Imaging and optical sensing of Aquatic environments (GLORIA) in situ dataset, the model achieved a prediction error of 40.52%, significantly outperforming conventional machine learning baselines. Case studies in California waters demonstrate the model's ecological utility, showing how probabilistic modeling can resolve fine-scale variability and flag high-uncertainty regions. This study presents a robust computational framework for leveraging spaceborne imaging spectroscopy in complex coastal and inland environments.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"94 ","pages":"Article 103677"},"PeriodicalIF":7.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147421598","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}