International journal of applied earth observation and geoinformation : ITC journal最新文献

{"title":"A transferable framework for roof vegetation mapping across tropical cities based on open-source multimodal data and morphology-aware UDASS","authors":"Jinhui Liu,&nbsp;Mi Zhang,&nbsp;Lina Tang,&nbsp;Guofan Shao,&nbsp;Huiwei Jiang,&nbsp;Xiangyun Hu,&nbsp;Yibing Xiong,&nbsp;Jiongwei Wang","doi":"10.1016/j.jag.2026.105490","DOIUrl":"10.1016/j.jag.2026.105490","url":null,"abstract":"<div><div>Green roofs play a vital role in urban sustainability, offering substantial environmental, ecological, and social benefits. However, automated roof vegetation extraction mostly relies on costly high-resolution data, with limited investigations in Global South cities. Existing methods insufficiently fuse multimodal priors and overlook domain adaptation, thereby constraining extraction accuracy and cross-regional generalization. This work presents a transferable open-source data-driven framework for cross-domain multimodal extraction and analysis of roof vegetation. Specifically, we introduce a multimodal transformer encoder to fuse Google imagery, Sentinel-2-derived products and OpenStreetMap building footprints at the feature level to improve extraction accuracy. We then propose an unsupervised domain-adaptive semantic segmentation (UDASS) model with morphology-aware pseudo-label correction to improve cross-domain generalization cost-effectively. Validated across six Southeast Asian cities, our framework achieves an F1-score of 81.71% for supervised source domain extraction and up to 70.85% for UDASS. Further validation on four additional cities spanning East Africa, South America, West Asia, and South Asia verifies its strong cross-regional generalization ability. Finally, approximately 297 ha of roof vegetation are detected across the six Southeast Asian cities, mainly on low-to-mid-rise buildings with distinct regional disparities and diverse vegetation intensity. This study delivers the first large-scale dataset on roof vegetation and green roofs for tropical Southeast Asia, establishes a cost-effective and scalable multimodal paradigm for cross-regional monitoring, and supports evidence-based green roof planning and analysis.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105490"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636353","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Self-supervised spectral unmixing using scene-aware deep priors","authors":"Soorya Suresh,&nbsp;Shri Charan Ramesh,&nbsp;Arun Pattathal Vijayakumar,&nbsp;Vandita Srivastava,&nbsp;Krishna Mohan Buddhiraju","doi":"10.1016/j.jag.2026.105483","DOIUrl":"10.1016/j.jag.2026.105483","url":null,"abstract":"<div><div>High spectral resolution of hyperspectral sensors facilitates the detailed identification of material, but the coarse spatial resolution affects accurate interpretation. In this regard, spectral unmixing techniques attempt to estimate the constituent materials (endmembers) and their fractional areas (abundances) within a pixel’s instantaneous field of view (IFOV). However, many existing methods optimize denoising, endmember detection, and abundance estimation separately, which can limit robustness under noise and spectral variability. The current study proposes a self-supervised deep learning framework for joint hyperspectral unmixing and denoising. A noise-aware probabilistic encoding strategy is adopted to model a structured latent space to implicitly detect the endmembers. Further, the proposed encoding scheme and regularizations ensure spatio-spectral fidelity to improve robustness to noise. Additionally, a contrastive learning mechanism is incorporated to improve latent separability and robustness to spectral variability. The framework also integrates available partial endmember priors when present and employs a policy-based adaptive refinement mechanism to update endmember estimates under complex noise and mixing conditions. In addition, a learnable loss-balancing strategy is introduced to reduce manual tuning and stabilize optimization. Experiments on synthetic and real hyperspectral datasets demonstrate strong and balanced performance across reconstruction accuracy, spectral fidelity, and clustering quality, indicating the effectiveness of the proposed framework for practical hyperspectral unmixing.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105483"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636356","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HeightCLIP: A vision-language network for height estimation from monocular remote sensing imagery","authors":"Tao Zhang,&nbsp;Furong Shi,&nbsp;Yuanping Zhu,&nbsp;Xin Huang","doi":"10.1016/j.jag.2026.105499","DOIUrl":"10.1016/j.jag.2026.105499","url":null,"abstract":"<div><div>Accurate height estimation from monocular remote sensing imagery is a fundamental task with broad applications in urban planning and management. However, existing vision-based methods rely solely on visual features, which limits their ability to fully exploit height-related semantic information. Vision-language models, e.g., CLIP, offer a promising alternative by leveraging multimodal information, yet directly adapting them to height regression remains challenging due to their weak perception of continuous numerical values. To overcome this limitation, we propose <strong>HeightCLIP</strong>, a novel vision-language framework tailored for height estimation from monocular remote sensing imagery. The task-adaptive textual prompts and image-side representations are optimized under height supervision to derive pixel-text similarity features for accurate height estimation. Specifically, we design an interval-based prompting strategy that integrates learnable context tokens and a fixed natural language template with parameterized interval values, allowing CLIP to better comprehend height-related semantics. In addition, we incorporate a reciprocal cross-attention fusion mechanism to facilitate bidirectional interaction between visual and textual features, thereby improving cross-modal feature integration. Finally, building upon this enriched feature space, we impose cross-modal consistency constraints to ensure pixel-level correspondence between height descriptions and visual features, providing more informative representations for subsequent height prediction. Extensive experiments on three benchmark datasets i.e., Vaihingen, Potsdam, and Data Fusion Contest 2019 (DFC2019), demonstrate that HeightCLIP achieves competitive or superior performance compared with existing state-of-the-art methods, and comprehensive ablation studies confirm the effectiveness of each proposed component. The source code will be made available at <span><span>https://github.com/songtaowhu/HeightCLIP</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105499"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148651560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"BQEL-Net: A bimodal quantitative ensemble learning network with uncertainty-aware for vectorized extraction of martian impact craters from bimodal catalogic-and-quantitative dataset (BCQDS)","authors":"Chang Li,&nbsp;Haoyan Ma,&nbsp;Yi He,&nbsp;Zhizhong Kang","doi":"10.1016/j.jag.2026.105494","DOIUrl":"10.1016/j.jag.2026.105494","url":null,"abstract":"<div><div>Current research on bimodal extraction of Martian impact craters, which integrates satellite imagery and digital elevation models (DEMs), remains limited to the object level, with no existing reports about addressing both pixel-level and irregularly vectorized extraction. Furthermore, there is a lack of open-source bimodal datasets that include statistics (such as crater count, area and volume) for Martian impact craters, and no deep learning approaches have been reported for simultaneous extraction and quantitative analysis of Martian impact craters. To address these gaps, this study is the first to create and release a bimodal catalogic and quantitative dataset (BCQDS), combining DEM and satellite imagery with topographic landscape pattern indices. We also propose an innovative bimodal quantitative ensemble learning network (BQEL-Net) with uncertainty-aware paradigm designed for pixel-level, vectorized and quantitative extraction for Martian impact craters from BCQDS. BQEL-Net uses an uncertainty-aware Bayesian ensemble learning strategy (UBELS) by combining multiple strong segmentors confidence and pixel-level confidence to enhance robustness and accuracy; the proposed bimodal coordinate channel attention (BC²AM) aligns bimodal features in the latent space, making them complementary; and a newly developed quantitative statistics module (QSM) with a pioneer quantitative task loss function, enabling the simultaneous extraction of up to 140 quantitative indices. Experiments conducted using MOLA data within the BCQDS demonstrated that BQEL-Net achieves a maximum MIoU of 0.858. Ablation studies confirmed that the bimodal structure, BC²AM, UBELS and QSM contribute MIoU gains. This work pioneers a crater catalog with 3D geometric morphology to support quantitative studies of Martian surface evolution.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105494"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636345","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Quantifying methane emissions from large coal mine using Sentinel-5 Precursor satellite data and Gaussian plume model","authors":"Yuanyuan Chen,&nbsp;Yi Li,&nbsp;Lixia Tan,&nbsp;Yufan Qian,&nbsp;Shuangxi Fang","doi":"10.1016/j.jag.2026.105491","DOIUrl":"10.1016/j.jag.2026.105491","url":null,"abstract":"<div><div>Coal mining methane is a major contributor to anthropogenic greenhouse gas emissions in China. However, point-source emission data for Chinese coal mining areas remain scarce. Current bottom-up emission inventories suffer from significant uncertainties due to incomplete or inaccurate activity data and emission factors. This study presents a satellite-based methodology to quantify methane emissions from large coal mines, utilizing Gaussian plume model and observations from the Copernicus Sentinel-5 Precursor TROPOspheric Monitoring Instrument (TROPOMI-S5P), in which mixed plume effects of multiple point sources are considered. Applied to four typical underground and open-pit coal mining areas in China, we identify 18 effective TROPOMI-S5P methane plume enhancements, with emission rates ranging from 49.15 ± 19.43 t/d (Xiheishan mining area in Xinjiang) to 694.07 ± 58.66 t/d (Haerwusu-Heidaigou coal mine cluster in Inner Mongolia). Overall, the interannual variability of the estimated methane emissions is generally consistent with that of the coal production in the mining area. Uncertainty in individual plume-derived estimates ranges from 8.45 % to 41.01 %, primarily attributable to wind field data. Our satellite-derived estimates align closely with calculations using regionally refined emission factors, yet fall substantially below global inventories like Emissions Database for Global Atmospheric Research v8.0 (EDGARv8.0) and Global Fuel Exploitation Inventory (GFEI). This discrepancy may stem from outdated emission factors, evolving mining practices, observational capabilities, and sampling bias. This study highlights the critical role of satellite monitoring in validating and refining emission inventories, providing actionable insights for methane emission quantification and targeted mitigation in the coal mining sector.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105491"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636354","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enabling maize mapping in double-cropping regions of Argentina and Brazil without ground reference data by leveraging multiple satellite platforms","authors":"Ryoungseob Kwon,&nbsp;Youngryel Ryu,&nbsp;Philippe Ciais,&nbsp;David Makowski,&nbsp;Changhyun Choi,&nbsp;Huaize Feng,&nbsp;Sungchan Jeong,&nbsp;Pengfei Tang,&nbsp;Kyungdo Lee","doi":"10.1016/j.jag.2026.105489","DOIUrl":"10.1016/j.jag.2026.105489","url":null,"abstract":"<div><div>Large-scale maize mapping is challenging due to the lack of field-level datasets, which were essential for training machine learning models. The United States (US) provides abundant in-situ labels and national maize maps while Argentina and Brazil offered limited data and double-cropping further complicates mapping. In this study, we propose a survey-free method for maize mapping in Argentina and Brazil, with cropping intensity considered. We first pre-trained a maize classifier on US cornfields using Global Ecosystem Dynamics Investigation (GEDI) Relative Height (RH) metrics and Sentinel-2 imagery, then transferred this model to Argentina and Brazil to produce candidate maize labels at peak growth. We distinguished maize from other crops using phenology-based empirical thresholds. These high-confidence labels were then used to train a Random Forest model. Finally, we validated the maps with Google Street View (GSV) imagery and benchmark maps. Our framework produced 18,481 maize and 15,725 non-maize labels in Argentina and 2,125 first-season maize, 7,406 s-season maize and 7,643 non-maize in Brazil. Evaluation of our maize maps against GSV images showed satisfactory agreement, with accuracies around 90 %. The maps also aligned well with benchmark maps, exhibiting 91 % accuracy for Argentina and 88 % for Brazil. Our proposed frameworks can enhance maize mapping where field surveys are challenging.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105489"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636355","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Earth Observation–Based Machine learning for operational forecasting of harmful algal blooms in the Arabian Gulf and Sea of Oman","authors":"Kaltham A. Ismail,&nbsp;Roqaya A. Ismail,&nbsp;Mohammed Salim","doi":"10.1016/j.jag.2026.105501","DOIUrl":"10.1016/j.jag.2026.105501","url":null,"abstract":"<div><div>Harmful algal blooms (HABs) pose increasing risks to coastal ecosystems, fisheries, and desalination infrastructure in the Arabian Gulf and Sea of Oman. Forecasting bloom occurrence remains challenging due to strong spatial variability in physical forcing and biological response. Here, we develop a physics-informed machine-learning framework to forecast HAB probability using satellite Earth observation data and reanalysis-derived environmental predictors over the period 2002–2024. Bloom events are defined using region-specific, seasonally varying chlorophyll thresholds derived from training data, and a Random Forest classifier is used to estimate bloom probability at 8-day resolution. Operational probability thresholds are calibrated using historical data (2002–2020) and applied unchanged to independent evaluation (2021–2022) and strictly forward (2023–2024) periods. The model demonstrated moderate classification skill during the independent evaluation period (F1 = 0.46–0.67) and strong forward operational performance, achieving F1 scores of 0.79–0.97 across the study regions. Permutation importance analysis showed that chlorophyll anomaly contributed most strongly to forecast performance across regions, indicating that deviations from seasonal biomass conditions are strongly associated with bloom detection skill, while physical variables provide complementary predictive information. These findings demonstrate that a biomass-informed, Earth observation–based framework enables robust and operationally stable detection of harmful algal blooms across hydrographically complex marginal seas, supporting its application as an early-warning system for HAB monitoring and management in the Arabian Gulf and Sea of Oman.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105501"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148651559","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fusing very high-resolution aerial imagery with Sentinel-1 and Sentinel-2 time series improves multi-year orchard type classification","authors":"Lior Fine,&nbsp;Tarin Paz-Kagan,&nbsp;Offer Rozenstein","doi":"10.1016/j.jag.2026.105495","DOIUrl":"10.1016/j.jag.2026.105495","url":null,"abstract":"<div><div>Accurate mapping of perennial orchards is essential for agricultural management but remains challenging due to spatially discontinuous canopies and subtle phenological differences. This study presents an operational, application-oriented framework that integrates annual Very High-Resolution (VHR) aerial imagery (0.2 m) with monthly Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical time series to classify twelve orchard types across Israel from 2018 to 2023. By utilizing a multi-year dataset rather than a single seasonal cycle, we specifically evaluate model generalization and temporal robustness against inter-annual variability in phenology and imaging conditions. The dataset includes 106,413 labeled parcel-year samples from 19,950 parcels. We evaluate three model families: (i) satellite-only, (ii) VHR-only, and (iii) a multimodal mid-fusion deep-learning model (integrating spatial-structural and spectral-temporal modalities), under the same-year and cross-year (leave-one-year-out) scenarios. The satellite stream is modeled with a Lightweight Temporal Attention Encoder (LTAE) to capture seasonal dynamics, while the VHR stream uses a Swin Transformer backbone to extract hierarchical spatial features from 0.2 m imagery. The fused model achieved the highest performance across all settings (same-year OA = 0.890 ± 0.009; cross-year OA = 0.881 ± 0.014), outperforming both satellite-only (0.83) and VHR-only models (same-year OA = 0.846; cross-year OA = 0.795). In the best-performing fused cross-year scenario, class-wise F1-scores ranged from 0.71 for apple to 0.99 for banana. Fusion provided the largest gains in early-stage orchards (ages 1–3), where sparse canopy cover weakens spectral signals, but VHR imagery reveals planting geometry. Per-class analyses show that fusion benefits decrease with increasing mean NDVI, highlighting strong gains for sparse or deciduous orchards and minimal gains for dense evergreens. Overall, the results demonstrate that phenological time series and VHR spatial structure are highly complementary, enabling robust, operational orchard-type mapping and informing cost-aware workflows that selectively prioritize VHR acquisitions for young or low-NDVI orchards.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105495"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636344","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A hierarchical framework employing voting mechanism for large-scale ALB and MBES point cloud registration","authors":"Dongdong Pu,&nbsp;Hongzhou Chai,&nbsp;Chao Dong,&nbsp;Yongzhong Ouyang","doi":"10.1016/j.jag.2026.105404","DOIUrl":"10.1016/j.jag.2026.105404","url":null,"abstract":"<div><div>Registration is essential for fusing airborne LiDAR bathymetry (ALB) and multibeam echo sounder (MBES) data. However, existing studies are limited, often overlooking the unique characteristics and heterogeneity of bathymetric point clouds, which complicates the establishment of accurate correspondences. This study thoroughly analyzes the key challenges in ALB and MBES point cloud registration and proposes a hierarchical framework based on a voting mechanism. The framework transforms coarse registration into a hierarchical matching process, progressing from global to local compatibility graphs. It employs weak and strong voting mechanisms to progressively refine inlier selection from relaxed to strict criteria. An ICP algorithm incorporating geometric consistency is then applied to achieve fine registration. The strong voting mechanism enhances the weak voting mechanism by integrating local information. These hierarchical and voting strategies effectively reduce graph sensitivity, balancing accuracy and efficiency. Additionally, the study adapts initial correspondence extraction based on terrain complexity and introduces adaptive inlier selection through Gaussian fitting of fusion scores, calculated using the coefficient of variation. This study validates the proposed method using two field datasets from the South China Sea, providing a comprehensive analysis of the key steps in the algorithm. The proposed method outperforms Comparison Methods 1 and 2, reducing overall error (OE) by 221.2% and 206.4% on Data I, and 223.1% and 202.6% on Data II. Furthermore, experiments demonstrate the method’s robustness in complex noisy environments.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"151 ","pages":"Article 105404"},"PeriodicalIF":8.2,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148289481","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An adaptive hybrid algorithm for retrieving particulate organic carbon in complex coastal waters","authors":"Tao Zhou,&nbsp;Wen Zhou,&nbsp;Weilang Xie,&nbsp;Cai Li,&nbsp;Zhantang Xu,&nbsp;Xiankun Yang","doi":"10.1016/j.jag.2026.105413","DOIUrl":"10.1016/j.jag.2026.105413","url":null,"abstract":"<div><div>The retrieval of particulate organic carbon (POC) from optical properties in coastal waters remains a significant challenge because of profound spatiotemporal heterogeneity in water composition. Here, we propose an adaptive hybrid algorithm for POC retrieval (AHA-POC). This algorithm integrates optical water cluster classification with cluster specific model optimization to deliver improved POC estimates in optically complex environments, as demonstrated for the coastal waters of Guangdong and eastern Hainan Island. The algorithm first defines bio-optical water clusters using K-means clustering based on inherent optical properties (IOPs) and derived composite features. This classification separated samples into optically distinct clusters characterized by systematic differences in particle composition, absorption and scattering properties, and bio-optical relationships, which reduce the influence of variability in aquatic particulate matter composition on subsequent POC retrieval algorithms. After cluster definition, cluster specific sensitive features and retrieval models were selected to account for variable POC–IOPs relationships across optical regimes. To avoid station level data leakage, model performance was evaluated using leave-one-station-out cross-validation (LOSO-CV), in which all samples from the same station were held out together during validation. The AHA-POC achieved stable retrieval performance (R² = 0.79, RMSE = 0.071 mg /L, MAPE = 32.38 %, Bias = -0.78 %), showing improved accuracy relative to both univariate retrieval models and threshold-based water classification schemes in optically complex water environments. The algorithm was further applied to high resolution in situ IOPs profiles to reconstruct POC sections at 1 m vertical resolution, revealing fine scale vertical heterogeneity associated with riverine plume influence and near bottom particle resuspension. These results demonstrate the potential of cluster specific optical modeling for improving regional POC estimation in optically complex coastal waters and for supporting high resolution assessments of coastal carbon variability.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"151 ","pages":"Article 105413"},"PeriodicalIF":8.2,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148289478","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}