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

{"title":"MFRXG: A multi-feature stacking ensemble RF-XGBoost model for total nitrogen retrieval in the Yellow River from Landsat 8/9 data","authors":"Yuanyuan Zhang,&nbsp;Shuanggen Jin","doi":"10.1016/j.jag.2026.105492","DOIUrl":"10.1016/j.jag.2026.105492","url":null,"abstract":"<div><div>Total nitrogen (TN) is a key indicator for assessing eutrophication in aquatic systems, and its concentration dynamics are influenced by a combination of hydrometeorological factors, terrain, and other pollution sources. However, traditional retrieval models based solely on remote sensing reflectance often suffer from limited generalization capability. This study develops a multi-feature stacked random forest (RF)-extreme gradient boosting (XGBoost) algorithm (MFRXG) based on field TN data from automatic stations along the Yellow River (YR). The MFRXG integrates Landsat 8/9 satellite data, ERA5 hydrometeorological data, and terrain data to estimate the TN concentration in YR. Using an independent validation dataset in 2025, MFRXG demonstrates robust prediction accuracy and stability, with a coefficient of determination (R²) of 0.75, a root mean square error (RMSE) of 0.53 mg/L, and a mean absolute error (MAE) of 0.31 mg/L, outperforming multiple linear regression (MLR), support vector regression (SVR), artificial neural networks (ANN), RF, and XGBoost. Subsequently, the spatiotemporal changes of TN concentration in YR are investigated from 2015 to 2025. Results show that spatially, TN concentrations exhibit a fluctuating pattern with a generally increasing gradient from the upstream to the downstream of the YR. Temporally, over the past decade, TN concentrations along the main stream show an overall slight upward trend, which is primarily attributable to changes in concentration contributions from upstream areas. Overall, the proposed modeling approach provides effective technical support for long-term and dynamic monitoring of TN concentration changes in YR and fills the gaps in spatial and temporal coverage and assessment efficiency of traditional monitoring methods.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105492"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635731","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":"Pioneering Test-Time training for robust Fine-Grained chlorophyll-a retrieval from Sentinel-2: Unveiling China’s lake eutrophication dynamics (2018–2025)","authors":"Zhen Zhang,&nbsp;Cuilin Yang,&nbsp;JiaJia Yuan,&nbsp;Lehao Huang,&nbsp;Zhongxi Ge,&nbsp;Zhitao Fu,&nbsp;Bo-Hui Tang","doi":"10.1016/j.jag.2026.105487","DOIUrl":"10.1016/j.jag.2026.105487","url":null,"abstract":"<div><div>Remote sensing retrieval of lake chlorophyll-a (Chl-a) concentration is essential for eutrophication assessment and water quality management, yet conventional models suffer severe degradation in large-scale, long-term monitoring due to domain shifts arising from inconsistent atmospheric correction (regional aerosol variability) and spatiotemporal heterogeneity in aquatic optical properties. This study pioneers Test-Time Training (TTT) for robust lake Chl-a retrieval. The TTT dual-task architecture—a primary regression head paired with a self-supervised spectral reconstruction head—enables real-time adaptation to test-domain distributions without retraining, effectively compensating for residual atmospheric errors and regional apparent optical properties (AOP) distortions. Based on the proposed algorithm, this study comprehensively analyzed the spatiotemporal evolution of Chl-a in 2,621 lakes (≥1 km²) in China from 2018 to 2025. The results showed the proposed model achieved superior performance in Chl-a retrieval (<em>R</em>² = 0.79, RMSE = 9.99 μg/L), outperforming conventional machine learning approaches by approximately 28 %-36 % and underscoring its robustness in optically complex inland waters. National Chl-a distribution exhibits a clear east-high, west-low pattern, with mean concentrations of 31.29 μg/L in the Eastern Plain versus 13.83 μg/L on the Tibetan Plateau. Regional analysis indicated that most regions exhibited a downward trend in Chl-a concentrations, suggesting an improvement in water quality. Bloom risk assessment, based on over 170,000 observations, indicates that the Eastern Plain lakes face “extremely high risk” (44.7 % bloom occurrence), requiring targeted management. The proposed lake Chl-a remote sensing retrieval algorithm provides a new paradigm for fine-grained global water quality monitoring and spatiotemporal analysis, holding landmark significance for global water environment surveillance and proactive management. The code is available at <span><span>https://github.com/KustAIRS/SR-TTA</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 105487"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635735","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":"Beyond fidelity: Rethinking realism through human behavior in VR city modeling LOD design","authors":"Yunlei Su,&nbsp;Guosheng Yang,&nbsp;Wufan Zhao,&nbsp;Cai Wu","doi":"10.1016/j.jag.2026.105484","DOIUrl":"10.1016/j.jag.2026.105484","url":null,"abstract":"<div><div>This study redefines realism in virtual reality (VR) city modeling by proposing a perception-driven Level of Detail (LOD) framework informed by human gaze behavior. Challenging the conventional assumption that higher fidelity equates to greater realism, the study systematically investigates the interplay of Feature Complexity (FC), Appearance (AP), and Semantics (SE) in achieving “perceptual realism”. Using eye-tracking technology integrated with VR, we conducted experiments in Guangzhou’s Xianfeng Community to identify LOD configurations that most closely align with real-world gaze distribution. Results reveal that the High Feature Complexity, Mid Appearance, Low Semantics (H-M-L) combination optimizes both global gaze patterns and perceptual coherence, outperforming higher semantic density configurations. The study also introduces an open-source tool for analyzing gaze behavior in urban VR environments, providing a scalable solution for immersive urban design research. These findings establish a foundation for advancing human-centric design in VR city modeling, emphasizing perceptual realism over technical fidelity to better align virtual environments with real-world human cognition. This work further positions VR-based perception-driven modeling as a key enabler for Urban Digital Twins, supporting the transition from static 3D models to dynamic, multi-source-data decision-making platforms.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105484"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636349","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":"SuperDove satellite detection of matter-loaded windrows with a focus on extreme environmental events","authors":"Dimitris Papageorgiou,&nbsp;Stefano Aliani,&nbsp;Andrés Cózar,&nbsp;Konstantinos Topouzelis","doi":"10.1016/j.jag.2026.105469","DOIUrl":"10.1016/j.jag.2026.105469","url":null,"abstract":"<div><div>Matter-loaded windrows are high-density submesoscale accumulation structures that form on the ocean surface and are often associated with extreme environmental events. This study presents a framework for windrow detection using a UNet deep-learning model applied to Planet SuperDove (PSD) 8-band satellite imagery (FLODAS). The model is trained on hand-annotated cases representing a variety of windrows and evaluated using complementary performance metrics, demonstrating consistent detection across independent scenes, while also revealing scene-specific patterns of false positives. Spectral analysis and classification based on the spectral angle distance metric is used to examine the potential for discriminating floating matter types using the PSD bands. Limited ground-truth data and the incomplete spectral characterisation of floating matter currently constrain classification performance. Application of the framework in the Aegean Sea enabled the identification of surface matter accumulations linked to storm-driven runoff and riverine discharge. The results demonstrate the potential of the framework to support in situ sampling and debris-removal operations, as well as short-term assessments of matter transport and windrow dynamics. Overall, the proposed framework represents a proof-of-concept step towards the development of operational detection systems for surface matter and marine debris accumulations.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105469"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636352","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":"Urban digital twin for environmentally sensitive mobility planning: conceptual framework and application in pilot region Leipzig","authors":"Thomas Trabert,&nbsp;Elmar Brockfeld,&nbsp;Alexander Sohr,&nbsp;Kerstin Krellenberg,&nbsp;Jakob Erdmann,&nbsp;Xiaoxu Bei,&nbsp;Carsten Dalaff,&nbsp;Timo Houben,&nbsp;Jan Bumberger","doi":"10.1016/j.jag.2026.105477","DOIUrl":"10.1016/j.jag.2026.105477","url":null,"abstract":"<div><div>This study presents the development and pilot implementation of a conceptual, practice-oriented Urban Digital Twin (UDT) used for environmentally sensitive mobility management within the pilot region of Leipzig (Germany). The UDT uses modular traffic simulations (SUMO-simulation of urban mobility), real-time sensor data and floating car data to create dynamic, near-real-time models of intermodal urban mobility systems and to evaluate mobility measures. Through a collaborative, co-production process with municipal stakeholders, the study focuses on scenario-based simulations to assess the potential impact of traffic scenarios under upcoming air pollution limits. The Leipzig case serves as a test environment for validating the technical feasibility of the conceptual UDT framework, applying a co-production process and discussing transferability of the approach. The results show the potential of UDTs to support responsive, data-driven decision-making using the developed workflow manager. In doing so, this study provides a transferable methodological basis for cities that aim at supporting urban sustainability transformations using digital twin technology.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105477"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636358","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":"DINO-Pheno-Cluster: Integrating few-shot foundation models and UAV time-series for spatiotemporal growth and functional characterization of alfalfa","authors":"Tao Cheng,&nbsp;Zhangru Gao,&nbsp;Ziyu Ding,&nbsp;Dong Han,&nbsp;Yaling Liu,&nbsp;Ruopu Zan,&nbsp;Xinyuan He,&nbsp;Shibo Yang,&nbsp;Feng Yuan,&nbsp;Weibo Ren,&nbsp;Dongyan Zhang","doi":"10.1016/j.jag.2026.105497","DOIUrl":"10.1016/j.jag.2026.105497","url":null,"abstract":"<div><div>Precise characterization of alfalfa growth dynamics is essential for breeding accessions with superior regrowth capacity, persistence, and yield stability. However, traditional plot level and coarse scale observations suffer from low signal to noise ratios particularly before canopy closure when phenotypic data are strongly affected by weeds and soil background. Moreover, existing studies rarely capture the dynamic mechanisms of crop development across the entire growth cycle. To address this, the DINO-Pheno-Cluster framework is introduced as a foundation model driven and mechanism decomposed phenotyping approach. This decoupled framework first utilizes DINO-XMem, a few-shot individual plant segmentation network based on DINOv3 and a dual memory mechanism. It subsequently applies parameterized dynamic modeling guided by growth process knowledge. Validation utilized high frequency Unmanned Aerial Vehicle (UAV) imagery from 12 time points across three growing seasons covering 127 alfalfa accessions. DINO-XMem achieved an 89.54% mean Intersection over Union (mIoU) under a 10-shot setting and maintained 86.77% under extreme 1-shot conditions. It successfully resolved dense canopy oversegmentation outperforming fully supervised baselines by 4.08% to 11.97% in mIoU. Crucially, the extracted high purity time series trajectories were parameterized into specific biological indicators including maximum growth rate, comprehensive regeneration index, and seasonal stability index. Gaussian Mixture Model (GMM) clustering based on these mechanistic traits identified four distinct functional ideotypes comprising High yield/High regrowth, Upright/Sparse, High stability/Persistent, and Short/Dense, all validated by ground measured biomass. This workflow establishes a precision screening tool for multi harvest crops advancing crop phenomics toward process level analysis.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105497"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148651500","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":"Generation of high-precision precipitable water vapor products from morphed integrated microwave imagery","authors":"Yuhao Wu,&nbsp;Bao Zhang,&nbsp;Yibin Yao,&nbsp;Zheng Du,&nbsp;Hongming Li,&nbsp;Quanyu Chen","doi":"10.1016/j.jag.2026.105485","DOIUrl":"10.1016/j.jag.2026.105485","url":null,"abstract":"<div><div>Water vapor is a key component in regulating Earth’s climate system, and precise monitoring of its variability is essential for advancing weather forecasting and climate research. While the Morphed Integrated Microwave Imagery at the Cooperative Institute for Meteorological Satellite Studies (MIMIC) provides near real-time, high-resolution estimates of Precipitable Water Vapor (PWV), its utility is constrained by notable inaccuracies and systematic biases. To address these limitations, this study introduces a hybrid calibration framework combining Spherical Cap Harmonic Analysis (SCHA) with Extreme Gradient Boosting (XGBoost) to enhance the accuracy of MIMIC-derived PWV. The resulting dataset offers high-precision PWV estimates over North America spanning 2016–2024. Validation against Global Navigation Satellite System (GNSS) PWV measurements demonstrates substantial improvements, with root mean square errors (RMSEs) reduced from 4.1 mm to 1.6 mm for the 0.25° resolution product and from 4.1 mm to 1.4 mm for the 0.02° product—representing accuracy gains of 61% and 66% over the original MIMIC PWV. The optimized PWV exhibits negligible bias relative to GNSS data and improved spatial and temporal consistency. Additionally, the hybrid approach effectively mitigates common artifacts in the original MIMIC products, including striping, shot noise, land–water boundary discontinuities, and excessive spatial smoothing. Further validation using radiosonde observations and ERA5 PWV yields RMSEs of 2.2–2.3 mm (0.25°) and 2.1–2.3 mm (0.02°), corresponding to 44–49% and 41–55% improvements. This work significantly enhances the reliability of MIMIC PWV data, providing a robust, high-quality resource for weather and climate applications.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105485"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635734","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":"Dual-polarization GNSS reflectometry for FYI–MYI sea ice type classification using microsatellite observations","authors":"Runchuan Li,&nbsp;Fei Guo,&nbsp;Wentao Yang,&nbsp;Weiqiang Li,&nbsp;Xiaohong Zhang,&nbsp;Estel Cardellach","doi":"10.1016/j.jag.2026.105502","DOIUrl":"10.1016/j.jag.2026.105502","url":null,"abstract":"<div><div>Accurate sea ice type classification is essential for polar climate monitoring, ocean–climate interaction studies, and safe navigation. Global Navigation Satellite System Reflectometry (GNSS-R) provides all-weather, high-temporal-resolution sea ice observations with low-power bistatic microwave measurements. The hypothesis of this study is that dual circular polarization GNSS-R observations from small satellites, which simultaneously capture left-hand circularly polarized (LHCP) and right-hand circularly polarized (RHCP) reflected signals, provide additional information for characterizing sea ice scattering properties. Using dual-polarization observations from the Tianmu-1 satellite constellation, we extract features from Delay Doppler Maps (DDMs) and perform sea ice type classification with an eXtreme Gradient Boosting (XGBoost) model. Results show that incorporating dual-polarization features improves performance compared with LHCP-only observations. In the Arctic, first-year ice (FYI) and multi-year ice (MYI) are classified with recall rates of 74.79 % and 69.04 %, with an overall accuracy of 74.29 %. In the Antarctic, FYI and MYI recall rates are 73.64 % and 74.85 %, respectively, with an overall accuracy of 73.71 %. MYI recall improves by approximately 1.7 % in the Arctic and 4.5 % in the Antarctic, while maintaining stable FYI performance. The performance in the Arctic is affected by attitude maneuvers in latitudes &gt; 80°N. These results demonstrate that dual circular polarization GNSS-R provides complementary information that enhances sea ice type classification and supports its potential for next-generation GNSS-R missions.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105502"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148636350","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":"Spatiotemporal dynamics of grasshopper habitats under future climate projections, vegetation growth, and land use change","authors":"Jing Guo,&nbsp;Wenjiang Huang,&nbsp;Yingying Dong,&nbsp;Qiaoyun Xie,&nbsp;Guofei Fang,&nbsp;Fangzheng Yue,&nbsp;Kun Wang,&nbsp;Zhuoqing Hao,&nbsp;Yan Zhang,&nbsp;Xiangzhe Cheng,&nbsp;Yue Shi,&nbsp;Bing Zhang","doi":"10.1016/j.jag.2026.105481","DOIUrl":"10.1016/j.jag.2026.105481","url":null,"abstract":"<div><div>Grasshopper outbreaks pose a persistent threat to grassland ecosystems and livestock production, yet large-scale monitoring of their suitable habitats under future climate change remains challenging due to complex interactions among climate, vegetation, and land use. Consequently, this study, taking China’s grasslands as the research area, explores the relationships between grasshoppers, host plants, and the environment and constructs habitat monitoring indicators using future climate, land use, and topographic data. Building on this, an ensemble model comprising Random Forest, eXtreme Gradient Boosting, and Light Gradient Boosting Machine is established to monitor the grasshopper suitable habitats under four Shared Socioeconomic Pathways (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) from 2021 to 2100. Results indicate that elevation, temperature seasonality, slope, and precipitation during the nymph and egg stages significantly influence grasshopper occurrence. Moreover, elevation, temperature seasonality, precipitation seasonality, slope, precipitation during the growth period, and isothermality are major contributors to host plant distribution. Suitable habitats are projected to remain concentrated in Inner Mongolia, northern Shaanxi, southern Gansu, Qinghai, and northern Xinjiang, with pronounced regional shifts over time. Low emission scenarios (SSP1-2.6) consistently possess the smallest and declining extent of most suitable habitats, whereas high emission pathways have the largest habitat, potentially creating environmental conditions that are more favorable for grasshopper persistence and expansion. The results provide spatially explicit insights for early warning, targeted monitoring, and sustainable grassland pest management in a warming world.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105481"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148651558","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 partially supervised joint segmentation method for farmland and rural roads from single-class annotated remote sensing datasets","authors":"Zhaoxiang Cao,&nbsp;Yuchun Huang,&nbsp;Yibo Zhou,&nbsp;Peng Zhou,&nbsp;Zhiqiang Qin,&nbsp;Shengyuan Zhang","doi":"10.1016/j.jag.2026.105509","DOIUrl":"10.1016/j.jag.2026.105509","url":null,"abstract":"<div><div>Accurate segmentation of co-existing land-cover objects in high-resolution remote sensing imagery is essential for precision agriculture and rural planning. Although multi-class datasets are increasingly available, many high-quality annotations remain task-specific, where only a single target class is labeled and other co-occurring objects are treated as background. Jointly exploiting such heterogeneous single-class datasets introduces supervision ambiguity, resulting in pseudo-background interference, feature conflicts, and unstable optimization. To address these challenges, we propose RSFRNet, a partially supervised joint segmentation framework for farmland and rural road extraction from mutually exclusive single-class annotations. The key idea of RSFRNet is to explicitly match different types of conflicts in this setting with corresponding network and optimization designs. First, considering that roads are dominated by thin linear structures whereas farmland is characterized by large homogeneous regions, we design a Filtering-based Feature-wise Linear Modulation module (F-FiLM) to adapt decoder features to class-specific structural patterns. Second, to reduce the interference caused by unlabeled co-existing objects being treated as background, we introduce a Learnable Class Codebook (L-Codebook), which uses class prototypes to refine encoder representations before decoding. Third, to alleviate inconsistent parameter updates caused by heterogeneous single-class supervision, we introduce PSL-PCGrad to correct conflicting gradients during joint training. In addition, task-aware loss constraints are used to preserve road connectivity and farmland regional integrity. In this way, RSFRNet addresses partially supervised farmland–road joint segmentation from feature adaptation, semantic purification, and optimization stabilization perspectives. Experiments on FarmSeg-VL and WHU-RuR + show that RSFRNet achieves the best evaluated performance, reaching 74.04 % mIoU with balanced farmland and road IoU. These results highlight the effectiveness of RSFRNet for partially supervised farmland–road joint segmentation under mutually exclusive single-class annotations and suggest its potential for related single-class annotation scenarios.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"152 ","pages":"Article 105509"},"PeriodicalIF":8.2,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148651557","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}