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

{"title":"3D-M2C-ResNet: A Multi-Scale feature enhancement and fusion model for Fine-Scale tree species classification in urban forests","authors":"Jushuang Qin ,&nbsp;Zhibo Chen ,&nbsp;Hao Lu ,&nbsp;Xiaohui Cui ,&nbsp;Zhenyao Wang ,&nbsp;Chao Mou ,&nbsp;Guangpeng Fan","doi":"10.1016/j.jag.2025.104874","DOIUrl":"10.1016/j.jag.2025.104874","url":null,"abstract":"<div><div>Forests play a crucial role in global carbon sequestration, and the varying carbon storage capacities of tree species underscore the need for accurate vegetation classification. This study introduces 3D-M²C-ResNet, a deep learning model for high-resolution, fine-scale tree species classification. The model leverages fused remote sensing inputs, combining Zhuhai-1 hyperspectral imagery with phenological parameters derived from Sentinel-2 time-series data. A Multi-Scale Cascaded Dilated Convolution (MCDC) module was developed to expand the receptive field through a three-branch architecture, enabling comprehensive spectral–spatial feature extraction. Additionally, a Multi-level Feature Enhancement Strategy (MFES) adaptively refines shallow and deep features, enhancing semantic–spatial integration across layers. The model was evaluated against support vector machine (SVM), VGG16, and ResNet50 on a test set of 22,380 pixels. 3D-M²C-ResNet achieved an overall accuracy of 98.08% and a Kappa coefficient of 97.88%, outperforming baseline methods. Ablation experiments confirmed the effectiveness of the MCDC and MFES modules. Notably, incorporating phenological information substantially improved classification performance, particularly for spectrally similar tree species. This approach provides a robust and scalable solution for detailed urban forest mapping, supporting ecological monitoring, carbon accounting, and sustainable forest management. Data and code are publicly available at: <span><span>https://github.com/qinjs123/3D-M2C-ResNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104874"},"PeriodicalIF":8.6,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145160040","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improved coseismic deformation detection via SBAS InSAR-Hyperbolic tangent step model integration","authors":"Hua Gao ,&nbsp;Mingsheng Liao ,&nbsp;Hui Lin ,&nbsp;Guangcai Feng ,&nbsp;Yuchao Zhong ,&nbsp;Xiaohui Zha","doi":"10.1016/j.jag.2025.104847","DOIUrl":"10.1016/j.jag.2025.104847","url":null,"abstract":"<div><div>The centimeter level deformation generated by small and medium-sized earthquakes contains rich information on tectonic activity, which is of great value for improving the coseismic deformation database and fault dynamics models. InSAR is an important means of observing coseismic deformation, but there is a problem of insufficient observation accuracy for centimeter level coseismic deformation in complex environments. Time-series InSAR advances achieve millimeter-scale deformation monitoring, yet detecting subtle coseismic signals from Mw 5.0–6.5 earthquakes remains challenging in complex environments by computational limits and step-like model discontinuities. We develop a method incorporating SBAS-InSAR and a hyperbolic tangent (tanh) step function to overcome these barriers. Simulations based on Sentinel-1 and MintPy demonstrate 15–55 % RMSE reductions (0.08–0.45 cm) in coseismic fields versus conventional DInSAR/Stacking methods. Applied to the 2021 Yangbi Mw5.9 event, our approach reveals: (1) The tanh-based model maintains phase continuity during abrupt deformation and decouples linear tectonic motion. (2) It enhances displacement field accuracy with 18–61 % noise suppression and the reliability of finite fault inversion. The model enables seconds deformation estimation for single earthquakes on standard hardware, advancing detection thresholds to sub-centimeter levels in high-coherence region. These breakthroughs expand InSAR’s capability in small-magnitude earthquake mechanics analysis. Future integration with advanced InSAR methodologies promises enhancements in seismic hazard system assessments.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104847"},"PeriodicalIF":8.6,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145160038","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Individual tree species prediction using airborne laser scanning data and derived point-cloud metrics within a dual-stream deep learning approach","authors":"Brent A. Murray ,&nbsp;Nicholas C. Coops ,&nbsp;Joanne C. White ,&nbsp;Adam Dick ,&nbsp;Ignacio Barbeito ,&nbsp;Ahmed Ragab","doi":"10.1016/j.jag.2025.104877","DOIUrl":"10.1016/j.jag.2025.104877","url":null,"abstract":"<div><div>Accurate tree species mapping is essential for effective forest management but is often constrained by manual, labour-intensive workflows that limit scalability. While airborne laser scanning (ALS) supports large-scale forest attribute prediction, species classification remains difficult in complex, multi-species forests. To address this, we propose an automated, data-driven dual-stream deep learning framework that integrates ALS data with point-cloud metrics to identify individual tree species. Our framework incorporates an automated approach to individual tree segmentation and species labelling using existing forest inventory and field data, resulting in a dataset of 16,269 labelled individual tree point-clouds of four species across a 630,000 ha boreal mixed species forest in Ontario, Canada. Our dual-stream deep learning model integrates a Point Extractor to generate feature representations from raw ALS point-clouds and a complementary Metrics Network to process the point-cloud metrics. Results, based on the split test set of 2441 trees, showed that the inclusion of the Metrics Network improved tree species classification accuracy by approximately 11 % compared to models that rely solely on the Point Extractor. A weighted F1-score of 0.70 and area under the receiver operating characteristic curve of 0.88 was achieved using this dual-stream approach, along with enhanced predictive probabilities for all species thus improving the reliability of the predicted results. This approach reduces the manual processing bottleneck of individual tree segmentation and labelling and demonstrates the value of combining raw point-clouds and point-cloud metrics into a deep learning framework, offering a scalable and operational solution for reliable species predictions.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104877"},"PeriodicalIF":8.6,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145160148","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Geospatial grid management: A comprehensive framework and systematic review of subdivision, encoding, indexing and storage","authors":"Yuanhao Su ,&nbsp;Daoye Zhu ,&nbsp;Boyong Xiao ,&nbsp;Shuang Li ,&nbsp;Tengteng Qu ,&nbsp;Weixin Zhai ,&nbsp;Chengqi Cheng","doi":"10.1016/j.jag.2025.104860","DOIUrl":"10.1016/j.jag.2025.104860","url":null,"abstract":"<div><div>With the rapid advancement of the Internet of Things (IoT), sensor technologies, and remote sensing, spatiotemporal data has emerged as a crucial data source across diverse industries, extensively utilized in environmental monitoring, intelligent transportation, socio-economic analysis, and other domains. Spatiotemporal data encompasses the locations, states, and interrelationships of objects within specific temporal and spatial contexts. It is characterized by dynamic properties, high dimensionality, and large data volumes, which pose significant challenges for storage, querying, and analysis. To address the challenges associated with managing large-scale spatiotemporal data, geospatial grid subdivision, grid encoding, grid indexing, and grid storage technologies offer essential support and have demonstrated remarkable effectiveness. In the past, existing reviews have typically focused on a single aspect, such as the fundamental methods of grid subdivision, the implementation details of encoding and indexing techniques, or solely on spatiotemporal databases. Although these reviews provide in-depth discussions of specific technologies, they lack a systematic analysis of the interrelationships among multiple technical modules, resulting in an inability to fully reveal the collaborative potential between modules. Additionally, current research provides limited comprehensive discussions on grid indexing technologies. This paper aims to address this gap by providing a systematic review of the development status of grid indexing technologies. Furthermore, it reviews and summarizes the key technologies, interrelationships, research advancements, and future directions of grid subdivision, grid encoding, grid indexing, and grid storage, thereby providing references for enhancing the storage and querying efficiency of spatiotemporal data.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104860"},"PeriodicalIF":8.6,"publicationDate":"2025-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-modal DEM super-resolution using relative depth: A new benchmark and beyond","authors":"Wenjun Huang,&nbsp;Qun Sun,&nbsp;Wenyue Guo,&nbsp;Qing Xu,&nbsp;Bowei Wen,&nbsp;Tian Gao,&nbsp;Anzhu Yu","doi":"10.1016/j.jag.2025.104865","DOIUrl":"10.1016/j.jag.2025.104865","url":null,"abstract":"<div><div>Learning-based Digital Elevation Model (DEM) super-resolution (SR) remains a challenge due to the complexity of real-world terrains. Existing approaches typically treat DEMs as digital grids or triangulated irregular networks, solving numerical fitting problems to densify points through learning models. However, these methods often overlook the spatial context and structural textures inherent in the terrain. To address this limitation, we propose utilizing relative depth maps derived from open-source remote sensing images by a foundational Depth Anything Model (DAM), which provide complementary structural information about the terrain and enhance the elevation details in DEMs. A novel DEMSR dataset, DEM-OPT-Depth SR, is constructed, pairing open-source remote sensing images, DEMs, and their corresponding relative depth maps. Additionally, we present a benchmark method, the Multi-modal Fusion Super-Resolution (MFSR) network, which extracts features through multi-branch pseudo-siamese networks and performs multi-scale feature fusion. Extensive experiments on the DEM-OPT-Depth SR dataset demonstrate a 24.63% improvement in RMSE-Elevation, a 22.05% improvement in RMSE-Slope, and an 11.44% improvement in RMSE-Aspect, showing the superiority and generalization capabilities of the MFSR model over previously proposed state-of-the-art baselines in DEMSR tasks. The code and dataset can be accessed at <span><span>https://github.com/hwj0711/MFSR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104865"},"PeriodicalIF":8.6,"publicationDate":"2025-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121238","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Modeling grassland parameters with hyperspectral satellite data: Comparison of sensors, acquisition times and spectral transformations","authors":"Christine I.B. Wallis ,&nbsp;Ann-Kathrin Holtgrave ,&nbsp;Daniel Prati ,&nbsp;Michael Förster ,&nbsp;Birgit Kleinschmit","doi":"10.1016/j.jag.2025.104857","DOIUrl":"10.1016/j.jag.2025.104857","url":null,"abstract":"<div><div>Recent advancements in hyperspectral satellite technology, including sensors like EnMAP, are promising for monitoring grassland ecosystems at the landscape scale. These developments include detailed vegetation analysis capabilities, crucial for understanding plant traits and species composition. However, technical restrictions of recent hyperspectral satellite missions can hinder comprehensive coverage of research areas resulting in a temporal mismatch of field measurements and satellite data.</div><div>Here we utilize hyperspectral data from the DESIS, PRISMA, and EnMAP mission, along with field measurements from 74 grassland plots of the German Biodiversity Exploratories in Schorfheide-Chorin and Hainich, collected in May 2020 and 2023. We focus on the impact of different satellite sensors and their acquisition timing grouped within five phenological seasons (early April to August) on the accuracy of biomass and species composition models using Partial Least Squares Regression (PLSR) and Procrustes randomization tests. Additionally, we are evaluating the effectiveness of two spectral transformations in improving model accuracy and reliability.</div><div>Our findings reveal significant differences in the relationship of hyperspectral satellite data with grassland biomass and species composition. Even though comparison of DESIS biomass models indicated that sensor data from the beginning of April achieved best results for biomass (R² = 0.48), sensor data covering the SWIR from late April and June showed slightly better modeling results (EnMAP: R² = 0.51, PRISMA: R² = 0.53). Species composition was significantly related to spectral composition, with sensor data from late April showing the strongest relationships. The performance of sensors, including VNIR and VNIR-SWIR, was almost equal (e.g., DESIS: R² = 0.54, PRISMA: R² = 0.59). Overall, the results highlight the benefit of SWIR bands for biomass modeling, while their importance was minor in relation to species composition. A trend of improved model performance was observed with mean normalization for EnMAP and PRISMA data, while continuum removal led to a decrease in performance. Our study underscores the critical role of temporal and spectral data selection in improving grassland models, suggesting potential pathways for refining remote sensing approaches in ecological monitoring and management.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104857"},"PeriodicalIF":8.6,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121240","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Geo-Mamba: A data-driven Mamba framework for spatiotemporal modeling with multi-source geographic factor integration","authors":"Zongwen Shi ,&nbsp;Chengyi Zhao ,&nbsp;Kaixin Wang ,&nbsp;Xiangkui Kong ,&nbsp;Jianting Zhu","doi":"10.1016/j.jag.2025.104854","DOIUrl":"10.1016/j.jag.2025.104854","url":null,"abstract":"<div><div>Earth science data exhibit inherent complexity characterized by heterogeneous spatiotemporal attributes, high collinearity among variables, and diverse input formats. Despite rapid advancements in deep learning, geographic modeling lacks unified frameworks for integrating heterogeneous spatiotemporal data and diverse factor types. While Mamba architecture has demonstrated efficiency in large language models, computer vision and remote sensing, its applicability to geographic modeling remains unexplored. This study introduces Geo-Mamba, a novel framework addressing these challenges through three key innovations. First, we propose a systematic geographical factor classification method that categorizes elements into dynamic, static, and categorical factors, enabling standardized integration of heterogeneous data within a unified paradigm. Second, we design a selective encoder module based on Mamba architecture that leverages its linear complexity and scanning mechanism to establish selective state spaces for geographical inputs, revealing intricate associations between diverse feature types. Third, we incorporate Kolmogorov-Arnold Network (KAN) layers as intermediate components replacing multilayer perceptron linear layers, enhancing numerical regression accuracy in geographical applications. Experimental validation across three tasks demonstrates Geo-Mamba’s effectiveness: in net ecosystem exchange modeling (R2 = 0.92, RMSE = 0.37 μmol·m^–2·s^–1), groundwater storage anomaly downscaling (R² = 0.95, RMSE = 1.916 cm), and land cover classification (accuracy = 88.12 %, F1-Score = 84.27 %). These results confirm Geo-Mamba as an efficient unified framework for complex Earth science modeling, while establishing its viability for geographical data processing and factor integration.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104854"},"PeriodicalIF":8.6,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Mapping invasive Opuntia stricta in Kenya’s Drylands using explainable machine learning with time-series remote sensing and geographic context","authors":"Jiayi Song ,&nbsp;Chang Zhao ,&nbsp;Kenneth T. Oduor ,&nbsp;Hao-Yu Liao ,&nbsp;Zhou Tang ,&nbsp;Igor L. Bretas ,&nbsp;Srikantnag A. Nagaraja ,&nbsp;José C.B. Dubeux ,&nbsp;Willis O. Owino ,&nbsp;Wei Shao","doi":"10.1016/j.jag.2025.104867","DOIUrl":"10.1016/j.jag.2025.104867","url":null,"abstract":"<div><div><em>Opuntia stricta</em> is a globally widespread invasive species that degrades dryland ecosystems and threatens pastoral livelihoods. Accurate, high-resolution distribution maps are essential for effective management, but its spectral similarity to native vegetation complicates remote sensing-based classification. We developed an interpretable Random Forest model, incorporating SHapley Additive exPlanations (SHAP), to map <em>Opuntia stricta</em> and co-occurring land cover types at 10  m resolution across heterogeneous arid and semi-arid lands in Laikipia County, Kenya. The model integrated monthly Sentinel-2 imagery, climate, topographic, landscape structural, and anthropogenic factors. Field surveys were combined with manual labeling using Google Maps and Street View to address annotation gaps in remote areas. Grid-based spatial blocking at multiple scales was used to reduce spatial autocorrelation and assess generalizability. The multi-temporal model achieved 0.91 overall accuracy and an F1-score of 0.92 for <em>Opuntia stricta</em> on the spatial validation set (100 m grid), and 0.86 accuracy with an F1-score of 0.85 on the independent test set, substantially outperforming single-month models (accuracy: 0.62–0.79; F1: 0.67–0.82), with February identified as the most informative single-time window. SHAP analysis identified July precipitation, population density and nighttime land surface temperatures as top predictors, linking invasion patterns to dry-season aridity, wet-season rainfall, and warm night conditions, underscoring the role of climate seasonality and human activity in shaping detectability and distribution. Invasion hotspots were concentrated near Dol-Dol and in degraded group ranches, with lower levels on private ranches and conservancies. Our findings highlight the potential of multi-temporal, context-integrated remote sensing for targeted invasive species management in dryland ecosystems.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104867"},"PeriodicalIF":8.6,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dual-task learning for dead tree detection and segmentation with hybrid self-attention U-Nets in aerial imagery","authors":"Anis Ur Rahman,&nbsp;Einari Heinaro,&nbsp;Mete Ahishali,&nbsp;Samuli Junttila","doi":"10.1016/j.jag.2025.104851","DOIUrl":"10.1016/j.jag.2025.104851","url":null,"abstract":"<div><div>Mapping standing dead trees is critical for assessing forest health, monitoring biodiversity, and mitigating wildfire risks, for which aerial imagery has proven useful. However, dense canopy structures, spectral overlaps between living and dead vegetation, and over-segmentation errors limit the reliability of existing methods. This study introduces a hybrid post-processing framework that refines deep learning-based tree segmentation by integrating watershed algorithms with adaptive filtering to enhance instance separation and boundary precision. Leveraging a dual-task learning architecture with a Self-Attention U-Net, the framework simultaneously predicts segmentation masks, centroid heatmaps, and hybrid boundary maps, optimizing for both pixel-level accuracy and instance-level detection. Tested on high-resolution aerial imagery from boreal forests, the framework, compared to the U-Net baseline, improved instance-level segmentation accuracy by 41.5% (Tree IoU of 0.3810 vs. 0.2694) and reduced positional errors by 57% (centroid error of 3.70 pixels vs. 5.10 pixels), demonstrating robust performance in the densely vegetated boreal forest regions tested. By balancing detection accuracy (F1-score of 0.5895) and over-segmentation artifacts, the method enabled the accurate identification of individual dead trees, which is critical for ecological monitoring. The framework’s computational efficiency supports scalable applications, such as wall-to-wall tree mortality mapping over large geographic regions using aerial or satellite imagery. These capabilities directly benefit wildfire risk assessment, carbon stock estimation, and precision forestry. This work advances tools for large-scale ecological conservation and climate resilience planning.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104851"},"PeriodicalIF":8.6,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121058","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing land use and land cover classification with deep learning-based satellite imagery segmentation","authors":"Tsion Fekadu Deressu ,&nbsp;Amanuel Kumsa Bojer ,&nbsp;Taye Girma Debelee ,&nbsp;Worku Gachena Negera ,&nbsp;Saralees Nadarajah ,&nbsp;Kena Wendimu Gebissa","doi":"10.1016/j.jag.2025.104839","DOIUrl":"10.1016/j.jag.2025.104839","url":null,"abstract":"<div><div>Semantic segmentation of satellite imagery plays a vital role in applications such as sustainable development, agriculture, forestry, urban planning, and climate change monitoring. Despite its importance, optimizing deep learning models for this task remains challenging. This study evaluates several advanced deep learning architectures UNet, LinkNet, DeepLabV3+, and a modified version, AE-DeepLabV3+ in combination with different backbone networks, including ResNet101, ResNet152, Xception, MobileNetV2 and, EfficientNetV2. The objective is to identify the most effective model for classifying satellite images into eight land cover categories: built-up areas, roads, water bodies, agricultural land, shrubland, forest, grassland, and others. A high-resolution dataset with corresponding segmentation masks was developed to support this analysis and serve as a resource for future research. Preprocessing steps included normalization and data augmentation techniques such as vertical and horizontal flipping and random brightness adjustments. Experimental results indicate that UNet with the Xception backbone, LinkNet with ResNet152, DeepLabV3+ with the Xception backbone, and AE-DeepLabV3+ with the Xception backbone achieved Dice coefficients of 85.7%, 86.7%, 90.4%, and 91.3%, respectively. Among these, AE-DeepLabV3+ with Xception demonstrated the highest segmentation accuracy. The findings are contextualized through comparison with recent studies, highlighting the model’s ability to generalize across diverse geographic regions. To enhance model transparency and interpretability, explainable AI (XAI) techniques Seg-Grad-CAM++ and Seg-Score-CAM are employed to visualize class-specific feature attributions and better understand the model’s decision-making process.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104839"},"PeriodicalIF":8.6,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121055","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}