Science of Remote Sensing最新文献

{"title":"Combining machine learning algorithms for bridging gaps in GRACE and GRACE Follow-On missions using ERA5-Land reanalysis","authors":"Jaydeo K. Dharpure ,&nbsp;Ian M. Howat ,&nbsp;Saurabh Kaushik ,&nbsp;Bryan G. Mark","doi":"10.1016/j.srs.2025.100198","DOIUrl":"10.1016/j.srs.2025.100198","url":null,"abstract":"<div><div>The Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GFO) missions have provided valuable data for monitoring global terrestrial water storage anomalies (TWSA) over the past two decades. However, the nearly one-year gap between these missions pose challenges for long-term TWSA measurements and various applications. Unlike previous studies, we use a combination of Machine Learning (ML) methods—Random Forest (RF), Support Vector Machine (SVM), eXtreme Gradient Boosting (XGB), Deep Neural Network (DNN), and Stacked Long-Short Term Memory (SLSTM)—to identify and efficiently bridge the gap between GRACE and GFO by using the best-performing ML model to estimate TWSA at each grid cell. The models were trained using six hydroclimatic variables (temperature, precipitation, runoff, evapotranspiration, ERA5-Land derived TWSA, and cumulative water storage change), as well as a vegetation index and timing variables, to reconstruct global land TWSA at 0.5° grid resolution. We evaluated the performance of each model using Nash-Sutcliffe Efficiency (NSE), Pearson's Correlation Coefficient (PCC), and Root Mean Square Error (RMSE). Our results demonstrate test accuracy with area weighted average NSE, PCC, and RMSE of 0.51 ± 0.31, 0.71 ± 0.23, and 4.75 ± 3.63 cm, respectively. The model's performance was further compared across five climatic zones, with two previously reconstructed products (Li and Humphrey methods) at 26 major river basins, during flood/drought events, and for sea-level rise. Our results showcase the model's superior performance and its capability to accurately predict data gaps at both grid and basin scales globally.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100198"},"PeriodicalIF":5.7,"publicationDate":"2025-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143327506","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":"Investigating the contribution of understory to radiative transfer simulations through reconstructing 3-D realistic temperate broadleaf forest scenes based on multi-platform laser scanning","authors":"Xiaohan Lin ,&nbsp;Ainong Li ,&nbsp;Jinhu Bian ,&nbsp;Zhengjian Zhang ,&nbsp;Xi Nan ,&nbsp;Limin Chen ,&nbsp;Yi Bai ,&nbsp;Yi Deng ,&nbsp;Siyuan Li","doi":"10.1016/j.srs.2025.100196","DOIUrl":"10.1016/j.srs.2025.100196","url":null,"abstract":"<div><div>Forests are complex, multi-layered ecosystems mainly comprising an overstory, understory, and soil. Radiative transfer simulations of these forests underpin the theoretical framework for retrieving forest parameters; however, the understory has often been neglected due to limitations in data acquisition technology. In this study, we assessed the contribution of the understory to canopy reflectance in a temperate broadleaf forest by comparing simulated bidirectional reflectance factor (BRF) differences between forest scenes with and without the understory. These scenes were reconstructed through voxel-based, boundary-based, and ellipsoid-based approaches respectively based on the multi-layered point cloud data acquired via combining unmanned aerial vehicle (UAV) and backpack laser scanning. The results show that the understory influences the simulated BRF across all three forest scene reconstruction approaches, suggesting that canopy reflectance signals can be used to evaluate the understory information, which provides a theoretical foundation for the feasibility of retrieving understory parameters via remote sensing. The understory increases BRF by 80% in shaded regions beneath the overstory in the red and NIR bands, and can increase BRF by 40% in the NIR band for voxel-based and ellipsoid-based forest scenes. Conversely, it reduces the simulated BRF in sunlit soil areas in the red band. Among the three forest reconstruction methods, the canopy reflectance simulation using the boundary-based model can consistently project the most understory information. Notably, the findings also indicate that the reflectance of the forest canopy definitely capture less understory vegetation information as the simulation resolution decreases, for instance, as the simulated resolution decreased from 1 m to 30 m, the absolute difference in the red band between the multi-layered BRF and L50 BRF decreased from 23.93% to 10.22% when using the boundary-based approach. It implies that higher resolution remote sensing observations are more advantageous for the retrieval of understory parameters. This study provides a successful case for modeling the multi-layered forest structure in natural temperate broadleaf forests, and even offers a theoretical reference for facilitating the retrieval of biochemical and biophysical information from the understory by remote sensing.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100196"},"PeriodicalIF":5.7,"publicationDate":"2025-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099577","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":"Using airborne LiDAR and enhanced-geolocated GEDI metrics to map structural traits over a Mediterranean forest","authors":"Aaron Cardenas-Martinez ,&nbsp;Adrian Pascual ,&nbsp;Emilia Guisado-Pintado ,&nbsp;Victor Rodriguez-Galiano","doi":"10.1016/j.srs.2025.100195","DOIUrl":"10.1016/j.srs.2025.100195","url":null,"abstract":"<div><div>The estimation of three-dimensional (3D) vegetation metrics from space-borne LiDAR allows to capture spatio-temporal trends in forest ecosystems. Structural traits from the <span>NASA</span> <span>Global</span> Ecosystem Dynamics Investigation (GEDI) are vital to support forest monitoring, restoration and biodiversity protection. The Mediterranean Basin is home of relict forest species facing the consequences of intensified climate change effects and whose habitats have been progressively shrinking over time. We used two sources of 3D-structural metrics, LiDAR point clouds and full-waveform space-borne LiDAR from GEDI to estimate forest structure in a protected area of Southern Spain, home of relict species in jeopardy due to recent extreme water-stress conditions. We locally calibrated GEDI spaceborne measurements using discrete point clouds collected by Airborne Laser Scanner (ALS) to adjust the geolocation of GEDI waveform metrics and to predict GEDI structural traits such as canopy height, foliage height diversity or leaf area index. Our results showed significant improvements in the retrieval of ecological indicators when using data collocation between ALS point clouds and comparable GEDI metrics. The best results for canopy height retrieval after collocation yielded an RMSE of 2.6 m, when limited to forest-classified areas and flat terrain, compared to an RMSE of 3.4 m without collocation. Trends for foliage height diversity (FHD; RMSE = 2.1) and leaf area index (LAI; RMSE = 1.6 m²/m²) were less consistent than those for canopy height but confirmed the enhancement derived from collocation. The wall-to-wall mapping of GEDI traits framed over ALS surveys is currently available to monitor Mediterranean sparse mountain forests with sufficiency. Our results showed that combining different LiDAR platforms is particularly important for mapping areas where access to in-situ data is limited and especially in regions with abrupt changes in vegetation cover, such as Mediterranean mountainous forests.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100195"},"PeriodicalIF":5.7,"publicationDate":"2025-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143094977","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":"ForestAlign: Automatic forest structure-based alignment for multi-view TLS and ALS point clouds","authors":"Juan Castorena ,&nbsp;L. Turin Dickman ,&nbsp;Adam J. Killebrew ,&nbsp;James R. Gattiker ,&nbsp;Rod Linn ,&nbsp;E. Louise Loudermilk","doi":"10.1016/j.srs.2024.100194","DOIUrl":"10.1016/j.srs.2024.100194","url":null,"abstract":"<div><div>Access to highly detailed models of heterogeneous forests, spanning from the near surface to above the tree canopy at varying scales, is increasingly in demand. This enables advanced computational tools for analysis, planning, and ecosystem management. LiDAR sensors, available through terrestrial (TLS) and aerial (ALS) scanning platforms, have become established as primary technologies for forest monitoring due to their capability to rapidly collect precise 3D structural information directly. Selection of these platforms typically depends on the scales (tree-level, plot, regional) required for observational or intervention studies. Forestry now recognizes the benefits of a multi-scale approach, leveraging the strengths of each platform while minimizing individual source uncertainties. However, effective integration of these LiDAR sources relies heavily on efficient multi-scale, multi-view co-registration or point-cloud alignment methods. In GPS-denied areas, forestry has traditionally relied on target-based co-registration methods (e.g., reflective or marked trees), which are impractical at scale. Here, we propose ForestAlign: an effective, target-less, and fully automatic co-registration method for aligning forest point clouds collected from multi-view, multi-scale LiDAR sources. Our co-registration approach employs an incremental alignment strategy, grouping and aggregating 3D points based on increasing levels of structural complexity. This strategy aligns 3D points from less complex (e.g., ground surface) to more complex structures (e.g., tree trunks/branches, foliage) sequentially, refining alignment iteratively. Empirical evidence demonstrates the method’s effectiveness in aligning TLS-to-TLS and TLS-to-ALS scans locally, across various ecosystem conditions, including pre/post fire treatment effects. In TLS-to-TLS scenarios, parameter RMSE errors were less than 0.75 degrees in rotation and 5.5 cm in translation. For TLS-to-ALS, corresponding errors were less than 0.8 degrees and 8 cm, respectively. These results, show that our ForestAlign method is effective for co-registering both TLS-to-TLS and TLS-to-ALS in such forest environments, without relying on targets, while achieving high performance.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100194"},"PeriodicalIF":5.7,"publicationDate":"2025-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143094967","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":"Comparative analysis of UAV-based LiDAR and photogrammetric systems for the detection of terrain anomalies in a historical conflict landscape","authors":"Marcel Storch ,&nbsp;Benjamin Kisliuk ,&nbsp;Thomas Jarmer ,&nbsp;Björn Waske ,&nbsp;Norbert de Lange","doi":"10.1016/j.srs.2024.100191","DOIUrl":"10.1016/j.srs.2024.100191","url":null,"abstract":"<div><div>The documentation of historical artefacts and cultural heritage using high-resolution data obtained from unmanned aerial vehicles (UAVs) is of paramount importance in the preservation of historical knowledge. This study compares three UAV-based systems for the detection of historically relevant terrain anomalies in a conflict landscape. Two laser scanners, a high-end (RIEGL miniVUX-1UAV) and a lower priced model (DJI Zenmuse L1), along with a cost-effective optical camera system (photogrammetry using Structure from Motion, SfM) were employed in two study sites with different densities of vegetation. In the study area with deciduous trees and little low vegetation, the DJI Zenmuse L1 system performs comparably to the RIEGL miniVUX-1UAV, with higher completeness but lower correctness. The SfM method demonstrated inferior performance with respect to correctness and the F1-score, yet achieved comparable or higher completeness values compared to the laser scanners (maximum 1.0, median 0.84). In the study area characterized by dense near-ground vegetation, the detection results are less optimal. However, the RIEGL miniVUX-1UAV system still demonstrates superior results in anomaly detection (F1-score maximum 0.61, median 0.53) compared to the other systems. The DJI Zenmuse L1 data showed lower performance (F1-score maximum 0.56, median 0.46). Both laser scanners exhibited enhanced results in comparison to the SfM approach, with a maximum F1-score of 0.12. Hence, the SfM method is viable under specific conditions, such as defoliated trees without dense low vegetation. Therefore, lower-cost systems can offer cost-effective alternatives to the high-end LiDAR system in suitable environments. However, limitations persist in densely vegetated areas.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100191"},"PeriodicalIF":5.7,"publicationDate":"2025-01-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099668","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":"Large-scale mapping of plastic-mulched land from Sentinel-2 using an index-feature-spatial-attention fused deep learning model","authors":"Lizhen Lu ,&nbsp;Yunci Xu ,&nbsp;Xinyu Huang ,&nbsp;Hankui K. Zhang ,&nbsp;Yuqi Du","doi":"10.1016/j.srs.2024.100188","DOIUrl":"10.1016/j.srs.2024.100188","url":null,"abstract":"<div><div>Accurate and timely mapping of Plastic-Mulched Land (PML) on a large-scale using satellite data supports precision agriculture and enhances understanding the PML's impacts on regional climate and environment. However, accurately mapping large-scale PML remains challenging due to the relatively small size and short lifespan of visible PML. In this paper, we demonstrated a large-scale PML mapping using Sentinel-2 data by combining the PML domain knowledge and the deep Convolutional Neural Network (CNN). We developed a dual-branch Index-Feature-Spatial-Attention fused Deep Learning Model (IFSA_DLM) for effectively acquiring and fusing multi-scale discriminative features and thus for accurately detecting PML. The proposed model was trained on one agricultural zone with 2019 Sentinel-2 data and evaluated across six agricultural zones in Xinjiang, China (span &gt;1500 km in dimension) for Sentinel-2 and Landsat 8 data acquired over 2019 and 2023 to examine the spatial, temporal and across-sensor transferability. Results show that the IFSA_DLM model outperforms three compared U-Net series models with 94.48% Overall Accuracy (OA), 87.69% mean Intersection over Union (mIoU) and 93.38% F1 score. The model's spatial, temporal and sensor transferability is demonstrated by its successful cross-region, cross-time and Landsat-8 applications. Large-scale maps of PML in Xinjiang in both 2019 and 2023 further confirmed the effectiveness of the proposed approach.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100188"},"PeriodicalIF":5.7,"publicationDate":"2025-01-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099567","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 human-on-the-loop approach for labelling seismic recordings from landslide site via a multi-class deep-learning based classification model","authors":"Jiaxin Jiang ,&nbsp;David Murray ,&nbsp;Vladimir Stankovic ,&nbsp;Lina Stankovic ,&nbsp;Clement Hibert ,&nbsp;Stella Pytharouli ,&nbsp;Jean-Philippe Malet","doi":"10.1016/j.srs.2024.100189","DOIUrl":"10.1016/j.srs.2024.100189","url":null,"abstract":"<div><div>With the increased frequency and intensity of landslides in recent years, there is growing research on timely detection of the underlying subsurface processes that contribute to these hazards. Recent advances in machine learning have introduced algorithms for classifying seismic events associated with landslides, such as earthquakes, rockfalls, and smaller quakes. However, the opaque, “black box” nature of deep learning algorithms has raised concerns of reliability and interpretability by Earth scientists and end-users, hesitant to adopt these models. Leveraging on recent recommendations on embedding humans in the Artificial Intelligence (AI) decision making process, particularly training and validation, we propose a methodology that incorporates data labelling, verification, and re-labelling through a multi-class convolutional neural network (CNN) supported by Explainable Artificial Intelligence (XAI) tools, specifically, Layer-wise Relevance Propagation (LRP). To ensure reproducibility, a catalogue of training events is provided as supplementary material. Evaluation from the French Seismologic and Geodetic Network (Résif) dataset, gathered in the Alps in France, demonstrate the effectiveness of the proposed methodology, achieving a recall/sensitivity of 97.3% for rockfalls and 68.4% for quakes.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100189"},"PeriodicalIF":5.7,"publicationDate":"2025-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099669","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":"Responses of spectral indices to heat and drought differ by tree size in Douglas-fir","authors":"Olivia J.M. Waite ,&nbsp;Nicholas C. Coops ,&nbsp;Samuel Grubinger ,&nbsp;Miriam Isaac-Renton ,&nbsp;Jonathan Degner ,&nbsp;Jacob King ,&nbsp;Alex Liu","doi":"10.1016/j.srs.2024.100193","DOIUrl":"10.1016/j.srs.2024.100193","url":null,"abstract":"","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100193"},"PeriodicalIF":5.7,"publicationDate":"2024-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099576","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":"Synthesizing field plot and airborne remote sensing data to enhance national forest inventory mapping in the boreal forest of Interior Alaska","authors":"Pratima Khatri-Chhetri ,&nbsp;Hans-Erik Andersen ,&nbsp;Bruce Cook ,&nbsp;Sean M. Hendryx ,&nbsp;Liz van Wagtendonk ,&nbsp;Van R. Kane","doi":"10.1016/j.srs.2024.100192","DOIUrl":"10.1016/j.srs.2024.100192","url":null,"abstract":"&lt;div&gt;&lt;div&gt;The boreal biome, the largest terrestrial biome on Earth, is increasingly vulnerable to climate change due to warming twice as rapidly as the global average. Climate change has increased the temperature, frequency, severity, and amount of area burned, which is leading to changes in the spatial extent of forest type and species range. These rapid ecological shifts necessitate fine-scale monitoring of forest type to detect potential type conversions and guide management interventions. In this study, we present a framework for forest type classification combining field plots and high-resolution remote sensing data using machine learning models in the boreal forest of Interior Alaska. For this purpose, we conducted forest type classification at three different levels, including 1. forest and nonforest, 2. hardwood, softwood, and nonforest, and 3. three dominant forest types, including paper birch, black spruce, white spruce, and nonforest. To achieve this goal, we compared the performance of two advanced modeling approaches, the convolutional neural network (CNN) and the XGBoost model. Our datasets included field and high-resolution topographic metrics including elevation, slope, aspect, and solar radiation and canopy height derived from lidar (1 m) and 44 vegetation indices derived from high-resolution (1 m) visible to near infrared (VNIR) hyperspectral data collected by NASA Goddard's Lidar, Hyperspectral and Thermal Imager (G-LiHT) sensor. The remote sensing data were collected under variable sky conditions (clear to overcast) throughout a 1-month growing-season period, and field data collected by United States Department of Agriculture (USDA) Forest Service, Forest Inventory and Analysis program (FIA). In this framework, we also studied the importance of topographic and remote sensing variables for the classification of forest types. We found the CNN model outperformed the XGBoost model in terms of overall accuracy and a macro average F1 score for all three different forest type classifications. The CNN model achieved an overall accuracy of 93.1% for forest or nonforest, 82.6% for hardwood, softwood, and nonforest, and 74.7% for three dominant forest types including paper birch, black spruce, and white spruce along with nonforest. Among the various topographic factors, we found that elevation was the most important factor for discriminating all forest types. In addition, we found that canopy height and vegetation indices including Photochemical Reflectance Index (PRI) (R531 &amp; R570), Pigment Specific Normalized Difference (PSND) (R635 &amp; R800), and Gitelson and Merzlyak (GM1) (R550 &amp; R750) were important for differentiating between hardwood and softwood while Anthocyanin Reflectance Index (ARI1) (R550 &amp; R700) was important for differentiating between forest and nonforest. The high-resolution forest type information can improve our ecological understanding of boreal forest dynamics, estimate above ground biomass, and carbon","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100192"},"PeriodicalIF":5.7,"publicationDate":"2024-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099568","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 comprehensive evaluation of deep vision transformers for road extraction from very-high-resolution satellite data","authors":"Jan Bolcek ,&nbsp;Mohamed Barakat A. Gibril ,&nbsp;Rami Al-Ruzouq ,&nbsp;Abdallah Shanableh ,&nbsp;Ratiranjan Jena ,&nbsp;Nezar Hammouri ,&nbsp;Mourtadha Sarhan Sachit ,&nbsp;Omid Ghorbanzadeh","doi":"10.1016/j.srs.2024.100190","DOIUrl":"10.1016/j.srs.2024.100190","url":null,"abstract":"<div><div>Transformer-based semantic segmentation architectures excel in extracting road networks from very-high-resolution (VHR) satellite images due to their ability to capture global contextual information. Nonetheless, there is a gap in research regarding their comparative effectiveness, efficiency, and performance in extracting road networks from multicity VHR data. This study evaluates 11 transformer-based models on three publicly available datasets (DeepGlobe Road Extraction Dataset, SpaceNet-3 Road Network Detection Dataset, and Massachusetts Road Dataset) to assess their performance, efficiency, and complexity in mapping road networks from multicity, multidate, and multisensory VHR optical satellite images. The evaluated models include Unified Perceptual Parsing for Scene Understanding (UperNet) based on the Swin transformer (UperNet-SwinT), and Multi-path Vision Transformer (UperNet-MpViT), Twins transformer, Segmenter, SegFormer, K-Net based on SwinT, Mask2Former based on SwinT (Mask2Former-SwinT), TopFormer, UniFormer, and PoolFormer. Results showed that the models recorded mean F-scores (mF-score) ranging from 82.22% to 90.70% for the DeepGlobe dataset, 58.98%–86.95% for the Massachusetts dataset, and 69.02%–86.14% for the SpaceNet-3 dataset. Mask2Former-SwinT, UperNet-MpViT, and SegFormer were the top performers among the evaluated models. The Mask2Former, based on the SwinT, demonstrated a strong balance of high performance across different satellite image datasets and moderate computational efficiency. This investigation aids in selecting the most suitable model for extracting road networks from remote sensing data.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100190"},"PeriodicalIF":5.7,"publicationDate":"2024-12-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099667","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}