Science of Remote Sensing最新文献

{"title":"Using low-resolution LIDAR on a ground-based agile robot to estimate height, DBH and crown volume of trees","authors":"Omar A. Lopez,&nbsp;Kasper Johansen,&nbsp;Dario Scilla,&nbsp;Mariana Elías-Lara,&nbsp;Victor Angulo,&nbsp;Samer Almashharawi,&nbsp;Matthew F. McCabe","doi":"10.1016/j.srs.2025.100296","DOIUrl":"10.1016/j.srs.2025.100296","url":null,"abstract":"<div><div>Measurements of tree structural properties, such as trunk diameter at breast height (DBH), tree height, crown diameter, and volume, are crucial for estimating aboveground biomass, carbon stocks, or for managing forestry and silviculture applications. Traditional manual surveys are time-consuming, inaccurate, inconsistent, and subject to observer bias. In this study, we explore the capacity of a ground-based robotic quadruped (“Spot”, from Boston Dynamics), equipped with an Enhanced Autonomy Payload (EAP) module and a Velodyne VLP-16 LIDAR sensor, to measure tree height, DBH, and crown volume. Here, we leverage the Spot EAP’s low-beam LIDAR for efficient data processing and maximizing payload capacity without compromising the EAP’s primary navigation function, leading to lower energy consumption. We developed a scanning method and pre-processing pipeline to generate high-quality point clouds for tree structural analysis. Focusing the study in an urban park with 58 trees (22 <em>Erythrina variegata</em> and 36 <em>Ficus altissima</em>), we collected tree height using a metric staff for reference data and also measurements for DBH. Crown volume reference data were derived by combining height measurements obtained from the metric staff with crown extent measurements captured by a LIDAR system mounted on an unmanned aerial vehicle (UAV). Implementing a multi-pose scanning strategy improved the vertical field of view from ±15 to ±60 degrees and increased the point cloud density by more than 800%, achieving a point cloud registration root mean square error (RMSE) of 2.09 cm. Efficient 2D-assisted segmentation, which combines a simplified delineation step with automated refinement, along with leveling, produced individual tree point clouds suitable for structural estimations. Height estimation based on the ground-based robot achieved an RMSE of 29.27 cm and a relative RMSE (rRMSE) of 5.23%. The algorithm for identifying bifurcated trees at breast height showed 100% accuracy. DBH estimates had an RMSE of 4.5 cm and a rRMSE of 18.9%. Crown volume estimation achieved a coefficient of determination (R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>) of 0.895, outperforming several existing methods. Overall, the study underscores the potential of agile ground-based robots for efficient and accurate tree structural analysis, with likely future improvements possible through automatic segmentation and parameter tuning.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100296"},"PeriodicalIF":5.2,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265184","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":"Towards 90 m resolution digital terrain model combining ICESat-2 and GEDI data: Balancing accuracy and sampling intensity","authors":"Petra Pracná ,&nbsp;Eliška Šárovcová ,&nbsp;Xiao Liu ,&nbsp;Anette Eltner ,&nbsp;Jana Marešová ,&nbsp;Kateřina Gdulová ,&nbsp;Mikhail Urbazaev ,&nbsp;Michele Torresani ,&nbsp;Giorgi Kozhoridze ,&nbsp;Vítězslav Moudrý","doi":"10.1016/j.srs.2025.100293","DOIUrl":"10.1016/j.srs.2025.100293","url":null,"abstract":"<div><div>This study employs Ice, Cloud and Elevation Satellite-2 (ICESat-2) and the Global Ecosystems Dynamics Investigation (GEDI) observations to generate gridded digital terrain models (DTMs). Specifically, we (1) compared how acquisition characteristics affect the accuracy of ICESat-2 and GEDI observations, (2) assessed the sampling intensity with respect to observation accuracy (0.25–50 m) and grid resolution (90, 300, and 1000 m), and (3) interpolated DTMs at 90 m resolution and compared their accuracy with a global digital elevation model (DEM) Copernicus GLO-90. ICESat-2 data consistently outperformed GEDI footprints in terms of terrain elevation accuracy across a range of conditions (terrain slope, landcover, beam strength, and day/night). Sampling intensity is strongly shaped by the trade-offs between observation accuracy and grid resolution, limiting the coverage at finer scales and stricter thresholds. However, combining ICESat-2 and GEDI boosted sampling intensity, with over 60 % of cells containing at least one observation, which enabled a 90 m DTMs interpolation. The spaceborne lidar DTMs at 90 m resolution achieved RMSEs between 9.9 and 14.7 m, comparable to Copernicus DEM (9.9–15.6 m). However, the local accuracy of the interpolated DTMs depended on both the number of input observations and their accuracy. Where at least 4–6 observations per a 90 m cell with vertical accuracy better than 5 m were available, spaceborne lidar DTMs outperformed the Copernicus DEM, with RMSEs of 3.7 m vs. 11.2 m in forests and 2.6 m vs. 3.1 m in non-forested areas. This demonstrates that spaceborne lidar-derived DTMs could replace global DEMs.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100293"},"PeriodicalIF":5.2,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145320132","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":"Automatic mapping of high-resolution impervious surfaces driven by hierarchical adaptive features","authors":"Linyilin Xu,&nbsp;Genyun Sun,&nbsp;Aizhu Zhang,&nbsp;Zheng Han,&nbsp;Zheng Li,&nbsp;Yuanhao Zhao","doi":"10.1016/j.srs.2025.100300","DOIUrl":"10.1016/j.srs.2025.100300","url":null,"abstract":"<div><div>The uncontrolled expansion of impervious surfaces across the Indochina Peninsula poses serious environmental and societal challenges. Accurate monitoring of their spatiotemporal dynamics is crucial for regional sustainable development. However, existing medium-to low-resolution datasets often suffer from systematic and regionally diverse rounding errors, which significantly undermine the reliability of monitoring efforts. Here, we propose a robust and computationally efficient framework built on Google Earth Engine to automate the production of large-scale, high-resolution impervious surface datasets. First, the Phenology Enhanced Vegetation Index (PEVI) is introduced to suppress noise in automated samples, enabling a de-manualized training. Then, hierarchical adaptive features are derived from multi-scale convolution and a hierarchical strategy to advance regional heterogeneous target representation, especially in complex scenarios. Accordingly, we develop ICPIS, the first-ever impervious surface dataset with a 5-m resolution for the Indochina Peninsula, spanning the period from 2016 to 2023. Accuracy assessments show its overall accuracies of 90.69 % and 91.32 % for 2016 and 2023, with Kappa coefficients of 0.813 and 0.826. In practical applications of spatial representation and temporal monitoring, ICPIS outperforms existing high-resolution and multi-temporal datasets. This study successfully reduces rounding errors in current impervious surface mapping for the Indochina Peninsula, and enables a provision of clearer data support for stakeholders.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100300"},"PeriodicalIF":5.2,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145320118","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":"Annual national tree canopy cover mapping: A novel workflow with temporal transferability and improved uncertainty quantification","authors":"Joshua Heyer ,&nbsp;Karen Schleeweis ,&nbsp;Bonnie Ruefenacht ,&nbsp;Ian Housman ,&nbsp;Yang Zhiqiang ,&nbsp;Daniel Ryerson ,&nbsp;Jaclyn Reischmann ,&nbsp;Kevin Megown ,&nbsp;M. Seth Bogle","doi":"10.1016/j.srs.2025.100301","DOIUrl":"10.1016/j.srs.2025.100301","url":null,"abstract":"<div><div>In 2023, a new generation of National Land Cover Database (NLCD) Tree Canopy Cover (TCC) data were produced that include annual 30m gridded time-series products for the Conterminous United States (CONUS). These NLCD TCC v2021.4 products have more frequent time steps, inter-annual coherency, lower product release latency, improved accuracy, uncertainty metrics, and documentation over previous versions. Our methodology leveraged Google Earth Engine (GEE) to generate annual time-series composites from Landsat and Sentinel-2 imagery that were used to predict annual pixel-wise TCC. Next, we used a moving window approach with a 5x5 window of equal-area tiles (480 × 480 km) to calibrate 54 random forest models on 2011 Forest Inventory and Analysis (FIA) reference data. Then we used model temporal transfer to apply each 2011 model to the window's center tile, with annual time-series predictors, to predict per pixel annual TCC and standard error (SE). To quantify model uncertainty, we simulated uncertainty for each TCC model to derive an uncertainty metric, Tau (τ), which enables users to incorporate confidence boundaries into environmental and ecological applications using the NLCD TCC v2021.4 data. An independent statistical assessment of the 2011 TCC map, conducted using 16,607 FIA observations, yielded a weighted Root Mean Square Deviation (RMSD_w) of 12.8 percent TCC and weighted Mean Absolute Error (MAE_w) of 8.0 percent TCC at the CONUS scale. This paper provides a detailed description of the methodology and example use cases of the NLCD TCC and United States Forest Service Science v2021.4 products, paving the way for robust and repeatable future TCC updates.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100301"},"PeriodicalIF":5.2,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145265187","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":"Near real-time monitoring reveals extensive recent forest disturbance in Ghana's protected areas","authors":"Luofan Dong ,&nbsp;Xiaojing Tang ,&nbsp;Foster Mensah ,&nbsp;Bashara Ahmed Abubakari ,&nbsp;Kelsee H. Bratley ,&nbsp;Pontus Olofsson ,&nbsp;Curtis E. Woodcock","doi":"10.1016/j.srs.2025.100299","DOIUrl":"10.1016/j.srs.2025.100299","url":null,"abstract":"<div><div>The Protected Areas (PAs) in Ghana play a critical role in preserving the abundant biodiversity of the West Africa Green Belt. But recent changes in policies and regulations have facilitated logging and mining activities, which have accelerated forest disturbances. While there is a consensus that PAs are undergoing destructive change, the extent, rate, and locations of forest disturbances are undocumented. In this study, we applied the fusion near real-time (FNRT) algorithm that utilizes Landsat, Sentinel-1, and Sentinel-2 data and sampling to monitor forests in the PAs of Ghana. The results reveal that 704.74 (±177.24) km² of forest in the PAs were lost in 2023 and 2024, representing 6 % (±1.5 %) of their forest area. Additionally, the forest disturbance rate in 2024 was estimated to be 91 % higher than the rate observed in 2023 (95 % CI: 34 %–172 %). Extensive forest disturbance areas were found in the PAs around Kumasi, such as the Tano Ofin (15.4 % ±3.9 %), Tinte Bepo (42 % ± 10.5 %), and Oda River (18.5 % ± 4.7 %) PAs. Native forests in these PAs are at high risk of further degradation or disappearance in the absence of effective conservation measures. We compared the FNRT results with other alert systems, including RADD and GLAD-L. The comparison demonstrates that multi-sensor fusion provides more timely and accurate detection of forest disturbances in the study area.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100299"},"PeriodicalIF":5.2,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145218955","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":"A perspective on the interpretability of poverty maps derived from Earth Observation","authors":"Gary R. Watmough ,&nbsp;Dan Brockington ,&nbsp;Charlotte L.J. Marcinko ,&nbsp;Ola Hall ,&nbsp;Rose Pritchard ,&nbsp;Tristan Berchoux ,&nbsp;Lesley Gibson ,&nbsp;Enrique Delamonica ,&nbsp;Doreen Boyd ,&nbsp;Reason Mlambo ,&nbsp;Seán Ó Héir ,&nbsp;Sohan Seth","doi":"10.1016/j.srs.2025.100298","DOIUrl":"10.1016/j.srs.2025.100298","url":null,"abstract":"<div><div>The use of Earth Observation Data and Machine Learning models to generate gridded micro-level poverty maps has increased in recent years, with several high-profile publications. producing some compelling results. Poverty alleviation remains one of the most critical global challenges. Earth Observation (EO) technologies represent a promising avenue to enhance our ability to address poverty through improved data availability. However, global poverty maps generated by these technologies tend to oversimplify the complex and nuanced nature of poverty preventing progression from proof-of-concept studies to technology that can be deployed in decision making. We provide a perspective on the EO4Poverty field with a focus on areas that need attention. To increase the awareness of what is possible with this technology and reduce the discomfort with model-based estimates, we argue that the EO4Poverty models could and should focus on explainability and operationalizability alongside accuracy and robustness. The use of raw imagery in black-box models results in predictions that appear highly accurate but that are often flawed when investigated in specific local contexts. These models will benefit for incorporating interpretable geospatial features that are directly linked to local context. The use of domain expertise from local end users could make model predictions accessible and more transferable to hard-to-reach areas with little training data.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100298"},"PeriodicalIF":5.2,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145219496","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 of anomalous C-band backscatter signals caused by subsurface scattering and their correlations with land surface characteristics over the Tibetan Plateau","authors":"Xiaojing Bai ,&nbsp;Donghai Zheng ,&nbsp;Xiangzhuo Liu","doi":"10.1016/j.srs.2025.100295","DOIUrl":"10.1016/j.srs.2025.100295","url":null,"abstract":"<div><div>Subsurface scattering is a rarely recognized phenomenon, which may cause abnormal backscatter (σ^o) signal and deteriorate soil moisture (SM) retrieval. Although this phenomenon has been recently detected by Wagner et al. (2024a) over the global scale, the cold regions such as the Tibetan Plateau (TP) widely covered by frozen soil were excluded for analysis. This study aims to fill the research gap to detect and map the σ^o anomalies and subsurface scattering on the TP, discuss their impacts on SM retrievals and explore their correlations with land surface characteristics in the warm seasons when surface soil is thawed. Pearson correlation coefficients (R) obtained between ASCAT or Sentinel-1A σ^o observations and in-situ or ERA5-Land SM data at local scale show that anomalous C-band σ^o signals are detected in the arid area of TP as indicated by negative R values. This directly leads to poorer performance of ASCAT SM retrievals over barren areas than over grasslands in the humid and semi-humid regions. By fitting two backscatter models without and with subsurface scattering term to above collocated σ^o and SM data pairs, the detected anomalous σ^o signals are found to be dominated by subsurface scattering. The results obtained with three indicators proposed to characterize the subsurface scattering further confirm above findings, which show that notable σ^o anomalies and strong subsurface scattering are predominately detected in the western part of TP and Qaidam Basin. Along with the dryness of SM, high level of soil pH index and sparse vegetation cover especially bare land are important factors that favor their occurrences. Moreover, the maps of above indicators with certain thresholds show great potential for masking the regions with high probability of anomalous σ^o signals caused by subsurface scattering and thus inaccurate ASCAT SM retrievals. These results demonstrate for the first time the potential of detecting σ^o anomalies and subsurface scattering and alleviate their effects on SM retrievals in cold and arid regions.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100295"},"PeriodicalIF":5.2,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145218956","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":"AQUAVis: Landsat-sentinel virtual constellation of remote sensing reflectance (Rrs) product for coastal and inland waters","authors":"Thainara M.A. Lima ,&nbsp;Vitor S. Martins ,&nbsp;Rejane S. Paulino ,&nbsp;Cassia B. Caballero ,&nbsp;Claudio C.F. Barbosa ,&nbsp;Akash Ashapure","doi":"10.1016/j.srs.2025.100294","DOIUrl":"10.1016/j.srs.2025.100294","url":null,"abstract":"<div><div>Operational Land Imager (OLI) onboard Landsat-8 and -9 (L89) and the MultiSpectral Instrument (MSI) onboard Sentinel-2A and -2B (S2A-B), as well as the recently launched Sentinel-2C, have been increasingly used for inland and coastal water monitoring. Integrating both data sensors into a single virtual constellation presents opportunities to enhance the revisit rate and capture the temporal variability of optically active water constituents. Since the NASA Harmonized Landsat-Sentinel-2 product is primarily focused on land applications, a new water-focused framework is needed for consistent integration of both OLI and MSI Remote Sensing Reflectance (<span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span>) products for aquatic studies. This study proposes a new globally validated framework, called AQUAVis (Aquatic Virtual Constellation for Landsat-Sentinel <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> Observations), designed to generate a harmonized L89-S2A/B dataset for aquatic applications. The AQUAVis workflow includes atmospheric, adjacency effect, and sunglint corrections, as well as spectral bandpass adjustment derived from a global dataset of more than 4000 water bodies to ensure seamless <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> integration. We acquired a total of 3568 L89 and 2986 S2A/B images (6.6 TB) between 2015 and 2023, and AQUAVis <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> retrievals were assessed using 2664 matchup observations with the AErosol RObotic NETwork-Ocean Color across 23 sites worldwide. Results demonstrated high agreement between AQUAVis-derived and <em>in situ</em> <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> retrievals, with spectral differences ranging from 0.39 % in the blue band to 21 % in the near-infrared band. The AQUAVis <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> product was further assessed for Chlorophyll-a estimation using the XGBoost model, which showed high spatial agreement between AQUAVis OLI and MSI matchup results in a wide range of Chl-a concentrations. These findings highlight the potential of AQUAVis to provide consistent <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub></mrow></math></span> products for large-scale water quality monitoring. By facilitating global aquatic data integration, AQUAVis supports applications in aquatic remote sensing, contributing to a broader understanding of water quality dynamics and trends.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100294"},"PeriodicalIF":5.2,"publicationDate":"2025-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145219495","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":"Rapid domain adaptation for disaster impact assessment: Remote sensing of building damage after the 2021 Germany floods","authors":"Victor Hertel ,&nbsp;Christian Geiß ,&nbsp;Marc Wieland ,&nbsp;Hannes Taubenböck","doi":"10.1016/j.srs.2025.100287","DOIUrl":"10.1016/j.srs.2025.100287","url":null,"abstract":"<div><div>The extent of building damage is a crucial indicator for guiding post-disaster relief strategies and rescue operations. However, diverse built environments and variations in imaging setups pose significant challenges for rapid, automated damage assessment from remote sensing data, leading to strong domain shifts and significantly reduced performance of pre-trained models. To align advanced domain adaptation techniques with the practical constraints of rapid mapping, we evaluate and propose techniques that effectively balance accuracy, resource efficiency, and operational applicability. By employing a Siamese multitask fusion network for semantic segmentation and change detection, we introduce a novel experimental approach that quantifies the influence of a priori information on domain adaptation performance. All strategies are benchmarked on a fully labeled dataset from the 2021 Germany floods. Our evaluation includes class-specific accuracy improvements, model tendencies toward over- or underestimation of damage, and resource requirements in terms of processing time, human capacity, and computational demands. Scenario-based recommendations are provided to assist in selecting the most suitable method for given conditions. All adopted techniques significantly improved model performance in a short time, achieving up to 86 % of the potential performance gain compared to supervised learning. Supervised domain adaptation with minimal annotations per class emerged as the most effective method for immediate action. Semi-supervised domain adaptation, coupled with an automatic labeling strategy based on hazard intensity, provided the highest performance improvements while maintaining low demands on time and human resources. Purely semi-supervised domain adaptation turned out time-consuming and computationally expensive, therefore advisable only under specific conditions with sufficient time or in the absence of human capacity.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100287"},"PeriodicalIF":5.2,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145320095","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":"Comparative evaluation of thirteen satellite-derived surface solar radiation products over China","authors":"Junmei He ,&nbsp;Liang Hong ,&nbsp;Bing Hu ,&nbsp;Wenjun Tang","doi":"10.1016/j.srs.2025.100291","DOIUrl":"10.1016/j.srs.2025.100291","url":null,"abstract":"<div><div>High-quality surface solar radiation (SSR) data are essential for assessing climate change impacts and quantifying solar energy potential. Satellite remote sensing is the primary method to obtain SSR data globally and regionally, especially in regions with sparse ground observation networks. This study validated thirteen satellite-derived SSR products over China using in situ data from the China Meteorological Administration (CMA). Results revealed that Himawari-8-based estimates (H8-ITP, H8-AIR, and GeoNEX) outperformed other products. CERES-SYN, GLASS, ISCCP-ITP, and CLARA-A3 showed moderately inferior performance, whereas BESS, the four MCD18 variants (MCD18A1.V61, MCD18A1.V62, MCD18C1.V61, MCD18C1.V62), and DSCOVR/EPIC demonstrated relatively poor performance in China. Across all stations, the mean bias error (MBE) for these products ranged from −16.9 (BESS) to 26.1 (DSCOVR/EPIC) W m⁻², and the root mean square error (RMSE) from 27.8 (H8-ITP) to 50.0 (DSCOVR/EPIC) W m⁻² at the daily scale. It should be noted that systematic biases (exceeding ±10 W m⁻²) were observed in H8-AIR, GeoNEX, BESS, and DSCOVR/EPIC. H8-AIR and GeoNEX displayed apparent overestimation with MBE values of 14.8 W m⁻² and 12.3 W m⁻², respectively. H8-ITP demonstrated the best overall performance with a minimum RMSE of 27.8 W m⁻², a smaller MBE of 3.9 W m⁻², and a higher R of 0.95 compared to other products. All products exhibit a pronounced accuracy decline in August, with DSCOVR/EPIC showing the sharpest summer deterioration, whereas H8-ITP and GeoNEX remain the most stable. Regionally, all products exhibited generally lower accuracy in the rugged southwestern plateau region and the cloudy southern part of China. These results would provide a valuable reference for selecting the most appropriate SSR product for specific needs, whether for solar energy assessments, climate studies, or surface radiative process analysis.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100291"},"PeriodicalIF":5.2,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145218958","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}