Peng Wang , Yongkang Chen , Bo Huang , Daiyin Zhu , Tongwei Lu , Mauro Dalla Mura , Jocelyn Chanussot
{"title":"MT_GAN: A SAR-to-optical image translation method for cloud removal","authors":"Peng Wang , Yongkang Chen , Bo Huang , Daiyin Zhu , Tongwei Lu , Mauro Dalla Mura , Jocelyn Chanussot","doi":"10.1016/j.isprsjprs.2025.04.011","DOIUrl":"10.1016/j.isprsjprs.2025.04.011","url":null,"abstract":"<div><div>Synthetic Aperture Radar (SAR) is an active microwave imaging and earth observation device capable of penetrating through clouds, rain, and fog, enabling it to operate effectively regardless of the weather conditions and throughout the day. However, speckle noise in SAR images can make them difficult to interpret, and optical images are often difficult to observe when they are covered by clouds. Therefore, after preprocessing, SAR images can be directly converted to optical images through end-to-end translation learning without optical images as auxiliary information, improving the interpretability of SAR images and realizing cloud removal. Due to the relatively simple structure design of the existing generator based on residual network, it is not perfect to capture and extract the feature information of the image, and the relationship between the features is not well connected, resulting in the existing SAR-optical translation results are not accurate enough. To mitigate this issue, we propose an image translation method utilizing a multilayer translation generative adversarial network (MT_GAN) for cloud removal. First, we design a despeckling module (DSM) to preprocess the speckle noise in SAR. Furthermore, a multilayer translation generator (MTG) is designed for SAR-to-optical (S-O) image translation. It can perform multi-scale image translation on different layers and combine them to enrich the semantic information of features and optimize the translation results. In addition, MTG combined with PatchGAN discriminator is used to compose the optical image generation sub-network (OGS) and SAR image regression sub-network (SRS). Finally, the SRS and OGS are used to establish the connection of cycle consistency loss and optimize the generated optical image. We prepare four datasets for experiments, two of which are used for image translation experiments and the other two for cloud removal experiments. The findings demonstrate that our proposed approach outperforms existing methods across all evaluation metrics and reaches 28.6140 and 0.7069 in PSNR and SSIM indicators, which surpass MS-GAN (28.3348, 0.6403) and DSen2-CR (28.3472, 0.6857), and effectively removes the cloud. The datasets and codes are available at <span><span>https://github.com/NUAA-RS/MT_GAN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 180-195"},"PeriodicalIF":10.6,"publicationDate":"2025-05-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143899564","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shaopeng Li , Bo Jiang , Shunlin Liang , Xiongxin Xiao , Jianghai Peng , Hui Liang , Jiakun Han , Xiuwan Yin
{"title":"Estimation of surface all-wave net radiation from MODIS data using deep residual neural network based on limited samples","authors":"Shaopeng Li , Bo Jiang , Shunlin Liang , Xiongxin Xiao , Jianghai Peng , Hui Liang , Jiakun Han , Xiuwan Yin","doi":"10.1016/j.isprsjprs.2025.04.035","DOIUrl":"10.1016/j.isprsjprs.2025.04.035","url":null,"abstract":"<div><div>Deep learning methods have demonstrated significant success in estimating land surface parameters from satellite data. However, these methods often require large sample sizes for optimal performance, which can be difficult to obtain. This study introduces a novel framework that combines transfer learning (TL) and data augmentation (DA) to improve the performance of a deep learning model, the residual neural network (ResNet), in estimating daily all-wave net radiation (<em>R<sub>n_daily</sub></em>) from Moderate Resolution Imaging Spectroradiometer (MODIS) top-of-atmosphere (TOA) observations using limited samples. The framework involves two main steps: first, constructing a robust base model using augmented training samples generated through image rotation in the source region; and second, fine-tuning this base model in target regions with limited local samples. The framework was tested in three regions: the Continental United States (US), Mainland China (MC), and the tropical zone (TR), all with limited ground measurement data. The US was selected as the source region due to its relatively better sample conditions. The results showed that: (1) the ResNet model trained in the US using augmented samples achieved a validated R<sup>2</sup> of 0.95, RMSE of 14.31, and bias of −0.28 Wm<sup>−2</sup>, which outperformed the multi-layer perceptron (MLP) and ResNet models trained with original samples by reducing the validated RMSEs of 2.77 Wm<sup>−2</sup> and 0.80 Wm<sup>−2</sup>, respectively; (2) the transferred base model also performed the best in MC and TR, with R<sup>2</sup> values of 0.86 and 0.66, RMSEs of 22.22 and 25.25 Wm<sup>−2</sup>, and biases of 0.22 Wm<sup>−2</sup> and −0.21 Wm<sup>−2</sup>, respectively, leading to a decrease in validated RMSE by 3.20, 1.87, and 1.14 Wm<sup>−2</sup> for MC and by 2.32, 1.12, and 0.55 Wm<sup>−2</sup> for TR compared to the MLP and ResNet model trained directly and the ResNet model trained using the augmented samples in these regions, respectively; and (3) the more comprehensive the pre-training sample, the better the framework’s performance in the target domain. However, challenges related to cloud cover and input window size need to be carefully addressed when applying the new framework. Overall, the results highlight the effectiveness of the proposed framework and provide a promising approach for applying deep learning methods with limited samples.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 131-143"},"PeriodicalIF":10.6,"publicationDate":"2025-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895449","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yunmeng Cao , Ian Hamling , Zhiwei Li , Chris Rollins
{"title":"Robust variance–covariance estimation of tropospheric turbulence improves InSAR capability for monitoring of small tectonic displacements","authors":"Yunmeng Cao , Ian Hamling , Zhiwei Li , Chris Rollins","doi":"10.1016/j.isprsjprs.2025.04.028","DOIUrl":"10.1016/j.isprsjprs.2025.04.028","url":null,"abstract":"<div><div>Evaluating uncertainties and errors in interferometric synthetic aperture radar (InSAR)-derived displacements is challenging but important for any interpretation or analysis. InSAR observation error mainly arises from tropospheric delays, particularly tropospheric turbulence which is hard to quantify in a deterministic way (e.g., using global atmospheric models). Here we propose a new variance–covariance estimation (VCE) method to robustly model the stochastic properties of the uncorrected tropospheric turbulence in time-series of SAR images. Based on this, we further propose 1) an improved linear estimator to quantify interseismic deformation, and 2) a new minimum variance optimization (MVO) to monitor coseismic displacements of small to moderate earthquakes. Synthetic experiments show that compared with the previous VCE approach, the new VCE method is more than 50 % more accurate in deriving SAR variance components, and estimates of interseismic and coseismic displacements are 40 % and 55 % more accurate than those using conventional time-series analysis. Applying the VCE-based linear estimator to real Sentinel-1 time-series over the North Anatolian Fault, Turkey, interseismic displacements have uncertainties reduced by 28 %, 18 %, and 6 % compared to ordinary least-squares approach, for 2, 3 and 4 years of data, respectively. Applying the MVO approach to the 19 September 2023 Mw 5.6 Mount Harper (New Zealand) earthquake, coseismic displacements have uncertainties reduced by 23 % on average compared to the stacking approach. Our results highlight the importance of considering the spatiotemporal variance of InSAR measurements when mapping small tectonic displacements.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 144-162"},"PeriodicalIF":10.6,"publicationDate":"2025-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Gaoxiang Yang , Xingrong Li , Yuan Xiong , Meng He , Lei Zhang , Chongya Jiang , Xia Yao , Yan Zhu , Weixing Cao , Tao Cheng
{"title":"Annual winter wheat mapping for unveiling spatiotemporal patterns in China with a knowledge-guided approach and multi-source datasets","authors":"Gaoxiang Yang , Xingrong Li , Yuan Xiong , Meng He , Lei Zhang , Chongya Jiang , Xia Yao , Yan Zhu , Weixing Cao , Tao Cheng","doi":"10.1016/j.isprsjprs.2025.04.031","DOIUrl":"10.1016/j.isprsjprs.2025.04.031","url":null,"abstract":"<div><div>Spatially explicit information on crop distribution over large areas and long timespans is essential for optimizing agricultural spatial allocation and promoting food security. Despite the emergence of numerous remote sensing-based approaches for crop type mapping in recent years, the generation of long-term and high-quality crop type maps still remains challenging due to the poor spatiotemporal scalability of existing algorithms in the absence of ground labels or satellite imagery sources. In this study, we proposed a knowledge-guided machine learning (KGML) approach for extracting year-to-year training data and producing long-term winter wheat products in China by integrating multi-source remote sensing and environmental datasets. Based on the crop development patterns, critical phenological domains were first retrieved by spectral or polarization variation characteristics, and then the corresponding spectral signatures were combined to strengthen the differentiation between crop types. Consequently, annual training samples were extracted and refined automatically from the candidate crop pixels and employed to train the machine learning classifier with harmonic features, thus producing winter wheat maps over China year by year. Based on the long-term dataset, the spatiotemporal dynamics of winter wheat planting areas at the national scale from 2000 to 2023 were revealed by spatial and trend analyses, and the driving forces were further quantified.</div><div>With the KGML, the first long-term winter wheat products at 30-m spatial resolution were produced over China (ChinaWheat30L). Independent validation suggested that the overall accuracy and F1-score of ChinaWheat30L were 0.929 and 0.906, respectively, with weak variations across years. The mapped areas of winter wheat aligned well with agricultural statistics at provincial and municipal levels (<em>R<sup>2</sup></em> = 0.93 and 0.84). Furthermore, the ChinaWheat30L exhibited minimal classification bias across various landscapes and demonstrated accuracy improvements of 4–10% compared with counterpart products. In general, the total winter wheat planting areas remained stable at the national scale over the past two decades. Nevertheless, significant declines were observed in mountainous, arid, and highly urbanized regions, while increases were mostly clustered in the plain regions with suitable climate conditions and concentrated cropland fields. This research delivers winter wheat products at a national scale robustly and automatically over long timespans without ground labels, thereby offering new insights for spatiotemporal dynamic and food security analyses.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 163-179"},"PeriodicalIF":10.6,"publicationDate":"2025-05-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143899563","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Liangzhi Li , Ling Han , Yuanxin Ye , Yuming Xiang , Tingyu Zhang
{"title":"Deep learning in remote sensing image matching: A survey","authors":"Liangzhi Li , Ling Han , Yuanxin Ye , Yuming Xiang , Tingyu Zhang","doi":"10.1016/j.isprsjprs.2025.04.001","DOIUrl":"10.1016/j.isprsjprs.2025.04.001","url":null,"abstract":"<div><div>Deep learning demonstrates significant potential in enhancing the techniques of remote sensing image (RSI) matching. The current review delves into the incorporation of deep learning in RSI matching methods. Four predominant strategies are elucidated: area-based matching, feature-based matching, regression-based matching, and unsupervised learning methods. Area-based strategies concentrate on the quantification of similarity among image regions through sophisticated deep networks. Conversely, feature-based strategies are designed to detect, describe, and correspond salient features via comprehensive end-to-end networks. Regression-based matching methods leverage labeled data to train networks to identify correspondences. Unsupervised methods directly learn matching transformations in an end-to-end manner without labels. For each approach, representative methods, network architectures, loss functions, and modules are analyzed. Current challenges and future directions are provided, including needs for unified datasets, cross-modal loss functions, and end-to-end matching networks. This review offers researchers and practitioners systematic insights into deep learning advances for RSI matching. The discussion of methods, techniques, and research directions provides valuable reference for future research and application development in this important area.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 88-112"},"PeriodicalIF":10.6,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Comparison of gap-filling methods for generating landsat-like land surface temperatures under all-weather conditions","authors":"Jiali Guo , Jinling Quan , Wenfeng Zhan , Zhongguan Wen","doi":"10.1016/j.isprsjprs.2025.04.029","DOIUrl":"10.1016/j.isprsjprs.2025.04.029","url":null,"abstract":"<div><div>Thermal infrared remote sensors provide cost-effective and widespread land surface temperatures (LSTs) but often with spatiotemporal gaps due to discrete sampling and synoptic disturbance, greatly limiting their reliability and application. Current gap-filling methods have been primarily developed and validated for medium- to low-resolution LSTs; however, with rising demand for spatiotemporally continuous, high-resolution (tens of meters like Landsat) LSTs across disciplines, there is an urgent need to assess these methods’ applicability and uncertainty at higher spatial resolutions under a unified framework. In this study, we apply eight typical and hybrid methods, including temporal interpolation, spatiotemporal interpolation, weight-based fusion, learning-based fusion, and four standard annual temperature cycle (ATC)-based hybrid reconstructions, to fill gaps in irregularly spaced Landsat series over Weishan, Huairou, and Yulin, China. These sites represent cropland in a sub-humid plain, forest in a sub-humid mountain region, and grassland in the semi-arid Loess Plateau. We evaluate their performance in terms of spatiotemporal pattern, statistical accuracy, sensitivity to input data quality and distribution, and adaptability to different synoptic and surface conditions based on cloudy Landsat data and in-situ measurements. Results reveal that the enhanced ATC (EATC) method is optimal among these methods, capturing all-weather spatiotemporal dynamics at the Landsat scale with superior accuracy and robustness under various input, cloud, and ground conditions. In addition, the ATC-based hybrid methods do not necessarily improve the statistical accuracy over their respective typical ones. This comprehensive evaluation provides valuable insights into the selection of appropriate gap-filling methods for generating Landsat-like LSTs under all-weather conditions and highlights the need for further advancements especially in addressing abrupt changes in land cover types and temporal sparsity in high-resolution LST observations to improve accuracy, stability, and generality.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 113-130"},"PeriodicalIF":10.6,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143888263","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pengfei Tang , Shanchuan Guo , Lu Qie , Xingang Zhang , Hong Fang , Liang Wan , Jocelyn Chanussot , Peijun Du
{"title":"Tidal flat topography mapping with Sentinel time series using cross-modal sample transfer and deep learning","authors":"Pengfei Tang , Shanchuan Guo , Lu Qie , Xingang Zhang , Hong Fang , Liang Wan , Jocelyn Chanussot , Peijun Du","doi":"10.1016/j.isprsjprs.2025.04.017","DOIUrl":"10.1016/j.isprsjprs.2025.04.017","url":null,"abstract":"<div><div>Tidal flats are crucial components of coastal geomorphic systems, where the ocean meets the land. Timely and accurate topographic maps of tidal flats are essential for sustainable coastal management and development. Although satellite imagery-based inversion methods offer a cost-effective solution for constructing large-scale intertidal topography, their accuracy remains heavily dependent on the availability and quality of satellite data. Frequent cloudy and rainy weather in coastal areas presents significant challenges for extracting waterlines from optical images. To address these challenges, this study developed an integrated framework that leverages the complementary strengths of optical and Synthetic Aperture Radar (SAR) imagery, providing an innovative solution to accurately map tidal flat topographies at high spatial resolution. By utilizing the high-precision spatiotemporal distribution results of tidal flats extracted from optical images and integrating tidal constraints and temporal conditions, a cross-modal sample transfer strategy for Optical-SAR imagery was designed, which automatically generates a pseudo-sample library for SAR images. To optimize the automatic extraction of tidal flats in complex SAR imagery environments, we constructed a hybrid semantic segmentation network, UCTCNet. UCTCNet combines the local feature extraction capabilities of convolutional neural networks with the global information focus provided by attention mechanisms. ICESat-2 data was used as altimetry input based on the relationship between tidal flat elevations and inundation frequencies, which was combined with derived inundation frequency maps, low-tide imagery, and spectral indices to accurately invert tidal flat elevations using a random forest algorithm. Experimental results showed that the UCTCNet model demonstrated high potential in processing single-channel, high-noise, weak-feature Sentinel-1 SAR imagery, achieving an IoU of over 0.90, indicating strong performance in extracting high-level semantic features of tidal flats. The elevation inversion framework was validated along the entire coastal region of Jiangsu, China, for multi-temporal and multi-scene analysis. Further validation using generated topographic maps from unmanned aerial vehicle photogrammetry showed superior performance (RMSE = 0.24 m) compared to existing public tidal flat elevation data. The framework was also applied to derive DEMs from 2019 to 2023, revealing significant spatial and elevation changes in the North Jiangsu Radial Sand Ridges. The results further demonstrated the influence of various features, including inundation frequency maps, low-tide imagery, and spectral indices, on elevation inversion accuracy. The integration of S1 SAR data not only improved inversion accuracy but also helped address the limitations associated with discrete frequency data. These findings demonstrate that our proposed framework offers novel insights into high-resolution, large-sca","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 69-87"},"PeriodicalIF":10.6,"publicationDate":"2025-04-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143876716","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Huanyu Zhang , Tian Hu , Bo-Hui Tang , Kanishka Mallick , Xiaopo Zheng , Mengmeng Wang , Albert Olioso , Vincent Rivalland , Darren Ghent , Agnieszka Soszynska , Zoltan Szantoi , Lluís Pérez-Planells , Frank M. Göttsche , Dražen Skoković , José A. Sobrino
{"title":"Deep learning coupled with split window and temperature-emissivity separation (DL-SW-TES) method improves clear-sky high-resolution land surface temperature estimation","authors":"Huanyu Zhang , Tian Hu , Bo-Hui Tang , Kanishka Mallick , Xiaopo Zheng , Mengmeng Wang , Albert Olioso , Vincent Rivalland , Darren Ghent , Agnieszka Soszynska , Zoltan Szantoi , Lluís Pérez-Planells , Frank M. Göttsche , Dražen Skoković , José A. Sobrino","doi":"10.1016/j.isprsjprs.2025.04.016","DOIUrl":"10.1016/j.isprsjprs.2025.04.016","url":null,"abstract":"<div><div>Land surface temperature (LST) is a fundamental parameter in environmental and climatic studies. Over the past decades, various clear-sky LST retrieval methods have been developed, among which the temperature-emissivity separation (TES) algorithm prevails due to its good accuracy and the simultaneous retrieval of LST and land surface emissivity (LSE). However, TES relies on complete atmospheric profiles and radiative transfer calculations for atmospheric correction, which accumulates large uncertainties and requires intensive computation. In this study, we integrated the physical mechanisms of the split window (SW) and TES algorithms into the deep learning (DL) model, constructing the DL-SW-TES framework. This new framework directly retrieves LST from easily accessible parameters without requiring any prior knowledge of LSE information and atmospheric profiles. The DL-SW-TES framework was evaluated using both the simulation dataset and high-resolution ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) observations. The simulation analysis showed that the DL-SW-TES method achieved a root mean squared error (RMSE) of 1.05 K in LST retrieval and appeared robust across various uncertainty conditions. The evaluation of the ECOSTRESS LST estimates at the six radiometer sites revealed that the DL-SW-TES method achieved a better performance with an overall RMSE of 1.56 K and a bias of −0.06 K compared to the official ECO2LTSE product (with an RMSE of 1.94 K and a bias of −0.25 K). The nighttime ground measurements from the twelve pyrgeometer sites reaffirms the accuracy improvements achieved by the new model, with bias reduced by 0.7 K and RMSE reduced by approximately 0.3 K. LST estimates from DL-SW-TES and the ECO2LTSE product also present good consistency in terms of spatial patterns. The demonstrated advantage of the developed DL-SW-TES method over the traditional TES is attributed to its simplified input parameters and robustness to uncertainties in these parameters. We conclude that DL-SW-TES achieves improved accuracy compared to the traditional TES algorithm with significantly simplified input parameters and enhanced computational efficiency, standing as a promising approach for mapping clear-sky high-resolution LST at large scales from the future thermal missions. The source code and data are available at <span><span>https://github.com/cas222huan/DLSWTES</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 1-18"},"PeriodicalIF":10.6,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864752","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kaixu Bai , Ke Li , Songyun Qiu , Zhe Zheng , Penglong Jiao , Yibing Sun , Liuqing Shao , Chaoshun Liu , Xinran Li , Zhengqiang Li , Jianping Guo , Ni-Bin Chang
{"title":"SCAGAT: A scene-aware ensemble graph attention network for global PM2.5 pollution mapping via land–atmosphere interactions","authors":"Kaixu Bai , Ke Li , Songyun Qiu , Zhe Zheng , Penglong Jiao , Yibing Sun , Liuqing Shao , Chaoshun Liu , Xinran Li , Zhengqiang Li , Jianping Guo , Ni-Bin Chang","doi":"10.1016/j.isprsjprs.2025.04.019","DOIUrl":"10.1016/j.isprsjprs.2025.04.019","url":null,"abstract":"<div><div>The sparse and uneven distribution of ground-based air quality monitoring stations poses significant challenges for large scale PM<sub>2.5</sub> pollution mapping. Spatially heterogenous land–atmosphere interactions often lead to large uncertainties in satellite-based PM<sub>2.5</sub> estimations from global modeling strategies. To enhance global PM<sub>2.5</sub> mapping accuracy, particularly in poorly monitored regions, we propose a novel ensemble learning framework called the SCene-Aware ensemble Graph ATtention network (SCAGAT), which integrates locally trained PM<sub>2.5</sub> prediction models across regions using a graph attention network and transfer learning concept. Unlike popular global modeling strategy, SCAGAT first constructs thousands of site-specific PM<sub>2.5</sub> estimation models at individual monitoring station using the random forest (RF) method. For each target grid, raw PM<sub>2.5</sub> estimates are predicted by the 32 site-specific RF models with the most similar geographic scene attributes, characterized by nine variables relevant to haze pollution levels, land cover, and climate characteristic. A graph attention network then aggregates these initial estimates to produce an optimal PM<sub>2.5</sub> prediction through ensemble learning. By taking advantage of the strength of SCAGAT, global daily gap-free PM<sub>2.5</sub> concentrations over land from 2000 to 2021 were finally mapped based on a long-term gap-filled aerosol optical depth dataset. Cross-validation shows that SCAGAT achieves high global PM<sub>2.5</sub> modeling accuracy, with a correlation coefficient of 0.909 and a root-mean-squared error of 9.87 μg m<sup>−3</sup>. Intercomparison results demonstrate SCAGAT’s superiority over other widely used global modeling methods, reducing PM<sub>2.5</sub> modeling bias by 44.2 %, 12.7 %, 32.4 %, 44.4 %, and 48.3 % in China, the USA, Europe, India, and a global product, respectively. Overall, SCAGAT provides a robust solution for large-scale air quality mapping and effectively resolves data imbalance related low accuracy in poorly monitored areas by accounting for geographic scene similarity. Furthermore, this method can be readily adapted to other data-driven Earth observing applications facing similar challenges.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 19-35"},"PeriodicalIF":10.6,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864649","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hua Su , Jianchen Teng , Feiyan Zhang , An Wang , Zhanchao Huang
{"title":"Can satellite observations detect global ocean heat content change with high resolution by deep learning?","authors":"Hua Su , Jianchen Teng , Feiyan Zhang , An Wang , Zhanchao Huang","doi":"10.1016/j.isprsjprs.2025.04.018","DOIUrl":"10.1016/j.isprsjprs.2025.04.018","url":null,"abstract":"<div><div>The development of <em>in situ</em> observations has significantly improved ocean heat content (OHC) estimation. However, high-resolution OHC data remain limited, hindering detailed studies on mesoscale oceanic warming variability. This study used a deep learning method-Densely Deep Neural Network (DDNN) to reconstruct a high-resolution (0.25° × 0.25°) global OHC dataset for the upper 2000m ocean from 1993 to 2023, named the Ocean Projection and Extension Neural Network 0.25° (OPEN0.25°) product. This deep ocean remote sensing approach integrates multi-source remote sensing data, including Sea Surface Temperature (SST), Absolute Dynamic Topography (ADT), and Sea Surface Wind (SSW), alongside spatiotemporal coordinates and <em>in situ</em> observations. The DDNN model was trained using Argo-based gridded data and EN4-profile data, initially undergoing pre-training to assimilate large-scale oceanic features, followed by fine-tuning to enhance its accuracy in capturing mesoscale thermal structures. Our results demonstrate that the DDNN model achieves high accuracy across various depths. Particularly, OPEN0.25° can effectively capture detailed thermal variations in regions with complex dynamics, as well as the heat transfer processes within the ocean interior, outperforming traditional methods in resolution. The research highlights that, influenced by strong El Niño-Southern Oscillation (ENSO) events, OHC in the upper 700m of the Pacific Ocean potentially far exceeding expectations over the past decade. Through this study, OPEN0.25° has demonstrated its critical role in detecting and monitoring long-term changes in global OHC at high resolution.</div></div>","PeriodicalId":50269,"journal":{"name":"ISPRS Journal of Photogrammetry and Remote Sensing","volume":"225 ","pages":"Pages 52-68"},"PeriodicalIF":10.6,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864551","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}