Science of Remote Sensing最新文献

{"title":"High spatial resolution crop type and land use land cover classification without labels: A framework using multi-temporal PlanetScope images and variational Bayesian Gaussian mixture model","authors":"Minh Tri Le ,&nbsp;Khuong H. Tran ,&nbsp;Phuong D. Dao ,&nbsp;Hesham El-Askary ,&nbsp;Tuyen V. Ha ,&nbsp;Taejin Park","doi":"10.1016/j.srs.2025.100264","DOIUrl":"10.1016/j.srs.2025.100264","url":null,"abstract":"<div><div>Previous studies often combined high spatial resolution data (e.g., PlanetScope) with wider spectral range data (e.g., Sentinel-2) and relied on supervised classification methods to produce land use and land cover (LULC) maps. This study proposed a new unsupervised framework to generate crop type and LULC maps at high spatial resolution (&lt;5 m) using available PlanetScope data solely without requiring ground truths. We used PlanetScope surface reflectance images and their derived spectral indices during growing seasons to create multi-temporal input features, which were fed into an unsupervised Variational Bayesian Gaussian Mixture Model (VBGMM). The VBGMM, unlike the traditional unsupervised classification methods, (1) first estimated optimal parameters that are most suitable based on the input features and then (2) assigned pixels to the cluster with maximum posteriori probability of a mixture of several Gaussian distributions. The crop type and LULC maps were then generated by labeling the derived clusters using the best possible assignment method, referring to the existing crop type or LULC products. We evaluated the produced PlanetScope-based crop type and LULC maps using true labels, corresponding reference maps, and other unsupervised classification methods. The results demonstrated the robustness and effectiveness of the proposed framework in mapping crop types and LULC at 3–5 m pixels across various ecosystems, climate zones, and human-managed landscapes. The spatial patterns of PlanetScope-based maps were (1) highly comparable with all the reference datasets at 10–30 m spatial resolution and (2) better than the traditional GMM and K-means clustering methods. The VBGMM produced classification maps with high confidence, yielding class probabilities above 0.9 for over 90 % of all study areas. The area percentage for all crop type and LULC classes agreed well with their reference maps, with R² of 0.95 and RMSE of 1.04 %. The confusion matrices using true labels indicated that PlanetScope-based maps achieved a higher overall accuracy of 84 % than the supervised referenced maps of 81 %. Besides, the entropy comparison showed that our framework-based maps were better at capturing fine-scale features such as developed areas within cities that commonly mix with open space and vegetation, deforestation and cropland conversion in South America, smallholder croplands in Africa and Asia, and generating homogeneous crop fields in North America. This study further highlighted the potential for future research to implement our proposed framework to generate timely and extensive annotated datasets, which can be used for operationally training machine learning models to map crop types and LULC, track deforestation, detect wildfires, and delineate flooded areas at larger scales using medium/coarse Earth observations.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100264"},"PeriodicalIF":5.2,"publicationDate":"2025-07-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144739324","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":"2023 activity of Nyamulagira volcano monitored by SAR interferometric coherence","authors":"Marco Polcari ,&nbsp;Emanuele Ferrentino ,&nbsp;Charles Balagizi ,&nbsp;Diego Coppola ,&nbsp;Sébastien Valade","doi":"10.1016/j.srs.2025.100261","DOIUrl":"10.1016/j.srs.2025.100261","url":null,"abstract":"<div><div>In this work the multi-temporal InSAR coherence is exploited to analyze several phenomena during the 2023 Nyamulagira volcano activity in the Democratic Republic of Congo. Starting from March 2023 a significant increasing of the activity was observed with several eruptions and a lava lake inside the northeastern crater pit. SAR data acquired from January to December 2023 by Sentinel-1 missions were exploited to follow the temporal evolution of the volcanic activity by the analysis of any InSAR coherence variations. The aim is to detect any lava flows inside the caldera and overflowing the flanks of the volcano. In addition, the time needed for the lava to solidify is tentatively estimated by studying the statistical behavior of the coherence pattern ascribable to a lava flow. Experimental results show a significant lava effusion inside Nyamulagira caldera from the beginning of March, while in mid-May the increase of the activity induced an overflowing from the caldera, with a lava flow observed along the NW flank of the volcano. Then, starting from the end of May-early June, the activity started to return to low levels. The statistical analysis of the probability distributions suggest that after the crisis of May, the lava emitted during the eruption, follows a solidification and likely cooling phase with an approximate time to complete the process estimated in 45/50 days. These results were synergistically used with other data to support the GVO local monitoring team in managing the May 2023 Nyamulagira volcano crisis.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100261"},"PeriodicalIF":5.7,"publicationDate":"2025-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144711284","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":"Characterizing the influence of varying functional traits from remotely sensed data on forest productivity acquired from selected NEON sites","authors":"Paige T. Williams ,&nbsp;Valerie A. Thomas ,&nbsp;Randolph H. Wynne ,&nbsp;Karl F. Huemmrich ,&nbsp;David J. Harding ,&nbsp;K. Jon Ranson ,&nbsp;Petya K. Campbell ,&nbsp;Elizabeth M. Middleton","doi":"10.1016/j.srs.2025.100262","DOIUrl":"10.1016/j.srs.2025.100262","url":null,"abstract":"<div><div>Gross primary productivity (GPP) describes total photosynthesis (carbon fixation) in an ecosystem and is key to the global land carbon budget. To reduce uncertainties in carbon accounting for different forest ecosystems, it is crucial to analyze the health and productivity of forested ecosystems. Plant functional traits, which are a combination of morphological, physiological, and environmental characteristics, have been shown to be predictive of forest ecosystem carbon dynamics. This study aimed to assess how well GPP can be predicted by remotely quantified functional traits across varying forested ecosystems. Airborne remote sensing observations and in situ flux tower measurements used in this analysis were acquired from selected forested sites from the National Ecological Observatory Network (NEON) data portal. We investigated hyperspectral indices and lidar derived products as proxies of remotely sensed plant functional traits. Average midday GPP around the date of flight was calculated by developing a relationship between night respiration and temperature and removing that component from the net surface-atmosphere CO₂ exchange (NSAE). We applied multiple linear regression with a best subset approach for three trait classes: morphological and environmental traits from lidar, physiological traits from hyperspectral data, and a combined functional trait model. The best-performing model, using lidar and hyperspectral traits, included CHM mean, DSM standard deviation, PRI standard deviation, and WBI mean producing a R² of 0.87, an adjusted R² of 0.84, a PRESS R² of 0.75 and RMSE of 3.48 μmol CO₂/m²/s. Results show that a combination of plant functional traits are important predictors of forest productivity.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100262"},"PeriodicalIF":5.2,"publicationDate":"2025-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144721302","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":"Remote sensing monitoring of fluorescent dissolved organic matter in Admiralty Bay: fusion of multi-source signal removal and machine learning","authors":"Ruiwu Zhang ,&nbsp;Ruru Deng ,&nbsp;Jun Ying ,&nbsp;Jiayi Li ,&nbsp;Yu Guo ,&nbsp;Junying Yang ,&nbsp;Cong Lei","doi":"10.1016/j.srs.2025.100260","DOIUrl":"10.1016/j.srs.2025.100260","url":null,"abstract":"<div><div>Fluorescent dissolved organic matter (fDOM), a fluorescent component of dissolved organic matter (DOM), plays a crucial role in tracing pollution pathways in marine environments. While remote sensing has been used to monitor fDOM changes, the impact of multi-source interference has often been overlooked, limiting the accuracy of inversion results. In this study, based on fDOM measurements from Admiralty Bay and from the perspective of optical physical mechanisms, we eliminated atmospheric effects, surface reflection, solar-induced fluorescence (SIF), Raman scattering, and particle absorption from remote sensing reflectance (<span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub><mrow><mo>(</mo><mi>λ</mi><mo>)</mo></mrow></mrow></math></span>). This preprocessing improved the stability of <span><math><mrow><msub><mi>R</mi><mrow><mi>r</mi><mi>s</mi></mrow></msub><mrow><mo>(</mo><mi>λ</mi><mo>)</mo></mrow></mrow></math></span>, enhancing the reliability of subsequent fDOM inversion. Based on the corrected reflectance, three sensitive wavelengths highly correlated with fDOM were selected. Five machine learning models—Random Forest (RF), XGBoost, Classification and Regression Trees (CART), Gradient Boosting Regression (GBR), and AdaBoost—were then applied for fDOM inversion, with XGBoost achieving the best performance. The inversion results revealed that fDOM concentrations in Admiralty Bay were highest in the western and coastal areas, gradually increasing toward the center, exhibiting a locally clustered distribution. This study demonstrates the effectiveness of combining physical and data-driven methods for fDOM inversion, providing a foundation for long-term monitoring of dissolved organic matter in polar marine environments.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100260"},"PeriodicalIF":5.7,"publicationDate":"2025-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144703011","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":"Estimation of aboveground biomass in Tajikistan based on upscaling extrapolation of UAV and Sentinel-2 multi-source data synergy","authors":"Lina Hao ,&nbsp;Huping Ye ,&nbsp;Shuang He ,&nbsp;Xinyu Zhang ,&nbsp;Dalai Bayin ,&nbsp;Mustafo Safarov ,&nbsp;Mekhrovar Okhonniyozov ,&nbsp;Xiaohan Liao","doi":"10.1016/j.srs.2025.100259","DOIUrl":"10.1016/j.srs.2025.100259","url":null,"abstract":"<div><div>Grasslands constitute the largest terrestrial ecosystem, currently sequestering significant amounts of atmospheric carbon and playing a critical role in climate change mitigation and the global carbon cycle. Tajikistan is a key representative of Central Asian, effective and accurate grassland Above-ground biomass (AGB) monitoring in Tajikistan is crucial for sustainable management. However, the relevant research remains markedly limited. Here we develop a dynamic sampling scale-up method for AGB estimation by integrating multi-source data from Sentinel-2 MSI, Unmanned Aerial Vehicle (UAVs), and ground observations, which enabled efficient and accurate AGB estimation across Tajikistan. Specifically, our analyses combine UAV and Sentinel-2 multispectral imagery with field-measured data to construct and optimize eight models for AGB estimation at different spatial scales, and apply the dynamic sampling scale-up method to enhance estimation accuracy. We find that: (1) At the UAV scale, the Extra Trees model achieves the highest accuracy (R² = 0.88, RMSE = 52.72 g ·m⁻²), whereas at the Sentinel-2 scale, the Support Vector Machine (SVM) model performs best (R² = 0.80, RMSE = 158.43 g ·m⁻²); (2) Texture features are the most crucial features for grassland AGB estimation; (3) The scale-up method improves the accuracy of Sentinel-2-derived AGB estimations, enabling more detailed spatial representation of AGB distribution. Our results demonstrate that coordinated multi-source monitoring elucidates the environmental controls on grassland AGB and provides a robust framework for conservation and sustainable management of grassland ecosystems under current and future climate scenarios.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100259"},"PeriodicalIF":5.7,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144703010","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":"Landsat Next current design for geological remote sensing: VNIR-SWIR-TIR data continuity and new opportunities","authors":"Bruno Portela,&nbsp;Harald van der Werff,&nbsp;Christoph Hecker,&nbsp;Mark van der Meijde","doi":"10.1016/j.srs.2025.100258","DOIUrl":"10.1016/j.srs.2025.100258","url":null,"abstract":"<div><div>Landsat Next, the proposed mission in NASA's Landsat program planned for 2031, is designed to extend the legacy of Landsat 8–9 and Sentinel-2 in the visible-near and shortwave infrared and to introduce operational thermal infrared capabilities comparable to ASTER. As the first multispectral spaceborne sensor to combine visible-near, shortwave, and thermal infrared coverage since ASTER, it presents a unique opportunity to reestablish long-term geological remote sensing continuity.</div><div>In this study, we assess whether Landsat Next, in its currently published design, can replicate or even improve geological information derived from Sentinel-2 and ASTER. We simulate Landsat Next imagery using airborne hyperspectral datasets acquired over two well-characterised mineral systems in the Yerington district, Nevada (USA), generating equivalent datasets for Sentinel-2 and ASTER to enable sensor-level comparison without environmental influences. By adapting established band ratios and applying spectral-only and spectral-spatial resampling when simulating Landsat Next data, we isolate the influence of Landsat Next's band configuration and resolution.</div><div>Our results confirm that Landsat Next replicates key mineralogical patterns observed in Sentinel-2 and ASTER products. Moreover, it enables enhanced discrimination in zones of spectrally overlapping alteration, especially where its higher spectral or spatial resolution improves mineral identification.</div><div>By replicating established band ratio products while enhancing the detection of key mineralogical features, Landsat Next represents the first spaceborne sensor since ASTER that can potentially deliver continuous multispectral information across the visible-near, shortwave, and thermal infrared ranges, supporting future geological remote sensing studies.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100258"},"PeriodicalIF":5.7,"publicationDate":"2025-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144655536","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":"An integrated object-based-deep learning approach applied for mapping armed conflict impacts and land scars","authors":"Bakhtiar Feizizadeh ,&nbsp;Mohammad Kazemi Garajeh ,&nbsp;Mohsen Makki ,&nbsp;Tobia lakes ,&nbsp;Murat Yakar ,&nbsp;Amin Naboureh ,&nbsp;Kolja Thestorf","doi":"10.1016/j.srs.2025.100257","DOIUrl":"10.1016/j.srs.2025.100257","url":null,"abstract":"<div><div>The increasing number of wars and armed conflicts worldwide demands the development of an efficient, cost-effective, and transferable methodological framework for mapping environmental impacts and land scars. This state-of-the-art research develops a novel and transferable data-driven approach for mapping war and armed conflict scars and impacts. A variety of war land scars were determined for the Iran-Iraq war using integrated object-based image analysis (OBIA) and deep learning convolutional neural networks (DL-CNNs). Then, the efficiency and transferability of the approach are examined for the Syrian civil war and the Karabakh (Azerbaijan-Armenia) conflict. To validate the results and examine the transferability of the approach and its efficiency, an integrated fuzzy synthetic evaluation (FSE) and quantity and allocation disagreement index (QADI) were applied. The overall result of an accuracy assessment using FSE-QADI yielded an accuracy of 0.96 for the obtained land scars in the case of the Iran-Iraq war, and the transferability of the developed OBIA-DL-CNNs was demonstrated with overall accuracies of 0.93 and 0.94 for the Syrian and Karabakh land scars, respectively. Recent progress in earth observation technology and the availability of a variety of products demand the development of efficient and transferable data-driven approaches. Thus, as a state of the art, the current study makes a significant contribution to the domain of remote sensing by developing an integrated transferable approach of OBIA-DL and FSE-QADI. The proposed approach will support authorities and stakeholders in detecting and mapping the impacts of armed conflicts in battle areas to enhance the environmental aspect of armed conflict operations, which may be helpful in preventing future conflicts.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100257"},"PeriodicalIF":5.7,"publicationDate":"2025-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144655537","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":"Temporal aggregation of point clouds improves permanent laser scanning of landslides in forested areas","authors":"Ronald Tabernig ,&nbsp;William Albert ,&nbsp;Hannah Weiser ,&nbsp;Patrick Fritzmann ,&nbsp;Katharina Anders ,&nbsp;Martin Rutzinger ,&nbsp;Bernhard Höfle","doi":"10.1016/j.srs.2025.100254","DOIUrl":"10.1016/j.srs.2025.100254","url":null,"abstract":"<div><div>Permanent laser scanning has recently developed as a technology to monitor landslides by repeatedly acquiring point clouds at short intervals (e.g., sub-hourly). In such areas, forests can hinder the capture of ground points and reduce the quantity and thus the spatial coverage of direct surface change information. The objective of this study is to evaluate the effectiveness of aggregating sequential point clouds for improved point cloud analysis. A forested landslide that primarily consists of <em>Pinus cembra</em> and <em>Picea abies</em> is investigated using permanent laser scanning data comprising 600 scans with an acquisition frequency of three hours. With our application of tree trunk tracking, we demonstrate how temporal aggregation improves tree trunk representation, thereby enabling quantification of landslide displacement. Corresponding trunks are matched and tracked throughout the full time series to compute the 3D displacements of the trunks. The effects of temporal aggregation are analysed by applying it to a digital replica of the study site, which is generated by virtual laser scanning. This targeted temporal aggregation approach increased the number of detectable trunks in forested areas by a factor of 5-6. Similarly, the number of trunk matches across the time series increased by a factor of 5. Performance gains plateaued after three aggregated scans. By using a permanently installed terrestrial laser scanner that repeatedly scans through the canopy, our method allows direct quantification of 3D landslide displacements in densely forested terrain. Our findings demonstrate that temporal aggregation of point clouds significantly increases the applicability and performance of continuous, long-range laser scanning-based monitoring of forested environments.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100254"},"PeriodicalIF":5.7,"publicationDate":"2025-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144655535","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":"Accounting for seasonal retrieval errors in the merging of multi-sensor satellite soil moisture products","authors":"Pietro Stradiotti,&nbsp;Alexander Gruber,&nbsp;Wolfgang Preimesberger,&nbsp;Wouter Dorigo","doi":"10.1016/j.srs.2025.100242","DOIUrl":"10.1016/j.srs.2025.100242","url":null,"abstract":"<div><div>ESA CCI soil moisture (SM) merges satellite microwave remote sensing datasets by means of their inverse-uncertainty weighted average. Estimates of uncertainty are produced with Triple Collocation Analysis (TCA) and assume a constant level of noise for the entire sensor period. However, errors in soil moisture retrievals vary throughout the year, since many impacting environmental parameters are characterized by a seasonality of their own. Here, we attempt to quantify this seasonal component and assess the impact of time-variant uncertainty estimates on the quality of merged soil moisture. We derive a long-term error variance estimate for three satellite products (from ASCAT, AMSR2, and SMAP) per day of year using a sliding window of 90 days. Merging weights climatologies are subsequently obtained as the inverse of such uncertainty. We analyse the impact of the modified approach by comparison with the merging based on stationary uncertainties/weights. The two key findings are that (i) the merged soil moisture estimates do not differ significantly between the stationary and the seasonal merging because seasonal uncertainty variations, e.g. caused by vegetation cover, usually affect all satellite missions in a similar way and thus cause only marginal changes in their relative weighting; yet, (ii) an evaluation against in situ data suggests that the estimated uncertainties of the new merged product are more representative of their seasonal behaviour. Based on these findings, we conclude that using a seasonal TCA approach can add value to merged products such as the ESA CCI SM by providing a more realistic characterization of dataset uncertainty – in particular its temporal variation.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100242"},"PeriodicalIF":5.7,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687106","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":"Early-season delineation of agricultural fields using a fully convolutional multi-task network and satellite images","authors":"Ghaith Amin ,&nbsp;Thomas Oberlin ,&nbsp;Valerie Demarez","doi":"10.1016/j.srs.2025.100256","DOIUrl":"10.1016/j.srs.2025.100256","url":null,"abstract":"<div><div>The accurate delineation of agricultural fields is essential for crop condition monitoring, yield estimation, and irrigation management. While satellite imagery offers a cost-effective solution for large-scale mapping, traditional boundary detection methods often face challenges such as false edges and suboptimal segmentation. To address these limitations, we present an operational multi-task deep learning approach using the ResUNet-a d7 model, leveraging freely available Sentinel-2 Level-3A data, which ensures enhanced temporal and spatial consistency for large-scale applications. Additionally, we introduce a novel post-processing method based on Gaussian Mixture Models (GMM) to refine boundaries between adjacent fields, enabling precise extraction of individual fields. Extensive assessments were conducted across 14 geographically diverse sites spanning two years of data, along with spatio-temporal experiments to evaluate the model's transferability to unseen regions and new acquisition year. The results demonstrate that the ResUNet-a d7 model achieves a high weighted F1 score of approximately 92 %, highlighting its strong performance and reliability. This study provides a scalable and robust solution for early-seasonagricultural field delineation, paving the way for operational applications in irrigation detection, crop type identification, and crop health monitoring.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"12 ","pages":"Article 100256"},"PeriodicalIF":5.7,"publicationDate":"2025-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144695030","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}