Remote Sensing of Environment最新文献

{"title":"TIF: A time-series-based image fusion algorithm","authors":"Kexin Song, Zhe Zhu, Shi Qiu, Pontus Olofsson, Christopher S.R. Neigh, Junchang Ju, Qiang Zhou","doi":"10.1016/j.rse.2025.115035","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115035","url":null,"abstract":"We developed a Time-series-based Image Fusion (TIF) algorithm to generate 10-m surface reflectance time series by synthesizing Landsats 8/9 and Sentinel-2 A/B data. Unlike traditional methods that rely on image pairs or thematic maps, TIF extracts all valid pixel-level observation pairs across time to build per-pixel linear regression models. This approach captures the spectral relationships between sensors while accounting for land surface dynamics. A temporal weighting scheme and an iterative refinement strategy improves the fusion process, yielding reusable coefficients that support efficient, scalable 10-m time-series generation. TIF was applied to all Landsat multispectral bands, using native 10-m Sentinel-2 bands (Blue, Green, Red) and resampled bands (NIR and SWIR1/2) for visual assessment, with quantitative accuracy evaluated at the original Sentinel-2 resolutions. Experiments across five U.S. sites show TIF consistently outperforms state-of-the-art methods like STARFM, FSDAF 2.0, Sen2Like, and ESRCNN. For instance, TIF demonstrated a reduction in RMSE by 24 % and an increase in SSIM by 6 % compared to FSDAF 2.0 and ESRCNN, and outclassed STARFM and Sen2Like, which showed weaker results across all metrics. In multi-date change detection, TIF-predicted images achieved a mean F1 score of 0.70 and a mean disagreement rate of 0.05 against reference maps. TIF offers a potential practical and efficient pathway for creating 10-m versions of NASA's HLS products, opening new opportunities for fine-scale, time-sensitive Earth observations.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"8 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145203975","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":"Backscatter-sensitive retrieval of iceberg areas from Sentinel-1 Extra Wide Swath SAR data","authors":"Henrik Fisser, Anthony P. Doulgeris, Wolfgang Dierking","doi":"10.1016/j.rse.2025.115042","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115042","url":null,"abstract":"We present the first systematic study on Arctic iceberg area retrieval from Sentinel-1 Extra Wide Swath SAR data. Our dataset contains 4014 Arctic icebergs in open water. To detect icebergs in Sentinel-1 images, we applied a constant false alarm rate (CFAR) algorithm. Icebergs were matched in co-located Sentinel-1 and optical Sentinel-2 acquisitions for a period from May to September in the years 2016 to 2023. The Sentinel-1 iceberg areas are moderately correlated with the Sentinel-2 reference areas, with Pearson’s r-values of 0.65 at HH- and 0.69 at HV-polarization. The CFAR algorithm mostly overestimates iceberg areas in both channels, but in some cases underestimates iceberg areas in the HV channel when the iceberg backscatter is low. The 10th and 90th percentiles of the relative error are -36% and +569% at HH-, and -37% and +349% at HV-polarization. In addition to a resolution-imposed size-dependency of the error, we find that the HH ocean clutter is moderately negatively correlated with the relative error (Pearson’s r-value: -0.64). Additionally, we observe weak correlations between the HH (HV) iceberg backscatter and the Sentinel-2 iceberg area, with a Pearson’s r-value of 0.34 (0.26). We utilize the analyzed relationships to predict iceberg areas from Sentinel-1 data, using a gradient boosting regression. Backscatter-sensitive retrieval models yield more accurate iceberg areas than backscatter-insensitive models. In the HH (HV) channel, the former models reduce the mean absolute error by 58% (49%). Backscatter-sensitive retrieval of iceberg areas will foster consistent SAR-based observations of Arctic iceberg areas during summer months. Future studies need to quantify the combined impact of iceberg area retrieval and iceberg detection performance on area distributions and total iceberg areas.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"99 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145195059","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":"APPLE-GO: Modeling high-spatial resolution forest canopy reflectance with effect of Adjacent Pixels using Path Length Extended Geometric Optical theory","authors":"Qunchao He, Siqi Yang, Naijie Peng, Wenjie Fan, Xihan Mu, Biao Cao, Dechao Zhai, Zhicheng Huang, Huazhong Ren, Guangjian Yan","doi":"10.1016/j.rse.2025.115043","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115043","url":null,"abstract":"Forests are the key component of terrestrial ecosystems, playing a vital role in the global carbon and water cycles as well as in climate change. Satellite remote sensing imagery has the advantage of quantitatively monitoring and assessing the health status of forest canopies at large scales. With the improvement in spatial resolution of satellite sensors, it has become feasible to conduct quantitative research at high spatial resolutions (< 10 m). However, classic physical models that are based on simplified assumptions and only account for the radiative transfer process within the target pixel face challenges in supporting quantitative analysis at high-resolution scales, as high-resolution pixels are subject to significant radiative influences from adjacent pixels. In this study, we propose a high-spatial resolution forest canopy reflectance model, APPLE-GO, which comprehensively considers the shading effect and cross-radiation caused by adjacent pixels. The two-dimensional path length distribution (2-PLD) method is used to calculate the area fractions of each component, while shading factors are introduced to quantitatively calculate the reductions in the area fractions of sunlit components due to adjacent pixels. Multiple scattering energy is calculated based on the spectral invariant theory and the eight-neighborhood convolution algorithm. The bi-directional reflectance factor (BRF) calculated by the APPLE-GO model was evaluated against the three-dimensional (3D) radiative transfer model LESS, yielding RMSEs/RRMSEs of 0.008/10.2 % and 0.054/15.9 % in the red and near-infrared (NIR) bands, respectively. The model was also validated with satellite observations, showing RMSEs below 0.01 (RRMSE <27 %) for larch forests and under 0.017 (RRMSE <35 %) for mixed forests in the visible bands. These results demonstrate that the proposed model can accurately calculate the BRF in the nadir viewing direction, highlighting its potential for extracting vegetation parameters from high-resolution remotely sensed imagery.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"102 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145195060","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":"Satellite data-driven estimation of daily and 500 m net ecosystem exchange over China during 2003–2020","authors":"Xian Wang, Yongqiang Zhang, Xuanze Zhang, Shaoyang He, Dongdong Kong, Jing Tian, Haoshan Wei, Longhao Wang, Yu Quan, Yufeng Zheng, Yingping Wang","doi":"10.1016/j.rse.2025.115047","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115047","url":null,"abstract":"The terrestrial biosphere plays a critical role in mitigating climate change by absorbing anthropogenic CO₂. However, accurately quantifying the net ecosystem exchange (NEE), which is a key indicator for monitoring carbon sequestration of terrestrial ecosystems, remains a major challenge. Widely used products from large-scale ecosystem models and atmospheric inversions operate at coarse resolutions (0.25° or greater), which hinders the ability to resolve the carbon dynamics of heterogeneous landscapes and poses a significant challenge to understanding the impact of climate and land-use changes. In this study, a remote sensing data-driven water‑carbon coupling model with the incorporation of terrestrial carbon cycle processes, Penman–Monteith–Leuning Version 2 Carbon (PML<img alt=\"single bond\" src=\"https://sdfestaticassets-us-east-1.sciencedirectassets.com/shared-assets/55/entities/sbnd.gif\" style=\"vertical-align:middle\"/>V2C), is developed for estimating daily NEE over China at a 500 m resolution. The parameters of PML-V2C model were well calibrated against observations from 41 eddy covariance (EC) flux tower sites across nine plant functional types (PFTs) over China, demonstrating a strong performance for daily NEE estimates (<em>r</em> = 0.71, RMSE = 1.85 g C m⁻² day⁻¹). The model is only slightly degraded when compared with independent global FLUXNET data across 157 sites, demonstrating its robustness and transferability across diverse climates and biomes. Applying the model from 2003 to 2020, our product revealed a significant enhancement of China's terrestrial carbon sink with an increasing trend of 0.041 Tg C yr⁻² (p &lt; 0.01). This enhancement was primarily driven by the increasing GPP from forests in Southern China, grasslands in Northern China, and croplands across the East and North China Plain. Our high-resolution, process-based NEE product driven by satellite data provides a new evaluation of the effectiveness of ecosystem restoration and land management policies, offering valuable insights for achieving national carbon neutrality goals.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"6 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145189091","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":"Unrevealing site-dependent relationship between solar-induced chlorophyll fluorescence and gross primary productivity using the terrestrial ecosystem carbon cycle simulator","authors":"Haoran Liu, Zoe Amie Pierrat, Hamid Dashti, Min Chen","doi":"10.1016/j.rse.2025.115052","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115052","url":null,"abstract":"Solar-induced fluorescence (SIF), a small light signal emitted during the photosynthetic process, is a powerful tool for tracking gross primary productivity (GPP) across scales, particularly in evergreen needleleaf forests, which are traditionally challenging to monitor with remote sensing. Terrestrial biosphere models (TBMs) that incorporate a SIF module can help address the spatiotemporal limitations of field and satellite observations and explain variations in the SIF-GPP relationship across different temporal scales and ecosystems. In this study, we developed TECs-SIF, a TBM that integrates the spectral invariant property-based radiative transfer model across leaf and canopy scales to simultaneously simulate canopy SIF emissions and GPP and investigate how the SIF-GPP relationship varies across forest ecosystems. We calibrated and validated TECs-SIF using data from three evergreen needleleaf forest and one deciduous broadleaf forest AmeriFlux sites: Southern Old Black Spruce (CA-Obs), Delta Junction (US-xDJ), Niwot Ridge Forest (US-NR1), and University of Michigan Biological Station AmeriFlux site (US-UMB). Our results show that TECs-SIF as a promising tool that accurately simulates SIF and GPP across various temporal scales (Hourly: SIF: R² = 0.48–0.87, Root Mean Squared Error (RMSE) = 0.03–0.12 W/m²/μm/sr; GPP: R² = 0.60–0.79, RMSE = 1.82–5.31 μmol/m²/s; Daily: SIF: R² = 0.64–0.91, RMSE = 0.02–0.09 W/m²/μm/sr; GPP: R² = 0.89–0.97, RMSE = 0.51–2.05 μmol/m²/s), captures nonlinear relationships at hourly intervals, and linear trends at daily and monthly scales. Meanwhile, SIF-GPP relationship is site-dependent across temporal scales, influenced by canopy structure (e.g., CI) and leaf traits.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"94 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145189035","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":"Fusing Sentinel-1 and Sentinel-2 data with diffusion models for cloud removal","authors":"Jiajun Cai, Bo Huang, Hao Liu","doi":"10.1016/j.rse.2025.115049","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115049","url":null,"abstract":"Cloud cover significantly hinders the use of optical remote sensing data, such as Sentinel-2, by obscuring critical information needed for environmental monitoring. This study introduces Enhanced Diffusion Model for Cloud Removal (EDM-CR), an enhanced cloud removal framework that integrates Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 optical imagery using efficient diffusion models. Our approach features three key novel aspects: (1) a forward diffusion process conditioned on both the previous timestep image and cloudy Sentinel-2 data, simulating cloud addition to improve the backward diffusion process; (2) a two-branch backward diffusion process conditioned on the current timestep image, cloudy Sentinel-2 data, and Sentinel-1 data, enhancing cloud removal fidelity and restoration efficiency; and (3) a modified Learned Perceptual Image Patch Similarity (LPIPS) loss function that incorporates all 13 Sentinel-2 spectral bands, ensuring comprehensive spatial information preservation. The framework is trained using paired and co-registered Sentinel-1 and Sentinel-2 images (both cloudy and cloud-free) from the SEN12MS-CR dataset and validated on a large set of unseen cloudy images. Experimental results demonstrate that our method outperforms five state-of-the-art cloud removal techniques. Furthermore, the cloud-removed Sentinel-2 images are used as year-round inputs for farmland segmentation in the Netherlands, providing temporal context that improves segmentation accuracy compared to using limited timesteps. These findings underscore the effectiveness of our diffusion model framework in integrating multi-sensor data for robust cloud removal and highlight the benefits of incorporating temporal information for accurate semantic segmentation of farmland.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"4 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145183168","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":"Breaking the limitations of scenes and sensors variability: A novel unsupervised domain adaptive instance segmentation framework for agricultural field extraction","authors":"Ren Wei, Lin Yang, Xiang Li, Chenxu Zhu, Lei Zhang, Jie Wang, Jie Liu, Liming Zhu, Chenghu Zhou","doi":"10.1016/j.rse.2025.115051","DOIUrl":"https://doi.org/10.1016/j.rse.2025.115051","url":null,"abstract":"Extraction of agricultural field parcels is of great importance for agricultural condition monitoring, farm management, and food security. Several methods have been developed to map the distribution of agricultural field parcels, among which deep learning-based supervised learning is increasingly employed. Nevertheless, advanced deep learning models face two major limitations: limited ability to generalize across different spatial，temporal and sensor contexts with varying scene and object characteristics, and high requirement for annotated datasets to support training and validation. To address this challenge, we introduce a novel unsupervised domain adaptation (UDA) framework (UDA-Field Teacher, UDA-FT) for agricultural field parcel instance segmentation, which is designed to transfer knowledge from labeled source domains to unlabeled target domains. UDA-FT is based on the Mask R-CNN framework and incorporates a target-oriented teacher model and a cross-domain student model. This cross-domain student model embeds an image adaptation module and an instance adaptation module, employing adversarial learning strategies to mitigate cross-domain distribution differences. Additionally, we propose a consistency mutual learning module based on soft pseudo-label technology, overcoming the limitations of traditional hard pseudo-labeling in confidence threshold selection and improving model robustness in the target domain. Furthermore, to address the difficulty in generating independent instance labels for densely packed agricultural field parcels and capturing spatial contextual relationships during soft pseudo-label generation, we propose two data augmentation methods, namely CutMatch (CM) and LeakyMask (LM). We adopted the proposed framework on cross-scene and cross-sensor datasets to evaluate its effectiveness and robustness under different scenes. Quantification and visualization results demonstrate our UDA-FT outperforms existing domain adaptation methods for cross-scene and cross-sensor agricultural field parcels across all metrics. Ablation studies highlight the substantial impact of strong data augmentation on model performance, emphasizing the importance of learning from out-of-distribution data. As an innovative application of unsupervised domain adaptation in agricultural field parcel instance segmentation, this research provides a novel method for domain shift in agricultural remote sensing imagery, enabling more accurate field instance segmentation with significant implications for global agriculture.","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"273 1","pages":""},"PeriodicalIF":13.5,"publicationDate":"2025-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145189034","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":"Mapping and early warning of hidden landslides under forests: A case in Lantau, Hong Kong","authors":"Ziyuan Li ,&nbsp;Guoqiang Shi ,&nbsp;Songbo Wu ,&nbsp;Tao Li ,&nbsp;Zhong Lu ,&nbsp;Xiaoli Ding","doi":"10.1016/j.rse.2025.115039","DOIUrl":"10.1016/j.rse.2025.115039","url":null,"abstract":"<div><div>The intensification of extreme rainfall has exacerbated widespread landslide hazards, particularly in tropic and subtropic regions. Hong Kong—the world's most densely populated city situated on steep, forested terrain—faces chronic landslide risks that are challenging to monitor with conventional Aperture Radar Interferometry (InSAR), as hillslope failures are typically small and hidden beneath dense canopy. This study develops a novel detection framework integrating: 1) HARMIE (Homogeneous Amplitude-phase RefineMent for local Inconsistent phase Estimation), which enhances localized phase variability for subtle displacement retention; and 2) a phase gradient-based detection approach, linking slope responses with extreme rainfall. Simulated and real-data experiments demonstrate that HARMIE outperforms conventional methods by better preserving localized phase detail and magnitude. Using high-resolution ascending and descending Lutan-1 (LT-1) SAR datasets (July 2023–October 2024) over Lantau Island, Hong Kong, the framework mapped widespread hillslope failures triggered by the October 2023 extreme rainfalls, achieving a 27 % higher recognition rate than amplitude-homogeneity-based detection, with notable improvements in capturing subtle failures as narrow as ∼10 m. Ten active landslides concealed beneath forests were also pinpointed. Beyond detection, our analysis reveals that prolonged antecedent rainfall drives seasonal progressive creep on minor slopes and, for certain slopes, may interact with extreme rainfall to accelerate destabilization. This study represents the first InSAR-based mapping of small, forest-covered landslides in Hong Kong using L-band SAR, offering new insights into hillslope destabilization in forested mountainous terrain and advancing the development of landslide early-warning systems in such regions.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"331 ","pages":"Article 115039"},"PeriodicalIF":11.4,"publicationDate":"2025-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145154042","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":"A practical angular spectrum method to estimate ocean surface wave directions and wavelengths using Sentinel-2 MSI imagery","authors":"Lifeng Wang ,&nbsp;Yingcheng Lu ,&nbsp;Chuanmin Hu","doi":"10.1016/j.rse.2025.115044","DOIUrl":"10.1016/j.rse.2025.115044","url":null,"abstract":"<div><div>High spatial-resolution Sentinel-2 Multi-Spectral Instrument (MSI) images often show image features due to surface waves, swells, fronts, internal waves, eddies, and currents under different sunglint and skyglint reflections. Here, MSI images are used to estimate wave direction and wavelength through a simple but practical method using Angular spectrum analysis (ASA). The method does not require the complex transfer function between the image domain and the physical domain, but is based on the MSI-derived Fast Fourier Transform (FFT) spectrum that reveals the wave direction, whose 180^o ambiguity is then removed by the inter-band time lag between the B04 and B08 bands (0.74 s). The wavelength of different sea waves (wind wave and swell) can also be estimated by quantifying the MSI FFT images. There is a statistically significant linear relationship between MSI-derived and buoy-measured wave direction and wavelength (<em>N</em> = 144) with low bias. Further analyses according to wave age criterion show a better statistical relationship for wind waves (<em>N</em> = 62) than swells (<em>N</em> = 82). Comparison with the hourly wind products of the fifth generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5) shows RMS differences of &lt;1.4 m/s and 14.4^o for wind speed and direction, respectively. These results demonstrate that, under certain observing conditions, high spatial-resolution optical remote sensing images can provide relatively accurate estimates of surface wave directions and wavelengths as well as wind speeds, while operational applications still require further work.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"331 ","pages":"Article 115044"},"PeriodicalIF":11.4,"publicationDate":"2025-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145154040","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":"Synergizing machine learning and interpolation methods: A Stacking framework for global-scale satellite soil moisture gap filling","authors":"Jiaming Rong ,&nbsp;Jiangyuan Zeng ,&nbsp;Kun-Shan Chen ,&nbsp;Hongliang Ma ,&nbsp;Pengfei Shi ,&nbsp;Husi Letu ,&nbsp;Xiang Zhang ,&nbsp;Xihui Gu ,&nbsp;Haiyun Bi ,&nbsp;Chunlin Zhang","doi":"10.1016/j.rse.2025.115040","DOIUrl":"10.1016/j.rse.2025.115040","url":null,"abstract":"&lt;div&gt;&lt;div&gt;Satellite-derived soil moisture (SM) products frequently encounter extensive data gaps that significantly limit their practical utility, necessitating the development of robust gap-filling techniques to generate SM datasets with enhanced accuracy and continuous spatiotemporal coverage. Existing studies have typically relied on single machine learning or interpolation methods to fill SM gaps at regional scales. Machine learning approaches excel at filling missing values in large regions but tend to smooth out important local SM features, while the interpolation methods perform well in areas with low levels of missing data, but exhibit significant uncertainty in regions with large amounts of continuously missing data. These two kinds of approaches show potential complementarity and could together contribute to a more robust gap-filling method, which however have rarely been investigated. To fill this research gap, we established a novel SM gap-filling method by synergizing the advantages of machine learning for large-scale gap filling and the excellent gap-filling performance of interpolation in localized areas using the Stacking method at a global scale. The proposed approach integrates four base models including three machine learning techniques namely Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Feed-forward Neural Network (FNN), and one interpolation method known as Natural Neighbor Interpolation (NNI), and employs the Least Absolute Shrinkage and Selection Operator (LASSO) as the meta model. We compared the Stacking method and individual approaches in filling ESA CCI missing SM data, and validated the gap-filled SM using extensive ground SM from 1086 sites worldwide. The results indicate: (1) RF performs the best among the six selected machine learning methods, and its overall accuracy at a global scale is higher than that of interpolation methods. The feature importance analysis by SHapley Additive exPlanations (SHAP) indicates ERA5 SM, NDVI, and Global Aridity Index have high importance in the RF gap-filling model; (2) NNI is the best performing approach among the four selected interpolation methods, and it demonstrates better performance than machine learning methods in localized areas where the original SM data is relatively abundant; (3) Stacking is an effective method for SM gap filling on a global scale, with an averaged ubRMSE of 0.017 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt;, RMSE of 0.022 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt;, Bias of 0.006 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt;, and &lt;em&gt;R&lt;/em&gt; of 0.87 against the original ESA CCI SM, and it reduces the RMSE by 0.009 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt;, ubRMSE by 0.006 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt;, and improves &lt;em&gt;R&lt;/em&gt; by 0.15 relative to the individual best-performing RF method; (4) The gap-filled SM shows an improved skill than the original ESA CCI SM against global distributed ground SM, with Stacking displaying the lowest ubRMSE of 0.057 m&lt;sup&gt;3&lt;/sup&gt;/m&lt;sup&gt;3&lt;/sup&gt; and the highest &lt;em&gt;R&lt;/em","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"331 ","pages":"Article 115040"},"PeriodicalIF":11.4,"publicationDate":"2025-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145154041","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}