Remote Sensing of Environment最新文献_第10页

A novel hybrid approach for mapping global surface solar radiation with DSCOVR/EPIC: Combining deep learning with physical algorithm 基于DSCOVR/EPIC的全球表面太阳辐射制图新方法：深度学习与物理算法相结合

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-17 DOI: 10.1016/j.rse.2025.115200

Wenjun Tang , Jinwen Qi , Junmei He , Fuxin Zhu , Chuanming Yuan , Jinyan Yang , Bing Hu

Surface solar radiation (SSR, also known as global radiation, R_g), is critical for Earth's energy, water, and carbon cycles, yet existing satellite-derived global R_g products suffer from spatial inconsistencies due to multi-source data fusion. To address this, we propose a novel hybrid approach integrating deep learning with physical algorithms using observations from the Deep Space Climate Observatory (DSCOVR/EPIC), positioned at the Sun-Earth Lagrange-1 point and continuously observed the entire portion of the Earth with sunshine. Unlike traditional physical algorithms or machine learning algorithms, this method estimates cloud transmittance via a DenseNet-based convolutional neural network (CNN), calculates clear-sky R_g using a physical parameterization scheme, and combines these to derive all-sky R_g. Meanwhile, the direct and diffuse components (R_dir and R_dif) are further separated from the estimated R_g using a Light Gradient Boosting Machine (LightGBM) model. The method was trained with in-situ observations from the Baseline Surface Radiation Network (BSRN), and further independently evaluated against in-situ observations from three networks of the Solar Radiation (SOLRAD), China Meteorological Administration (CMA) radiation stations and Global Energy Balance Archive (GEBA). Independent evaluation demonstrates that our hybrid method exhibits excellent spatial scalability. Comparative validation against the product of Hao et al. (2020) derived from DSCOVR/EPIC observations demonstrates our method can generate more accurate global products of R_g, R_dir and R_dif. The innovation of our method lies in integrating machine learning with physical algorithms to leverage their complementary strengths, while overcoming the limitations of high uncertainty associated with cloud optical property retrievals from DSCOVR/EPIC observations. This approach will contribute to the mapping of global spatially consistent radiation products, overcoming the limitations of geostationary and polar-orbiting satellites.

地表太阳辐射（SSR，也称为全球辐射，Rg）对地球的能量、水和碳循环至关重要，但由于多源数据融合，现有的卫星衍生全球太阳辐射产品存在空间不一致性。为了解决这个问题，我们提出了一种新的混合方法，将深度学习与物理算法结合起来，利用深空气候观测站（DSCOVR/EPIC）的观测数据，该观测站位于太阳-地球拉格朗日-1点，连续观测地球的整个部分。与传统的物理算法或机器学习算法不同，该方法通过基于densenet的卷积神经网络（CNN）估计云透射率，使用物理参数化方案计算晴空Rg，并将这些结合起来得出全天Rg。同时，使用光梯度增强机（Light Gradient Boosting Machine, LightGBM）模型将直接分量和漫射分量（Rdir和Rdif）与估计的Rg进一步分离。利用基线地面辐射网（BSRN）的现场观测数据对方法进行了训练，并进一步与太阳辐射网（SOLRAD）、中国气象局（CMA）辐射站和全球能量平衡档案（GEBA）三个网络的现场观测数据进行了独立评估。独立评价表明，我们的混合方法具有良好的空间可扩展性。与Hao et al.（2020）从DSCOVR/EPIC观测中得到的产品进行比较验证表明，我们的方法可以生成更准确的Rg， Rdir和Rdif的全球产品。该方法的创新之处在于将机器学习与物理算法相结合，以利用它们的互补优势，同时克服了从DSCOVR/EPIC观测中检索云光学特性的高不确定性的局限性。这种方法将有助于绘制全球空间一致辐射产品的地图，克服地球静止卫星和极轨卫星的局限性。

{"title":"A novel hybrid approach for mapping global surface solar radiation with DSCOVR/EPIC: Combining deep learning with physical algorithm","authors":"Wenjun Tang , Jinwen Qi , Junmei He , Fuxin Zhu , Chuanming Yuan , Jinyan Yang , Bing Hu","doi":"10.1016/j.rse.2025.115200","DOIUrl":"10.1016/j.rse.2025.115200","url":null,"abstract":"<div><div>Surface solar radiation (SSR, also known as global radiation, R<sub>g</sub>), is critical for Earth's energy, water, and carbon cycles, yet existing satellite-derived global R<sub>g</sub> products suffer from spatial inconsistencies due to multi-source data fusion. To address this, we propose a novel hybrid approach integrating deep learning with physical algorithms using observations from the Deep Space Climate Observatory (DSCOVR/EPIC), positioned at the Sun-Earth Lagrange-1 point and continuously observed the entire portion of the Earth with sunshine. Unlike traditional physical algorithms or machine learning algorithms, this method estimates cloud transmittance via a DenseNet-based convolutional neural network (CNN), calculates clear-sky R<sub>g</sub> using a physical parameterization scheme, and combines these to derive all-sky R<sub>g</sub>. Meanwhile, the direct and diffuse components (R<sub>dir</sub> and R<sub>dif</sub>) are further separated from the estimated R<sub>g</sub> using a Light Gradient Boosting Machine (LightGBM) model. The method was trained with in-situ observations from the Baseline Surface Radiation Network (BSRN), and further independently evaluated against in-situ observations from three networks of the Solar Radiation (SOLRAD), China Meteorological Administration (CMA) radiation stations and Global Energy Balance Archive (GEBA). Independent evaluation demonstrates that our hybrid method exhibits excellent spatial scalability. Comparative validation against the product of <span><span>Hao et al. (2020)</span></span> derived from DSCOVR/EPIC observations demonstrates our method can generate more accurate global products of R<sub>g</sub>, R<sub>dir</sub> and R<sub>dif</sub>. The innovation of our method lies in integrating machine learning with physical algorithms to leverage their complementary strengths, while overcoming the limitations of high uncertainty associated with cloud optical property retrievals from DSCOVR/EPIC observations. This approach will contribute to the mapping of global spatially consistent radiation products, overcoming the limitations of geostationary and polar-orbiting satellites.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115200"},"PeriodicalIF":11.4,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145785694","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}

引用次数: 0

GPP-net: a robust high-resolution GPP estimation network for Sentinel-2 using only surface reflectance and photosynthetically active radiation GPP-net: Sentinel-2的高分辨率GPP估计网络，仅使用表面反射率和光合有效辐射

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-16 DOI: 10.1016/j.rse.2025.115198

Shaoyu Wang , Youngryel Ryu , Benjamin Dechant , Helin Zhang , Huaize Feng , Jeongho Lee , Changhyun Choi

High-resolution gross primary productivity (GPP) estimation is crucial for ecological and agricultural applications that require fine spatial details to capture GPP heterogeneity. Satellite-based GPP estimation usually relies on land cover and meteorological data. However, the misclassification of land cover data and coarse resolution of meteorological data greatly increase the uncertainty. Here, we propose a robust high-resolution GPP estimation deep learning (DL) network, named GPP-net, using only satellite surface reflectance (SR) from Sentinel-2 and photosynthetically active radiation (PAR). Specifically, GPP-net is based on a fully 1-D convolutional encoder-decoder network combined with a spectral band importance estimation module. To enhance the generalization of GPP-net, we ran the soil-canopy energy balance radiative transfer (SCOPE) model, and then combined these SCOPE-simulated reflectance data with GPP and PAR data extracted from FLUXNET2015 to pre-train GPP-net. Compared to benchmark models including near-infrared reflectance of vegetation multiplied by incoming sunlight (NIRvP), partial least squares (PLS) and random forest (RF), GPP-net improved half-hourly and daily GPP retrieval across seven plant functional types (PFTs) including four forest types, cropland, grassland and wetland. Owing to its robust nonlinear feature learning capabilities, GPP-net also facilitated robust GPP estimation across both C3 and C4 vegetation. We found that GPP-net could reliably estimate GPP under drought and heatwave conditions, with minimal improvement from including vapor pressure deficit (VPD) as a predictor. Furthermore, GPP-net demonstrated great robustness to soil effects in GPP mapping, and had strong ability in capturing inter-annual variability of GPP. The pretraining paradigm enabled us to fully leverage historical data, and the DL framework ensured that the model generalization continually improves as new data is integrated. Our model dispenses with land cover data and minimizes the requirements of coarse-resolution meteorological data for high-resolution GPP estimation, which could support future efforts in global high-resolution GPP mapping.

高分辨率的总初级生产力（GPP）估算对于需要精细空间细节来捕捉GPP异质性的生态和农业应用至关重要。基于卫星的GPP估算通常依赖于土地覆盖和气象数据。然而，土地覆被数据的错误分类和气象数据的粗分辨率大大增加了不确定性。在这里，我们提出了一个鲁棒的高分辨率GPP估计深度学习（DL）网络，命名为GPP-net，仅使用来自Sentinel-2的卫星表面反射率（SR）和光合有效辐射（PAR）。具体来说，GPP-net是基于一个全一维卷积编码器-解码器网络，并结合了一个频谱频带重要性估计模块。为了提高GPP-net的泛化能力，我们运行了土壤-冠层能量平衡辐射传输（SCOPE）模型，然后将SCOPE模拟的反射率数据与FLUXNET2015提取的GPP和PAR数据相结合，对GPP-net进行了预训练。与包括植被近红外反射率乘以入射阳光（NIRvP）、偏最小二乘（PLS）和随机森林（RF）在内的基准模型相比，GPP-net提高了包括四种森林类型、农田、草地和湿地在内的七种植物功能类型（PFTs）的半小时和每日GPP检索。由于其鲁棒的非线性特征学习能力，GPP-net还可以实现C3和C4植被的鲁棒GPP估计。我们发现GPP-net可以可靠地估计干旱和热浪条件下的GPP，而将蒸汽压差（VPD）作为预测因子的改进很小。此外，GPP-net在GPP制图中对土壤效应具有较强的鲁棒性，具有较强的捕捉GPP年际变化的能力。预训练范式使我们能够充分利用历史数据，DL框架确保随着新数据的集成，模型泛化能力不断提高。我们的模型省去了土地覆盖数据，并最大限度地减少了对高分辨率GPP估算的粗分辨率气象数据的要求，这可以支持未来全球高分辨率GPP制图的努力。

{"title":"GPP-net: a robust high-resolution GPP estimation network for Sentinel-2 using only surface reflectance and photosynthetically active radiation","authors":"Shaoyu Wang , Youngryel Ryu , Benjamin Dechant , Helin Zhang , Huaize Feng , Jeongho Lee , Changhyun Choi","doi":"10.1016/j.rse.2025.115198","DOIUrl":"10.1016/j.rse.2025.115198","url":null,"abstract":"<div><div>High-resolution gross primary productivity (GPP) estimation is crucial for ecological and agricultural applications that require fine spatial details to capture GPP heterogeneity. Satellite-based GPP estimation usually relies on land cover and meteorological data. However, the misclassification of land cover data and coarse resolution of meteorological data greatly increase the uncertainty. Here, we propose a robust high-resolution GPP estimation deep learning (DL) network, named GPP-net, using only satellite surface reflectance (SR) from Sentinel-2 and photosynthetically active radiation (PAR). Specifically, GPP-net is based on a fully 1-D convolutional encoder-decoder network combined with a spectral band importance estimation module. To enhance the generalization of GPP-net, we ran the soil-canopy energy balance radiative transfer (SCOPE) model, and then combined these SCOPE-simulated reflectance data with GPP and PAR data extracted from FLUXNET2015 to pre-train GPP-net. Compared to benchmark models including near-infrared reflectance of vegetation multiplied by incoming sunlight (NIRvP), partial least squares (PLS) and random forest (RF), GPP-net improved half-hourly and daily GPP retrieval across seven plant functional types (PFTs) including four forest types, cropland, grassland and wetland. Owing to its robust nonlinear feature learning capabilities, GPP-net also facilitated robust GPP estimation across both C3 and C4 vegetation. We found that GPP-net could reliably estimate GPP under drought and heatwave conditions, with minimal improvement from including vapor pressure deficit (VPD) as a predictor. Furthermore, GPP-net demonstrated great robustness to soil effects in GPP mapping, and had strong ability in capturing inter-annual variability of GPP. The pretraining paradigm enabled us to fully leverage historical data, and the DL framework ensured that the model generalization continually improves as new data is integrated. Our model dispenses with land cover data and minimizes the requirements of coarse-resolution meteorological data for high-resolution GPP estimation, which could support future efforts in global high-resolution GPP mapping.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115198"},"PeriodicalIF":11.4,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145771648","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}

引用次数: 0

Hyperspectral OCI/PACE observations of the Atlantic Sargassum 大西洋马尾藻的高光谱OCI/PACE观测

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-15 DOI: 10.1016/j.rse.2025.115185

Lin Qi , Menghua Wang , Chuanmin Hu , Yuyuan Xie , Brian B. Barnes

Since the first appearance of the annually recurrent Great Atlantic Sargassum Belt (GASB) in 2011, satellite remote sensing has been used as a primary technique to monitor and track the pelagic Sargassum fluitans/natans in the Atlantic Ocean. The Ocean Color Instrument (OCI) on the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE, 2024–present) mission is a first-ever operational hyperspectral sensor designed to measure the surface ocean's biological and biogeochemical properties at global scale on a near-daily basis, which is expected to provide improved performance over traditional multi-band polar-orbiting ocean color sensors. Here, we evaluate the capacity of OCI in detecting and quantifying the Atlantic Sargassum, referenced against heritage multi-band satellite sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS on Aqua) and Visible Infrared Imaging Radiometer Suite (VIIRS on NOAA-20). Our research involved development of a shared deep-learning Sargassum detection algorithm, fine-tuned for each sensor. As such, we found that OCI not only provided 47% more valid observations (# of valid pixels) than MODIS in the central west Atlantic for the study period of May–August 2024, but from the common valid pixels OCI also detected 51% more Sargassum-containing pixels. While the former is mostly due to the OCI's ability to avoid sun glint, the latter appears to be due to band binning and fewer artifacts around clouds. Using VIIRS as a bridge, it is further found that such a difference between OCI and MODIS is not due to the reported MODIS sensor degradation after 2023, but is inherent to sensor and measurement characteristics. On the other hand, VIIRS also showed more valid observations and detected more Sargassum pixels than MODIS, attributed to the larger swath and a finer spatial resolution, respectively. For these reasons, both OCI and VIIRS detected substantially more Sargassum than MODIS at daily, weekly, and monthly scales, although the spatial distributions and temporal changes of Sargassum revealed by the three sensors are similar. Finally, because of the hyperspectral capability, OCI is the only sensor that can spectrally discriminate Sargassum pixels without ambiguity. Such improved performance will make OCI a unique sensor to map both macroalgae mats and microalgae scums at global scale in both near real-time and retrospective analyses.

自2011年每年周期性的大大西洋马尾藻带首次出现以来，卫星遥感一直被用作监测和跟踪大西洋中上层马尾藻的主要技术。浮游生物、气溶胶、云、海洋生态系统（PACE， 2024年至今）任务上的海洋颜色仪器（OCI）是有史以来第一个可操作的高光谱传感器，旨在几乎每天测量全球范围内海洋表面的生物和生物地球化学特性，预计将提供比传统多波段极轨道海洋颜色传感器更好的性能。在这里，我们评估了OCI在探测和量化大西洋马尾藻方面的能力，参考了传统的多波段卫星传感器，如Aqua上的中分辨率成像光谱仪（MODIS）和NOAA-20上的可见红外成像辐射计套件（VIIRS）。我们的研究涉及开发共享的深度学习马尾藻检测算法，并对每个传感器进行微调。因此，我们发现，在2024年5月至8月的研究期间，OCI不仅比MODIS在大西洋中西部提供了47%的有效观测值（有效像元数），而且从普通有效像元中，OCI检测到的含sarg假设的像元也多51%。前者主要是由于OCI有能力避免太阳的闪光，而后者似乎是由于带化和云周围的人工制品较少。利用VIIRS作为桥梁，进一步发现OCI和MODIS之间的这种差异不是由于2023年之后报道的MODIS传感器退化，而是传感器和测量特性固有的。另一方面，VIIRS也比MODIS显示了更有效的观测结果，并检测到更多的马尾藻像素，这分别归功于更大的条带和更精细的空间分辨率。由于这些原因，OCI和VIIRS在日、周、月尺度上对马尾藻的探测明显多于MODIS，尽管这三种传感器所揭示的马尾藻的空间分布和时间变化相似。最后，由于具有高光谱特性，OCI是唯一能够准确分辨马尾藻像素的传感器。这种改进的性能将使OCI成为一种独特的传感器，可以在近实时和回顾性分析中绘制全球范围内的大藻垫和微藻浮渣。

{"title":"Hyperspectral OCI/PACE observations of the Atlantic Sargassum","authors":"Lin Qi , Menghua Wang , Chuanmin Hu , Yuyuan Xie , Brian B. Barnes","doi":"10.1016/j.rse.2025.115185","DOIUrl":"10.1016/j.rse.2025.115185","url":null,"abstract":"<div><div>Since the first appearance of the annually recurrent Great Atlantic <em>Sargassum</em> Belt (GASB) in 2011, satellite remote sensing has been used as a primary technique to monitor and track the pelagic <em>Sargassum fluitans/natans</em> in the Atlantic Ocean. The Ocean Color Instrument (OCI) on the Plankton, Aerosol, Cloud, ocean Ecosystem (PACE, 2024–present) mission is a first-ever operational hyperspectral sensor designed to measure the surface ocean's biological and biogeochemical properties at global scale on a near-daily basis, which is expected to provide improved performance over traditional multi-band polar-orbiting ocean color sensors. Here, we evaluate the capacity of OCI in detecting and quantifying the Atlantic <em>Sargassum</em>, referenced against heritage multi-band satellite sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS on Aqua) and Visible Infrared Imaging Radiometer Suite (VIIRS on NOAA-20). Our research involved development of a shared deep-learning <em>Sargassum</em> detection algorithm, fine-tuned for each sensor. As such, we found that OCI not only provided 47% more valid observations (# of valid pixels) than MODIS in the central west Atlantic for the study period of May–August 2024, but from the common valid pixels OCI also detected 51% more <em>Sargassum</em>-containing pixels. While the former is mostly due to the OCI's ability to avoid sun glint, the latter appears to be due to band binning and fewer artifacts around clouds. Using VIIRS as a bridge, it is further found that such a difference between OCI and MODIS is not due to the reported MODIS sensor degradation after 2023, but is inherent to sensor and measurement characteristics. On the other hand, VIIRS also showed more valid observations and detected more <em>Sargassum</em> pixels than MODIS, attributed to the larger swath and a finer spatial resolution, respectively. For these reasons, both OCI and VIIRS detected substantially more <em>Sargassum</em> than MODIS at daily, weekly, and monthly scales, although the spatial distributions and temporal changes of <em>Sargassum</em> revealed by the three sensors are similar. Finally, because of the hyperspectral capability, OCI is the only sensor that can spectrally discriminate <em>Sargassum</em> pixels without ambiguity. Such improved performance will make OCI a unique sensor to map both macroalgae mats and microalgae scums at global scale in both near real-time and retrospective analyses.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115185"},"PeriodicalIF":11.4,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145759817","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}

引用次数: 0

A synergistic super-ellipsoidal particle shape and ice cloud optical thickness retrieval method based on satellite polarimetric observations 基于卫星极化观测的超椭球粒子形状和冰云光学厚度协同反演方法

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-15 DOI: 10.1016/j.rse.2025.115172

Yizhen Meng , Lei Bi , Wei Han

Ice clouds, composed of irregular ice crystals, play a critical role in the Earth's radiative balance and climate regulation. Satellite polarimetric observations, such as those from the POLarization and Directionality of Earth Reflectance-3 (POLDER-3), exhibit high sensitivity to particle characteristics, making them valuable for deriving ice cloud microphysical properties. Conventional ice cloud remote sensing methods typically rely on single-particle models, which assume a prior particle shape across entire regions, thereby neglecting the inherent spatial heterogeneity. Under this context, the super-ellipsoidal particle model was developed, enabling continuous variation in surface morphology through three parameters (i.e., aspect ratio, roundness, and surface roughness), thus facilitating the retrieval of particle shape variations. To comprehensively consider the spatial heterogeneity of ice crystals and assess the effectiveness of the super-ellipsoidal multi-particle model, a synergistic retrieval of particle shape parameters and ice cloud optical thickness (IOT) was conducted across six tropical cyclone (TC) and cloud cases. The retrieval framework was built upon vector radiative transfer simulations derived from the adding-doubling model, linking POLDER-3 observations with the super-ellipsoidal particle models and IOT. The retrieved particle shapes and IOT were validated by comparing re-simulated radiance with satellite observations. The findings indicate an order-of-magnitude enhancement over single-particle models in retrieval performance, with root mean square errors (RMSEs) for normalized radiance decreasing from [0.0371, 0.1063] to [0.0023, 0.0042], and for polarized radiance reducing from [0.0036, 0.0061] to [0.0008, 0.0018]. The proposed novel method offers substantial improvements in retrieving IOT, contributing valuable insights for advancing ice cloud remote sensing techniques.

由不规则冰晶组成的冰云在地球的辐射平衡和气候调节中起着至关重要的作用。卫星极化观测，例如来自地球反射率-3 （POLDER-3）偏振和方向性的观测，对粒子特征表现出很高的灵敏度，这使得它们对推导冰云微物理特性很有价值。传统的冰云遥感方法通常依赖于单粒子模型，该模型假设整个区域的先验粒子形状，从而忽略了固有的空间异质性。在此背景下，建立了超椭球粒子模型，通过三个参数（即纵横比、圆度和表面粗糙度）实现表面形貌的连续变化，从而便于颗粒形状变化的检索。为了综合考虑冰晶的空间异质性，评估超椭球多粒子模型的有效性，在6个热带气旋（TC）和云的情况下，对粒子形状参数和冰云光学厚度（IOT）进行了协同检索。检索框架建立在从添加加倍模型导出的矢量辐射传输模拟的基础上，将POLDER-3观测与超椭球粒子模型和物联网联系起来。通过将重新模拟的辐射与卫星观测结果进行比较，验证了检索到的粒子形状和物联网。结果表明，与单粒子模型相比，该模型的检索性能提高了一个数量级，归一化辐射的均方根误差（rmse）从[0.0371,0.1063]降至[0.0023,0.0042]，极化辐射的均方根误差（rmse）从[0.0036,0.0061]降至[0.0008,0.0018]。提出的新方法在检索物联网方面提供了实质性的改进，为推进冰云遥感技术提供了有价值的见解。

{"title":"A synergistic super-ellipsoidal particle shape and ice cloud optical thickness retrieval method based on satellite polarimetric observations","authors":"Yizhen Meng , Lei Bi , Wei Han","doi":"10.1016/j.rse.2025.115172","DOIUrl":"10.1016/j.rse.2025.115172","url":null,"abstract":"<div><div>Ice clouds, composed of irregular ice crystals, play a critical role in the Earth's radiative balance and climate regulation. Satellite polarimetric observations, such as those from the POLarization and Directionality of Earth Reflectance-3 (POLDER-3), exhibit high sensitivity to particle characteristics, making them valuable for deriving ice cloud microphysical properties. Conventional ice cloud remote sensing methods typically rely on single-particle models, which assume a prior particle shape across entire regions, thereby neglecting the inherent spatial heterogeneity. Under this context, the super-ellipsoidal particle model was developed, enabling continuous variation in surface morphology through three parameters (i.e., aspect ratio, roundness, and surface roughness), thus facilitating the retrieval of particle shape variations. To comprehensively consider the spatial heterogeneity of ice crystals and assess the effectiveness of the super-ellipsoidal multi-particle model, a synergistic retrieval of particle shape parameters and ice cloud optical thickness (IOT) was conducted across six tropical cyclone (TC) and cloud cases. The retrieval framework was built upon vector radiative transfer simulations derived from the adding-doubling model, linking POLDER-3 observations with the super-ellipsoidal particle models and IOT. The retrieved particle shapes and IOT were validated by comparing re-simulated radiance with satellite observations. The findings indicate an order-of-magnitude enhancement over single-particle models in retrieval performance, with root mean square errors (RMSEs) for normalized radiance decreasing from [0.0371, 0.1063] to [0.0023, 0.0042], and for polarized radiance reducing from [0.0036, 0.0061] to [0.0008, 0.0018]. The proposed novel method offers substantial improvements in retrieving IOT, contributing valuable insights for advancing ice cloud remote sensing techniques.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115172"},"PeriodicalIF":11.4,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145752888","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}

引用次数: 0

A review of forward modelling and retrieval approaches for forest soil moisture and vegetation optical depth using L-band radiometry 森林土壤水分和植被光学深度的l波段辐射正演模拟与反演方法综述

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-15 DOI: 10.1016/j.rse.2025.115158

Andreas Colliander , Mike Schwank , Yiwen Zhou , Mehmet Kurum , Cristina Vittucci , Leung Tsang , Alex Roy , Aaron Berg

Forests are a critical component of the Earth system, accounting for approximately one-third of global photosynthetic activity and carbon storage. They also provide essential habitats for countless species and vital resources for human activities. Low-frequency (L-band; 1–2 GHz) microwave radiometry enables the measurement of forest soil moisture (SM) and L-band vegetation optical depth (L-VOD), offering valuable insights into processes such as tree growth, water infiltration, soil fertility, fuel moisture, carbon stocks, wildfire vulnerability, and biodiversity dynamics. These measurements also support the study of carbon and water fluxes, tree responses to hydrological stress (e.g., drought), and fuel moisture estimation. However, existing algorithms for retrieving SM and L-VOD were primarily developed for low-biomass vegetation types (e.g., grasslands and croplands), differing structurally from forests. This motivates the present review to evaluate the current retrieval approaches, their performance assessment methods, and available validation resources. The review found that systematic uncertainties persist in forest retrievals, despite the demonstrated sensitivity of L-band brightness temperature (TB) to forest SM and L-VOD. Moreover, the focus on non-forest ecosystems has led to a lack of suitable ground truth and reference data for validating forest SM and L-VOD products, and current validation techniques remain underdeveloped. To fully harness the potential of L-band radiometry in forest monitoring, new retrieval algorithms that account for the unique structural and compositional characteristics of forests are required. Additionally, validation efforts must be enhanced both quantitatively and qualitatively—particularly for L-VOD—to improve confidence in these remote sensing products.

森林是地球系统的重要组成部分，约占全球光合作用活动和碳储量的三分之一。它们还为无数物种提供了重要的栖息地和人类活动的重要资源。低频（l波段；1-2 GHz）微波辐射测量能够测量森林土壤水分（SM）和l波段植被光学深度（L-VOD），为树木生长、水分渗透、土壤肥力、燃料水分、碳储量、野火脆弱性和生物多样性动态等过程提供有价值的见解。这些测量还支持对碳和水通量、树木对水文压力（如干旱）的反应以及燃料水分估算的研究。然而，现有的SM和L-VOD检索算法主要是针对低生物量植被类型（如草地和农田）开发的，在结构上与森林不同。这促使本综述对当前的检索方法、它们的性能评估方法和可用的验证资源进行评估。研究发现，尽管l波段亮度温度（TB）对森林SM和L-VOD具有敏感性，但森林反演的系统不确定性仍然存在。此外，对非森林生态系统的关注导致缺乏适合森林SM和L-VOD产品验证的地面真值和参考数据，目前的验证技术仍不发达。为了充分利用l波段辐射测量在森林监测中的潜力，需要考虑到森林独特结构和组成特征的新检索算法。此外，必须加强定量和定性的验证工作，特别是l - vod，以提高对这些遥感产品的信心。

{"title":"A review of forward modelling and retrieval approaches for forest soil moisture and vegetation optical depth using L-band radiometry","authors":"Andreas Colliander , Mike Schwank , Yiwen Zhou , Mehmet Kurum , Cristina Vittucci , Leung Tsang , Alex Roy , Aaron Berg","doi":"10.1016/j.rse.2025.115158","DOIUrl":"10.1016/j.rse.2025.115158","url":null,"abstract":"<div><div>Forests are a critical component of the Earth system, accounting for approximately one-third of global photosynthetic activity and carbon storage. They also provide essential habitats for countless species and vital resources for human activities. Low-frequency (L-band; 1–2 GHz) microwave radiometry enables the measurement of forest soil moisture (SM) and L-band vegetation optical depth (L-VOD), offering valuable insights into processes such as tree growth, water infiltration, soil fertility, fuel moisture, carbon stocks, wildfire vulnerability, and biodiversity dynamics. These measurements also support the study of carbon and water fluxes, tree responses to hydrological stress (e.g., drought), and fuel moisture estimation. However, existing algorithms for retrieving SM and L-VOD were primarily developed for low-biomass vegetation types (e.g., grasslands and croplands), differing structurally from forests. This motivates the present review to evaluate the current retrieval approaches, their performance assessment methods, and available validation resources. The review found that systematic uncertainties persist in forest retrievals, despite the demonstrated sensitivity of L-band brightness temperature (TB) to forest SM and L-VOD. Moreover, the focus on non-forest ecosystems has led to a lack of suitable ground truth and reference data for validating forest SM and L-VOD products, and current validation techniques remain underdeveloped. To fully harness the potential of L-band radiometry in forest monitoring, new retrieval algorithms that account for the unique structural and compositional characteristics of forests are required. Additionally, validation efforts must be enhanced both quantitatively and qualitatively—particularly for L-VOD—to improve confidence in these remote sensing products.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115158"},"PeriodicalIF":11.4,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145759818","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}

引用次数: 0

Highest quality remote sensing reflectance database compiled from 20+ years of MODIS-aqua measurements 从20多年的MODIS-aqua测量中编译的最高质量的遥感反射率数据库

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-13 DOI: 10.1016/j.rse.2025.115195

Longteng Zhao, Zhongping Lee, Xiaolong Yu, Tianhao Wang, Daosheng Wang, Shaoling Shang

Remote sensing reflectance (R_rs) is a fundamental property in satellite ocean color remote sensing, which is critical for retrieving optical-biogeochemical properties and data-driven atmospheric correction algorithms. In this study, with three criteria applicable to ∼91% of the global ocean, we compiled a database of the highest quality R_rs (HQ_MODISA-R_rs) of oceanic waters based on 20+ years of ocean color measurements by the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Aqua satellite. While removing a large number of daily “standard” data products, our evaluation showed that the criteria for the highest-quality R_rs (CHQR) improved MODIS R_rs data consistency with benchmark in situ R_rs datasets, such as those from MOBY and AERONET-OC. After applying CHQR, analysis of imagery products in the South Pacific Ocean revealed that the coefficient of variation (CV) of R_rs among pixels reduced from 0.042 (standard quality control) to 0.030, along with enhanced temporal consistency, which indicates that this approach effectively filters abnormal data products. While such a dataset played a key role in the development of the cross-satellite atmospheric correction algorithm (Lee et al., 2024), we here further demonstrate that applications of HQ_MODISA-R_rs have ∼21.0% of oceanic areas between 50°S and 50°N showing reversed long-term trends of R_rs compared to the trend based on the standard R_rs product. We anticipate that this highest-quality R_rs database would not only improve our evaluation and understanding of long-term changes in various R_rs-derivative bio-optical properties of the global ocean, but also help to obtain consistent products among various satellite ocean color missions.

遥感反射率（Rrs）是卫星海洋颜色遥感的一项基本属性，它对于检索光学-生物地球化学属性和数据驱动的大气校正算法至关重要。在这项研究中，根据适用于全球海洋约91%的三个标准，我们基于Aqua卫星上的中分辨率成像光谱仪（MODIS） 20多年的海洋颜色测量数据，编制了一个最高质量的海水Rrs数据库（HQMODISA-Rrs）。在删除大量日常“标准”数据产品的同时，我们的评估表明，最高质量Rrs （CHQR）标准提高了MODIS Rrs数据与基准原位Rrs数据集（如MOBY和AERONET-OC）的一致性。应用CHQR后，对南太平洋地区影像产品的分析表明，像素间rrr的变异系数（CV）从0.042（标准质量控制）降至0.030，且时间一致性增强，表明该方法能够有效过滤异常数据产品。虽然这样的数据集在跨卫星大气校正算法的发展中发挥了关键作用（Lee et al., 2024），但我们在这里进一步证明，与基于标准Rrs产品的趋势相比，HQMODISA-Rrs的应用在50°S和50°N之间的海洋区域中显示出相反的Rrs长期趋势。我们预计，这个最高质量的Rrs数据库不仅可以提高我们对全球海洋各种Rrs衍生生物光学特性长期变化的评估和理解，而且还有助于在各种卫星海洋颜色任务中获得一致的产品。

{"title":"Highest quality remote sensing reflectance database compiled from 20+ years of MODIS-aqua measurements","authors":"Longteng Zhao, Zhongping Lee, Xiaolong Yu, Tianhao Wang, Daosheng Wang, Shaoling Shang","doi":"10.1016/j.rse.2025.115195","DOIUrl":"10.1016/j.rse.2025.115195","url":null,"abstract":"<div><div>Remote sensing reflectance (<em>R</em><sub><em>rs</em></sub>) is a fundamental property in satellite ocean color remote sensing, which is critical for retrieving optical-biogeochemical properties and data-driven atmospheric correction algorithms. In this study, with three criteria applicable to ∼91% of the global ocean, we compiled a database of the highest quality <em>R</em><sub><em>rs</em></sub> (HQ<sub>MODISA</sub>-<em>R</em><sub><em>rs</em></sub>) of oceanic waters based on 20+ years of ocean color measurements by the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Aqua satellite. While removing a large number of daily “standard” data products, our evaluation showed that the criteria for the highest-quality <em>R</em><sub><em>rs</em></sub> (CHQR) improved MODIS <em>R</em><sub><em>rs</em></sub> data consistency with benchmark <em>in situ R</em><sub><em>rs</em></sub> datasets, such as those from MOBY and AERONET-OC. After applying CHQR, analysis of imagery products in the South Pacific Ocean revealed that the coefficient of variation (CV) of <em>R</em><sub><em>rs</em></sub> among pixels reduced from 0.042 (standard quality control) to 0.030, along with enhanced temporal consistency, which indicates that this approach effectively filters abnormal data products. While such a dataset played a key role in the development of the cross-satellite atmospheric correction algorithm (Lee et al., 2024), we here further demonstrate that applications of HQ<sub>MODISA</sub>-<em>R</em><sub><em>rs</em></sub> have ∼21.0% of oceanic areas between 50°S and 50°N showing reversed long-term trends of <em>R</em><sub><em>rs</em></sub> compared to the trend based on the standard <em>R</em><sub><em>rs</em></sub> product. We anticipate that this highest-quality <em>R</em><sub><em>rs</em></sub> database would not only improve our evaluation and understanding of long-term changes in various <em>R</em><sub><em>rs</em></sub>-derivative bio-optical properties of the global ocean, but also help to obtain consistent products among various satellite ocean color missions.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115195"},"PeriodicalIF":11.4,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145731548","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}

引用次数: 0

Retrieval of global surface phytoplankton community structure using a minimal set of predictors 基于最小预测因子的全球表层浮游植物群落结构检索

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-13 DOI: 10.1016/j.rse.2025.115174

Xueyin Li , Shuguo Chen , Junwei Wang , Qing Zhu

Phytoplankton underpin marine food webs and drive essential biogeochemical processes. Their pigment composition serves as a vital indicator of community structure and physiological state. In this study, we developed a multilayer perceptron (MLP)-based model to estimate the concentrations of 26 pigments and infer phytoplankton community structure in global surface waters. By integrating prior knowledge and quantifying through SHapley Additive exPlanations (SHAP), we selected a physically meaningful subset of 10 input features to construct pigment inversion models, including environmental parameters such as sea surface temperature (SST), salinity (SSS), and sea surface height (SSH), as well as remote sensing reflectance (R_rs) at seven spectral bands. For 24 pigments, the model achieved high predictive accuracy and strong generalization performance, with an average coefficient of determination (R²) of 0.76 and median absolute percentage difference (MAPD) of 12.01 % under random cross-validation, and slightly lower accuracy under temporal cross-validation (R² = 0.67; MAPD = 14.35 %). Feature analysis revealed that R_rs, particularly spectral regions encompassing absorption peaks and adjacent gradients, dominated pigment predictions. Moreover, the model captured nonlinear thermal responses of phytoplankton to SST, consistent with known ecophysiological patterns, and reflected synergistic interactions between SST and SSS affecting pigment variability. The inferred phytoplankton community structures, estimated via Diagnostic Pigment Analysis (DPA), showed strong agreement with previous studies, validating the model's ecological reliability. Additionally, Hydrolight simulations based on in situ a_ph spectra demonstrated that the model performs reliably under water conditions with suspended particulate matter (SPM) ≤ 1 g m⁻³ and colored dissolved organic matter (CDOM) absorption at 440 nm is ≤0.25 m⁻¹, maintaining MAPD below 25 %. Our research demonstrates that neural networks operate in a physically informed manner rather than as purely data-driven models, enabling them to represent complex interactions between phytoplankton and ecological environment. This underscores the necessity of designing ecologically informed and physically interpretable input features in remote sensing applications, offering valuable guidance for future biogeochemical modeling efforts.

浮游植物支撑着海洋食物网，驱动着重要的生物地球化学过程。它们的色素组成是群落结构和生理状态的重要指标。在这项研究中，我们建立了一个基于多层感知器（MLP）的模型来估计26种色素的浓度，并推断全球地表水浮游植物的群落结构。通过整合先验知识并通过SHapley加性解释（SHAP）进行量化，我们选择了10个具有物理意义的输入特征子集来构建色素反演模型，包括环境参数，如海面温度（SST）、盐度（SSS）、海面高度（SSH），以及7个光谱波段的遥感反射率（Rrs）。对于24种色素，该模型具有较高的预测准确率和较强的泛化性能，随机交叉验证的平均决定系数（R2）为0.76，中位数绝对百分比差（MAPD）为12.01%，时间交叉验证的准确率略低（R2 = 0.67, MAPD = 14.35%）。特征分析表明，Rrs，特别是包含吸收峰和相邻梯度的光谱区域，主导了色素预测。此外，该模型捕获了浮游植物对海温的非线性热响应，与已知的生态生理模式一致，并反映了海温和海温之间影响色素变异的协同相互作用。通过诊断色素分析（DPA）推测的浮游植物群落结构与先前的研究结果非常吻合，验证了该模型的生态可靠性。此外，基于原位aph光谱的水光模拟表明，该模型在悬浮颗粒物(SPM)≤1 g m−3、彩色溶解有机质（CDOM）在440 nm吸收≤0.25 m−1的水条件下运行可靠，MAPD保持在25%以下。我们的研究表明，神经网络以物理知情的方式运作，而不是纯粹的数据驱动模型，使它们能够代表浮游植物与生态环境之间复杂的相互作用。这强调了在遥感应用中设计生态信息和物理可解释的输入特征的必要性，为未来的生物地球化学建模工作提供了有价值的指导。

{"title":"Retrieval of global surface phytoplankton community structure using a minimal set of predictors","authors":"Xueyin Li , Shuguo Chen , Junwei Wang , Qing Zhu","doi":"10.1016/j.rse.2025.115174","DOIUrl":"10.1016/j.rse.2025.115174","url":null,"abstract":"<div><div>Phytoplankton underpin marine food webs and drive essential biogeochemical processes. Their pigment composition serves as a vital indicator of community structure and physiological state. In this study, we developed a multilayer perceptron (MLP)-based model to estimate the concentrations of 26 pigments and infer phytoplankton community structure in global surface waters. By integrating prior knowledge and quantifying through SHapley Additive exPlanations (SHAP), we selected a physically meaningful subset of 10 input features to construct pigment inversion models, including environmental parameters such as sea surface temperature (SST), salinity (SSS), and sea surface height (SSH), as well as remote sensing reflectance (<em>R</em><sub><em>rs</em></sub>) at seven spectral bands. For 24 pigments, the model achieved high predictive accuracy and strong generalization performance, with an average coefficient of determination (<em>R</em><sup><em>2</em></sup>) of 0.76 and median absolute percentage difference (MAPD) of 12.01 % under random cross-validation, and slightly lower accuracy under temporal cross-validation (<em>R</em><sup><em>2</em></sup> = 0.67; MAPD = 14.35 %). Feature analysis revealed that <em>R</em><sub><em>rs</em></sub>, particularly spectral regions encompassing absorption peaks and adjacent gradients, dominated pigment predictions. Moreover, the model captured nonlinear thermal responses of phytoplankton to SST, consistent with known ecophysiological patterns, and reflected synergistic interactions between SST and SSS affecting pigment variability. The inferred phytoplankton community structures, estimated <em>via</em> Diagnostic Pigment Analysis (DPA), showed strong agreement with previous studies, validating the model's ecological reliability. Additionally, Hydrolight simulations based on <em>in situ a</em><sub><em>ph</em></sub> spectra demonstrated that the model performs reliably under water conditions with suspended particulate matter (SPM) ≤ 1 g m<sup>−3</sup> and colored dissolved organic matter (CDOM) absorption at 440 nm is ≤0.25 m<sup>−1</sup>, maintaining MAPD below 25 %. Our research demonstrates that neural networks operate in a physically informed manner rather than as purely data-driven models, enabling them to represent complex interactions between phytoplankton and ecological environment. This underscores the necessity of designing ecologically informed and physically interpretable input features in remote sensing applications, offering valuable guidance for future biogeochemical modeling efforts.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115174"},"PeriodicalIF":11.4,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732128","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}

引用次数: 0

Long-term forest structure trends in the peninsular Spain from lidar-optical sensors synergies 从激光雷达-光学传感器协同效应看西班牙半岛森林结构的长期趋势

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-13 DOI: 10.1016/j.rse.2025.115196

M. Tanase , J.P. Martini , P. Miranda , D. Garcia , V. Wilke , S. Miguel , C. Mihai , J. Diez , S. Natal , D. San Martin , P. Ruiz-Benito

Information on forest structure is needed for many management aspects, from carbon stock evaluation to fire hazard prediction. Such information is increasingly available from remote sensing data, including light detection and ranging (lidar) sensors. However, continuous forest monitoring is hindered by the limited temporal availability of lidar acquisitions. This study focused on integrating temporally consistent optical acquisitions with temporally sparse lidar acquisitions to provide long term information on forest structural characteristics across the peninsular Spain. We evaluated different modeling approaches including machine learning and advanced deep learning models to ascertain their limitations for long-term forest monitoring. Subsequently, annual estimates of forest structural attributes were generated from 1985 to 2024. Across all forests, height (H), canopy cover (FCC), and above ground biomass (AGB) increased by 34 %, 29 % and respectively 17.5 % from the first (1985–1989) to the last lustrum (2019–2024). By biome, the increase in H, FCC and AGB was larger over the Mediterranean (35 %, 33 %, and 18 % respectively) when compared to the Atlantic (33 %, 10 %, and 15 % respectively) forests. For the model training year, deep learning models improved estimation accuracy by 27 ± 2.6 % (mean value across regions and variables) when compared to machine learning models. However, when applied to data from other years (temporal inference), model precision was similar except for height in the Mediterranean were deep learning models improved RMSE estimates by 7 %. Overall, the deep learning models presented significant computational drawbacks when applied to large geographic extents and extensive temporal ranges, substantially increasing both training and prediction times without providing a clear improvement in accuracy compared to machine learning across years. Therefore, the long-term forest database was generated using machine learning models. This database provides a spatially explicit understanding of forest structural changes over nearly four decades, offering a valuable resource for forest management, ecological assessments, and climate-related policy-making.

从碳储量评价到火险预测，森林结构信息在许多管理方面都是需要的。这些信息越来越多地来自遥感数据，包括光探测和测距（激光雷达）传感器。然而，持续的森林监测受到激光雷达获取的有限时间可用性的阻碍。本研究的重点是整合时间一致的光学采集与时间稀疏的激光雷达采集，以提供关于整个西班牙半岛森林结构特征的长期信息。我们评估了不同的建模方法，包括机器学习和高级深度学习模型，以确定它们在长期森林监测中的局限性。随后，从1985年到2024年，对森林结构属性进行了年度估算。所有森林的高度(H)、冠层盖度（FCC）和地上生物量（AGB）从第一季（1985-1989年）到末季（2019-2024年）分别增加了34%、29%和17.5%。从生物群系来看，地中海森林的H、FCC和AGB分别增加了35%、33%和18%，而大西洋森林的H、FCC和AGB分别增加了33%、10%和15%。在模型训练年，与机器学习模型相比，深度学习模型的估计精度提高了27±2.6%（跨区域和变量的平均值）。然而，当应用于其他年份的数据（时间推断）时，模型精度相似，除了地中海的高度，深度学习模型将RMSE估计提高了7%。总的来说，深度学习模型在应用于大地理范围和大时间范围时存在显著的计算缺陷，与机器学习相比，深度学习模型大大增加了训练和预测时间，但在准确性方面却没有明显的提高。因此，使用机器学习模型生成长期森林数据库。该数据库提供了近40年来森林结构变化的空间明确理解，为森林管理、生态评估和气候相关决策提供了宝贵的资源。

{"title":"Long-term forest structure trends in the peninsular Spain from lidar-optical sensors synergies","authors":"M. Tanase , J.P. Martini , P. Miranda , D. Garcia , V. Wilke , S. Miguel , C. Mihai , J. Diez , S. Natal , D. San Martin , P. Ruiz-Benito","doi":"10.1016/j.rse.2025.115196","DOIUrl":"10.1016/j.rse.2025.115196","url":null,"abstract":"<div><div>Information on forest structure is needed for many management aspects, from carbon stock evaluation to fire hazard prediction. Such information is increasingly available from remote sensing data, including light detection and ranging (lidar) sensors. However, continuous forest monitoring is hindered by the limited temporal availability of lidar acquisitions. This study focused on integrating temporally consistent optical acquisitions with temporally sparse lidar acquisitions to provide long term information on forest structural characteristics across the peninsular Spain. We evaluated different modeling approaches including machine learning and advanced deep learning models to ascertain their limitations for long-term forest monitoring. Subsequently, annual estimates of forest structural attributes were generated from 1985 to 2024. Across all forests, height (H), canopy cover (FCC), and above ground biomass (AGB) increased by 34 %, 29 % and respectively 17.5 % from the first (1985–1989) to the last lustrum (2019–2024). By biome, the increase in H, FCC and AGB was larger over the Mediterranean (35 %, 33 %, and 18 % respectively) when compared to the Atlantic (33 %, 10 %, and 15 % respectively) forests. For the model training year, deep learning models improved estimation accuracy by 27 ± 2.6 % (mean value across regions and variables) when compared to machine learning models. However, when applied to data from other years (temporal inference), model precision was similar except for height in the Mediterranean were deep learning models improved RMSE estimates by 7 %. Overall, the deep learning models presented significant computational drawbacks when applied to large geographic extents and extensive temporal ranges, substantially increasing both training and prediction times without providing a clear improvement in accuracy compared to machine learning across years. Therefore, the long-term forest database was generated using machine learning models. This database provides a spatially explicit understanding of forest structural changes over nearly four decades, offering a valuable resource for forest management, ecological assessments, and climate-related policy-making.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115196"},"PeriodicalIF":11.4,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145752945","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}

引用次数: 0

Temporal attention multi-resolution fusion of satellite image time-series, applied to Landsat-8/9 and Sentinel-2: all bands, any time, at best spatial resolution 应用于Landsat-8/9和Sentinel-2的卫星图像时间序列时间关注多分辨率融合：所有波段、任何时间、最佳空间分辨率

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-11 DOI: 10.1016/j.rse.2025.115159

Julien Michel, Jordi Inglada

This paper introduces a general formulation for the fusion of Satellite Image Time Series (SITS) of variable length from several sensors at different spatial resolutions and acquisition times over the same geographical area. In this formulation, all the spectral bands from all the input sensors are predicted at the best input spatial resolution, and at any observed or non-observed acquisition time requested. To address this general problem, an advanced Masked Auto-Encoder training strategy is proposed, utilising two new loss functions: a Linear-Regression Learned Perceptual Image Similarity term to favor high spatial frequency details, and a mask-contrastive term to ignore clouds and other non-informative areas in the input data. This strategy is applied to the training of Temporal Attention Multi-Resolution Fusion of Satellite Image Time-Series (TAMRF-SITS), a novel Deep Learning architecture designed to implement the proposed general formulation. Experiments with joint Landsat-8/9 and Sentinel-2 time-series were conducted on four different tasks from the literature and demonstrate that a single pre-trained TAMRF is on par with or better than existing ad-hoc methods. For instance, TAMRF provides a gain of 0.01 surface reflectance Root Mean Square Error on a Spatio-Temporal Fusion task when compared to competing algorithms, while showing the highest spatial frequency content. Moreover, the proposed method relaxes unrealistic assumptions routinely found in the literature, including: same or similar spectral bands in different sensors, same-day acquisitions, and scale-invariance of the relationship between high and low resolution images. To the best of our knowledge, our method is the first to achieve this range of capabilities with a single model, without making any of these assumptions. The complete source code for training and experiments is available here: https://github.com/Evoland-Land-Monitoring-Evolution/tamrfsits.

本文介绍了同一地理区域内不同空间分辨率、不同采集时间、不同传感器的变长卫星图像时间序列融合的一般公式。在该公式中，所有输入传感器的所有光谱带都是在最佳输入空间分辨率下预测的，并且在任何观测或非观测的采集时间都是如此。为了解决这个普遍问题，提出了一种先进的掩模自编码器训练策略，利用两个新的损失函数：线性回归学习感知图像相似性项，以支持高空间频率细节，掩模对比项，以忽略输入数据中的云和其他非信息区域。该策略应用于卫星图像时间序列时间注意力多分辨率融合（tamrf - sit）的训练，这是一种新的深度学习架构，旨在实现所提出的通用公式。利用联合Landsat-8/9和Sentinel-2时间序列对文献中的四种不同任务进行了实验，并证明单个预训练的TAMRF与现有的ad-hoc方法相当或更好。例如，与竞争算法相比，TAMRF在时空融合任务中提供了0.01的表面反射率均方根误差，同时显示了最高的空间频率内容。此外，所提出的方法放宽了文献中常见的不切实际的假设，包括：不同传感器的相同或相似的光谱带，同一天采集，以及高低分辨率图像之间关系的尺度不变性。据我们所知，我们的方法是第一个用单个模型实现这个范围的功能，而不做任何这些假设。完整的训练和实验源代码可在这里获得：https://github.com/Evoland-Land-Monitoring-Evolution/tamrfsits。

{"title":"Temporal attention multi-resolution fusion of satellite image time-series, applied to Landsat-8/9 and Sentinel-2: all bands, any time, at best spatial resolution","authors":"Julien Michel, Jordi Inglada","doi":"10.1016/j.rse.2025.115159","DOIUrl":"10.1016/j.rse.2025.115159","url":null,"abstract":"<div><div>This paper introduces a general formulation for the fusion of Satellite Image Time Series (SITS) of variable length from several sensors at different spatial resolutions and acquisition times over the same geographical area. In this formulation, all the spectral bands from all the input sensors are predicted at the best input spatial resolution, and at any observed or non-observed acquisition time requested. To address this general problem, an advanced Masked Auto-Encoder training strategy is proposed, utilising two new loss functions: a Linear-Regression Learned Perceptual Image Similarity term to favor high spatial frequency details, and a mask-contrastive term to ignore clouds and other non-informative areas in the input data. This strategy is applied to the training of Temporal Attention Multi-Resolution Fusion of Satellite Image Time-Series (TAMRF-SITS), a novel Deep Learning architecture designed to implement the proposed general formulation. Experiments with joint Landsat-8/9 and Sentinel-2 time-series were conducted on four different tasks from the literature and demonstrate that a single pre-trained TAMRF is on par with or better than existing ad-hoc methods. For instance, TAMRF provides a gain of 0.01 surface reflectance Root Mean Square Error on a Spatio-Temporal Fusion task when compared to competing algorithms, while showing the highest spatial frequency content. Moreover, the proposed method relaxes unrealistic assumptions routinely found in the literature, including: same or similar spectral bands in different sensors, same-day acquisitions, and scale-invariance of the relationship between high and low resolution images. To the best of our knowledge, our method is the first to achieve this range of capabilities with a single model, without making any of these assumptions. The complete source code for training and experiments is available here: <span><span>https://github.com/Evoland-Land-Monitoring-Evolution/tamrfsits</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115159"},"PeriodicalIF":11.4,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145731843","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}

引用次数: 0

Began+: Leveraging bi-temporal SAR-optical data fusion to reconstruct clear-sky satellite imagery under large cloud cover 开始+：利用双时相sar -光学数据融合重建大云量下的晴空卫星图像

IF 11.4 1区地球科学 Q1 ENVIRONMENTAL SCIENCES

Remote Sensing of Environment

Pub Date : 2025-12-11 DOI: 10.1016/j.rse.2025.115171

Yu Xia , Wei He , Liangpei Zhang , Hongyan Zhang

In recent years, optical remote sensing imagery has played an increasingly vital role in Earth observation, but cloud contamination exists as an inevitable degradation. Combining synthetic aperture radar (SAR) and optical data with machine learning offers a promising solution for reconstructing clear-sky satellite imagery. Nevertheless, several challenges persist, including insufficient attention to large cloud cover, difficulties in restoring temporal changes, and limited practicality of deep models. To address these issues, this paper introduces a novel deep learning-based cloud removal framework, termed Began+, which integrates bi-temporal SAR-optical data to deal with cloudy images with high cover ratios. The Began+ framework comprises two primary components: a deep network and a flexible post-processing step, combining the strengths of data-driven models for restoring change information and traditional gap-filling algorithms for mitigating radiance discrepancies. First, a bi-output enhanced generative adversarial network, abbreviated as Began, is designed for image synthesis, featuring an enhanced channel-wise fusion block (ECFB) and a multi-scale depth-wise convolution residual block (MDRB). By applying the dual-tasking optimization and co-learning strategy, the Began model identifies potential change areas from bi-temporal SAR and pre-temporal optical inputs, guiding the synthesis of target optical images. Second, a range of cloud masking and gap-filling techniques can be optionally employed to effectively reduce radiometric discrepancies between the synthesized images and the cloudy data, ultimately yielding high-quality, clear-sky imagery. To meet the big data requirements of deep learning, we constructed two globally distributed cloud removal datasets, named BiS1L8-CR and BiS1S2-CR. Supported by these datasets, extensive experiments demonstrated that the Began+ framework effectively captures bi-temporal change features, reconstructing precise surface information in both Landsat-8 and Sentinel-2 satellite images under large cloud cover. Compared to the latest solutions and algorithms, our proposed Began+ framework exhibits significant advantages from both qualitative and quantitative perspectives in both simulated and real experiments. Furthermore, without strict constraints on input timing, the Began+ framework enables accurate reconstruction of large-scale dual-sensor imagery under high-ratio cloud cover, effectively restoring changing surfaces and improving the quality of unsupervised vegetation extraction.

近年来，光学遥感影像在对地观测中发挥着越来越重要的作用，但云污染的存在是不可避免的。将合成孔径雷达（SAR）和光学数据与机器学习相结合，为重建晴空卫星图像提供了一种很有前途的解决方案。然而，一些挑战仍然存在，包括对大云量的关注不足，恢复时间变化的困难，以及深度模型的实用性有限。为了解决这些问题，本文引入了一种新的基于深度学习的云去除框架，称为begin +，它集成了双时相sar光学数据来处理高覆盖率的多云图像。begin +框架包括两个主要组成部分：深度网络和灵活的后处理步骤，结合了数据驱动模型的优势，用于恢复变化信息和传统的空白填充算法，以减轻亮度差异。首先，设计了一个双输出增强生成对抗网络（简称Began）用于图像合成，具有增强的通道智能融合块（ECFB）和多尺度深度智能卷积残差块（MDRB）。该模型通过双任务优化和共同学习策略，从双时相SAR和前时相光学输入中识别出潜在的变化区域，指导目标光学图像的合成。其次，可以选择性地采用一系列云掩蔽和间隙填充技术来有效地减少合成图像与云数据之间的辐射差异，最终产生高质量的晴空图像。为了满足深度学习的大数据需求，我们构建了两个全球分布式的去云数据集，分别命名为BiS1L8-CR和BiS1S2-CR。在这些数据集的支持下，大量的实验表明，Began+框架有效地捕获了双时相变化特征，在大云量下重建了Landsat-8和Sentinel-2卫星图像中的精确地表信息。与最新的解决方案和算法相比，我们提出的begin +框架在模拟和真实实验中从定性和定量的角度都表现出显著的优势。此外，在没有严格限制输入时间的情况下，begin +框架能够在高云量下精确重建大尺度双传感器图像，有效地恢复变化的地表，提高无监督植被提取的质量。

{"title":"Began+: Leveraging bi-temporal SAR-optical data fusion to reconstruct clear-sky satellite imagery under large cloud cover","authors":"Yu Xia , Wei He , Liangpei Zhang , Hongyan Zhang","doi":"10.1016/j.rse.2025.115171","DOIUrl":"10.1016/j.rse.2025.115171","url":null,"abstract":"<div><div>In recent years, optical remote sensing imagery has played an increasingly vital role in Earth observation, but cloud contamination exists as an inevitable degradation. Combining synthetic aperture radar (SAR) and optical data with machine learning offers a promising solution for reconstructing clear-sky satellite imagery. Nevertheless, several challenges persist, including insufficient attention to large cloud cover, difficulties in restoring temporal changes, and limited practicality of deep models. To address these issues, this paper introduces a novel deep learning-based cloud removal framework, termed Began+, which integrates bi-temporal SAR-optical data to deal with cloudy images with high cover ratios. The Began+ framework comprises two primary components: a deep network and a flexible post-processing step, combining the strengths of data-driven models for restoring change information and traditional gap-filling algorithms for mitigating radiance discrepancies. First, a bi-output enhanced generative adversarial network, abbreviated as Began, is designed for image synthesis, featuring an enhanced channel-wise fusion block (ECFB) and a multi-scale depth-wise convolution residual block (MDRB). By applying the dual-tasking optimization and co-learning strategy, the Began model identifies potential change areas from bi-temporal SAR and pre-temporal optical inputs, guiding the synthesis of target optical images. Second, a range of cloud masking and gap-filling techniques can be optionally employed to effectively reduce radiometric discrepancies between the synthesized images and the cloudy data, ultimately yielding high-quality, clear-sky imagery. To meet the big data requirements of deep learning, we constructed two globally distributed cloud removal datasets, named BiS1L8-CR and BiS1S2-CR. Supported by these datasets, extensive experiments demonstrated that the Began+ framework effectively captures bi-temporal change features, reconstructing precise surface information in both Landsat-8 and Sentinel-2 satellite images under large cloud cover. Compared to the latest solutions and algorithms, our proposed Began+ framework exhibits significant advantages from both qualitative and quantitative perspectives in both simulated and real experiments. Furthermore, without strict constraints on input timing, the Began+ framework enables accurate reconstruction of large-scale dual-sensor imagery under high-ratio cloud cover, effectively restoring changing surfaces and improving the quality of unsupervised vegetation extraction.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"334 ","pages":"Article 115171"},"PeriodicalIF":11.4,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732131","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}

引用次数: 0