Cloud detection is crucial for remote sensing image analysis. Presently, deep learning methods are the preferred choice for cloud detection. Nevertheless, the prevalence of luminous ice and snow substrates in polar regions introduces significant complexities for cloud detection in medium-resolution remote sensing imagery. Consequently, conventional deep learning methods often misclassify clouds in intricate regions. Considering these challenges, we introduce a deep convolutional neural network (DCNN) model called DLACD-Net. This model is purpose-built for cloud detection and segmentation in mediumresolution, multispectral (VIS+SWIR) polar remote sensing images. DLACD-Net incorporates a nested U-shaped encoderdecoder structure and proposes DSACRSU at each encodingdecoding stage. These enhancements significantly improve the model's feature extraction capability. Additionally, we propose the deformed large kernel attention into the decoding stage. It can capture finer details and boundary information of the target object, autonomously determining the level of attention required in different regions based on the task. To address class imbalance in training samples, we introduce α-balanced Focal Loss, which focuses the model on challenging minority samples, leading to improved overall performance. Leveraging prior research, we adopt semi-supervised learning techniques that combine visible light (VIS) and short-wave infrared (SWIR) band data to construct the FY-3D MERSI-II dataset specifically for polar cloud detection. Within the dataset, our proposed model achieves an impressive overall classification accuracy of 95.97% and a mean intersection over union (MIoU) of 90.98%. Experimental results demonstrate that our proposed model outperforms baseline models primarily due to superior feature extraction. Furthermore, we systematically analyze the contribution of each added substructure.

The extensive existence of high-brightness ice and snow underlying surfaces in polar regions presents notable complexities for cloud detection in remote sensing imagery. To elevate the accuracy of cloud detection in polar regions, a novel polar cloud detection algorithm is proposed in this paper. Employing the MOD09 surface reflectance product, we compiled a database of monthly composite surface reflectance in the shortwave infrared bands specific to polar regions. Through the forward simulation of the correlation between the apparent reflectance and surface reflectance across diverse conditions using the 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) radiative transfer model, we established a dynamic cloud detection model for the shortwave infrared channels. In contrast to a machine learning algorithm and the widely used MOD35 cloud product, the algorithm introduced in this study demonstrates enhanced congruence with the authentic cloud distribution within cloud products. It precisely distinguishes between the cloudy and clear-sky pixels, achieving rates surpassing 90% for both, while maintaining an error rate and a missing rate each under 10%. The algorithm yields positive results for cloud detection in polar regions, effectively distinguishing between ice, snow, and clouds. It provides robust support for comprehensive and long-term cloud detection efforts in polar regions.

Arctic sea ice variability is a key indicator of global climate change and is closely related to navigation safety. In this study, we propose a sea ice identification method based on fuzzy c-means (FCM) clustering, combined with a two-stage cloud detection strategy, and quantitatively evaluate its performance using bias and root mean square error (RMSE). The cloud detection first performs an initial screening, then refines the cloud mask through scene-based spectral analysis of cloud-free pixels, effectively reducing cloud interference. This method enables accurate discrimination between sea ice and water surfaces, including water in leads and broken ice areas, and is applied to medium-resolution optical observations from the Medium Resolution Spectral Imager III (MERSI-III) onboard the Fengyun-3F (FY-3F) satellite. Validation of the MERSI-derived sea ice concentration (SIC) using Landsat 8/9 imagery at 30 m spatial resolution shows an average bias of approximately 5.5% and an RMSE of about 8.5%. Comparisons with MODIS sea ice products (MOD29/MYD29) further demonstrate reduced cloud-ice misclassification and improved representation of fine-scale sea ice features.

Accurate radiometric calibration consistency is fundamental for quantitative remote sensing retrievals using multi-source sensors. Previous studies have mainly evaluated inter-sensor calibration consistency (e.g., MODIS and VIIRS) but lacked systematic corrections. Such inconsistencies can propagate into long-term climate data records, leading to systematic uncertainties in temperature and radiative flux retrievals. Therefore, developing a robust correction framework to ensure multi-sensor consistency remains an urgent challenge. This study corrected calibration biases in the thermal infrared split-window channels of FY-3D/MERSI-II, FY-3E/MERSI-LL, and Aqua/MODIS using long-term observations over Dome C, Antarctica (2019–2024). Taking advantage of the homogeneous surface characteristics and minimal atmospheric interference at Dome C, Automatic Weather Station (AWS) temperature data were used as a proxy to calculate the fluctuation trends of relative biases for assessing calibration stability and ensuring the comparability of near-nadir satellite observations acquired at different times. This study reveals stable long-term seasonal variations in low-end calibration biases over years, showing a consistent linear relationship with brightness temperatures from 190 to 250 K. After correction, inter-sensor biases were reduced to near-zero levels with minimal temporal drift (within ± 0.02 K), significantly improving the reliability of multi-sensor data integration for quantitative applications in global climate studies.

Recently, the high-order tensor Singular Value Decomposition (t-SVD) and the t-SVD rank has achieved great success in tensor completion. However, the t-SVD rank lacks the flexibility to capture the correlations between different modes of a high-order tensor. In addition, the Tensor Nuclear Norm (TNN), which is a convex surrogate of the t-SVD rank, applies the soft-thresholding operator to the singular value tensor in the transform domain, which will result in a bias. To overcome the mentioned shortcomings in the Low Rank Tensor Completion (LRTC) problem, we first propose a new high-order tensor average rank and its surrogate tensor φ norm. Then, we define the multi-directional tensor average rank and its surrogate multi-directional Tensor φ Norm (MDTN) by the tensor moveaxis operator to characterize the correlations between different modes of a tensor. ADMM based algorithm is designed and experiments on different visual data are carried out, experimental results show that our method outperforms the competing methods.