Guest Editors Marija Strojnik, Wen Chen, Sarath Gunapala, Joern Helbert, Esteban Vera, and Eric Shirley introduce the Special Section on Advanced Infrared Technology and Remote Sensing Applications II.

The experimental Indian Nano-Satellite (INS)-2TD acquires data in a long-wave infrared (7 to 16 μm) region with a fairly good spatial resolution of 175 m. Our study focuses on the retrieval of land surface temperature (LST) using a physics-based generalized single-channel (GSC) algorithm for the INS-2TD observations. A total of 597,240 at-sensor radiance simulations were carried out using moderate resolution atmospheric transmittance 5.3 radiative transfer model for varying conditions pertaining to surface, atmosphere, and sensor geometry to develop and validate the GSC algorithm for broadband INS-2TD sensor. The result from simulated test dataset shows the algorithm’s consistent performance with root-mean-square error (RMSE) of 2.87 K and 0.97 R2. Pixel-to-pixel intercomparison of retrieved LST and standard LST product of Indian National Satellite (INSAT)-3D indicates a good agreement with 0.99 R2 and range of RMSE from 1.17 to 4.78 K over the six selected datasets of South-Asia. The results reveal that the retrieved INS-2TD LST products perform very well, except having a hot bias of around 4.78 K compared to INSAT-3D LST over the Himalayan mountains due to the topographic effect. These results show the overall reasonable accuracy of the retrieved LST over heterogeneous surfaces and highly dynamic atmospheric conditions.

By leveraging the characteristics of different optical sensors, infrared and visible image fusion generates a fused image that combines prominent thermal radiation targets with clear texture details. Existing methods often focus on a single modality or treat two modalities equally, which overlook the distinctive characteristics of each modality and fail to fully utilize their complementary information. To address this problem, we propose an end-to-end infrared and visible image fusion model based on shared-individual multi-scale feature decomposition. First, to extract multi-scale features from source images, a symmetric multi-scale decomposition encoder consisting of nest connections and a multi-scale receptive field network is designed to capture small, medium, and large-scale features. Second, to sufficiently utilize complementary information, common edge feature maps are introduced to the feature decomposition loss function to decompose extracted features into shared and individual features. Third, to aggregate shared and individual features, a shared-individual self-augmented decoder is proposed to take the individual fusion feature maps as the main input and the shared fusion feature maps as the residual input to assist the decoding process and the reconstruct the fused image. Finally, through comparing subjective evaluations and objective metrics, our method demonstrates its superiority compared with the state-of-the-art approaches.

Artificial intelligence has been widely applied to water depth retrieval across various environments, deemed essential for habitat modeling, hydraulic structure design, and watershed management. However, most of these models have been developed for deep waters, with the critical impact of the gradient descent algorithm often not evaluated. To address this gap in current research, this study adopted the artificial neural network with seven gradient descent methods, including step, momentum, quick propagation, delta-bar-delta, conjugate gradient, Levenberg–Marquardt, and resilient backpropagation (RProp), for shallow water depth modeling. Shallow water depths in Taiwan’s mountainous rivers were then modeled using multispectral imagery taken by drone and vegetation indices. From our results, it was revealed that methods optimizing weight updates were outperformed by those based on gradient information, such as RProp. The selection of gradient descent algorithm was identified as pivotal; an inappropriate selection might even result in performance inferior to a traditional linear regression model. In the sensitivity analysis, near-infrared and normalized difference water index were classified as highly sensitive. By leveraging multispectral data and vegetation indices with ANN, the optimal gradient descent algorithm and the critical model input for shallow water modeling were successfully identified, offering invaluable insights for future studies.

Infrared (IR) imaging systems have sensor and optical limitations that result in degraded imagery. Apart from imperfect optics and the finite detector size being responsible for introducing blurring and aliasing, the detector fixed-pattern noise also adds a significant layer of degradation in the collected imagery. Here, we propose a single-shot super-resolution method that compensates for the nonuniformity noise of long-wave IR imaging systems. The strategy combines wavefront modulation and a reconstruction methodology based on total variation and nonlocal means regularizers to recover high-spatial frequencies while reducing noise. In simulations and experiments, we demonstrate a clear improvement of up to 16× in image resolution while significantly decreasing the fixed-pattern noise in the reconstructed images.

Deep neural network-based synthetic aperture radar (SAR) image change detection algorithms are affected by coherent speckle noise in the original image. Existing denoising methods have predominantly focused on generating binary images based on the pre-classification of original pixels, which is insufficient in removing interfering noise. Herein, to further reduce the noise points generated in the clustering algorithm, we combined the characteristics of the fuzzy clustering algorithm, demonstrating the obvious advantages of the proposed fast and flexible denoising convolutional neural network (FFDNet-F) method. An FFDNet was used to reduce noise interference in real SAR images and improve the detection accuracy and robustness of the method. Difference operators were then drawn from the weak noise images, and fuzzy local information C-means clustering was applied for analysis to generate the change detection results. The experimental results from two real datasets and the comparative analysis with other network models demonstrated the effectiveness of this method. Simultaneously, Gaofen-3 satellite images were used to verify and analyze surface flood disasters in Zhengzhou, China. The findings of this study demonstrate a significant improvement in detection accuracy using the proposed method compared with that of other algorithms.

Due to the complexity of hyperspectral data and the scarcity of labeled samples, unsupervised clustering segmentation has become a hot spot of interest in remote sensing. Sparse subspace clustering (SSC) is the most common clustering approach at the moment, although its computational cost restricts its use on big remote sensing datasets. Furthermore, SSC’s neglect of spatial information and limited recognition ability hinder the spatial homogeneity of clustering results. Hence, this work proposes a fast spectral clustering algorithm for local cosine similarity graphs. First, the fuzzy simple linear iterative clustering superpixel method is introduced into the SSC framework to treat superpixels as homogeneous entities and obtain global similarity maps using very low computational and spatial overheads. Then, a cosine similarity measure that combines spectral information and spatial information is used to obtain a local similarity graph, which enhances the accuracy of the final classification and suppresses noise. Extensive testing demonstrates the value of the proposed method. Compared to state-of-the-art SSC-based algorithms, it offers superior classification performance, noise immunity, and very little computational overhead.

We investigate the reliability of satellite river width (SRW) measurements to estimate the river discharge and its sensitivity to various hydro-geomorphological features. The study encompasses SRW extents at 141 in-situ hydrological observation stations, across seven tropical basins in India, with a mean annual discharge ranging from 2351 m3/s to less than 1 m3/s. Integrating optical (Sentinel-2, Landsat) and synthetic-aperture radar (SAR; Sentinel-1) data in the Google Earth Engine (GEE), 63,885 images are processed in the GEE to generate a dense time series of the SRW. Results demonstrate a good correlation (>0.50) between the SRW and in-situ discharge at 61 stations, primarily in the Godavari and Mahanadi basins. Furthermore, SRW-based rating curves exhibit reliable predictive capabilities at 44 stations, highlighting the potential to develop SRW rating curves in sparsely gauged basins. Investigations on the possible impact of different hydro-geomorphological features on the performance of the SRW to estimate the river discharge revealed optimal conditions in river reaches at lower elevations with substantial temporal variations in the discharge and associated variation in the river width along with a history of maximum water spread. Consequently, the Surface Water and Ocean Topography satellite’s river networks in the region are classified based on these findings, with 3567 out of 6132 river reaches identified as suitable for reliable SRW-based discharge estimation.

Extracting roads from complex remote sensing images is a crucial task for applications, such as autonomous driving, path planning, and road navigation. However, conventional convolutional neural network-based road extraction methods mostly rely on square convolutions or dilated convolutions in the local spatial domain. In multi-directional continuous road segmentation, these approaches can lead to poor road connectivity and non-smooth boundaries. Additionally, road areas occluded by shadows, buildings, and vegetation cannot be accurately predicted, which can also affect the connectivity of road segmentation and the smoothness of boundaries. To address these issues, this work proposes a multi-directional spatial connectivity network (MDSC-Net) based on multi-directional strip convolutions. Specifically, we first design a multi-directional spatial pyramid module that utilizes a multi-scale and multi-directional feature fusion to capture the connectivity relationships between neighborhood pixels, effectively distinguishing narrow and scale different roads, and improving the topological connectivity of the roads. Second, we construct an edge residual connection module to continuously learn and integrate the road boundaries and detailed information of shallow feature maps into deep feature maps, which is crucial for the smoothness of road boundaries. Additionally, we devise a high-low threshold connectivity algorithm to extract road pixels obscured by shadows, buildings, and vegetation, further refining textures and road details. Extensive experiments on two distinct public benchmarks, DeepGlobe and Ottawa datasets, demonstrate that MDSC-Net outperforms state-of-the-art methods in extracting road connectivity and boundary smoothness. The source code will be made publicly available at https://github/LYY199873/MDSC-Net.

Reliable rice yield information is critical for global food security. Optical vegetation indices (OVIs) are important parameters for rice yield estimation using remote sensing. Studies have shown that radar vegetation indices (RVIs) are correlated with OVIs. However, research on the implementation of RVIs in rice yield prediction is still in its early stages. In addition, existing deep learning yield prediction models ignore the contribution of temporal features at each time step to the predicted yield and lack the extraction of higher-level features. To address the above issues, this study proposed a rice yield prediction workflow using RVIs and a multiscale one-dimensional convolutional long- and short-term memory network (MultiscaleConv1d-LSTM, MC-LSTM). Sentinel-1 vertical emission and horizontal reception of polarization vertical emission and vertical reception of polarization data and county-level rice yield statistics covering Guangdong Province, China, from 2017 to 2021 were used. The experimental results show that the performance of the RVIs is comparable to that of the OVIs. The proposed MC-LSTM model can effectively improve the accuracy of rice yield prediction. For early rice yield prediction based on RVIs, the optimal accuracy of MC-LSTM [coefficient of determination R2 of 0.67, unbiased root mean square error (ubRMSE) of 217.77 kg/ha] was significantly better than that of the LSTM model (R2 of 0.61, ubRMSE of 229.52 kg/ha). For late rice yield prediction based on RVIs, the optimal accuracy of MC-LSTM (R2 of 0.61 and ubRMSE of 456.54 kg/ha) was significantly better than that of the LSTM model (R2 of 0.55 and ubRMSE of 486.76 kg/ha). The above results show that the proposed method has excellent application prospects in crop yield prediction. This work can provide a new feasible scheme for synthetic-aperture radar data to serve agricultural monitoring and improve the efficiency of rice yield monitoring in a large area.

Early disease detection is required, considering the impacts of diseases on crop yield. However, current methods involve labor-intensive data collection. Thus, unsupervised anomaly detection in time series imagery was proposed, requiring high-resolution unmanned aerial vehicle (UAV) imagery and sophisticated algorithms to identify unknown anomalies amidst complex data patterns to cope with within season crop monitoring and background challenges. The dataset used in this study was acquired by a Micasense Altum sensor on a DJI Matrice 210 UAV with a 4 mm resolution in Gottingen, Germany. The proposed methodology includes (1) date selection for finding the date sensitive to sugar beet changes, (2) vegetation index (VI) selection for finding the one sensitive to sugar beet and its temporal patterns by visual inspection, (3) sugar beet extraction using thresholding and morphological operator, and (4) an ensemble of bottom-up, Kernel, and quadratic discriminate analysis methods for unsupervised time series anomaly detection. The study highlighted the importance of the wide-dynamic-range VI and morphological filtering with time series trimming for accurate disease detection while reducing background errors, achieving a kappa of 76.57%, comparable to deep learning model accuracies, indicating the potential of this approach. Also, 81 days after sowing, image acquisition could begin for cost and time efficient disease detection.

Land surface temperature (LST) serves as a crucial parameter in scientific investigations pertaining to resource management, climate change, and terrestrial ecosystems. The implementation of large-scale water conservancy projects influences the water areas of the water source, thereby affecting regional climate, vegetation, and consequently, the LST. At present, widely used moderate resolution imaging spectroradiometer temperature products often exhibit data loss or distortion due to atmospheric interference or technical challenges. We introduce an advanced interpolation of the mean anomaly method, grounded in the digital elevation model, to address these missing values. Moreover, we utilize indicators such as temperature difference (TD) and project effect change intensity (PECI) to scrutinize the influence of the aforementioned water diversion project on the study area’s LST. Our findings indicate that: (1) The R2 values for the improved interpolation of mean anomalies based on the digital elevation model method for LST reconstruction consistently exceed 0.902, demonstrating higher accuracy in regions with significant elevation differences, especially during nighttime. (2) The TD is correlated with the proximity and size of water bodies. (3) The PECI is notably higher during wet seasons and nighttime. There is a minimal variation in the prevalence of extremely strong impact areas across different periods. (4) Subsequent to the middle route of the South-to-North Water Diversion Project, both warming and cooling effects have been observed in the study area. Notably, the direction of the warming effect is divergent in the northern mountainous regions compared with the central and southern plains.

In a previous study, we developed a method for assessing building damage at the district level in markedly damaged areas using satellite-derived synthetic aperture radar (SAR) data and conducted a coherence analysis of urban areas. We previously evaluated the relationship between the average coherence within a 200-m mesh and degree of building damage in Mashiki, Kumamoto Prefecture, Japan, using L-band ALOS-2 PALSAR-2 data obtained before and after the Kumamoto earthquake. Our findings indicated a significant negative correlation between these parameters. Currently, C-band Sentinel-1 provides global coverage at a high observation frequency. However, previous studies have not sufficiently evaluated the effects of differences in acquisition conditions. Therefore, we aimed to investigate how acquisition conditions affect the relationship between coherence and the building damage rate using images captured by Sentinel-1 C-SAR during the Kumamoto earthquake. The results showed that coherence tended to decrease as the damage rate increased but the correlation was not significant. The difference in coherence per 10% of the building damage rate was less pronounced compared with the results obtained from ALOS-2. Our findings also indicated that the effects of the pre-earthquake acquisition dates and the baseline were less than those of the orbit direction and incidence angle.

We propose a method for optimizing satellite constellations and a series of process strategies to achieve the desired revisit performance for each target in the area of interest. Using a repeat ground track orbit of periodic characteristics, a tailored regional-satellite-constellation method is developed to achieve the desired revisit performance for each target. The study consists of four parts representing four programs: finding orbit elements through optimization techniques, producing an access matrix between targets and satellites, deriving the number of satellites for each target for the desired revisit performance, and deploying satellite groups by minimizing the orbital plane. Then, the superiority of this optimization method is verified through numerical comparisons and contour distributions for revisit performance compared with the existing conventional Walker method.

Road information is a crucial type of geographic information. The extraction of road information from remote sensing images has been widely applied in various fields such as mapping, transportation, and navigation. However, due to the obstruction of buildings, trees, and shadows, or the spectral similarity between roads and buildings, road extraction remains a challenging research topic. Most current methods focus only on the spatial domain, neglecting the information contained in the image frequency domain. Therefore, this work proposes a remote sensing image road extraction model, frequency domain attention encoder-decoder network (FDAENet). This model mainly consists of three parts. First, the encoder is composed of frequency domain transformer modules (FDTMs). The gnConv in the FDTM includes depthwise separable convolution and phase and magnitude (PM) filters, where the PM filter contains a global filter and phase and amplitude filters located in two parallel layers, used to extract feature information of road remote sensing images from the frequency domain. Then, a multi-scale context extraction module is proposed, which introduces appropriate road context information to enhance inference capability. Finally, a stripe convolution module is introduced to capture long-distance context information from four different directions. Experiments on public road datasets show that FDAENet performs excellently in terms of F1, intersection over union, average path length similarity, and other indicators. Visualization results show that FDAENet performs better in extracting complex roads and can effectively extract roads from high-resolution remote sensing images.

To solve the problem of thin-cloud interference of remote sensing (RS) images under hardware-constrained environments, we present an end-to-end cloud-noise-robust lightweight convolution neural network model, 2DDSRU-MobileNet, based on MobileNetV3-small. We first propose a denoised module, which notably improves the robustness of lightweight convolution neural network under a thin-cloud environment. Then using the idea of a deep residual shrinkage network, we construct a two-dimensional deep shrinkage residual unit (2DDSRU). Then we introduce MobileNetV3-small as a background model to solve the problem of RS obstructed by thin cloud; this enables our method to obtain a higher efficiency compared with other methods. Moreover, deformable convolution supplants conventional dilated convolution, leading to agile capturing of contextual information in the thin-cloud image. Finally, the model’s efficiency and accuracy are evaluated using the RSSCN-cloud and UCM-cloud datasets. The experimental results demonstrate the cloud-noise-robust performance of our model in the thin-cloud environment. Under a thin-cloud environment, the proposed 2DDSRU-MobileNet model surpasses classic models, such as AlexNet, VGG16, ResNet50, MobileNet series, ShuffleNet-v1, Efficientnet-b3, GhostNetV2, and MobileViT, in terms of runtime computational complexity, network parameters, and classification accuracy.

During the excavation of the Alaskan Way Viaduct replacement tunnel in Seattle, Washington, a 17.5 m diameter tunnel boring machine (TBM) nicknamed “Big Bertha” was damaged after encountering unexpected subsurface conditions. Significant dewatering of multiple aquifers was required to reach the TBM for repairs. Groundwater drawdown and soil consolidation associated with dewatering created a 0.4 km2 region of initial subsidence with maximum vertical settlements exceeding 2.5 cm between August and December 2014. Dewatering wells remained operational until January 2016 and likely contributed to observed groundwater drawdown in areas outside the region of initial subsidence. To determine how an urban landscape with complex and poorly constrained geologic and hydrologic conditions responds to an extended period of dewatering within multiple aquifers, the rate, duration, spatial extent, and magnitude of dewatering-related displacements were analyzed by combining three paths of Sentinel-1 interferometric synthetic aperture radar data spanning November 2014 to October 2019 into a time series of vertical surface deformation using the minimum acceleration algorithm. Our results show that post-dewatering ground rebound within this complex hydrogeologic system occurred at faster rates and with more significant spatial deformation variability than initial subsidence, reaching rates of up to 17 cm/year coupled with potentially hazardous differential rebounds across short distances. In addition, prolonged groundwater pumping at depths greater than 60 m appears to have induced delayed subsidence over a larger area of ∼20 km2, reaching magnitudes of up to 3 cm and lasting for over 3 years after the cessation of pumping.

The increasing prevalence of nuisance benthic algal blooms in freshwater systems has led to water quality monitoring programs based on the presence and abundance of algae. Large blooms of the nuisance filamentous algae, Cladophora glomerata, have become common in the waters of the Upper Clark Fork River in western Montana. To aid in the understanding of algal growth dynamics, unoccupied aerial vehicle (UAV)-based hyperspectral images were gathered at three field sites along the length of the river throughout the growing season of 2021. Select regions within images covering the spectral range of 400 to 850 nm were labeled based on a combination of professional judgment and spectral profiles and used to train a random forest classifier to identify benthic algal growth across several classes, including benthic growth dominated by Cladophora (Clado), benthic growth dominated by growth forms other than Cladophora (non-Clado), and areas below a visually detectable threshold of benthic growth (bare substrate). After classification, images were stitched together to produce spatial distribution maps of each river reach while also calculating the average percent cover for each reach, achieving an accuracy of approximately 99% relative to manually labeled images. Results of this analysis showed strong variability across each reach, both temporally (up to 40%) and spatially (up to 46%), indicating that UAV-based imaging with high-spatial resolution could augment and therefore improve traditional measurement techniques that are spatially limited, such as spot sampling.

Winter wheat is one of the major food crops in China. It is of great importance to timely and accurately monitor winter wheat cultivation for the formulation of agricultural policies. Therefore, to accurately and effectively calculate the planting information of winter wheat, we proposed a method for extracting winter wheat from short-time series synthetic aperture radar (SAR) data. It combines three new SAR indices, SAR backscatter features, and coherence features for extracting the winter wheat based on Sentinel-1 images from October 2021 to June 2022. First, the SAR backscatter coefficients were counted, and the newly constructed SAR indices and coherence features were introduced to increase the distinction between the winter wheat and other ground objects. Subsequently, the artificial neural network (ANN), support vector machine (SVM), and random forest (RF) classifiers were used to identify the main ground objects. The spatial distribution of winter wheat was obtained, and the accuracy was verified. Finally, the planting area of different growth stages for winter wheat was compared, and the relative errors between the extraction area of winter wheat and official statistics data also were calculated. The results showed that: (1) The accuracy of the RF classifier is better than SVM and ANN in extracting winter wheat, the overall accuracy is 95.653%, the Kappa coefficient is 0.933, and producer accuracy and user accuracy are 97.68% and 98.19%, respectively. (2) A more accurate thematic map of winter wheat can be obtained by combining the SAR backscatter features, new SAR indices, and coherence features. (3) By comparing the extraction results of different growth stages, the accuracy of the tassel stage is the highest. The short-time series SAR data of the tassel stage are constructed to replace the time series SAR data of the complete growth stage. It provides basic data for carrying out acreage and yield estimation of winter wheat before maturity.

It is important to detect wintertime green vegetated cover (WGVC), since it includes cover crop information, which is one of the most important agricultural best management practices used today. Related to this, cover crop area, which is part of WGVC, has been estimated using survey methods traditionally, but remote sensing can be used as a more time- and cost-effective assessment. Previously developed pixel-based methods to assess cover crops using remote sensing can have a salt-and-pepper effect, which lowers classification accuracy. Therefore, object-based classification was applied to estimate the spatial distribution of WGVC across the entire state of Delaware. To reduce the financial burdens of fee-based software products, the workflow was formalized only with open-source remote sensing software and publicly available imagery. WGVC in this study was defined as any vegetation planted or surviving during winter on field crop areas. Obviously, the WGVC area estimated in this study was far more extensive than the surveyed area of conventionally defined cover crops, which had a narrower definition. Applying this methodology across Delaware, the total WGVC was estimated to be 137,297 ha between 12/26/2021 and 04/30/2022. The classification accuracy of each date was evaluated using samples collected from pan-sharpened Landsat 8 and 9 images, and the accuracies were higher than 85%. Kappa statistics were above 74% in all cases. The workflow in this study may improve time, labor, and cost efficiency in other areas.

Self-supervised learning has been widely applied in the field of remote sensing image change detection (CD). Traditional self-supervised learning approaches, on the other hand, are limited to image-level classification tasks, do not provide pixel-level feature learning, and require a certain amount of labeled data for fine-tuning. We offer a self-supervised change detection method that does not require fine-tuning and allows for the acquisition of change result images without the use of labeled data. The suggested method is built on a contrastive learning framework that employs an UNet network to facilitate pixel-level feature reconstruction by introducing guided filtering, resulting in superior edge fit detection results than prior pixel-level self-supervised CD methods. Furthermore, the technique increases the accuracy of CD by incorporating a global contrast module. Simulation experiments were conducted on three public datasets, the Onera Satellite Change Detection dataset, the GF-2 Satellite Change Detection dataset, and the Jiangsu dataset, and compared with state-of-the-art CD methods. The result shows the effectiveness of the proposed algorithms based on evaluation metrics, such as overall accuracy, kappa coefficient, and F1 score.

Passive microwave remote sensing is a valuable tool for snow depth estimation. However, accurate retrieval is limited by nonlinear relationships between the snow depth and passive microwave brightness temperature (TB) that are caused by snow physical properties, underlying surface type, and topographical factors. Our study aims to enhance snow depth estimation in Northern Xinjiang (NX), China, utilizing Advanced Microwave Scanning Radiometer 2 TB data (with a resolution of 0.1 deg) and fractional snow cover products through a combination of wavelet transform and two artificial neural network (ANN) models: feedforward neural network (FFNN) and generalized regression neural network (GRNN). The hybrid models were trained and validated using in situ snow depth observations from 44 stations across NX. Results indicate that applying wavelet transform reduces the root-mean-square error (RMSE) by 28.88% for FFNN. In the snow season of 2013 to 2014, Wavelet-GRNN (RMSE: 7.36 cm, NSE: 0.59, R: 0.78, bias: 1.68 cm) outperforms Wavelet-FFNN (RMSE: 8.26 cm, NSE: 0.48, R: 0.75, bias: 1.69 cm) by 10.90%. However, Wavelet-FFNN exhibits superior performance, up to 13.78% than Wavelet-GRNN in complex topographic areas like Xiaoquzi station. In addition, spatial–temporal estimations demonstrate that the hybrid models surpass three well-known snow depth products and alleviate issues of excessively high or low values in NX. These findings underscore the effectiveness of hybrid models combining wavelet transform and ANNs, integrating passive microwave remote sensing and auxiliary data, for accurate snow depth estimation in mountainous regions.

The southern hilly region of China boasts abundant forest resources, which are crucial for maintaining ecological stability. However, the complex vegetation structure and fragmented terrain in this area lead to intricate and disorderly forest types, resulting in semantic confusion among vegetation in remote sensing images. Consequently, accurately classifying forest types poses significant challenges. We propose a semantic segmentation model with multiple attention mechanisms using convolutional neural networks. We enhance the U-Net model’s encoder with a deeper convolutional network to expand the receptive field without significant computation increase. Furthermore, we integrate spatial attention within the U-Net’s skip connections and multiscale feature fusion. Experimentally, the multiple attention mechanism U-Net model outperforms the original, averaging 90.67% intersection over union, 94.33% pixel accuracy, and 96.00% classification accuracy for 0.5 m resolution forest type classification. These improvements are 8.00%, 4.33%, and 5.00%, respectively. The model accurately distinguishes forest types in the southern hilly region, enabling precise information-based forest supervision.

The use of remote sensing imagery has become indispensable for long-term crop mapping. Synthetic aperture radar can provide high-resolution polarization information on crop time-series canopy changes, making large-scale, long-time-series crop mapping possible. This study presents a method for precise crop mapping using Sentinel-1 data time series from 2015 to 2020. The proposed method, known as the temporal-spatial fusion U-Net (TSFUNet) model, makes use of the temporal characteristics of the time series to improve crop mapping precision. The model consists of a temporal channel fusion module for combining the temporal characteristics of multiple channels. In addition, an improved U-Net with a cascade feature fusion module is introduced to fuse spatial contextual information at different scales for the generation of final crop segmentation. Experimental results demonstrate that the TSFUNet model for temporal Sentinel-1 sequence-based crop mapping outperforms other state-of-the-art methods. For crop mapping at different times, a multi-stage learning (MSL) training strategy is proposed to improve the model’s generalization ability. In addition, the chaotic-Jaya algorithm is employed to optimize the selection of learning rate hyperparameters in MSL and further enhance the model’s high-accuracy crop mapping performance.

Agricultural greenhouses have a negative impact on the ecological environment while bringing huge economic and social benefits. Therefore, it is of great significance to obtain greenhouse information in a timely and accurate manner. Due to the complex spectral characteristics and dense spatial distribution characteristics of greenhouses, although the extraction of greenhouses based on a single semantic segmentation model can extract the area with high precision, the segmentation process has a serious problem of boundary adhesion between greenhouses, which makes it difficult to accurately obtain the quantity of greenhouses. To address this, our study proposes a method for greenhouse extraction that integrates semantic segmentation and edge constraints, using high-spatial-resolution remote sensing images to accurately extract the area and quantity of greenhouses. This method employs an improved semantic segmentation model (AtDy-D-LinkNet) to extract the greenhouse area, which embeds a convolutional attention module into the D-LinkNet and adopts a dynamic upsampling strategy, achieving precise greenhouse extraction. Experiments demonstrate that the improved model increased the recall, precision, F1 score, and intersection over union by 1.68%, 2.27%, 1.93%, and 3.54%, respectively, compared to the original model. To address the significant edge adhesion issue in semantic segmentation and accurately extract the quantity of greenhouses, we developed an edge constraint approach. This approach uses an edge detection model to extract greenhouse boundaries, further constrains the greenhouse surfaces, separates adhered greenhouses, and outputs vector patches representing individual greenhouses, thereby achieving precise greenhouse quantity extraction. The experiments show that this method effectively combines the advantages of semantic segmentation and edge detection. It not only ensures the accuracy of greenhouse area extraction but also effectively solves the boundary adhesion issue, significantly improving quantity extraction accuracy, resulting in vector patches that align with the actual area, quantity, and spatial distribution of greenhouses. This can provide a data foundation for greenhouse management and planning in agriculture.

Spatial and temporal land-use patterns in the Songhua River Basin (SRB) over the past 20 years were analyzed; the influence of natural geographic, socioeconomic, and anthropogenic factors was considered. Using spatial analysis and geodetector modeling, we assessed various indicators to comprehensively analyze land-use changes in the SRB in a long time series (2001 to 2021). Our goal was to determine the extent to which each factor influences land-use change and the mechanisms of interaction. We found that natural geographic factors and anthropogenic factors, particularly elevation and population density, had a greater influence on land-use changes than climatic and socio-economic factors. Despite a positive trend in land use indicated by the composite index, the SRB is experiencing a decrease in undeveloped land resources annually. We also identified that interactions between factors had varying effects, with the superposition of multiple factors potentially exacerbating conflicts between different land-use types. These findings provide valuable insights for strategic planning, policy formulation, and optimization of land resources in the Songhua River Basin.

Hyperspectral unmixing (HU) in hyperspectral image (HSI) processing is a crucial step. However, the accuracy of unmixing methods is limited by the variability in endmember and the complexity of the HSI structure found in natural scenes. Endmember variability refers to the variations or differences exhibited by endmembers in different locations or under varying conditions within a hyperspectral remote sensing scene. Therefore, to enhance the accuracy of unmixing results, it is crucial to fully leverage spectral, geometric, and spatial information within HSIs, comprehensively exploring the spectral characteristics of endmembers. We present a cascaded dual-constrained transformer autoencoder (AE) for HU with endmember variability and spectral geometry. The model utilizes a transformer AE network to extract the global spatial features in the HSI. Additionally, it incorporates the minimum distance constraint to account for the geometric information of the HSI. Given the similarity in shape exhibited by endmembers of each individual material, with the primary endmember variability being expressed through overall intensity fluctuations, an abundance-weighted constraint method for endmember spectral angle distance is proposed. During training, the architecture utilizes two cascaded networks to preserve the detailed information in the HSI. We evaluate the proposed model using three real datasets. The experimental results indicate that the proposed method achieves superior performance in abundance estimation and endmember extraction. Furthermore, the effectiveness of the two constraint methods was verified through ablation experiments.

To address the problem of large point cloud data and low segmentation accuracy in indoor and outdoor scenes, we propose a point cloud plane segmentation algorithm based on multiscale region fusion. When using traditional region growing methods for point cloud plane segmentation, the segmentation results are often unstable, and adjacent planes are difficult to separate. In our method, we convert the point cloud data in space into a three-dimensional grid model, and then use the proposed tensor voting method for accurate normal vector estimation. By introducing a multiscale weighted similarity measure to optimize the seed point growing rules, we enhance the accuracy of plane segmentation. Specifically, the algorithm first performs custom grid downsampling and filtering on the original point cloud data. Then different region growing parameters are applied to segment the super-voxel point cloud data at each scale, resulting in multiple segmentation results. The tensor voting is used for estimating the normal vectors of the point cloud, which achieves good estimation performance even at sharp edges. Finally, weights are assigned to similar feature regions, and the similarity measure optimizes the seed point growing rules. With accurate normal vector estimation and appropriate seed point selection, the plane segmentation is more complete. The final segmentation result is obtained by edge preservation and refinement of multiple segmentation results. Experimental results on public and self-collected datasets demonstrate that our proposed method can effectively improve segmentation accuracy while ensuring the real-time performance.

Cloud cover is a persistent challenge in remote sensing imagery, hindering accurate interpretation and analysis. Existing research focuses on supervised approaches to removing clouds from satellite images. Due to the highly challenging task of acquiring paired data for the area of interest with and without obstructions, we propose an unsupervised approach for thin cloud removal using the latest image-to-image translation generative adversarial network (GAN) called the UNet vision transformer cycle-consistent GAN (UVCGAN), enhanced with variational mode decomposition (VMD). Thin cloud removal from satellite images is adopted as an image-to-image translation task in this approach. VMD is used to enhance the cloud-covered input image by retaining most of the image-specific features by reconstructing the enhanced image from modes that have the most image-specific features, quantitatively identified by modes with the highest entropy, contrast, and energy. The enhanced image is taken as input by the UVCGAN model to generate an image without thin clouds. The proposed methodology is compared against the latest methods, and quantitative evaluations indicate superior performances in terms of both full- and no-reference metrics, affirming the reliability and robustness of our approach.

The classification of high-resolution remote sensing images finds widespread applications, yet achieving accurate classification of land types often relies heavily on labeled samples. However, obtaining labeled samples is a challenging and time-consuming task. To mitigate the algorithm’s dependence on labeled samples and reduce computational time and resource consumption, we propose an approach combining a context-aggregated transformer-based network (TSNet) with a differentiable feature clustering method for unsupervised remote sensing image classification. The algorithm comprises an end-to-end network model consisting of a feature extractor (TSNet), classifier, and argmax function. It introduces a loss function that incorporates feature similarity loss and spatial continuity loss of differentiable feature clustering to address the limitations of fixed boundaries, eliminating the need for training data. TSNet combines the strengths of both convolutional neural network (CNN) and transformer. It utilizes the transformer module to obtain hierarchical features, employs a CNN decoder to aggregate context, and integrates a multi-branch prediction header with classifiers from three CNNs to generate the classification map, enhancing supervision in deeper layers. The proposed approach has been tested on a public dataset (LoveDA), and compared with UNet-Net, CNN-Net, KMeans, and iterative self-organizing data analysis technique algorithm, the experimental results show that our method achieves the best results in terms of adjusted Rand index, adjusted mutual information, FMI, Dice, Jaccard, and frequency weighted intersection over union metrics.

The high resolution remote sensing images are characterized by rich surface details and diverse features, and the single-modality high-resolution images suffer from limited expressive ability in the earth object segmentation application scenarios. We propose a multi-modal remote sensing image segmentation method based on attention-driven dual-branch encoding framework. The method involves parallel encoding of multi-modal remote sensing data to thoroughly extract features from each modality. Furthermore, multistage multi-modal features are fused by attention-driven feature fusion modules to generate high-quality multi-modal feature representation. Extensive experiments are carried out on the International Society for Photogrammetry and Remote Sensing Vaihingen and Potsdam 2D semantic labeling datasets. The datasets include both RGB/IRRG images and digital surface model (DSM) images. Experimental results demonstrate that: (1) the elevation information of DSM images can bring obvious benefits to the earth objects with significant heights, and introducing DSM images properly can improve the segmentation accuracy compared to using only RGB/IRRG images; (2) the attention-driven feature fusion module outperforms traditional feature fusion methods in capturing cross-modal complementary features, leading to outstanding segmentation accuracy for each earth object.

A synthetic aperture radar (SAR) system is a notable source of information, recognized for its capability to operate day and night and in all weather conditions, making it essential for various applications. SAR image formation is a pivotal step in radar imaging, essential for transforming complex raw radar data into interpretable and utilizable imagery. Nowadays, advancements in SAR sensor design, resulting in very wide swaths, generate a massive volume of data, necessitating extensive processing. Traditional methods of SAR image formation often involve resource-intensive and time-consuming postprocessing. There is a vital need to automate this process in near-real-time, enabling fast responses for various applications, including image classification and object detection. We present an SAR processing pipeline comprising a complex 2D autofocus SARNet, followed by a CNN-based classification model. The complex 2D autofocus SARNet is employed for image formation, utilizing an encoder–decoder architecture, such as U-Net and a modified version of ResU-Net. Meanwhile, the image classification task is accomplished using a CNN-based classification model. This framework allows us to obtain near real-time results, specifically for quick image viewing and scene classification. Several experiments were conducted using real-SAR raw data collected by the European remote sensing satellite to validate the proposed pipeline. The performance evaluation of the processing pipeline is conducted through visual assessment as well as quantitative assessment using standard metrics, such as the structural similarity index and the peak-signal-to-noise ratio. The experimental results demonstrate the processing pipeline’s robustness, efficiency, reliability, and responsivity in providing an integrated neural network-based SAR processing pipeline.

Geometric calibration is an essential technique that improves the absolute positioning accuracy of synthetic aperture radar (SAR) imagery. Geometric calibration of SAR has recently been moving toward a field-free approach. Field-free geometric calibration leverages the consistency in positioning conjugate image points across multiple images to determine ground coordinates and geometric calibration parameters. Notably, when the number of available images is limited, field-free geometric calibration may present instability in determining calibration parameters. We conduct a detailed analysis of the various sources of error that can affect SAR geometric positioning. In addition, we introduce the concept of dilution of precision (DOP) to assess the accuracy of the obtained calibration parameters. When the DOP is less than 5, we believe the geometric calibration parameters to be accurate and reliable. We analyze the impact of satellite geometry distribution and calibrated image quantity on DOP. Out of the 18 YG-13 satellite images captured in the Songshan area of Henan province, 3 and 4 images were chosen for geometric calibration, resulting in 816 and 3060 potential combinations, respectively. Calibrated solutions are obtained for each combination, and combinations with inadequate assessment accuracy are filtered out based on the DOP. After getting the geometric calibration parameters from the solution, we compensate them in the positioning equation and compare the accuracy of the two checkpoints before and after screening. The findings demonstrate that the post-screening combination surpasses the pre-screening combination and achieves comparable performance to field-free calibration and cross calibration concerning image localization accuracy. The effectiveness of DOP in evaluating the accuracy of geometric calibration parameter estimation has been corroborated.

Radar receivers are vital components in modern radar systems, and their reliable operation is crucial for accurate target detection and tracking. However, degrading receiver components can lead to reduced gain, increased noise levels, and decreased probability of detection affecting the overall radar performance. We present an efficient real-time prognostic framework for a radar receiver. The effect of the performance degradation of critical devices on the radar receiver is analyzed. A prognostic framework is developed based on the relationship between device health and receiver performance. Subsequently, an improved prognostic model based on the integration of Weibull distribution and long short-term memory network is developed and trained to accurately estimate the remaining useful life (RUL) of the receiver. Integrating survival analysis and deep learning techniques offers a robust solution for accurate RUL estimation, which can significantly enhance maintenance strategies. The proposed framework facilitates transitioning from traditional reactive maintenance practices to a predictive maintenance approach, thereby reducing downtime and improving the overall availability of radar receivers.

Inaccurate solar vector orientation knowledge can considerably deteriorate calibration results for the Visible Infrared Imaging Radiometer Suite (VIIRS). We develop a methodology to use the Suomi National Polar-orbiting Partnership (SNPP) VIIRS solar diffuser stability monitor (SDSM) sun view data to assess the knowledge accuracy of the solar angles that reside in the onboard calibrator intermediate product (OBCIP) files used for on-orbit radiometric calibration. We applied an initial version of this methodology in 2013 and found that the solar declination angle had a relative error that varied between ∼0 deg to 0.17 deg. The relative error is referenced to the error at the SNPP satellite yaw maneuver time that occurred on February 15 to 16, 2012. Our mission long results from the current methodology show that the solar vector angular knowledge error occurred from the early mission until mission day 1129 (November 30, 2014). The error undulates yearly with the largest error in the solar declination angle increasing from ∼0.17 deg in the first year to 0.19 deg in the third year, agreeing with the solar vector error root cause understanding realized in early 2014. With the reprocessed OBCIP files, we find the solar vector declination and azimuth angular knowledge errors have near zero biases. The detection limit of this methodology strongly depends on how finely the solar angle is sampled by the SDSM detectors. With the SDSM sun view data collected when the SDSM operated once per day, this methodology yields detection standard deviations of 0.013 deg and 0.024 deg for the solar declination and azimuth angles. With a 3-sigma criterion, at the detection limits, the solar orientation errors result in a calibration error of 0.088%. This method can be applied to other Earth-orbiting sensors.

Over the last several decades, large wildfires have become increasingly common across the United States causing a disproportionate impact on forest health and function, human well-being, and the economy. Here, we examine the severity of large wildfires across the Contiguous United States over the past decade (2011 to 2020) using a wide array of meteorological, land cover, and topographical features in a deep neural network model. A total of 4538 wildfire incidents were used in the analysis covering 87,305 square miles of burned area. We observed the highest number of large wildfires in California, Texas, and Idaho, with lightning causing 43% of these incidents. Importantly, results indicate that the severity of wildfire occurrences is highly correlated with the weather, land cover, and elevation of the study area as indicated from their SHapley Additive exPlanations values. Overall, different variants of data-driven models and their results could provide useful guidance in managing landscapes for large wildfires under changing climate and disturbance regimes.

The multisource satellite observation data have been widely used in carbon cycle research owing to their long-term and large-scale characteristics. However, the sparse sampling density of satellite observation data often results in incomplete spatiotemporal coverage at certain time intervals, which hinders the accurate representation of global carbon dioxide (CO2) concentration variations and is inadequate for supporting research applications with different precision requirements. To address this issue, a new multiscale fixed rank kriging is proposed to generate long-term daily scale column-averaged dry-air mole fraction of CO2 (XCO2) products from 2016 to 2019 over the globe on grids of 1°, for which the XCO2 data from Orbiting Carbon Observatory-2, Orbiting Carbon Observatory-3, and Greenhouse gases Observing SATellite are applied. Experimental results show that the dataset has a high spatiotemporal resolution and coverage validated by the Total Carbon Column Observing Network data to effectively fill gaps in satellite observation data, with cross-validation of R2=0.93 and root mean square error = 1.06 ppm. Moreover, we analyze the spatial distribution and seasonal variation characteristics of global and Chinese XCO2 from 2016 to 2019, with XCO2 presenting an obvious latitudinal gradient and seasonal periodicity in space. The proposed method establishes a foundational research dataset for the analysis of spatiotemporal variation characteristics of CO2 concentration at global and regional scales, as well as investigations on carbon sources and sink.

We provide an innovative methodology for detecting small objects in remote sensing imagery. Our method addresses challenges related to missed and false detections caused by the limited pixel representation of small objects. It integrates super-resolution technology with dynamic feature fusion to enhance detection accuracy. We introduce a cross-stage local feature fusion module to improve feature extraction. In addition, we propose a super-resolution network with soft thresholding to refine small object features, resulting in improving resolution of feature maps while reducing redundancy. Furthermore, we embed a dynamic fusion module based on feature space relationships into a dual-branch network to strengthen the role of the super-resolution branch. Experimental validation on DIOR and NWPU VHR-10 datasets shows mAP improvements to 73.9% and 93.7%, respectively, with FLOPs of 24.89G and 22.33G. Our method outperforms existing approaches regarding accuracy and number of parameters, effectively addressing challenges in small object detection in remote sensing imagery.

The 20-year spatio-temporal variability of atmospheric methane (CH4) and its trends are presented using the CH4 volume mixing ratio (CH4 VMR) retrieved from Atmospheric Infrared Sounder. These data have a 1 deg×1 deg spatial resolution and cover pressure levels between 600 and 200 hPa, latitudes 67°S–67°N, and the time interval from March 2003 to February 2023. The vertical distribution of the CH4 VMR showed a definite decrease with altitude and a long-term increase in time. There was an obvious gradient with latitude, which increased from south to north. The horizontal distribution released strong regional signals. In the Northern Hemisphere, the high values of the CH4 VMR were mainly distributed in the northern temperate zone and most of the northern subtropic, and the low values were mainly observed in tropical Africa. The spatial distributions of seasonal growth rates presented clear climatic zone characteristics for all four seasons. Empirical orthogonal function analysis further illustrated a significant increase of CH4 VMR during the past 20 years and predicted a continued increase in the next few years. The comparative analysis between the Oceanic Niño Index and ΔCH4 VMR indicated that CH4 emissions are restrained under the influence of El Niño-Southern Oscillation.