The Gravity Recovery and Climate Experiment (GRACE) satellite provides an unprecedented tool for monitoring large-scale terrestrial water storage (TWS) changes. Yet, its coarse resolution restricts its effectiveness in areas with complex hydrogeological environments, such as southwestern China. To address this limitation, we propose a novel method to improve the spatial resolution of GRACE observations. Our approach leverages a deep learning downscaling model that integrates generative adversarial networks (GANs) and transformer attention mechanisms to derive the spatial patterns of TWS variations. The model incorporates the estimated total water storage changes from GRACE and some hydrological variables—including the digital elevation model (DEM), soil moisture, evapotranspiration, temperature, and precipitation—to enhance the resolution and accuracy of GRACE data. By implementing this method, we successfully increased the spatial resolution of GRACE observations from 0.25° to 0.05°. The advanced neural network downscaling model can accurately characterize local water storage variations, with Nash–Sutcliffe efficiency (NSE) values ranging from 0.58 to 0.92. Moreover, this model not only significantly increases the spatial resolution but also maintains the spatial distribution, offering valuable insights for regional water resources management and fostering small-scale hydrological research. The results have profound implications for sustainable water resources management and climate change assessment.

Hyperspectral imagery is valuable for accurate detection of land-cover changes within a consistent area across time. However, current training paradigms for hyperspectral change detection (CD) are usually in supervised form which are not consistent to heavy labeling cost reality. Moreover, current deep learning methods face limitations due to insufficient temporal dependencies’ sharing and inadequate densely bandwise spatial position dependencies. To tackle these challenges, we introduce an innovative semi-supervised training paradigm called Conflict HyperMatch and a deep learning model called temporal weight-sharing and bandwise spatial interacting transformer (TWBSIT) for hyperspectral CD. The key contributions of this study are as follows: 1) introduction of the Conflict HyperMatch training schedule, which relies on representation discrepancy and weak-to-strong prediction consistency, to improve the sample leveraging of both labeled and unlabeled data; 2) weight-sharing interacting temporal attention (WITA) module is contributed to capture shared temporal interactions and resemblances; and 3) bandwise cross-spatial attention (BCSA) module is introduced to achieve a more extensive spatial perception of the specific central image patch. Extensive experiments conducted on three real datasets validate the effectiveness of the proposed TWBSIT model based on Conflict HyperMatch in leveraging both labeled and unlabeled samples for hyperspectral CD. This method notably decreases the requirement for labeled training samples and surpasses the performance of many existing hyperspectral CD methods.

Hyperspectral image (HSI) denoising is a crucial step for its subsequent applications. In this article, we propose TDSAT, a 3-D spatial-spectral attention Transformer model designed to effectively remove noise in HSI processing while preserving essential spectral and spatial information. The primary objective of this model is to utilize the 3-D Transformer to explore the global spectral-spatial features in HSI, learn the relationships among different bands, and preserve high-quality spectral and spatial information for denoising. The proposed method consists of three main components: the multihead spectral attention (MHSA) module, the gated-dconv feedforward network (GDFN) module, and the spectral enhancement (SpeE) module. The MHSA module learns the relationships among different bands and emphasizes the local spatial information. The GDFN module explores more expressive and discriminative spectral features. The SpeE module enhances the perception of subtle differences between different spectrums. Moreover, unlike the previous Transformer denoising method that can only handle fixed bands, the proposed method combines 3-D convolution and spectral-spatial attention Transformer blocks, enabling the denoising of HSI with an arbitrary number of bands. Experimental results demonstrate that TDSAT outperforms compared methods. The code is available athttps://github.com/Featherrain/TDSAT.

Deep learning-based methods have made great progress in hyperspectral target detection. Unfortunately, the insufficient utilization of spatial information in most methods leaves deep learning-based methods to confront ineffectiveness. To ameliorate this issue, an attention-based multiscale spectral-spatial detector (AMSSD) for hyperspectral target detection is proposed. Firstly, the AMSSD leverages the Siamese structure to establish a similarity discrimination network, which can enlarge intraclass similarity and interclass dissimilarity to facilitate better discrimination between the target and the background. Secondly, 1D CNN and vision Transformer are used combinedly to extract spectral-spatial features more feasibly and adaptively. The joint use of spectral-spatial information can obtain more comprehensive features, which promotes subsequent similarity measurement. Finally, a multiscale spectral-spatial difference feature fusion module is devised to integrate spectral-spatial difference features of different scales to obtain more distinguishable representation and boost detection competence. Experiments conducted on two HSI datasets indicate that the AMSSD outperforms seven compared methods.

At present, hyperspectral image classification (HSIC) technology based on deep learning has been widely explored. However, the time and labor cost of obtaining enough labeled samples are expensive. To obtain higher classification performance with a few number of labeled samples, a coarse-to-fine semi-supervised classification learning (CFSSL) method is proposed in this letter. First of all, the CFSSL performs coarse-grained classification with a few number of labeled samples, and the breaking-ties (BT) criterion is introduced to sample the coarse-grained classification results to ensure that the samples with high confidence are selected to generate pseudo-labels. Then, the pseudo-labels and their corresponding unlabeled samples are sent to the feature extraction network for fine-grained classification, so as to obtain more advanced classification results. Finally, in the fine-grained classification stage, a multi-scale convolution kernel attention aggregation network (A2-MCKN) is designed to simultaneously extract the spatial-spectral features of the image and ensure clear texture boundaries of ground objects. Experimental results on two public datasets show that the CFSSL can obtain better accuracy than other methods with a few number of labeled samples.

Benefiting from powerful feature extraction capabilities, convolutional neural networks (CNNs) have gained prominence in hyperspectral image (HSI) classification. Nevertheless, with restricted receptive fields of convolution kernels, CNN-based methods fail to learn complex characteristics of long-range sequences. Meanwhile, vision transformer allows us to learn long-range dependencies in a global view, but local region features are ignored. To overcome these limitations, we propose a novel method entitled global-local three-dimensional convolutional transformer network (GTCT), where 3-D convolution is embedded in a dual-branch transformer to simultaneously capture global-local associations in both spectral and spatial domains. In particular, the global-local spectral convolutional transformer (GECT) is designed to exploit global spectral sequence signatures and local spectral relationships between bands. Symmetrically, the global-local spatial convolutional transformer (GACT) is devised to exploit local spatial context features and global interactions among different pixels. In addition, multiscale global-local spectral-spatial information is adaptively fused with trainable weights by the weighted multiscale spectral-spatial feature interaction (WMSFI) module. It is worth noting that a spectral-spatial global attention mechanism (SSGAM) is incorporated into multi-head convolutional attention to further integrate discriminative spectral-spatial information. Extensive experiments on four HSI datasets, including GF-5 and ZY1-02D satellite hyperspectral images, demonstrate the superiority of the proposed GTCT method over other state-of-the-art algorithms with fewer parameters and lower floating-point operations (FLOPs) in practical applications.

Improving the spatial resolution of hyperspectral (HS) images is of great significance for the subsequent applications.1As the multispectral (MS) image can provide abundant complementary land-cover spatial information, hyperspectral and multispectral image fusion (HMF) have become a mainstream to generate HS images with both high spatial and spectral resolution. HMF has witnessed rapid progress by leveraging dictionary learning technique. However, existing approaches are highly sensitive to the image registration accuracy, and the reconstruction performance of the non-overlapped spectral bands between HS and MS image are extremely limited. To alleviate the effect of image misregistration and enrich the spectral information of non-overlapped bands, a general HMF dictionary learning framework which considers non-overlapped spectral bands reconstruction and image misregistration is proposed in this paper. For registration error, the proposed method is rectified by the improved dictionary learning, which can solve the problem of the spectral information matching gap existing in traditional HMF methods between HS image with MS image. Meanwhile, for non-overlapped spectral bands reconstruction, a novel coefficient optimization strategy is adopted to improve the non-overlapped bands reconstruction. Therefore, the registration error can be avoided to greatest extent and the accuracy of non-overlapped bands reconstruction can be effectively improved. Experiments both on simulated and real-world datasets demonstrate that the proposed method can effectively tackle the registration error problem and increase HMF accuracy with different spectral range. Meanwhile, the proposed framework provides guidance significance for the dictionary learning based HMF methods with various constrains to improve the non-overlapped bands reconstruction accuracy.

With the development of remote sensing technology, high-resolution (HR) remote sensing optical images have gradually become the main source of change detection data. Albeit, the change detection for HR remote sensing images still faces challenges: 1) in complex scenes, a region contains a large amount of semantic information, which makes it difficult to accurately locate the boundaries between different semantics in the feature maps and 2) due to the inability to maintain consistent conditions such as light, weather, and other factors when acquiring bitemporal images, confounding factors such as the style of bitemporal data that are not related to change detection can cause detection difficulties. Therefore, a change detection method based on feature interaction and multitask learning (FMCD) is proposed in this article. To improve the ability to detect changes in complex scenes, FMCD models the context information of features through a multilevel feature interaction module, so as to obtain representative features, and to improve the sensitivity of the model to changes, the interaction between two temporal features is realized through the mix attention block (MAB). In addition, to eliminate the influence of weather and other factors, FMCD adopts a multitask learning strategy, takes domain adaptation as an auxiliary task, and maps the features of bitemporal images to the same space through the feature relationship adaptation module (FRAM) and feature distribution adaptation module (FDAM). Experiments on three datasets show that the proposed method is superior to other state-of-the-art methods.

Traditional spatiotemporal fusion methods utilize remote sensing images from two or three adjacent dates to predict missing images. Thus, temporal information in longtime data sets is not fully utilized, increasing the costs of long time series construction in terms of time and labor. Here, we propose a synchronous long time-series completion method using a 3-D fully convolutional neural network (LTSC3D). This method can be used to convert long time-series high-temporal-low-spatial (HTLS)- and high-spatial-low-temporal (HSLT)-resolution remotely sensed images into 4-D data sets. From these data sets, we can then extract and fuse spatiotemporal features to produce synchronous long time-series high-temporal-high-spatial (HTHS)-resolution predictions. The model was tested using simulated and real data sets and compared with four representative traditional spatiotemporal fusion methods. The results demonstrated the high accuracy and high efficiency of our method.

Despite success in the same-scene hyperspectral image classification (HSIC), for the cross-scene classification, samples between source and target scenes are not drawn from the independent and identical distribution, resulting in significant performance degradation. To tackle this issue, a novel unsupervised domain adaptation (UDA) framework toward multilevel features and decision boundaries (ToMF-B) is proposed for the cross-scene HSIC, which can align task-related features and learn task-specific decision boundaries in parallel. Based on the maximum classifier discrepancy, a two-stage alignment scheme is proposed to bridge the interdomain gap and generate discriminative decision boundaries. In addition, to fully learn task-related and domain-confusing features, a convolutional neural network (CNN) and Transformer-based multilevel features extractor (generator) is developed to enrich the feature representation of two domains. Furthermore, to alleviate the harm even the negative transfer to UDA caused by task-irrelevant features, a task-oriented feature decomposition method is leveraged to enhance the task-related features while suppressing task-irrelevant features, and enabling the aligned domain-invariant features can be contributed to the classification task explicitly. Extensive experiments on three cross-scene HSI benchmarks have validated the effectiveness of the proposed framework.

Since hyperspectral remote sensing (HRS) came in the middle 1980s, a number of hyperspectral imagers (e.g., EO-1 Hyperion) have been developed all over the world. China, as one of the pioneers in HRS technology development, also has been active and contributed significantly to the HRS community. This paper updates recent advances and future plans in Chinese developments in spaceborne imaging spectroscopy. Particularly, two powerful civilian micro/nano-hyperspectral mini-satellites (Spark01 and Spark02) as well as China's first Carbon monitoring satellite (called TanSat), all newly launched at the end of 2016 were introduced.

Remote sensing of top-of-canopy (TOC) long-term sun-induced chlorophyll fluorescence (SIF) is necessary to better understand the SIF-photosynthesis relationship. Statistical methods provide an alternative to TOC SIF retrieval, as they are independent of synchronous irradiance measurements and may better describe actual irradiance. This letter aims to evaluate the feasibility of using statistical methods for time series TOC SIF retrieval in the absence of synchronous irradiance measurements. Results show that the training set should include nonfluorescent radiance spectra under a variety of solar zenith angles, and that water vapor is an important contributor of spectral variation within 717-745 nm. On the diurnal scale, atmospheric features trained from irradiance spectra can be used to retrieve SIF values from high-frequency upwelling radiance spectra. Features independently trained from nonfluorescent radiance spectra measured on one day can be used for SIF retrieval on a different day within a relatively short period. Our results show that statistical methods have the potential to simplify ground-based SIF measurements and data processing.

Basing on the Aerosol Product data from the AERONET website, We analyzed the 550nm aerosol optical depth(AOD), Angstrom exponent(AE), aerosol volume concentration and aerosol size distribution from 2003 to 2013, excluding the year 2008 due to lacking of data. The research shows: high AOD often occurred in spring and summer. The coarse particle such as dust particle dominates in the spring and the fine particle dominates in the summer. On the interannual variation, there is an obvious trend: From 2003 to 2007, the AOD was often above 1.5 due to the dust pollution from north areas in spring and took the second place in summer. Before and after the Beijing Olympic Games, the atmospheric condition improved. Nevertheless, the AOD rised again till 2012 and 2013. In addition, fine mode aerosol optical depth is closed concerned with PM2.5 and the fine particle volume concentration affects the AOD more. In a word, the study of aerosol changing rule in Beijing from 2003 to 2013 can reflect some environment issues and contribute to atmospheric improvement as a related reference.

Real-time target detection requires immediate decision-making. For fast implementation of the entire detection process, offline band selection is applied. Because target signature is usually the only prior knowledge before running causal real-time detectors, traditional band selection methods based on statistics of the entire data set cannot be used before detection. This paper develops a band selection method based on target spectrum alone, extracting bands that can preserve main features of the target. It is expected that the selected bands can remain the detection accuracy unchanged and speed up the detection process. To substantiate the utility of proposed method, causal real-time constrained energy minimization (RT-CEM) is tested on a real hyperspectral data set for experiments.

This method presented in this paper was tested using CASI&SASI data collected from a site at the Huailai County, Hebei Province, China. Various methods for retrieving AOT and CWV specific to this region were assessed. Results show that retrieval of AOT from the remote sensing data required establishing empirical relationships between 465.6nm, 659nm and 2105nm augmented by ground-based reflectance validation data, and minimizing the merit function based on optimization of AOT@550nm. The paper also extends the SODA method using Powell's method to optimize the retrieval of CWV. The resultant CWV image shows relatively less residual surface features compared with the standard methods. Comparison of derived remote sensing surface reflectance with ground spectra of comparable vegetation and soil targets show significant correlations (R2) is 0.942 and 0.786, with RMSE being 0.0387 and 0.0406 respectively. Therefore, the method proposed in this paper is reliable enough for integrated atmospheric correction and surface reflectance retrieval from hyperspectral remote sensing data. This case provides a good reference to the surface reflectance inversion under the condition of lack of synchronized atmospheric parameters.

Locality-preserving projection (LPP) is a typical manifold-based dimensionality reduction (DR) method, which has been successfully applied to some pattern recognition tasks. However, LPP depends on an underlying adjacency graph, which has several problems when it is applied to hyperspectral image (HSI) processing. The adjacency graph is artificially created in advance, which may not be suitable for the following DR and classification. It is also difficult to determine an appropriate neighborhood size in graph construction. Additionally, only the information of local neighboring data points is considered in LPP, which is limited for improving classification accuracy. To address these problems, a modified version of the original LPP called MLPP is proposed for hyperspectral remote-sensing image classification. The idea is to select a different number of nearest neighbors for each data point adaptively and to focus on maximizing the distance between nonnearest neighboring points. This not only preserves the intrinsic geometric structure of the data but also increases the separability among ground objects with different spectral characteristics. Moreover, MLPP does not depend on any parameters or prior knowledge. Experiments on two real HSIs from different sensors demonstrate that MLPP is remarkably superior to other conventional DR methods in enhancing classification performance.

Sunlit vegetation and shaded vegetation are inseparable parts for most remotely sensed images, and the presence of shadows affects high spatial resolution remote sensing and multiangle remote sensing data. Shadows can lead to either a reduction in or a total loss of information in an image. This can potentially lead to the corruption of biophysical parameters derived from pixel values, such as vegetation indexes (VIs). VIs are widely used in remote sensing inversion applications. If the effects of shadows are not properly accounted for, retrieval may be uncertain when using a VI to calculate vegetation parameters. One of the major reasons that the effects of shadows are easy to be ignored in remote sensing is the spatial resolution of the measurement. High spatial and spectral resolutions are typically difficult to achieve simultaneously, and images that have one tend to not have the other. A ground-based imaging spectrometer brings a turning point to solve this problem as it can obtain both high spatial and high spectral resolutions to obtain feature and shadow images simultaneously. The resolution of the system used here was 1 mm at a height of 1 m, and the spectral resolution was better than 5 nm. For each pixel, the spectral curve of the image was almost a pure-pixel spectral curve, which allowed the differentiation of sunlit and shaded areas. To investigate the effects of shadows on different indexes, 14 hyperspectral VIs were calculated. Moreover, the vegetation fractional coverage calculated using the same 14 VIs was compared. The results show that shadows affect not only each narrowband of a VI but also vegetation parameters.

Independent component analysis (ICA) has been recently applied into hyperspectral unmixing as a result of its low computation time and its ability to perform without prior information. However, when applying ICA for hyperspectral unmixing, the independence assumption in the ICA model conflicts with the abundance sum-to-one constraint and the abundance nonnegative constraint in the linear mixture model, which affects the hyperspectral unmixing accuracy. In this paper, we consider an abundance matrix composed of Np-dimensional variables, and we propose a new hyperspectral unmixing approach with an abundance characteristic-based ICA model. Two characteristics of the abundance variables are explored, and the model is constructed by these characteristics. A corresponding gradient descent algorithm is also proposed to solve the proposed objective function. Both the synthetic and real experimental results demonstrate that the proposed method performs better than the other state-of-the-art methods in abundance and endmember extraction.

Advances in remote sensing paved a way to understand the landuse and landcover features from the eye of space borne sensors. Spectral indices are used to analyze these features. This study focuses on the two very important orthogonal indices for water mapping - Universal Pattern Decomposition Method (UPDM) and Tasseled Cap Transformation (TCT) by considering Landsat 8 data. Results are compared with another very famous water index Modified Normalized Difference Water Index (MNDWI). It was found that wetness index of TCT didn't give good visual interpretation of water body in highly dense vegetative areas. So, greenness index of TCT was used for water delineation. And then results were compared with MNDWI. For good visual interpretation, UPDM seems better than TCT.

Spectral index method was widely applied to the inversion of crop chlorophyll content. In this study, PSR3500 spectrometer and SPAD-502 chlorophyll fluorometer were used to acquire the spectrum and relative chlorophyll content (SPAD value) of winter wheat leaves on May 2nd 2013 when it was at the jointing stage of winter wheat. Then the measured spectra were resampled to simulated TM multispectral data and Hyperion hyperspectral data respectively, using the Gaussian spectral response function. We chose four typical spectral indices including Normalized Difference Vegetation Index (NDVI), Triangle Vegetation Index (TVI), the ratio of Modified Transformed Chlorophyll Absorption Ratio Index(MCARI) to Optimized Soil Adjusted Vegetation Index(OSAVI) (MCARI/OSAVI) and Vegetation Index Based on Universal Pattern Decomposition Method (VIUPD), which were constructed with the feature bands sensitive to the vegetation chlorophyll. After calculating these spectral indices based on the resampling TM and Hyperion data, the regression equation between spectral indices and chlorophyll content was established. For TM, the result indicates that VIUPD has the best correlation with chlorophyll (R2=0.8197) followed by NDVI (R2=0.7918), while MCARI/OSAVI and TVI also show a good correlation with R2higher than 0.5. For the simulated Hyperion data, VIUPD again ranks first with R2=0.8171, followed by MCARI/OSAVI (R2=0.6586), while NDVI and TVI show very low values with R2lesser than 0.2. It is demonstrated that VIUPD has the best accuracy and stability to estimate chlorophyll of winter wheat whether using simulated TM data or Hyperion data, which reaffirms that VIUPD is comparatively sensor independent.

Traditional hyperspectral classification methods based on per-pixel spectral or texture features fail to take account of spatial structure and spatial correlation characteristics. In order to overcome this problem, a mixed classification method is proposed which incorporates spatial information by fusion of object-based segmentation with pixel-wise classifier. This paper tentatively assesses two mixed classification strategies: (1) Combine multi-resolution segmentation algorithm which based on Fractal Net Evolution Approach with the use of Support Vector Machine (MSVM); (2) Combine multi-scale watershed segmentation with Support Vector Machine (WSVM). The two methods were applied to Tiangong-01 hyperspectral urban data and the results showed that the proposed methods improve the classification accuracy effectively which not only avoid the spectral confusion to some extent but also mitigate the land fragmentation problem.

This study introduced the vegetation index based on the universal pattern decomposition method (VIUPD) and then applied on a new sensor - Landsat 8 Operational Land Imager (OLI). VIUPD is a valuable sensor-independent spectral analysis method. Each pixel is described as the linear mixture of standard spectral patterns for water, vegetation, soil and supplementary patterns included when necessary. In the present paper, processing procedure about the data acquisition, radiometric calibration and atmospheric correction have been elaborated. The normalized reflectance (P) of four standard samples resampled to OLI has been listed. For validation of the results, Normalized Difference Vegetation Index (NDVI) and VIUPD have been calculated for comparison. The results showed that VIUPD is more sensitive to the vegetation amount change even in the high vegetation coverage, while the NDVI is more rapidly saturated in high vegetation cover area. In addition, VIUPD is more sensitive to the soil background than NDVI.

Land surface temperature (LST) is an important parameter in the physics of land surface processes, but a large number of pixels are often filled as zero due to cloud, heavy aerosols, and so on in the Moderate-resolution Imaging Spectroradiometer (MODIS) LST product. This letter presents the spectral angle distance (SAD)-weighting reconstruction (SADWER) method of reconstruction of zero-filled pixels of the MODIS LST product. It relies on the hypothesis that pixels with the same land-cover type have nearly the same LST in a localized area. SAD can measure the similarity of land-cover types of different pixels, and pixels with higher land-cover similarity can contribute more to the reconstruction using the weighting method. The result shows that the reconstruction ratio could be as high as 95% using only the SADWER method and nearly 100% after spatial filter postprocessing. The reconstruction accuracy is validated using artificially generated 20-, 50-, and 80-km-diameter concentrically filled areas in both forest and crop land-cover types. The statistical result shows that the standard deviations of the reconstruction errors are less than 2 Kelvin.

Water delineation during flooding is an important research issue in the context of different background regions for protecting both people and agriculture. This paper compares two very important water indices for practical water mapping especially in the context of river flood management. Desert Flood Index (DFI) is a new index which has been introduced as an modification to famous MNDWI for flooding in the river basin having both desert and vegetation areas. In this study disastrous Indus River flooding of 2010 in Pakistan is considered as a case study by using MODIS and Landsat TM data. Images for both Pre-flooding and flooding are analyzed for the most significant part of river flooding. Results proved that DFI appeared to be more efficient than MNDWI in delineating water body from its background features by enhancing contrast.

The traditional look up table (LUT) scheme may have some problems when applied in some cases. To avoid these problems, a general LUT with 1 nm spectral resolution is presented in this work. At first, the atmospheric optical parameters were derived by running MODTRAN4 twice. Then, these datasets were loaded into SQLite for accessing rapidly and then the 6-D linear interpolation was used for extracting desired datasets. Finally, some experiments were carried out to test the performance of the proposed technique. Results show that it is very fast to access data from SQLite database, and the proposed technique has high performance and is convenient in most of the cases.