We compare the performance of three different filters that identify solar artifacts in the Geostationary Lightning Mapper (GLM) data. These filters differ in their complexity and intended use. We use the most appropriate filters to document the severity of GLM solar contamination and to compute the frequency of extreme Americas lightning. Solar artifacts occur at an average rate of one every ∼8  min. Due to their continuous emissions across a portion of the GLM CCD array, these events disproportionately contribute to the GLM event and group rates. Up to 40% of all events recorded between 10:00 and 12:00 local time during 2018 were solar artifacts. The frequency-domain solar filter also reveals a carrier wave at 50 to 60 Hz in some lightning flashes and solar episodes that appears to be anthropogenic in origin. Removing the relatively frequent solar artifact cases reveals rare cases of extreme lightning flashes. Exceptional flashes on the order of 500+ km in length and 10+ s in duration only occur a few times per year and are produced by storms that are predisposed for exceptional lightning rather than as single random occurrences.

Image navigation and registration (INR) processing for the Geostationary Lightning Mapper (GLM) assigns geographic coordinates to lightning events using orbit and attitude telemetry and a geometric calibration that matches coastline features from the GLM background scene against a digital map. Required performance is expressed as an optical angle at the aperture of the instrument of 112  μrad (3σ), equivalent to 4 km at the subsatellite point. This is only one-half the linear dimension of the ground footprint of a detector element at the center of the field of view and is challenging to both attain and validate. Our validation approach uses imagery from the Advanced Baseline Imager (ABI) Band 3 (B03), the ABI channel closest to the GLM spectrally, with 1-km pixel resolution at the subsatellite point, as an INR reference. The finer spatial resolution and the high-accuracy INR (<28  μrad, 3σ) of ABI make it well suited for this application. Since both instruments are on the same platform, no parallax correction is required. We measure positions of feature templates extracted from the GLM background scene relative to the ABI reference image to assign spatial coordinates to the center-points of each template. Almost any feature, except clear-sky ocean scenes, can be used for matching, which allows for spatially dense measurement of INR errors in the forward navigation calculation of a GLM pixel. Cloud motion is compensated during the time between the acquisition of the ABI reference image and the GLM background scene, which further improves validation accuracy. We show that GLM on both GOES-16 and -17 spacecraft satisfy their INR requirements. The spatial and temporal density of validation measurements permits examination of smaller systematic errors, including thermally driven alignment offsets, a discontinuity between focal plane hemispheres arising from a readout artifact, the optical distortion of the GLM lens, and a calibration of the telescope focal length.

Image navigation and registration (INR) for the GOES-R series advanced baseline imager (ABI) is performed in ground processing of the raw image data. A Kalman filter processes ABI star measurements to compute time-varying optical misalignment parameters that allow real-time alignment of image samples to the GOES-R Fixed Grid. For the system to meet INR requirements, the filter measurements noise covariance (R) and process noise covariance (Q) matrices must closely approximate behavior of the true system. Prelaunch estimates of the Q and R parameters did not match actual GOES-16 performance. Several attempts to manually determine improved Q&R parameters were not completely successful. Instead, the Kalman filter was calibrated (tuned) using maximum likelihood (ML) estimation of the parameters: computed utilizing partial derivatives of the Kalman equations, Gauss–Newton iteration, and nonlinear optimization logic. The updated parameters significantly improved performance and are currently operational for GOES-16 and 17. ML estimation of Kalman filter parameters has been used successfully for other systems but has not widely been documented. Greater use is encouraged. We also include a review of ABI INR theory and implementation.

An image navigation (NAV) and registration (INR) performance assessment tool set (IPATS) was developed to assess the US Geostationary Operational Environmental Satellite R-series (GOES-R) Advanced Baseline Imager (ABI) and Geostationary Lightning Mapper (GLM) INR performance. IPATS produces five INR metrics for level 1B ABI images: navigation, channel-to-channel registration, frame-to-frame registration, swath-to-swath registration, and within-frame registration. IPATS also produces one INR metric for GLM: navigation of background images. The high-precision INR metrics produced by IPATS are critical to INR performance evaluation and long-term monitoring. IPATS INR metrics also provide feedback to INR engineers for tuning the navigation algorithms and parameters to further refine INR performance. IPATS utilizes a modular algorithm design to allow the user-selectable data processing sequence and configuration parameters. We first describe the algorithmic design and the implementation of IPATS. Next, it describes the investigation of the optimization of the configuration parameters to reduce measurement errors. Finally, sample INR performance is presented, including GOES-16 and GOES-17 ABI NAV performance from postlaunch test to November 2019 and the comparison of example 24-h INR performance against the mission performance requirements. The INR assessment results show that both GOES-R ABIs are in compliance with the mission INR requirements.

The Geostationary Lightning Mapper (GLM) is a new geostationary lightning detection and location instrument, developed for the current generation of Geostationary Operational Environmental Satellite-R Series (GOES-R, S, T, U) satellites. Detection efficiency (DE) and false alarm rate (FAR) are evaluations of what an instrument detects and what it misses, compared with other well-characterized measurements. These are critical parameters to be assessed for the GLM. We now have enough data to produce preliminary DE and FAR assessments of the prelaunch predicted DE and FAR performance specifications for GLM. However, one of the challenges of doing this assessment is that all available ground truth data sets have much smaller coverage, and in some places, lower DE than GLM. In an attempt to overcome these limitations, we have created a “virtual” lightning network by clustering flashes from several different ground-based networks, using a technique similar to the clustering of GLM pixels. All ground truth sources that are within 330 ms and 16.5 km of any other ground truth source are added to the current cluster. Treating this as a single, “virtual” network, we calculated the relative DE and FAR of GLM. We found that the GLM easily meets the DE spec of better than 70% averaged across the whole field-of-view, over 24 h. It meets the FAR spec in most of CONUS, and we suspect it will improve with updated processing software.

NOAA has both geostationary (GEO) and polar orbiting (LEO) satellites that are procured through different vendors. For long-term environmental studies, it is essential to establish consistent radiometric calibration among satellite instruments irrespective of the satellite platforms. The Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (S-NPP) and NOAA-20 satellites has been providing critical weather and climate-related data. With the launch of NOAA-20 in November 2017, the global coverage of VIIRS has doubled. With more than 7 years in space, S-NPP VIIRS has been rigorously calibrated and validated. We aim to quantify the radiometric consistency between the two VIIRS instruments through double differencing using GOES-16 Advanced Baseline Imager (ABI) as a reference instrument. We also provide insight into the temporal radiometric consistency between VIIRS and ABI. Since no direct simultaneous nadir overpasses (SNOs) exist between SNPP and NOAA-20 for intercomparison, GEO-LEO intercalibration is performed using SNOs between ABI and VIIRS instruments. We show that the NOAA-20 VIIRS measured top-of-atmosphere (TOA) reflectance is consistently lower than that of the S-NPP, mostly on the order of 2% to 3%, consistent with the past studies. GOES-16 ABI-observed reflectance is higher than both VIIRS instruments. SNOs over the all-sky tropical ocean are analyzed to quantify the bias for both NOAA-20 and S-NPP VIIRS relative to ABI. The impact on bias due to spectral differences is accounted for using spectral band adjustment factors estimated using hyperspectral measurements from SCIAMACHY. Uncertainties exist mainly due to calibration uncertainties in VIIRS and ABI, the time difference between the VIIRS and ABI observations, differences in cloud contamination, spectral response function differences, and lack of in-situ hyperspectral data to account for the spectral bias.

The resolution of an imaging system can be quantified using the modulation transfer function (MTF) derived from an image with an edge feature. For Earth-viewing satellites, imaging an appropriate terrestrial edge feature is difficult when optical blur from the atmosphere contaminates the MTF. With a sharp edge and no atmosphere, our Moon has provided researchers with an alternative by deriving the line spread function from the lunar edge. The Moon’s surface, however, has features, e.g., craters and seas, that depart from clean step-like behavior, making curve fitting to the data difficult and diminishing the effectiveness of the method. We demonstrate an improvement in which a reflectance (albedo) map of the Moon, created from a mosaic of the U.S. Department of Defense/NASA Clementine images, is used to flatten the surface features before the edge spread function is measured. The objective is to improve the reliability of MTF measurements for on-orbit calibration of the National Oceanic and Atmospheric Administration/NASA Geostationary Operational Environmental Satellite (GOES) weather satellites and similar missions. The technique is applied to the reflective bands of the Advanced Baseline Imager on GOES-16 and GOES-17 weather monitoring satellites.

The large focal plane arrays used in the new generation of the Geostationary Operational Environmental Satellites (GOES) advanced baseline imager (ABI) introduce new calibration challenges compared with the heritage GOES imagers. The increased number of detectors allows for increased spatial, spectral, and temporal performance, but as a trade-off, it has an increased risk of image striping. We detail the development of a new postlaunch relative calibration capability for ABI reflective solar bands that utilizes ABI special scans to generate a set of relative gains that can be applied to improve image quality and reduce image striping. Results demonstrate that the method reduces image striping in the ABI solar reflective bands over varying scene content and time, both diurnally and over an extended period. This methodology ensures a calibration strategy that is consistent with heritage approaches yet adapts to the new postlaunch validation challenges presented by the new class of operational imagers in the GOES-R series. The developed approach is ready for operational use, as needed, and can be easily implemented into operations to support the operational production of geostationary imagery of the Earth.

This guest editorial introduces the Special Section on Representation Learning and Big Data Analytics for Remote Sensing.

River segmentation is an important aspect of remote sensing image analysis. In the actual remote sensing images of rivers, the background is mostly complex and heterogeneous, the traditional detection methods cannot identify the small tributaries and the edge information is rough. To solve the aforementioned problems, we propose a separable attention network based on different size fusion. In this method, residual neural network is used as the backbone network to obtain the information features of rivers, and the deep feature information is fused with the shallow feature information through attention modules of different scales. The shallow feature and large-scale attention module are used to locate the main position of the river, while the deep feature and small-scale attention module are responsible for the fine segmentation of the river edge, so as to accurately extract the river from the background. The experimental results show that, compared with previous methods, the accuracy and detection speed of proposed method are significantly improved. The mean intersection over union (MIoU) on the river segmentation data set reaches 97.07%. Meanwhile, this model has excellent generalization performance, and the MIoU of the public data set CamVid reaches 69.4%.

Due to the problem that current hyperspectral image tensor classification methods cannot make full use of the multiple spectral–spatial features of hyperspectral images (HSIs), a HSI spectral–spatial classification method based on multikernel support tensor machine is proposed. In the tensor structural space, using the 3-D Gabor wavelet and morphological attribute filters to obtain multiple texture features and multiple attribute structural features, respectively, our study combines multiple spectral–spatial features to classify the HSIs in the multikernel learning framework. By forming the multiple kernel function with the linear weighted combination of the single kernel functions, the proposed method calculates the weights using the iterative method to effectively utilize the multicomplementary features of hyperspectral imagery to perform collaborative classification. The experiments are performed on the widely used Pavia University and Salinas hyperspectral data sets. The classification results indicate that the proposed method can effectively integrate complementary spatial–spectral features, which have higher classification accuracy and better spatial continuity classification mapping when applied to the classification images.

Nearest regularized subspace (NRS) has been recently proposed for hyperspectral image (HSI) classification. The NRS outperforms both collaborative representation classification and sparse representation-based techniques because the NRS makes use of the distance-weighted Tikhonov regularization to ensure appropriate representation from similar samples within-class. However, typical NRS only considers Euclidean distance, which may be suboptimal to resolve the problem of sensitivity in the absolute magnitude of a spectrum. An NRS-Manhattan distance (MD) strategy is proposed for HSI classification. The proposed distance metric controls over magnitude change and emphasizes the shape of the spectrum. Furthermore, the MD metric uses the entire information of the spectral bands in full dimensionality of the HSI pixels, which makes NRS-MD a more efficient pixelwise classifier. Validations are done with several hyperspectral data, i.e., Indian Pines, Botswana, Salinas, and Houston. Results demonstrate that the proposed NRS-MD is superior to other state-of-the-art methods.

There is a growing concern over surface water dynamics due to an increased understanding of water availability and management with current climate trends. Remote sensing has now become an effective means of water extraction due to the availability of an enormous amount of data with diverse spatial, spectral, and temporal resolutions. However, water extraction from optical remote sensing data is associated with several major difficulties, such as the applicability of the extraction method over large areas and complex environments; shadow contamination from clouds, buildings, and mountains; and disclosure of shadowed water and exclusion of floating and submerged plants. To address these difficulties, a learning vector quantization (LVQ) neural network-based method was proposed and implemented to extract water using Landsat 8 imageries. This method is capable of separating water from clouds, build-up areas, shadows, and shadowed water by the ideal input of bands 1 to 7 and normalized difference vegetation index. This model learns water across Sri Lanka. Eight OLI scenes were tested, and the performance was compared with five widely used machine learning algorithms: support vector machine, K-nearest neighbor, discriminant analysis, combination of modified normalized difference water index and modified fuzzy clustering method, and K-means clustering methods. This method performed the best, achieving overall accuracies and the kappa coefficients between 97.8% and 99.7% and between 0.96 and 0.99, respectively. Results have demonstrated robustness, consistency, and preciseness in various dark surfaces, noisiest water environments, and highly water scarce scenes. LVQ revealed a good generalizing ability to detect all types of water with less amount of training samples. This method can be easily adaptable for other sensors and global water to support water resource studies.

Local binary pattern (LBP) is a simple yet efficient image feature description algorithm that has been widely used in applications related to image classification. To more effectively make use of the local features of a hyperspectral image (HSI), a tensor-based spatial–spectral local binary pattern (TSSLBP) that can fully utilize the local spatial-spectral feature of an HSI is proposed. We specifically make efforts in three aspects. First, we construct a third-order tensor model for HSI to combine the local spatial and spectral features. Second, a dimension reduction method based on tensor decomposition is proposed to reduce redundant information of the HSI tensor model. Third, a joint spatial-spectral LBP coding algorithm is implemented based on the HSI tensor model to calculate the local spatial–spectral feature vector of HSI. To verify the effectiveness of the proposed TSSLBP, we apply it to the task of HSI classification and compare it with six existing algorithms. Experimental results on the three popular HSI datasets show that TSSLBP can fully utilize rich information in the local spatial–spectral domain and can significantly improve the classification accuracy of HSI, especially in the case of scarce training samples.

Stacked denoising autoencoder (SDAE) model has a strong feature learning ability and has shown great success in the classification of remote sensing images. However, built-up area (BUA) information is easily interfered with by broken rocks, bare land, and other features with similar spectral features. SDAEs are vulnerable to broken and similar features in the image. We propose a multiscale SDAE model to overcome these problems, which can extract BUA features in different scales and recognize the type of land object from multiple scales. The model effectively improves the recognition rate of BUA. The experimental results show that our algorithm can resist the disturbance information, and the classification accuracies are better than support vector machine, backpropagation, random forests, and SDAE. Then we investigate an application in Wuhan (China) metropolitan area analysis with the classification results of our algorithm. The range of the metropolitan area is 1.5-h isochronous circle calculated by Tencent map big data and is divided into three layers: core metropolitan area, subcore metropolitan area, and daily metropolitan. Finally, from the comprehensive statistical data and traffic data, we know that the Wuhan metropolitan area has a “target-shaped” distribution structure radiating outward from the core metropolitan area. It includes five metropolitan development corridors: Wuhan–Huanggang, Wuhan–Xiaogan–Suizhou, Wuhan–Ezhou–Huangshi, Wuhan–Xiantao–Tianmen, and Wuhan–Xianan–Chibi. The corridor is of great significance to the development of metropolitan areas.

Remote sensing data are often adversely affected by noises due to atmospheric water absorption, transmission errors, sensor sensitivity, and saturation. Therefore, the performance of the classical principal component analysis (PCA) is negatively affected when dealing with such noisy data. To overcome this problem, a robust deflated PCA via multiple instance factorings (RDPCA-MIF) for dimension reduction in remote sensing images is proposed. In contrast to the traditional PCA, a penalty is imposed on the instance space through instance factoring technique to suppress the effect of noise in pursuing projections. To achieve this, we use two strategies, total distance, and cosine similarity metrics. Both strategies discriminate between noise and authentic data points by iteratively learning the relationship between instances and projections. Moreover, a deflation strategy that enables multiple relationships learning between instances and projections is applied to thoroughly evaluate the importance of instances in relation to all projections. Extensive experimental evaluations of the proposed RDPCA-MIF on Salinas-A dataset and shallow and deep features from the NWPU-RESISC45 remote sensing dataset prove its superiority over six comparative methods in dimension reduction and classification tasks.

Cloud/snow recognition technology has application significance in meteorological detection, aviation control, remote sensing disaster prevention, and mitigation. At present, the method of labeling cloud and snow area manually is time- and labor-consuming. Shallow learning methods not only have limited ability to extract cloud/snow semantic features but also have wake learning and expressive ability. These problems lead to the low accuracy of cloud/snow recognition in shallow learning methods. Deep learning methods extract cloud and snow semantic features layer-by-layer, and this feature extraction improves the accuracy of cloud/snow recognition. However, due to the complexity of cloud/snow texture features and the high similarity of cloud/snow spectral features, it is difficult to obtain satisfactory results for the classification of cloud and snow in plateau areas, by the existing deep learning models. In order to solve the above problems, a multiscale fusion attention network is proposed to recognize cloud and snow areas in plateau remote sensing images. In the proposed model, the main network is DenseNet25, which enhances the propagation and reuse of cloud/snow features in the network. A multiscale fusion is proposed to extract more complex cloud/snow texture and spectral features from spatial dimension. The high weight attention mechanism is introduced to obtain dynamic features based on input, and it is able to improve the discriminating ability of cloud and snow features. The experimental results demonstrate that the proposed model can extract and utilize cloud/snow feature information better than existing models and improve the accuracy and generalization of cloud/snow recognition.

Satellite remote sensing has wide applications in many fields. However, the quality of the observed images captured from the satellite sensors exhibits significant variances and most images are low resolution. Therefore, they adversely affect the system performance in a variety of real-world applications such as object recognition and analysis. In order to enhance the resolution of remote sensing images, we propose a super-resolution neural network called progressive residual depth neural network (PRDNN). The progressive residual structure used by PRDNN can gradually discover the feature information of satellite images at different levels and different receptive fields, thus providing more detailed features for reconstructing super-resolution satellite images. The experimental results of the DOTA satellite image database demonstrate that the proposed method is superior to the most advanced super-resolution algorithm in recent years.

A large amount of remote sensing data can be easily acquired due to the increase in the advances in sensor’s technologies. The sensors can generate high-dimensional data in a lower time producing problems related to big data such as management and organization. Since the acquired data is characterized by a large dimension and lack of structure, the information analysis becomes harder. Therefore, an organization stage should structure the data reducing the dimension while maintaining the main properties to enable further analysis. The feature extraction and selection methods can achieve this task. Consequently, we aim to explore various pixel-wise feature extraction and selection algorithms to manage the organization stage of big data for hyperspectral images. Our work covers the comparison between feature vectors computed using the discrete Fourier transform, discrete cosine transform (DCT), and stationary wavelet transform. Moreover, spectral angle mapper, Jeffries–Matusita distance, spectral information divergence, and linear discriminant analysis (LDA) were implemented as feature selectors. Feature extraction and selection methods were combined and evaluated in terms of algorithm complexity, reduction efficiency, and classification accuracy with the aid of a support vector machine and a maximum likelihood classifier. The analysis shows that some linear transformations can perform better in natural landscapes and others in urban images. Furthermore, the study found that the combination of DCT and LDA, which achieves high classification rates with an efficient dimension reduction, can be suitable for the organization stage of a big data remote sensing application of hyperspectral images.

Deep learning techniques have been leveraged in numerous applications and across different data modalities over the past few decades, more recently in the domain of remotely sensed imagery. Given the complexity and depth of convolutional neural network (CNN) architectures, it is difficult to fully evaluate performance, optimize the hyperparameters, and provide robust solutions to a specific machine learning problem that can be applied to nontraditional real-world feature extraction and automation tasks. Ursa Space Systems Inc. develops machine learning approaches to build custom solutions and extract answers from synthetic aperture radar satellite data fused with other remote sensing data sets. One application is identifying the orientation of nontexture linear features in imagery, such as an inlet pipe on top of a cylindrical oil storage tank. We propose a two-phase approach for determining this orientation: first an optimized CNN is used in a nontraditional way to probabilistically determine a coarse location and orientation of the inlet pipe, followed by a maximum likelihood voting scheme to automatically extract the orientation of the angular feature within 7.5 deg. We use a known hyperparameter optimization technique to determine the best deep learning CNN architecture for our specific problem and under user-defined optimization and accuracy constraints, by optimizing model hyperparameters (number of layers, size of the input image, and data set preprocessing) using a manual and grid search approach. The use of this systematic approach for hyperparameter optimization yields increased accuracy for our angular feature extraction and orientation finding algorithm from 86% to 94%. Additionally, this proposed algorithm shows how machine learning can be used to improve real-world remote sensing workflows.

Extraction of information about individual trees using remotely sensed data is essential to supporting ecological and commercial applications in forest environments. Data acquired by consumer-grade cameras onboard unmanned aerial vehicles (UAV) offer an affordable option of high-spatial resolution imagery that can be used to extract forest structural information at a tree level. The aim of this work is to investigate the potential and accuracy of UAV time-series data to automatically detect and delineate tree crowns across an entire woodland. The workflow (presented in a step-by-step manner) involves the construction of a canopy height model (CHM) based on digital elevation models derived from the UAV photogrammetric point clouds. A watershed-based approach is modified to automatically detect and delineate the tree crowns, based on the CHM and the brightness information from the UAV orthomosaics. The accuracy of the proposed method was evaluated by comparing its results against manually delineated tree crowns. The results show an overall accuracy of 63%, where conifer species were more accurately delineated (up to 80%), while broadleaf species returned lower accuracies (<50  %  ). Continued research is necessary to improve the confidence of automated individual tree crown detection and delineation, especially over complex forest structures.

Urbanization in India has been rapid over the past few decades, which results in a substantial replacement of the natural surfaces into built-up lands. We present a performance evaluation of Sentinel-2B, Landsat-8 multispectral, and AVIRIS-NG hyperspectral imagery for extraction of road and roof surfaces using proposed spectral index-based and other conventional algorithms. The new road extraction index (NREI) and new building extraction index (NBEI) are developed for extraction of road and roof surfaces, respectively. Moreover, existing spectral angle mapper (SAM), spectral information divergence (SID), matched filter (MF), and support vector machine (SVM) are utilized as angle, information, filtering, and machine learning-based algorithms, respectively, for detection of both the surfaces. The results of our study suggest that the performance of AVIRIS-NG sensor is the best in comparison to aforesaid multispectral sensors, whereas Sentinel-2B performs better in comparison to Landsat-8 for extraction of road and roof surfaces. The comparison of various algorithms suggests that proposed indices, MF, and SVM produce the best results for extraction of road and roof surfaces, while SAM and SID are superior algorithms for extraction of both the surfaces in AVIRIS-NG imagery. Further, NREI and MF performed well for extraction of roads followed by NBEI and SAM for roofs in Landsat-8. Finally, NREI, SAM, and SID are found to be efficient for extraction of roads subsequently NBEI, SAM, and SID for roofs in Sentinel-2B imagery.

Building footprint segmentation from satellite and aerial images is an essential and challenging step for high-resolution building map generation. In urban management applications, such as building monitoring, infrastructure development, smart three-dimensional cities, and building change detection, building footprints are required to generate precise multiscale building maps. An efficient deep learning-based segmentation approach is proposed for multiscale building footprint extraction, and the results are presented for the most important challenges in photogrammetry and remote sensing, including shadows and occluded areas, vegetation covers, complex roofs, dense building areas, oblique images, and the generalization capability in different locations. The proposed method includes new dilated convolutional blocks containing kernels with different sizes to learn spectral–spatial relationships in multiscale satellite and aerial images with a high level of abstraction. The quantitative assessments of multiscale images from different locations with different spatial resolutions and spectral details show that the average F1 score and the average intersection over union for extracted footprints are about 86% and 76%, respectively. Compared with the state-of-the-art approaches, the proposed method has outstanding generalization capability and provides better performance for building footprint segmentation from multisensor single images.

Atmospheric correction can introduce errors in surface spectral reflectance, and hence induces errors in plant water estimation from remote sensing water indices. We intend to develop water indices that are less impacted by atmospheric effects for plant water content estimation based on the 970-nm water absorption feature. A simulation study using the PROSAIL and 6S models showed that uncertainty in atmospheric water vapor (WV) content can induce large variation in existing 970-nm water indices, such as WI, NWI-1, and NWI-3. An attempt was made to incorporate atmospheric WV absorption at 940 nm to correct for the perturbation due to atmospheric WV variability, leading to the development of improved indices, named as ARWI, NARWI-1, and NARWI-3. The performance of these indices was evaluated using the simulated and field spectral reflectance data, as well as Hyperion and GF5 satellite data. Results showed that the new indices were resistant to uncertainty of WV and could be used to deliver improved estimation of canopy water content, with a smaller root-mean-square-error (ARWI: 7.4  mg/cm², NARWI-1: 8.3  mg/cm², and NARWI-3: 8.8  mg/cm²) compared to that obtained using the traditional water indices (WI: 8.9  mg/cm², NWI-1: 9.4  mg/cm², and NWI-3: 16.6  mg/cm²). The water indices developed in this study, although needing further assessment in wide application scenarios, have great potential for monitoring of vegetation water status using satellite hyperspectral data with reflectance measurement around 970 nm.

The present state of the art technologies for flood mapping are typically tested on small geographical regions due to limitation of resources, which hinders the implementation of real-time flood management activities. We proposed a unified framework (GEE4FLOOD) for rapid flood mapping in Google Earth Engine (GEE) cloud platform. With the unexpected spells of extreme rainfall in August 2018, many parts of Kerala state in India experienced a major disastrous flood. Therefore, we tested the GEE4FLOOD processing chain on August 2018 Kerala flood event. GEE4FLOOD utilizes multitemporal Sentinel-1 synthetic aperture radar images available in GEE catalog and an automatic Otsu’s thresholding algorithm for flood mapping. It also utilizes other remote sensing datasets available in GEE catalog for permanent water body mask creation and result validation. The ground truth data collected during the Kerala flood indicates promising accuracy with 82% overall accuracy and 78.5% accuracy for flood class alone. In addition, the entire process from data fetching to flood map generation at a varying geographical extent (district to state level) took ∼2 to 4 min.

We describe an algorithm based on a convolutional neural network that detects cloud formations and snow cover in satellite images using textures. Herein, multispectral satellite images, received from a multizone scanning instrument used for hydrometeorological support and installed on the Russian satellite Electro-L No. 2, are used as input data. The problem of snow and cloud classification in the absence of a spectral channel in the range of 1.4 to 1.8  μm, which is necessary for their accurate separation, is considered. The developed algorithm can produce cloud and snow cover masks for an area limited by the values of the solar zenith angle in the range of 0 deg to 80 deg for daytime. Algorithm accuracy was evaluated using machine learning metrics and comparing its results with ground truth masks segmented manually by an experienced interpreter. In addition, we compared the resulting masks with a similar cloud mask product from the European Organisation for the Exploitation of Meteorological Satellites based on the data of the Spinning Enhanced Visible and Infrared Imager (SEVIRI) instrument installed on the Meteosat-8 satellite. According to the results of this comparison, we conclude that the cloud masks produced by the proposed convolutional neural network-based algorithm have a lower probability of false detection than products based on the SEVIRI data. The proposed algorithm is fully automatic, and it works in any season of the year during the daytime.

Organic matter (OM), iron (Fe), and zinc (Zn) in black soil are crucial to ensure high-quality production of agriculture, and hyperspectral technology is an effective approach to achieve a rapid estimation of these soil compositions. Eighty black soil samples were collected in Nehe city, Heilongjiang province, China. With indoor spectral data, the correlation between six spectral reflectance, which includes the original and five other transformed reflectance, and the contents of OM, Fe, and Zn on soil were analyzed. Then with the correlation coefficient significance test (bilateral) calculated at α  =  0.01 level to extract sensitive bands, the kernel principal component analysis (KPCA) algorithm was adopted and combined with random forest (RF) and support vector machine (SVM). The combined models were applied for quantitative inversion of soil OM, Fe, and Zn contents and compared them with the models without KPCA dimension reduction. The results show that the determination coefficients and residual prediction deviation for prediction samples of KPCA-RF model (Rp2=0.805 and RPD  =  2.329) that adopted to estimate soil OM content are higher than those of RF model (Rp2=0.681 and RPD  =  1.820), and the root-mean-square errors for prediction samples of KPCA-RF model (RMSEP  =  0.182) are lower than those of RF model (RMSEP  =  0.232). Meanwhile, the accuracy of the KPCA-RF model for estimating soil Fe and Zn contents is also higher with Rp2=0.731, 0.710, RMSEP  =  0.189, 0.003, and RPD  =  1.980, 1.905, respectively. Similarly, the accuracy of the KPCA-SVM model for estimating soil OM, Fe, and Zn contents is higher with Rp2=0.687, 0.609, and 0.585; RMSEP  =  0.230, 0.228, and 0.004; and RPD  =  1.840, 1.642, and 1.592, separately. Therefore, the machine learning models combined with KPCA are more promising in the quantitative inversion of soil composition contents and can be regarded as an effective approach.

In many areas of the globe, the installation of artificial drains on naturally poorly drained soils is a necessary part of farm management. Identifying the location of artificially drained areas is an important step in achieving environmentally sustainable agricultural production. However, in many regions, data on the presence or the distribution of artificial drainage systems are rare. We outline an approach to identify artificially drained soils using Earth observation (EO) satellite imagery and digital elevation data. The method exploits the contrasting phenology of grass during a peak growth stage to identify artificially drained and undrained soils. Two machine-learning techniques, support vector machine and random forest, were tested. Classification accuracy up to 91% was achieved using photointerpreted accuracy points using higher resolution satellite imagery. Additional investigations would be required to establish whether the drained conditions identified were a result of artificial drainage or from naturally well-drained soils occurring within larger soil units. Herein, the Republic of Ireland is used as a test case. Based on our findings, the area of artificially drained grassland within the study area could be revised upward, with 44% (or ∼345  ,  000  ha) of pasture currently classed as “poorly drained” identified as “artificially drained.” At one location, a change in the modeled drainage condition at field level was demonstrated following drain installation. The presented method demonstrates the ability of EO satellites to quickly and accurately map field drainage status at farm management scales over a wide area. This has the potential to improve management decisions at local scales, but also has implications in terms of national policy development and regulation in areas such as water quality and climate change mitigation.

Recently, many convolutional neural networks (CNN) have been successfully employed in bitemporal synthetic aperture radar (SAR) image change detection. However, most of the existing networks are too heavy and occupy a large volume of memory for storage and calculation. Motivated by this, we propose a lightweight neural network to reduce the computational and spatial complexity and facilitate the change detection on an edge device. We replace normal convolutional layers with bottleneck layers that keep the same number of channels between input and output. Next, we employ dilated convolutional kernels with a few non-zero entries that reduce the running time in convolutional operators. Comparing with the conventional convolutional neural network, our lightweight neural network will be more efficient with fewer parameters. We verify our light-weighted neural network on four sets of bitemporal SAR images. The experimental results show that the proposed network can obtain better performance than the conventional CNN and has better model generalization, especially on the challenging datasets with complex scenes.

In conventional inverse synthetic aperture radar (ISAR) imaging of complex moving aircrafts, a micro-Doppler effect is generated because of the high-speed rotating parts such as propellers. The rotational parts introduce a sinusoidal frequency-modulated (SFM) signal in the radar echoes, which can significantly affect the quality of the ISAR imaging. To solve this problem, the Doppler frequency forms for the rigid body and rotating parts of an aircraft are analyzed, and a method based on the discrete sinusoidal frequency-modulated transform (DSFMT) is proposed. The SFM signal can be transformed into the DSFMT domain, and the parameters can be estimated by the global maximum. The rigid body is separated after removing the SFM signals using successive elimination technology. This method is not only applicable to targets in uniform motion but also to those in accelerating motion. The results of simulations and a measured data test are given to verify the effectiveness of the proposed method.

Using satellite data to study small water bodies (SWB) and medium-sized water bodies (MSWB) is extremely useful for understanding their status, how to conserve them as water reservoirs, and their vulnerability to climate variability. The images studied in our work correspond to different-sized lagoons located in areas with high and low topography in a tropical region of Chiapas, Mexico. Our research project delineates SWB and MSWB. For this analysis, we considered water bodies to be uniform regions in a synthetic aperture radar image. The robustness of the method was determined based on an analysis of the morphologies of 23 lagoons. Several methods, including Hermite transform, were analyzed and compared with other image denoising methods used to improve speckle reduction. To obtain additional spatial information for image classification, we analyzed texture using the gray-level co-occurrence matrix. The results indicate that the Hermite filter is the best method for identifying water bodies. The advantage of this filter is the identification of local patterns such as edges and lines. It also preserves and improves aspects related to the homogeneity of water bodies, using the Hermite coefficient selection criteria for local pattern feature selection/extraction. The lake water extent products demonstrate that Sentinel-1 is useful for identifying SWB in this study area. The results show very high detection of water bodies, with adequate detection for water bodies larger than 2 ha, and an area accuracy of 80%.

Following the objectives of the mission, Sentinel-2 (S2) makes a significant contribution to land monitoring, climate change, emergency management and security, and related problems. To increase the spatial resolution, researchers have been challenged to obtain a synthesized panchromatic band with a fine resolution for all S2 bands. The capacity to collect satellite imagery with a short revisit time at different spatial resolutions increases the ability for more accurate data fusion applications. Our work presents an investigation of different synthesized panchromatic bands for producing high-resolution S2 data. We produced a synthesized environment from PlanetScope (PS) data to evaluate the performance of different panchromatic bands for the enhancement of S2 10- and 20-m bands. For this, imagery over three different study areas with different characteristics was chosen: Pakistan, North Macedonia, and Turkey. After the three different synthesized panchromatic bands were produced, we fused the panchromatic bands with the 10- and 20-m S2 bands with three different state-of-the-art pansharpening techniques. Experimental comparison between the three newly produced panchromatic bands indicates that all synthesized bands can enhance the spatial resolution of the original multispectral bands. Since the difference in spatial resolution may be critical for more detailed image classification, for future studies, we recommend investigation of the fused data for different land cover applications.

Spectral variability is one of the most limiting factors in hyperspectral unmixing, so it is important to further study the characteristics of spectral variability to improve the accuracy of unmixing. After conducting simulations under varying irradiation conditions, a linear mixed model combining endmember and band is proposed by introducing a band scaling factor to the endmember scaled spectrum. The total variation constraint is used to smooth the spatial distribution of both endmember and band scaling factors and then alternating iterative optimization is applied to solve the optimization problem. Experiments conducted with both simulated and real hyperspectral data sets indicate that the proposed algorithm is effective in hyperspectral unmixing and is superior to other state-of-the-art algorithms based on spectral variability.

The heavy-tailed properties of modulation transfer function (MTF) typically introduce noise and an aliasing effect in the MTF compensation (MTFC) effort to improve the spatial quality of optical satellite images. These degradative effects compromise the image’s signal-to-noise ratio (SNR). Consequently, users must evaluate the relative importance of image sharpness versus SNR for their applications to decide whether MTFC processing is appropriate. We propose a high-fidelity MTFC method that executes an optimal trade-off between noise regularization and detail preservation in the image restoration process to address this problem. To this end, we exploit the merit of image prior characteristics in both local smoothness and nonlocal self-similarity properties of an image in a hybrid domain (viz., space spatial and frequency) to design effective regularization terms that reflect these image properties. Furthermore, we establish a simple joint statistical model in the curvelet domain to combine these two properties. To make this regularization-based MTFC method tractable and robust, we employ the multiobjective bilevel optimization approach to solve the severely ill-posed inverse problem of MTFC. We conduct extensive experiments to evaluate the proposed regularization-based MTFC method using synthetically blurred images simulated from the level 2A product of IKONOS and real satellite images with unknown blur. Quantitative measurements of image quality reveal that the proposed method produces competitive restoration results with minimum computational complexity and exhibits a good convergence property. These experimental results show that the proposed method can find a compromise between regularizing noise and preserving image fidelity.

Dimensionality reduction is an important step in processing hyperspectral images (HSIs) to overcome the dimensionality problem. Linear dimensionality reduction methods such as independent component analysis (ICA) and linear discriminant analysis (LDA) are commonly employed to reduce the dimensionality of HSI. These methods fail to capture nonlinear dependency in the HSI data, as data lies in the nonlinear manifold. To handle this, nonlinear transformation techniques based on kernel methods were introduced for dimensionality reduction of HSI. However, the kernel methods involve cubic computational complexity while computing the kernel matrix, thus its potential cannot be explored when the number of pixels (samples) is large. In literature, a fewer number of pixels are randomly selected to partially overcome this issue; however, this suboptimal strategy might neglect important information in the HSI. We propose randomized solutions to the ICA and LDA dimensionality reduction methods using random Fourier features, and we call them randomized ICA and randomized LDA. Our proposed method overcomes the scalability issue and handles the nonlinearities present in the data more efficiently. Experiments conducted with three real-world hyperspectral datasets demonstrate that our proposed randomized methods outperform the conventional kernel ICA and kernel LDA overall and in terms of per-class accuracies, and computational time.

The projection of objects on the earth’s surface caused by the sunlight produces shadows. They are inevitable in high-spatial-resolution satellite remote sensing images and reduce the accuracy of change detection, land cover classification, target recognition, and many more applications. Dark-colored land covers in these satellite images, such as water bodies, road, and soil, appear with similar spectral properties as those of shadows and often result in difficulty in shadow detection, especially in complex urban settings. We propose an object-oriented building shadow extraction method and tested it using six selected study areas from TripleSat-2 satellite imagery with 3.2-m spatial resolution. The method’s main steps include (1) selecting six image features that can highlight the shadow information and then segment the image based on edge; (2) extracting shadow region based on multiple object features; and (3) masking nonbuilding shadow regions by the shadow and dark object separation index, image features including spectral, textural, and geometric features, and contextual information. The average precision, recall, and F1-score of the shadow detection were 85.6%, 88.6%, and 87.0%, respectively, and the ranges were 73.0% to 91.0%, 76.6% to 94.1%, and 74.7% to 91.2%, respectively. Compared with multiscale segmentation, edge-based segmentation is more efficient and helpful to completely and accurately extract shadows.

Graph embedding (GE) frameworks are used for extracting the discriminative features of hyperspectral images (HSIs). However, it is difficult to select a proper neighborhood size for graph construction. To overcome this difficulty, a semisupervised feature extraction (FE) method, called semisupervised collaborative representation graph embedding (SCRGE), is proposed. The proposed algorithm utilizes collaborative representation (CR) to obtain the collaborative coefficients of labeled and unlabeled samples. Then, a semisupervised graph is constructed using the collaborative coefficients of the labeled samples within the same class and the collaborative coefficients of the unlabeled samples, and an interclass graph is constructed using the collaborative coefficients of the labeled samples in different classes. Finally, a projection matrix for FE is obtained by embedding these graphs into a low-dimensional space. SCRGE not only inherits the merits of CR to reveal the collaborative reconstructive properties of data but also enhances intraclass compactness and interclass separability to improve the discriminating power for classification. Experimental results on three real HSIs datasets demonstrate that SCRGE outperforms other state-of-the-art FE methods in terms of classification accuracy.

In order to achieve accurate locations of multiple targets on an airborne optoelectronic platform, we propose a multitarget location system based on the digital elevation model and a multitarget location model based on the target vector. Through the target detection algorithm, the pixel coordinates of each target in the field of view can be obtained and the visual axis vector of each target is constructed. By integrating the measurement data of each sensor in an unmanned aerial vehicle and the photoelectric platform, the geographical location information of each target can be calculated. In the condition of a flight height of 3000 m, the location error for the main target is 16.47 m, and the location error for the secondary target is about 27.36 m. The effect of the improved filtering model is analyzed. After filtering, the main target location error is reduced to 7 m, and the secondary target location error is reduced to 11 m. The proposed method has the advantages of high real-time performance, high location accuracy, and convenient engineering application.

Images from high-resolution inverse synthetic aperture radar (ISAR) can provide more information about the targets. Multiband fusion imaging techniques can achieve higher range resolution without increasing hardware costs. A multiband fusion imaging algorithm based on variational Bayesian inference (VBI) is proposed to improve the range resolution of ISAR images. First, a multiband fusion ISAR imaging model is established based on sparse representation. Second, the scattering coefficients and noise are assumed to be the Laplacian scale mixture distribution and the complex Gaussian distribution, respectively. Finally, the fusion image is directly reconstructed in the complex domain by the VBI based on Laplace approximation method. The effectiveness and robustness of the proposed algorithm are verified by the experimental fusion results of one-dimensional signals and two-dimensional ISAR images.