Satellite aerosol optical depth (AOD) is a quantitative parameter frequently used to estimate ground-level fine particulate matters (PM2.5)at regional to global scales. In this article, Himawari-8 apparent reflectance (top-of-atmosphere reflectance) data were used to estimate the hourly ground-level PM2.5 concentrations (Ref-PM2.5) using deep neural networks (DNNs), and comparison was conducted with the AOD-based PM2.5 estimation method (AOD-PM2.5). In high-density site areas, the Ref-PM2.5 method was closer to the actual situation and more capable of PM2.5 estimation compared with the AOD-PM2.5 method. The PM2.5 samples used in the AOD-PM2.5 method were less than one-half of the Ref-PM2.5 method due to unavailability of AOD observations, which might be due to strict surface assumptions, cloud detection, and error in the aerosol scheme used in the AOD inversion method. This led to many missing values of AOD-derived PM2.5 in the spatial distribution map of a single day. Moreover, similar hourly variations in PM2.5 were observed for both the methods, and the highest concentration of PM2.5 appeared at the junction of Jiangsu, Anhui, and Shandong at 08:00, 09:00, and 10:00 in local time, which gradually decreased at 11:00 and reached to a minimum value at 16:00 and 18:00.
The application of empirical mode decomposition (EMD) in the analysis and processing of lightning electric field waveforms acquired by the low-frequency e-field detection array (LFEDA) in China has significantly improved the capabilities of the low-frequency/very-low-frequency (LF/VLF) time-of-arrival technique for studying the lightning discharge processes. However, the inherent mode mixing and the endpoint effect of EMD lead to certain problems, such as an inadequate noise reduction capability, the incorrect matching of multistation waveforms, and the inaccurate extraction of pulse information, which limit the further development of the LFEDA’s positioning ability. To solve these problems, the advanced ensemble EMD (EEMD) technique is introduced into the analysis of LF/VLF lightning measurements, and a double-sided bidirectional mirror (DBM) extension method is proposed to overcome the endpoint effect of EMD. EEMD can effectively suppress mode mixing, and the DBM extension method proposed in this article can effectively suppress the endpoint effect, thus greatly improving the accuracy of a simulated signal after a 25–500-kHz bandpass filter. The resulting DBM_EEMD algorithm can be used in the LFEDA system to process and analyze the detected electric field signals to improve the system’s lightning location capabilities, especially in terms of accurate extraction and location of weak signals from lightning discharges. In this article, a 3-D image of artificially triggered lightning obtained from an LF/VLF location system is reported for the first time, and methods for further improving the location capabilities of the LF/VLF lightning detection systems are discussed.
Changing a logic switch threshold in the linear fit sulfur dioxide (LFSO2) algorithm improves the performance based on the evaluation of the NOAA operational atmospheric SO2 near-real-time (NRT) retrieval. The LFSO2 is used to create estimates from measurements made by the Suomi NPP (S-NPP) Ozone Mapping and Profiler Suite (OMPS). We evaluate the LFSO2 and compare the results to those from a principal component analysis (PCA) offline algorithm. Twenty independent volcanic scenarios and one environmental disaster scenario spread over eight years are selected for this comparison. More than three months of Kilauea volcanic activity in 2018 are monitored and are included in this evaluation and comparison. We found that the operational LFSO2 retrievals at lower troposphere (TRL), mid-troposphere (TRM), and lower stratosphere (STL) exhibited a discontinuity and have a saturation-like relationship if compared with PCA results. Using the new retrieval logic, the discontinuity in LFSO2 retrievals and the saturation appearance in comparisons vanished and a close to a linear relationship with the matchup data from the PCA retrieval products is demonstrated. The minimum detectable values for all three SO2 layer products and the planetary boundary layer (PBL) products are estimated with the updated LFSO2 algorithm. Results for a volcanic cloud over Colombia for the updated LFSO2 for OMPS and a Differential Optical Absorption Spectroscopy (DOAS) algorithm for the Tropospheric Monitoring Instrument (TROPOMI) measurements are also examined. Similar SO2 total mass estimates over the region are obtained from the two products.
Since its launch in October 2011, the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (SNPP) satellite has provided high quality global ocean color products, which include normalized water-leaving radiance spectra $nL_{w}$ ( $lambda $ ) of six moderate (M) bands (M1–M6) at the wavelengths of 410, 443, 486, 551, 671, and 745 nm with a spatial resolution of 750-m, and one imagery (I) band at a wavelength of 638 nm with a spatial resolution of 375-m. Because the high-resolution I-band measurements are highly correlated spectrally to those of M-band data, it can be used as a guidance to super-resolve the M-band $nL_{w}$ ( $lambda $ ) imagery from 750- to 375-m spatial resolution. Super-resolving images from coarse spatial resolution to finer ones have been a field of very active research in recent years. However, no previous studies have been applied to satellite ocean color remote sensing, in particular, for VIIRS ocean color applications. In this study, we employ the deep convolutional neural network (CNN) technique to glean the high-frequency content from the VIIRS I1 band and transfer to super-resolved M-band ocean color images. The network is trained to super-resolve each of the VIIRS six M-bands $nL_{w}$ ( $lambda $ ) separately. In our results, the super-resolved (375-m) $nL_{w}$ ( $lambda $ ) images are much sharper and sh-
w finer spatial structures than the original images. Quantitative evaluations show that biases between the super-resolved and original $nL_{w}$ ( $lambda $ ) images are small for all bands. However, errors in the super-resolved $nL_{w}$ ( $lambda $ ) images are wavelength-dependent. The smallest error is found in the super-resolved $nL_{w}$ (551) and $nL_{w}$ (671) images, and error increases as the wavelength decreases from 486 to 410 nm. The results show that the networks have the capability to capture the correlations of the M-band and the I1 band images to super-resolved M-band images.
As interferometric synthetic aperture radar (InSAR) data improve in their global coverage and temporal sampling, studies of ground deformation using InSAR are becoming feasible even in heavily vegetated regions such as the American Pacific Northwest (PNW) and Sumatra. However, ongoing forest disturbance due to logging, wildfires, or disease can introduce time-variable signals which could be misinterpreted as ground displacements. This study constrains the error introduced into InSAR time series in the presence of time-variable forest disturbance using synthetic data. For satellite platforms with randomly distributed orbital positions in time (e.g., Sentinel-1), mid-time series forest disturbance results in random error on the order of 0.2 and 10 cm/year for 1-year secular and time-variable velocities, respectively. If the orbital positions are not randomly distributed in time (e.g., ALOS-1), a biased error on the order of 10 cm/year is introduced to the inferred secular velocity. A time series using real ALOS-1 data near Eugene, OR, USA, shows agreement with the bias estimated by synthetic models. Mitigation of time-variable land cover change effects can be achieved if their timing is known, either through independent observations of surface properties (e.g., Landsat/Sentinel-2) or through the use of more computationally expensive, nonlinear inversions with additional terms for the timing of height changes. Inclusion of these additional terms reduces the potential for misinterpretation of InSAR signals associated with land surface change as ground deformation.
Imbalanced training sets are known to produce suboptimal maps for supervised classification. Therefore, one challenge in mapping land cover is acquiring training data that will allow classification with high overall accuracy (OA) in which each class is also mapped onto similar user’s accuracy. To solve this problem, we integrated local adaptive region and box-and-whisker plot (BP) techniques into an iterative algorithm to expand the size of the training sample for selected classes in this article. The major steps of the proposed algorithm are as follows. First, a very small initial training sample (ITS) for each class set is labeled manually. Second, potential new training samples are found within an adaptive region by conducting local spectral variation analysis. Lastly, three new training samples are acquired to capture information regarding intraclass variation; these samples lie in the lower, median, and upper quartiles of BP. After adding these new training samples to the ITS, classification is retrained and the process is continued iteratively until termination. The proposed approach was applied to three very high-resolution (VHR) remote-sensing images and compared with a set of cognate methods. The comparison demonstrated that the proposed approach produced the best result in terms of OA and exhibited superiority in balancing user’s accuracy. For example, the proposed approach was typically 2%-10% more accurate than the compared methods in terms of OA and it generally yielded the most balanced classification.
In this article, we present a multipolarimetric estimation approach for two model-based vegetation structure parameters (shape ${A}_{P}$ and orientation distribution ${psi }$ of the main canopy elements). The approach is based on a reduced observation set of three incoherent (no phase information) polarimetric backscatter intensities ( $| {S}_{mathrm{ HH}} |^{2}$ , $| {S}_{mathrm{ HV}} |^{2}$ , and $| {S}_{mathrm{ VV}} |^{2}$ ) combined with a two-parameter ( ${A}_{P}$ and ${psi }$ ) discrete scatterer model of vegetation. The objective is to understand whether this confined set of observations contains enough information to estimate the two vegetation structure parameters from the L-band radar signals. In order to disentangle soil and vegetation scattering influences on these signals and ultimately perform a vegetation-only retrieval of vegetation shape ${A}_{P}$ and orientation distribution ${psi }$ , we use the subpixel spatial heterogeneity expressed by the covariation of co- and cross-polarized backscatter ${Gamma }_{{{mathrm{PP-PQ}}}}$ of the neighboring cells and assume it is indicative for the amount of a vegetation-only co-to-cross-polarized backscatter ratio ${mu }_{{{mathrm{PP-PQ}}}}$ . The ratio-based retrieval approach enables a relative (no absolute backscatter) estimation of the vegetation structure parameters which is more robust compared to retrievals with absolute terms. The application of the developed algorithm on global L-band Soil Moisture Active Passive (SMAP) radar data acquired from April to July 2015 indicates the potential and limitations of estimating these two parameters when no fully polarimetric data are available. A focus study on six different regions of interest, spanning land cover from barren land to tropical rainforest, shows a steady increase in orientation distribution toward randomly oriented volumes and a continuous decrease in shape arriving at dipoles for tropical vegetation. A comparison with independent data sets of vegetation height and above-ground biomass confirms this consistent and meaningful retrieval of ${A}_{P}$ and ${psi }$ . The retrieved shapes and orientation distributions represent the main vegetation elements matching the literature results from model-based decompositions of fully polarimetric L-band data at the SMAP spatial resolution. Based on our findings, ${A}_{P}$ and ${psi }$ can be directly applied for parameterizing the vegetation scattering component of model-based polarimetric decompositions. This should facilitate decomposition into ground and vegetation scattering components and improve the retrieval of soil parameters (moisture and roughness) under vegetation.
A new inverse synthetic aperture radar (ISAR) imaging framework is proposed to obtain high cross-range resolution under sparse aperture conditions, which is a challenge when the signal-to-noise ratio is low. Motivated by the sparsity and low rank of target’s 2-D distribution, the imaging problem is converted to the simultaneously sparse and low-rank signal matrix reconstruction problem under multiple measurement vector (MMV) model, and a novel reconstruction method based on joint constraints of sparsity and low rank is proposed. Due to the over-relax problem, the traditional convex optimization method cannot achieve a better performance using joint structures than exploiting just one of the constraints. As such, a nonconvex penalty function is introduced. To avoid the local minima, the convexity of the cost function should be ensured when constructing the nonconvex penalty function. The adaptive filtering framework, which is a powerful way to recovery the sparse low-rank matrix accurately from its noisy observation, is adopted as a reconstruction algorithm. Furthermore, the optimal step size formula and the idea of smoothed zero norm are used to enhance the convergence and the ability to suppress noise. The newly proposed method is verified by the simulation experiment, which has a better performance in image quality, robustness to noise, and imaging speed.
Buried landmines and unexploded remnants of war are a constant threat for the population of many countries that have been hit by wars in the past years. The huge amount of casualties has been a strong motivation for the research community toward the development of safe and robust techniques designed for landmine clearance. Nonetheless, being able to detect and localize buried landmines with high precision in an automatic fashion is still considered a challenging task due to the many different boundary conditions that characterize this problem (e.g., several kinds of objects to detect, different soils and meteorological conditions, etc.). In this article, we propose a novel technique for buried object detection tailored to unexploded landmine discovery. The proposed solution exploits a specific kind of convolutional neural network (CNN) known as autoencoder to analyze volumetric data acquired with ground penetrating radar (GPR) using different polarizations. This method works in an anomaly detection framework, indeed we only train the autoencoder on GPR data acquired on landmine-free areas. The system then recognizes landmines as objects that are dissimilar to the soil used during the training step. Experiments conducted on real data show that the proposed technique requires little training and no ad hoc data preprocessing to achieve accuracy higher than 93% on challenging data sets.
In this study, as part of an effort to study snowfall characteristics and quantify winter precipitation in East China, we investigated the microphysical properties of snowfall, including size, shape, density, and terminal velocity using a 2-D video disdrometer (2-DVD) and a weighing precipitation gauge in Nanjing (NJ), East China during the winters of 2015–2019. We obtained larger snow density and terminal velocity values than those reported in the literature for this region. Higher snow density could account for higher snowflake terminal velocity, after removing the effects of observation altitude and surface temperature. We then fit the snow particle size distributions (PSDs) to the gamma model and explored the interrelationships among the model parameters and snowfall rate (SR). The relationship between radar reflectivity factor ( $Z_{e}$ ) and SR was derived based on snow PSD measurements and the snow density relation. Using this $Z_{e}-mathrm {SR}$ relationship, the estimated liquid-equivalent SRs are obtained from S-band NJ radar data collected during several snowfall events. Radar-inferred SRs showed reasonable agreement with those measured on the ground, with a mean absolute error of 16% for the collected snowfall events in NJ.
Since tree roots are important to ecosystems, particularly in the context of global climate change, better understanding of their organization is necessary. Ground-penetrating radar (GPR) appears a useful tool to that effect. In this contribution, a novel processing procedure to reconstruct 3-D root architectures from GPR data in heterogeneous environments is proposed, involving three main steps: 1) noise-related information is removed using singular value decomposition (SVD); 2) a modified version of randomized Hough transform (RHT) yields the soil dielectric constant; and 3) a matched-filter technique combined with Hilbert transform then operates as wave migration. Viability is first studied from comprehensive numerical simulations carried out with the gprMax software on a realistic root model in a 3-D heterogeneous environment. The heterogeneous soil effect is studied carefully through a number of simulations involving six different soil types. Then, controlled laboratory measurements are conducted on a root prototype using a bistatic GPR system involving folded complementary bowtie antennas in the frequency range of 300 MHz to 3.3 GHz. The 3-D results from both simulations and experiments show the good performance and potential of the proposed processing.
In this article, a method is presented for automatic scaling of the F-layer from ionograms based on an image processing technique for the extraction of curvilinear structures. The algorithm obtains the ordinary and extraordinary traces and determines the F2 critical frequency. The performance was tested using a wide data set of ionograms recorded by the Advanced Ionospheric Sounder/Istituto Nazionale di Geofisica e Vulcanologia (AIS/INGV) ionosonde located at Universidad Nacional de Tucumán, Tucuman, Argentina, and the results are compared with manual scaling and also with Autoscala method. Results from these tests show that the method is feasible and can be the seed for the development of a robust automated scaling system.
The spatially and temporally correlated sea clutter with phase information is valuable for marine radar applications. The major difficulty of coherent sea clutter modeling is the generation of the continuous phases. This article presents a new phase retrieval approach for modeling the correlated complex sea clutter based on unsupervised neural networks. The unsupervised short-term and long-term neural networks have been developed for the phase retrieval on different term scales. Both these networks have the same input layer and feature extraction module, and however, the number of output neurons is different. The amplitude sea clutter series and the desired Doppler spectrum are fed into the network in parallel, and their features are extracted by two parallel bidirectional long short-term memory (Bi-LSTM) networks which sufficiently utilize the correlations of sea clutter data. These features are concatenated and fused by a residual network (ResNet). The phases can be successfully obtained by constraining to the desired Doppler spectrum and the given amplitudes of sea clutter series. This proposed approach has been verified by the measured Ice Multiparameter Imaging X-Band (IPIX) radar data, and it can precisely model the complex sea clutter with specified statistic characteristics and Doppler properties. The amplitude root mean square error (RMSE) between the obtained and measured Doppler spectra is only 1.5065 with the interval between adjacent frames equals to 32. The RMSE of Doppler central frequency and spectrum width is 6.9306 and 1.2293 Hz, respectively. It shows robustness with the change of range resolution and interval.
It has been observed that the lab-based calibration coefficients of a satellite instrument differ in the actual in-orbit operation due to different environmental conditions. The lab-based coefficients are measured when all the components of the satellite instrument are in thermal equilibrium, while during in-orbit operations, there may be significant variations in temperature between various components as well as the diurnal temperature gradients of subsystems, especially in the geostationary platform due to midnight sunray intrusion. This article proposes a new technique to measure the in-orbit calibration coefficients using the collocated matchup data at different electrooptical module temperatures of the INSAT-3D/3DR imager counts and hyperspectral sounder radiances such as Advanced InfraRed Sounder (AIRS) and Infrared Atmospheric Sounding Interferometer (IASI) that are considered as the reference instruments. The radiances generated by new calibration coefficients show less bias and standard deviations for all infrared (IR) channels of INSAT-3D and INSAT-3DR when compared with the convolved radiances generated from the AIRS and IASI radiances using the procedure established by the Global Space-based Intercalibration System (GSICS).
Synthetic aperture radar (SAR) sea ice imagery is a promising source of data for sea ice data assimilation. Classification of SAR sea ice imagery into ice and water is of particular relevance due to its relationship with ice concentration, a key variable in sea ice data assimilation systems. With increasing volumes of SAR data, automated methods to carry out these classifications are of particular importance. Although several automated approaches have been proposed, none look at the impact of including an estimate of uncertainty of the model parameters and input features on the classification output. This article uses an established database of SAR image features to train a multilayer perceptron (MLP) neural network to classify pixel locations as either ice, water, or unknown. The classification accuracies are benchmarked using a recently developed logistic regression approach for the same database. The two methods are found to be comparable. The MLP approach is then enhanced to allow uncertainty to be estimated at each pixel location. Following methods proposed in the deep learning community, two kinds of uncertainty are considered. The first, epistemic uncertainty, is that due to uncertainty in the MLP weights. The second kind of uncertainty, aleatoric uncertainty, is that which cannot be explained by the model, and is therefore associated with the input data. It is found that including these uncertainties in the MLP models reduces their accuracies slightly, but also reduces misclassification rates. This is of particular importance for data assimilation applications, where misclassifications could severely degrade the analysis.
The traditional nonlocal estimations have been demonstrated to be effective and widely used in polarimetric synthetic aperture radar interferometry (PolInSAR) data. However, there still exist some problems about two key steps: 1) in the homogeneous pixels selection step, the regular square (RS) patches matching strategy shows the limited performance in textured area and 2) in the central pixel value estimation from the selected pixels, the well-known Lee estimator, which only uses the intensity statistic, tends to be unstable. To overcome these restrictions, we put forward two robust strategies and then propose an improved PolInSAR complex coherence nonlocal estimation: 1) the shape-adaptive (SA) patch is utilized for flexibly matching the similar pixels in a large search window, which is constructed by combining the likelihood ratio test (LRT) and the region growing (RG) algorithm and 2) the trace-moment-based nonlocal reduced bias (TMB-NLRB) estimator is employed, which considers the interchannel correlations and evaluates more accurately the homogeneity level between the selected pixels. The denoising effect of both strategies is quantitatively analyzed on the simulated data set, and the proposed algorithm is compared with classical estimation algorithms on a TerraSAR-X/TanDEM-X PolInSAR data set. These experimental results show that the proposed method provides better performance in speckle reduction, detail preservation, and complex coherence estimation.
Synthetic aperture radar (SAR) images are inherently affected by speckle noise, for which deep learning-based methods have shown good potential. However, the deep learning-based methods proposed until now directly map low-quality images to high-quality images, and they are unable to characterize the priors for all the kinds of speckle images. The variational method is a classic model optimization approach that establishes the relationship between the clean and noisy images from the perspective of a probability distribution. Therefore, in this article, we propose the recursive deep convolutional neural network (CNN) prior model for SAR image despeckling (SAR-RDCP). First, the data-fitting term and regularization term of the SAR variational model are decoupled into two subproblems, i.e., a data-fitting block and a deep CNN prior block. The gradient descent algorithm is then used to solve the data-fitting block, and a predenoising residual channel attention network based on dilated convolution is used for the deep CNN prior block, which combines an end-to-end iterative optimization training. In the experiments undertaken in this study, the proposed model was compared with several state-of-the-art despeckling methods, obtaining better results in both the quantitative and qualitative evaluations.
This article proposes a novel synthetic aperture radar visual-inertial odometry (SAR-VIO) consisting of an SAR and an inertial measurement unit (IMU), which aims to enable the observation platform to complete successfully a continuous observation mission in the context of low-cost demand and lack of enough navigation information. First, we establish the observation models of the SAR in a continuous observation process based on the SAR frequency-domain imaging algorithm and the SAR time-domain imaging algorithm, respectively. With the preintegrated IMU data, we then propose a method for estimating the geographic locations of the matched targets in the SAR images and verify the condition and correctness of the method. The optimization of the track and the locations of the targets is achieved by bundle adjustment according to the minimum reprojection error criterion, and a sparse point-cloud map can be obtained. Finally, these methods and models are organized into a complete SAR-VIO framework, and the feasibility of the framework is verified through experiments.
Holographic synthetic aperture radar (HoloSAR) tomography is an attractive imaging mode that can retrieve the 3-D scattering information of the observed scene over 360° azimuth angle variation. To improve the resolution and reduce the sidelobes in elevation, the HoloSAR imaging mode requires many passes in elevation, thus decreasing its feasibility. In this article, an imaging method based on iterative adaptive approach (IAA) and generalized likelihood ratio test (GLRT) is proposed for the HoloSAR with limited elevation passes to achieve super-resolution reconstruction in elevation. For the elevation reconstruction in each range-azimuth cell, the proposed method first adopts the nonparametric IAA to retrieve the elevation profile with improved resolution and suppressed sidelobes. Then, to obtain sparse elevation estimates, the GLRT is used as a model order selection tool to automatically recognize the most likely number of scatterers and obtain the reflectivities of the detected scatterers inside one range-azimuth cell. The proposed method is a super-resolving method. It does not require averaging in range and azimuth, thus it can maintain the range-azimuth resolution. In addition, the proposed method is a user parameter-free method, so it does not need the fine-tuning of any hyperparameters. The super-resolution power and the estimation accuracy of the proposed method are evaluated using the simulated data, and the validity and feasibility of the proposed method are verified by the HoloSAR real data processing results.
The water body detection in high-resolution synthetic aperture radar (SAR) images is a challenging task due to the changing interference caused by multiple imaging conditions and complex land backgrounds. Inspired by the excellent adaptability of deep neural networks (DNNs) and the structured modeling capabilities of probabilistic graphical models, the cascaded fully-convolutional network (CFCN) is proposed to improve the performance of water body detection in high-resolution SAR images. First, for the resolution loss caused by convolutions with large stride in traditional convolutional neural network (CNN), the fully-convolutional upsampling pyramid networks (UPNs) are proposed to suppress this loss and realize pixel-wise water body detection. Then considering blurred water boundary, the fully-convolutional conditional random fields (FC-CRFs) are introduced to UPNs, which reduce computational complexity and lead to the automatic learning of Gaussian kernels in CRFs and the higher boundary accuracy. Furthermore, to eliminate the inefficient training caused by imbalanced categorical distribution in the training data set, a novel variable focal loss (VFL) function is proposed, which replaces the constant weighting factor of focal loss with the frequency-dependent factor. The proposed methods can not only improve the pixel accuracy and boundary accuracy but also perform well in detection robustness and speed. Results of GaoFen-3 SAR images are presented to validate the proposed approaches.
