Towards Unsupervised Flood Mapping Generation Using Automatic Thresholding and Classification Aproaches

The purpose of this study is to analyze the impact of upper ocean dynamics on velocity bunching, represented by azimuthal cutoff wavelength (i.e., sea surface wind, wave, and current). In this study, over 1400 dual-polarized (vertical-vertical (VV) and vertical-horizontal (VH)) Sentinel-1 (S-1) synthetic aperture radar (SAR) images collected in tropical cyclones (TC) are utilized. These images are combined with wind and rain observations from the stepped-frequency microwave radiometer (SFMR), wave simulations conducted using a third-generation numerical wave model, WAVEWATCH-III (WW3), and SAR-derived wind information collected from CyclObs winds. The WW3-simulated significant wave height (SWH) is validated against measurements from HY-2B altimeter taken in August and September 2021, yielding a root mean square error (RMSE) of 0.48 m and a correlation coefficient (COR) of 0.88. The SAR-based azimuthal cutoff wavelengths in VV polarization, which quantitatively represent the effect of velocity bunching, are compared with theoretical values calculated using WW3-simulated SWH. A notable relationship is observed between the difference in azimuthal cutoff wavelength and SAR-derived wind speed and WW3-simulated SWH. Analysis results show that the correlation between SAR-based azimuthal cutoff wavelength and SWH is stronger than that with wind and current. Finally, a machine learning algorithm is used to develop an algorithm aimed at simulating the azimuthal cutoff wavelength in TCs, including wind, wave, and incidence angle. This method yields an RMSE of 8.90 m, a COR of 0.91, and a scatter index of 0.04 for VV-polarization SAR.

In this paper, some algorithms for the rain rate from Sentinel-1 (S-1) synthetic aperture radar (SAR) images are proposed to increase the accuracy of rainfall retrieval during a tropical cyclone (TC). The proposed algorithms contain the extreme gradient boosting (XGBoost), random forest (RF), and support vector regression (SVR). The SAR rain rate retrieval results using the XGBoost are validated against the Global Precipitation Measurement (GPM) data, yielding a root mean square error (RMSE) of 2.43 mm h–1, correlation (COR) of 0.92, and scatter index (SI) of 0.65, which are better than the 5.3 mm h–1 RMSE, 0.51 COR, and 1.38 SI achieved using the re-tuned empirical model CRAIN2_S1, the RF values of 2.50 RMSE, 0.91 COR, and 0.67 SI, and the SVR model with a RMSE of 4.18, a COR of 0.73, and a SI of 1.11. It is concluded that the trained XGBoost is recommended for use in SAR rain rate retrieval during TCs.

This study develops a deep learning (DL) model to extract the ship size from Sentinel-1 synthetic aperture radar (SAR) images, named SSENet. We employ a single shot multibox detector (SSD)-based model to generate a rotatable bounding box (RBB) for the ship. We design a deep-neural-network (DNN)-based regression model to estimate the accurate ship size. The hybrid inputs to the DNN-based model include the initial ship size and orientation angle obtained from the RBB and the abstracted features extracted from the input SAR image. We design a custom loss function named mean scaled square error (MSSE) to optimize the DNN-based model. The DNN-based model is concatenated with the SSD-based model to form the integrated SSENet. We employ a subset of the OpenSARShip, a data set dedicated to Sentinel-1 ship interpretation, to train and test SSENet. The training/testing data set includes 1500/390 ship samples. Experiments show that SSENet is capable of extracting the ship size from SAR images end to end. The mean absolute errors (MAEs) are under 0.8 pixels, and their length and width are 7.88 and 2.23 m, respectively. The hybrid input significantly improves the model performance. The MSSE reduces the MAE of length by nearly 1 m and increases the MAE of width by 0.03m compared to the mean square error (MSE) loss function. Compared with the well-performed gradient boosting regression (GBR) model, SSENet reduces the MAE of length by nearly 2 m (18.68%) and that of width by 0.06 m (2.51%). SSENet shows robustness on different training/testing sets.

Accurate detection of sea ice leads is essential for safe navigation in polar regions. In this paper, a shape-aware (SA) network, SA-DeepLabv3+, is proposed for automatic lead detection from synthetic aperture radar (SAR) images. Considering the fact that training data are limited in the task of lead detection, we construct a dataset fusing dual-polarized (HH, HV) SAR images from the C-band Sentinel-1 satellite. Taking the DeepLabv3+ as the baseline network, we introduce a shape-aware module (SAM) to combine multi-scale semantic features and shape information and, therefore, better capture the shape characteristics of leads. A squeeze-and-excitation channel-position attention module (SECPAM) is designed to enhance lead feature extraction. Segmentation loss generated by the segmentation network and shape loss generated by the shape-aware stream are combined to optimize the network during training. Postprocessing is performed to filter out segmentation errors based on the aspect ratio of leads. Experimental results show that the proposed method outperforms the existing benchmarking deep learning methods, reaching 96.82% for overall accuracy, 93.01% for F1-score, and 91.48% for mIoU. It is also found that the fusion of dual-polarimetric SAR channels as the input could effectively improve the accuracy of sea ice lead detection.

<p><em>Flood inundation forecast maps provide an essential tool for disaster management teams to aid planning and preparation ahead of a flood event in order to mitigate the impacts of flooding on the community. Evaluating the accuracy of forecast flood maps is essential for model development and improving future flood predictions and can be achieved by comparison with flood maps derived from remote-sensing observations. Conventional, quantitative binary verification measures typically provide a domain averaged score, at grid level, of forecast skill. This score is dependent on the magnitude of the flood and the spatial scale of the flood map. Binary scores have limited physical meaning and do not indicate location specific variations in forecast skill that enable targeted model improvements to be made. A new, scale-selective approach is presented to evaluate forecast flood inundation maps against Synthetic Aperture Radar (SAR)-derived observed flood extents. We evaluate forecast flood maps out to 10-days lead time for the Rivers Wye and Lugg (UK) during Storm Dennis, February 2020. A neighbourhood approach based on the Fraction Skill Score is applied to assess the spatial scale at which the forecast becomes skilful at capturing the observed flood. This skilful scale varies with location and when combined with a contingency map creates a novel categorical scale map, a valuable visual tool for model evaluation and development. The impact of model improvements on forecast flood map accuracy skill scores are often masked by large areas of correctly predicted flooded/unflooded cells. To address this, the accuracy of the flood-edge location is evaluated: this provides a physically meaningful verification measure of the forecast flood-edge discrepancy. Representation errors are introduced where remote sensing observations capture the flood extent at different spatial resolutions in comparison with the model. We evaluate the sensitivity of the verification measures to the resolution of the SAR-derived flood map. </em></p><p><em>An ensemble of forecast flood inundation maps has the potential to represent the uncertainty in the flood forecast and provides a location specific, probabilistic, likelihood of flooding. This gives valuable information to flood forecasters, flood risk managers and insurers and will ultimately benefit people living in flood prone areas. We apply a scale selective approach to evaluate the spatial predictability of forecast ensemble flood maps. An ensemble forecast of flooding of the Brahmaputra in the Assam region, August 2017, is evaluated using flood extents derived from Sentinel-1 SAR images. The results are presented on a Spatial Spread-Skill (SSS) map, indicating where the flood map ensemble is over-, under- or well-spread. Overall, emphasis on scale, rather than domain-average score, means that comparisons can be made across different flooding scenarios and forecast systems and between forecasts at different spatial scales.</em></p>

Iceberg information in polar regions is crucial for various applications. Synthetic aperture radar (SAR) satellites provide high-resolution remote sensing images without being affected by weather conditions, which are widely used for iceberg monitoring. Most current studies track icebergs by shape similarity and use the distance between the centroids of the icebergs as the offset between the two temporal images. However, it is difficult to characterize icebergs comprehensively with a single shape similarity, the matching performance of the traditional shape similarity-based iceberg tracking method depends on the profile extraction accuracy in the iceberg detection. Furthermore, the limited coverage of satellite remote sensing images prevents the full capture of all icebergs. The drift of these icebergs in the time-series image exhibits shape variations, leading to a mismatch in the centroids, affecting the accuracy of drift velocity calculations. To address these issues, this letter proposes an object-oriented rotated intensity matching (OORIM)-based iceberg drift method that considers both the boundary shape and the intensity similarity of iceberg objects and obtains both offset information and rotation angle simultaneously. We successfully tracked 21 icebergs from Sentinel-1 SAR images near the Weddell Sea, and the results demonstrate the superiority of this method compared with other methods in terms of the accuracy of iceberg tracking. Specifically, the correct tracking accuracies of the centroid distance histogram (CDH), angle distance vector (ADV), and proposed OORIM methods are 90.5%, 66.7%, and 100%, respectively.

The movement of glaciers plays a crucial role in environmental and geological processes, significantly influencing the formation and dynamics of ice bodies. This study leverages feature tracking technology to analyze optical and Synthetic Aperture Radar (SAR) remote sensing imagery, specifically GF-1 optical images and GF-3, Sentinel-1 SAR images, captured during the 2020 to 2021 ablation season in Gongar. The aim was to quantify glacier surface velocities and to evaluate the comparative effectiveness of different remote sensing modalities in capturing these dynamics. Our findings indicate a strong consistency in the spatial distribution of glacier surface velocities derived from diverse remote sensing data sources, with high-precision optical imagery (GF-1) yielding the most accurate velocity measurements, followed by Sentinel-1 SAR data. Notably, large glaciers in Gongar exhibited rapid movements, with an average velocity of 0.16 m/d, primarily at elevations between 4,500 and 6,500 m. The fastest velocities were recorded at approximately 4,500 m elevation. Glaciers with inclines ranging from 10° to 60° displayed the highest velocities within the 20°–30° slope range. It was observed that glaciers on the southeast slope moved faster, exhibiting the highest average surface velocity, in contrast to those on the west slope, which moved more slowly. The surface velocity of the ice tongue region of Krayaylak Glacier that the largest glacier in Pamir, was observed to be lower than 0.6 m/d, indicating a slow movement speed. The study also reveals that the effectiveness of different remote sensing data in detecting glacier velocity in Gongar, with high-resolution data more accurately capturing surface velocities in melting areas or those with slower movement. This study underscores the importance of multi-source remote sensing data in understanding glacier dynamics and contributes valuable insights into the mechanisms driving glacier movements.

Synthetic Aperture Radar (SAR) interferometry is currently providing displacement measurements over large areas and recently also maps of water vapor in atmosphere. However, the interferometric phase can also capture tiny contributions due to soil moisture. In this work the results obtained in a 1-year experiment with in-situ measurements of soil moisture in agricultural area, in the flat land around the Tagus river, Lisbon, Portugal, are presented. We analyze two interferometric models for the estimation of soil moisture from SAR images. Sentinel-1 SAR images in the C-band are used due to their regular acquisition plan and the short revisiting time. We present examples of soil maps from Sentinel-1 data, characterized by a high spatial resolution and short updating time and comments on their use in precision farming.

Copernicus Sentinel-1 Synthetic Aperture Radar (SAR) mission systematically acquires data in Interferometric Wide swath mode over European land and water. This study investigates the potential of this data processed into Level-1 (L1) Single Look Complex (SLC) to compute meaningful image cross-spectra over the ocean to retrieve ocean surface waves parameters. First, each L1 SLC is processed relative to a "tile'', a prescribed spatial division of the burst (a unitary acquisition sequence of Sentinel-1 TOPS mode) and SAR image cross-spectra are defined within the burst and between two adjacent bursts. Then, using a neural network approach based on SAR image cross-spectra and on the WaveWatch III® wave model, an algorithm is proposed to provide estimates of significant wave height, fractions of wave height due to wind sea, and mean wave period. As obtained, the algorithm yields to promising performances when validated against in situ and satellite data.

Burned area (BA) mapping of a forest after a fire is required for its management and the determination of the impacts on ecosystems. Different remote sensing sensors and their combinations have been used due to their individual limitations for accurate BA mapping. This study analyzes the contribution of different features derived from optical, thermal, and Synthetic Aperture Radar (SAR) images to extract BA information from the Turkish red pine (Pinus brutia Ten.) forest in a Mediterranean ecosystem. In addition to reflectance values of the optical images, Normalized Burn Ratio (NBR) and Land Surface Temperature (LST) data are produced from both Sentinel-2 and Landsat-8 data. The backscatter of C-band Sentinel-1 and L-band ALOS-2 SAR images and the coherence feature derived from the Interferometric SAR technique were also used. The pixel-based random forest image classification method is applied to classify the BA detection in 24 scenarios created using these features. The results show that the L-band data provided a better contribution than C-band data and the combination of features created from Landsat LST, NBR, and coherence of L-band ALOS-2 achieved the highest accuracy, with an overall accuracy of 96% and a Kappa coefficient of 92.62%.

Supervised and unsupervised classification procedures are developed and applied to synthetic aperture radar (SAR) polarimetric images in order to identify their various Earth terrain components. For supervised classification processing, the Bayes technique is used to classify fully polarimetric and normalized polarimetric SAR data. Simpler polarimetric discriminates, such as the absolute and normalized magnitude response of the individual receiver channel returns, in addition to the phase difference between the receiver channels, are also considered. Another processing algorithm, based on comparing general properties of the Stokes parameters of the scattered wave to that of simple scattering models, is also discussed. This algorithm, which is an unsupervised technique, classifies terrain elements based on the relationship between the orientation angle and handedness of the transmitting and receiving polarization states. These classification procedures have been applied to San Francisco Bay and Traverse City SAR images, supplied by the Jet Propulsion Laboratory. It is shown that supervised classification yields the best overall performance when accurate classifier training data are used, whereas unsupervised classification is applicable when training data are not available.

The availability of synthetic aperture radar (SAR) data offers great potential for environmental monitoring due to the insensitiveness of SAR imagery to atmospheric and sun- light-illumination conditions. In addition, the short revisit time provided by future SAR-based missions will allow a huge amount of multitemporal SAR data to become systematically available for monitoring applications. In this paper, the problem of detecting the changes that occurred on the ground by analyzing SAR im- agery is addressed by a completely unsupervised approach, i.e., by developing an automatic thresholding technique. The image- ratioing approach to SAR change detection is adopted, and the Kittler and Illingworth minimum-error thresholding algorithm is generalized to take into account the non-Gaussian distribution of the amplitude values of SAR images. In particular, a SAR-specific parametric modeling approach for the ratio image is proposed and integrated into the thresholding process. Experimental results, which confirm the accuracy of the method for real X-band SAR and spaceborne imaging radar C-band images, are presented.

Extreme learning machine (ELM) as well as its variants have been widely used in many fields for its good generalization performance and fast learning speed. Though distributed ELM can sufficiently process large-scale labeled training data, the current technology is not able to process partial labeled or unlabeled training data. Therefore, we propose a new unified distributed ELM with supervised, semi-supervised and unsupervised learning based on MapReduce framework, called U-DELM. The U-DELM method can be used to overcome the existing distributed ELM framework’s lack of ability to process partially labeled and unlabeled training data. We first compare the computation formulas of supervised, semi-supervised and unsupervised learning methods and found that the majority of expensive computations are decomposable. Next, MapReduce framework based U-DELM is proposed, which extracts three different matrices continued multiplications from the three computational formulas introduced above. After that, we transform the cumulative sums respectively to make them suitable for MapReduce. Then, the combination of the three computational formulas are used to solve the output weight in three different learning methods. Finally, by using benchmark and synthetic datasets, we are able to test and verify the efficiency and effectiveness of U-DELM on learning massive data. Results prove that U-DELM can achieve unified distribution on supervised, semi-supervised and unsupervised learning.

Co-registering the Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 optical data of the European Space Agency (ESA) is of great importance for many remote sensing applications. However, we find that there are evident misregistration shifts between the Sentinel-1 SAR and Sentinel-2 optical images that are directly downloaded from the official website. To address that, this paper presents a fast and effective registration method for the two types of images. In the proposed method, a block-based scheme is first designed to extract evenly distributed interest points. Then, the correspondences are detected by using the similarity of structural features between the SAR and optical images, where the three-dimensional (3D) phase correlation (PC) is used as the similarity measure for accelerating image matching. Lastly, the obtained correspondences are employed to measure the misregistration shifts between the images. Moreover, to eliminate the misregistration, we use some representative geometric transformation models such as polynomial models, projective models, and rational function models for the co-registration of the two types of images, and we compare and analyze their registration accuracy under different numbers of control points and different terrains. Six pairs of the Sentinel-1 SAR L1 and Sentinel-2 optical L1C images covering three different terrains are tested in our experiments. Experimental results show that the proposed method can achieve precise correspondences between the images, and the third-order polynomial achieves the most satisfactory registration results. Its registration accuracy of the flat areas is less than 1.0 10 m pixel, that of the hilly areas is about 1.5 10 m pixels, and that of the mountainous areas is between 1.7 and 2.3 10 m pixels, which significantly improves the co-registration accuracy of the Sentinel-1 SAR and Sentinel-2 optical images.

More than 69,000 people died following the magnitude 8.0 Wenchuan earthquake of May 12, 2008. Bad weather hampered relief efforts, and in some cases rescuers had to trek into the disaster area on foot and search for trapped survivors by hand as roads were blocked by debris. Due to travel difficulties, spatial information needs to be extracted in the disaster area by remote sensing techniques. The main problem focused on in this paper is how to use the all-weather and all-day/night capability of Synthetic Aperture Radar (SAR) to extract primary seismic disaster information after the earthquake. Using air- and space-borne SAR images with different bands, polarizations and incidence angles, including multi-polarization X-band air-borne data, C-band polarimetric Radarsat-2, X-band TerraSAR-X with high resolution, multi-polarization X-band COSMO-SkyMed and L-band multi-polarization ALOS-PalSAR space-borne data, we perform image characteristics analysis of landslides. Obvious differences can be recognized between old and new landsides in SAR images with different bands. Multi-polarization SAR can play an important role in landslide discrimination. Two SAR images at different incidence angles do not provide much more information if the difference between the two angles is small. Landslide recognition accuracy strongly depends on the direction of view, especially for large incidence angles, in which case the characteristic difference for landslide recognition is great. There are different polarization responses between a landslide and its surroundings that can be used to recognize the landslide. Interferometric SAR images, on the other hand, do not provide good recognition capability due to temporal decorrelation and resolution. Meanwhile, information extraction of barrier lakes using different resolution and incidence angle SAR images is analyzed in this paper; small incidence angles and high resolutions improve the object recognition and information extraction of barrier lakes.