Search

With the rapid increase of the world population, climate changes, and the slow expansion of cultivated areas, only precision agriculture (PA) can provide enough food or resources. PA requires flexible instruments for measuring the spectral signatures of the crops to understand their conditions. Unfortunately, the high initial costs of multispectral (MS) cameras reduce the implementation of PA in small farms, which constitute a large portion of arable land in Europe and contribute with social bonds, local know-how, and cultural legacy. With the objective to speed up the use of MS imaging, in this article, we present a novel low-cost imaging device consisting of an MS camera with nine bands and a thermal imager, whose price is several times lower than commercially available ones. This article describes the design and the calibration of the imaging device based on the off-the-shelf components: Raspberry Pi, dedicated quad camera kit, thermal core, and multiband optical filters. The spectral reconstruction accuracy has a high average R2 score of 0.986. Finally, images from multiple sensors are aligned using phase-only correlation and dense optical flow, providing a method that can be implemented on all platforms. The presented solution is open source, permitting one to modify and expand the capabilities of the described device and adapt to specific needs. Moreover, the device is flexible as the thermal camera can be removed to reduce the total system cost if its usage is not required. Even if its primary application is PA, the proposed solution can be used for other applications.

Abstract Lunar surface chemistry is essential for revealing petrological characteristics to understand the evolution of the Moon. Existing chemistry mapping from Apollo and Luna returned samples could only calibrate chemical features before 3.0 Gyr, missing the critical late period of the Moon. Here we present major oxides chemistry maps by adding distinctive 2.0 Gyr Chang’e-5 lunar soil samples in combination with a deep learning-based inversion model. The inferred chemical contents are more precise than the Lunar Prospector Gamma-Ray Spectrometer (GRS) maps and are closest to returned samples abundances compared to existing literature. The verification of in situ measurement data acquired by Chang'e 3 and Chang'e 4 lunar rover demonstrated that Chang’e-5 samples are indispensable ground truth in mapping lunar surface chemistry. From these maps, young mare basalt units are determined which can be potential sites in future sample return mission to constrain the late lunar magmatic and thermal history.

Although graph convolutional networks have found application in polarimetric synthetic aperture radar (PolSAR) image classification tasks, the available approaches cannot operate on multiple graphs, which hinders their potential to generalize effective feature representations across different datasets. To overcome this limitation and achieve robust PolSAR image classification, this paper proposes a novel end-to-end cross-level interaction graph U-Net (CLIGUNet), where weighted max-relative spatial convolution is proposed to enable simultaneous learning of latent features from batch input. Moreover, it integrates weighted adjacency matrices, derived from the symmetric revised Wishart distance, to encode polarimetric similarity into weighted max-relative spatial graph convolution. Employing end-to-end trainable residual transformers with multi-head attention, our proposed cross-level interactions enable the decoder to fuse multi-scale graph feature representations, enhancing effective features from various scales through a deep supervision strategy. Additionally, multi-scale dynamic graphs are introduced to expand the receptive field, enabling trainable adjacency matrices with refined connectivity relationships and edge weights within each resolution. Experiments undertaken on real PolSAR datasets show the superiority of our CLIGUNet with respect to state-of-the-art networks in classification accuracy and robustness in handling unknown imagery with similar land covers.

Up-to-date canopy height model (CHM) estimates are of key importance for forest resource monitoring and disturbance analysis. In this article, we present a study on the potential of deep learning (DL) for the regression of forest height from TanDEM-X bistatic interferometric synthetic aperture radar (InSAR) data. We propose a novel fully convolutional neural network framework, trained in a supervised manner using reference CHM measurements derived from the LiDAR LVIS airborne sensor from NASA. The reference measurements were acquired during the joint NASA–ESA 2016 AfriSAR campaign over five sites in Gabon, Africa, characterized by the presence of different kinds of vegetation, spanning from tropical primary forests to mangroves. Together with the DL architecture and training strategy, we present a series of experiments to assess the impact of different input features on the network estimation accuracy (in particular of bistatic InSAR-related ones). When tested on all the considered sites, the proposed DL model achieves an overall performance of 1.46-m mean error, 4.2-m mean absolute error, and 15.06% mean absolute percentage error. Furthermore, we perform a spatial transfer analysis aimed at deriving preliminary insights on the generalization capability of the network when trained and tested on datasets acquired over different locations, combining different kinds of tropical vegetation. The obtained results are promising and already in line with state-of-the-art methods based on both physical-based modeling and data-driven approaches, with the remarkable advantage of requiring only one single TanDEM-X acquisition at inference time.

This article presents a novel system that produces multiyear high-resolution irrigation water demand maps for agricultural areas, enabling a new level of detail for irrigation support for farmers and agricultural stakeholders. The system is based on a scalable distributed deep learning (DL) model trained on dense time series of Sentinel-2 images and a large training set for the first year of observation and fine tuned on new labeled data for the consecutive years. The trained models are used to generate multiyear crop type maps, which are assimilated together with the Sentinel-2 dense time series and the meteorological data into a physically based agrohydrological model to derive the irrigation water demand for different crops. To process the required large volume of multiyear Copernicus Sentinel-2 data, the software architecture of the proposed system has been built on the integration of the Food Security thematic exploitation platform (TEP) and the data-intensive artificial intelligence Hopsworks platform. While the Food Security TEP provides easy access to Sentinel-2 data and the possibility of developing processing algorithms directly in the cloud, the Hopsworks platform has been used to train DL algorithms in a distributed manner. The experimental analysis was carried out in the upper part of the Danube Basin for the years 2018, 2019, and 2020 considering 37 Sentinel-2 tiles acquired in Austria, Moravia, Hungary, Slovakia, and Germany.

Modeling and large-scale mapping of forest aboveground biomass (AGB) is a complicated, challenging, and expensive task. There are considerable variations in forest characteristics that create functional disparity for different models and needs comprehensive evaluation. Moreover, the human-bias involved in the process of modeling and evaluation affects the generalization of models at larger scales. In this article, we present an automated machine learning framework for modeling, evaluation, and stacking of multiple base models for AGB prediction. We incorporate a hyperparameter optimization procedure for automatic extraction of targeted features from multitemporal Sentinel-2 data that minimizes human-bias in the proposed modeling pipeline. We integrate the two independent frameworks for automatic feature extraction and automatic model ensembling and evaluation. The results suggest that the extracted target-oriented features have an excessive contribution of red-edge and short-wave infrared spectrum. The feature importance scale indicates a dominant role of summer-based features as compared to other seasons. The automated ensembling and evaluation framework produced a stacked ensemble of base models that outperformed individual base models in accurately predicting forest AGB. The stacked ensemble model delivered the best scores of R2cv = 0.71 and RMSE = 74.44 Mgha−1. The other base models delivered R2cv and RMSE ranging between 0.38–0.66 and 81.27–109.44 Mg ha−1, respectively. The model evaluation metrics indicated that the stacked ensemble model was more resistant to outliers and achieved a better generalization. Thus, the proposed study demonstrated an effective automated modeling pipeline for predicting AGB by minimizing human-bias and deployable over large and diverse forest areas.

Multisensor data analysis allows exploiting heterogeneous data regularly acquired by the many available remote sensing (RS) systems. Machine- and deep-learning methods use the information of heterogeneous sources to improve the results obtained by using single-source data. However, the state-of-the-art methods analyze either the multiscale information of multisensor multiresolution images or the time component of image time series. We propose a supervised deep-learning classification method that jointly performs a multiscale and multitemporal analysis of RS multitemporal images acquired by different sensors. The proposed method processes very-high-resolution (VHR) images using a residual network with a wide receptive field that handles geometrical details and multitemporal high-resolution (HR) image using a 3-D convolutional neural network that analyzes both the spatial and temporal information. The multiscale and multitemporal features are processed together in a decoder to retrieve a land-cover map. We tested the proposed method on two multisensor and multitemporal datasets. One is composed of VHR orthophotos and Sentinel-2 multitemporal images for pasture classification, and another is composed of VHR orthophotos and Sentinel-1 multitemporal images. Results proved the effectiveness of the proposed classification method.

With many platforms and sensors continuously observing the earth surface, the large amount of remote sensing data presents a big data challenge. While remote sensing data acquisition capability can fully meet the requirements of many application domains, there is still a need to further explore how to efficiently mine the useful information from remote sensing big data (RSBD). Many researchers in the remote sensing community have introduced deep learning in the process of RSBD, and deep learning-based methods have achieved better performance compared with traditional methods. However, there are still substantial obstacles to the application of deep learning in remote sensing. One of the major challenges is the generation of pixel-level labels with high quality for training samples, which is essential to deep learning models. Weakly supervised deep learning (WSDL) is a promising solution to address this problem as WSDL can utilize greedily labeled datasets that are easy to collect but not ideal to train the deep networks. In this review, we summarize the achievements of WSDL-driven cost-efficient information extraction from RSBD. We first analyze the opportunities and challenges of information extraction from RSBD. Based on the analysis of the theoretical foundations of WSDL in the computer vision (CV) domain, we conduct a survey on the WSDL-based information extraction methods under the data characteristic and task demand of RSBD in four different tasks: (i) scene classification, (ii) object detection, (iii) semantic segmentation and (iv) change detection. Finally, potential research directions are outlined to guide researchers to further exploit WSDL-based information extraction from RSBD.

The Yangtze River Basin (YZRB) and the Yellow River Basin (YRB), which are crucial for ecology and economy in China, face growing challenges to ecosystem service (ES) functions due to global population growth, urbanization, and climate change. This study assessed the spatiotemporal dynamics of ESs in the YZRB and the YRB between 2001 and 2021, comprehensively encompassing essential aspects such as water yield (WY), carbon sequestration (CS), soil conservation (SC), and habitat quality (HQ) while also analyzing the trade-offs and synergies among these ESs at the grid cells. The GeoDetector was employed to ascertain individual or interactive effects of natural and anthropogenic factors on these ESs and their trade-offs/synergies. The results showed that (1) from 2001 to 2021, the four ESs exhibited significant spatial disparities in the distribution within two basins, with the overall trend of ESs mainly increasing. YZRB consistently exhibited substantially higher ES values than the YRB. (2) Complex trade-offs and synergies among these ESs were apparent in both basins, characterized by distinct spatial heterogeneity. The spatial relationships of WY–CS, WY–SC, CS–SC, and CS–HQ were mainly synergistic. (3) Precipitation, potential evapotranspiration, elevation, land use and land cover (LULC), and slope influenced ESs in both basins. Notably, interactive factors, particularly the interactions involving LULC and other factors, demonstrated more robust explanatory power for ESs and their trade-offs/synergies than individual drivers. These findings significantly affect the refined ecosystem management and sustainable development decision-making in large rivers or regions.

This article presents an operational system for the automatic production of high-resolution (HR) large-scale land cover (LC) maps in a fast, efficient, and unsupervised manner. This is based on a scalable and parallelizable tile-based approach, which does not require the collection of new training data. The method leverages the complementary information provided by the existing LC maps and recent acquisitions of HR Earth observation (EO) images to identify map units that have the highest probability of being correctly associated with their labels, and exploit the obtained “weak” training set to produce an updated HR LC map by classifying the recently acquired EO data. Both steps, performed at tile level, can be implemented on a high-performance computing (HPC) environment, which simultaneously process all required tiles (independently of each other) for the entire study area. The method was tested considering the publicly available 2018 Corine LC map having a minimum mapping unit of 25 ha and the Sentinel-2 images to generate an HR LC map of Italy. The obtained map has a spatial resolution of 10 m and presents the nine major LC types (i.e., “artificial land,” “bareland,” “grassland,” “cropland,” “broadleaves,” “conifers,” “snow,” “water,” and “shrubland”). Validation was performed using the 2018 European Land Use and Coverage Area Frame Survey database made up of in situ data. The overall accuracy achieved for the Northern, Southern, and Central part of Italy and the Italian Islands is 91.29%, 91.63%, 92.21%, and 91.06%, respectively.

Availability of multitemporal (MT) images, such as the sentinel-2 (S2) ones, offers accurate spatial, spectral and temporal information to effectively monitor vegetation, more specifically agriculture. Agricultural practices can benefit from temporally dense satellite image time series (SITS) for accurate understanding of the phenological evolution and behavior of crops. Developing techniques that deal with high spatial correlation and high temporal resolution requires a shift in the processing paradigm and poses new challenges in terms of data processing and methodology. This article presents an automatic approach to large-scale precise mapping of small agricultural fields based on the analysis of S2-SITS at Country level. The approach deals with a flexible and automatic processing chain for massive data and was tested at Country level. The large-scale application requires to consider: the management of big amount of data with particular attention to download and pre-processing of S2-SITS; and MT fine characterization of crop fields accounting for the strong variability in size and phenological behaviors when mapping at large scale. Both challenges are addressed in an automatic way by exploiting and/or updating state-of-the-art methodologies. Promising results have been obtained and validated over 2017 and 2018 agrarian years for Italy.

Aboveground biomass (AGB) is an important forest attribute directly linked to the forest carbon pool. The use of satellite remote sensing (RS) data has increased for AGB prediction due to their large footprint and low-cost availability. However, they have been limited due to saturation effect that leads to low prediction precision. In this article, we propose an innovative and dynamic architecture based on generative neural network that extracts target oriented generative features for predicting forest AGB using satellite RS data. These features are more resilient to mixed forest types and geographical conditions as compared to the traditional features and models. The effectiveness of the proposed features was assessed by experiments performed using multispectral, synthetic aperture radar, and combined dual-source datasets. The proposed model achieved best performance in terms of precision, model agreement, and overfitting as compared to the other conventional models for all analyzed datasets. The t-distributed stochastic neighbor embedding scatterplots of the generative features clearly show one dimension of the feature space associated with the target AGB. Feature importance analysis indicated that the produced generative features were more significant than the conventional analytical features. Also, the model provided a robust framework for homogeneous fusion of multisensor features from satellite RS data for predicting AGB.

A comprehensive 3-D structural mapping of stem is essential for an accurate 3-D crown modeling and tree parameter estimation. Terrestrial laser scanning (TLS) is an effective technology for a comprehensive collection of individual tree level data, compared to destructive and costly field measurements. The performance of 3-D stem modeling techniques is adversely affected by laser shadowing and point-density variations in TLS data. In addition, most of the state-of-the-art techniques perform stem modeling using regular geometric shapes, such as circle, ellipse, and cylinder, which cannot accurately capture the complex 3-D stem geometric shapes. This results in stem modeling errors. In this article, we propose a 3-D stem modeling approach for both single- and multiscan TLS data that: 1) does not make any prior assumption on the stem geometry and 2) is minimally affected by undesirable stem shadowing and point-density variations. The proposed approach accurately models 3-D stem and localizes branch-knots for both coniferous and deciduous trees by mapping the void volume formed within the stem due to the opaqueness of tree stem to laser. The modeling performance was evaluated on both single- and multiscan data obtained for pine, spruce, and birch species. The low estimation errors associated with the stem diameter at breast height and branch-knot location, compared to the reference methods, prove the ability of the proposed method to both accurately model 3-D stem and localize branch-knots.

This special issue of the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS) contains 11 papers both from the extended outcomes of the Multitemp 2019 presented papers, and from the submissions by following a general Call-for-Papers of this special issue. These papers focus on the interesting and relevant topics in the multitemporal data analysis.

This article presents TimeSen2Crop, a pixel-based dataset made up of more than 1 million samples of Sentinel 2 time series (TSs) associated to 16 crop types. This dataset, publicly available, aims to contribute to the worldwide research related to the supervised classification of TSs of Sentinel 2 data for crop type mapping. TimeSen2Crop includes atmospherically corrected images and reports the snow, shadows, and clouds information per labeled unit. The provided TSs represent an agronomic year (i.e., period from one year's harvest to the next one for agricultural commodity) ranging from September 2017 to August 2018. To generate the dataset, the publicly available Austrian crop type map based on farmer's declarations has been considered. To ensure the selection of reliable labeled units from the map (i.e., pure pixels correctly associated to their labels), an automatic procedure for the extraction of the training set based on a multitemporal deep learning model has been defined. TimeSen2Crop also includes a TS of Sentinel 2 images acquired in the following agronomic year (i.e., from September 2018 to August 2019). These data are provided with the aim of attract more research activities for solving a typical challenge of the crop type mapping task: adapting multitemporal deep learning models to different year (domain adaptation). The design of the dataset is described along with a benchmark comparison of deep learning models for crop type mapping.

Bringing together a number of cutting-edge technologies that range from storing extremely large volumes of data all the way to developing scalable machine learning and deep learning algorithms in a distributed manner and having them operate over the same infrastructure poses unprecedented challenges. One of these challenges is the integration of European Space Agency (ESA)'s Thematic Exploitation Platforms (TEPs) and data information access service platforms with a data platform, namely Hopsworks, which enables scalable data processing, machine learning, and deep learning on Copernicus data, and development of very large training datasets for deep learning architectures targeting the classification of Sentinel images. In this article, we present the software architecture of ExtremeEarth that aims at the development of scalable deep learning and geospatial analytics techniques for processing and analyzing petabytes of Copernicus data. The ExtremeEarth software infrastructure seamlessly integrates existing and novel software platforms and tools for storing, accessing, processing, analyzing, and visualizing large amounts of Copernicus data. New techniques in the areas of remote sensing and artificial intelligence with an emphasis on deep learning are developed. These techniques and corresponding software presented in this article are to be integrated with and used in two ESA TEPs, namely Polar and Food Security TEPs. Furthermore, we present the integration of Hopsworks with the Polar and Food Security use cases and the flow of events for the products offered through the TEPs.

High-resolution (HR) snow cover maps derived by remotely sensed images are an asset for data assimilation in hydrological models. However, the current satellite missions do not provide daily HR multispectral observations suitable for an accurate snow monitoring in alpine environments. On the contrary, low-resolution (LR) sensors acquire daily information of snow cover fraction (SCF) but at an inappropriate spatial scale. This article proposes a novel approach that combines multisource and multiscale acquisitions to infer the daily HR snow cover area (SCA) for mountainous basins. The approach builds on the assumption that interannual snow patterns are both affected by the local geomorphometry and meteorology. We derive these patterns through a hierarchical multistep approach based on historical statistical analyses on a long and sparse HR time-series. At each step, we obtain HR snow cover maps with higher number of reconstructed pixels but decreasing level of confidence. Historical data are used to estimate the probability that a HR pixel is covered by snow according to two possible multiscale strategies: 1) HR gap-filling, or 2) LR downscaling. These analyses lead to the identification of the patterns that regularly appear given certain conditions. When no systematic patterns are observed, we reinforce the inference of the pixel class by a generalized additive model that exploits not only the historical data, but also explicit geomorphometric, global snow, and multitemporal properties. The proposed approach has been validated on a catchment in the Sierra Nevada, USA, for three hydrological years (2017–2019) showing an average overall accuracy of 92%.

The combination of data acquired by Landsat-8 and Sentinel-2 earth observation missions produces dense time series (TSs) of multispectral images that are essential for monitoring the dynamics of land-cover and land-use classes across the earth's surface with high temporal resolution. However, the optical sensors of the two missions have different spectral and spatial properties, thus they require a harmonization processing step before they can be exploited in remote sensing applications. In this work, we propose a workflow-based on a deep learning approach to harmonize these two products developed and deployed on an high-performance computing environment. In particular, we use a multispectral generative adversarial network with a U-Net generator and a PatchGan discriminator to integrate existing Landsat-8 TSs with data sensed by the Sentinel-2 mission. We show a qualitative and quantitative comparison with an existing physical method [National Aeronautics and Space Administration (NASA) Harmonized Landsat and Sentinel (HLS)] and analyze original and generated data in different experimental setups with the support of spectral distortion metrics. To demonstrate the effectiveness of the proposed approach, a crop type mapping task is addressed using the harmonized dense TS of images, which achieved an overall accuracy of 87.83% compared to 81.66% of the state-of-the-art method.

A general framework for change detection is proposed to analyze multimodal remotely sensed data utilizing the Kronecker product between two data representations (vectors or matrices). The proposed method is sensor independent and provides comparable results to techniques that exist for specific sensors. The proposed fusion technique is a pixel-level approach that incorporates inputs from different modalities, rendering enriched multimodal data representation. Thus, the proposed hybridization procedure helps to assimilate multisensor information in a meaningful manner. A novel change index ($\zeta$) is defined for the general multimodal case. This index is then used to quantify the change in bitemporal remotely sensed data. This article explores the usability, consistency, and robustness of the proposed multimodal fusion framework, including the change index, with proper validation on two multimodal cases: 1) the dual-frequency ($C$- and $L$-band) fully polarimetric Danish EMISAR data and 2) the dual-polarimetric synthetic aperture radar and Sentinel-2 multispectral data. Detailed analysis and validation using extensive ground-truth data are presented to establish the proposed framework.

Using Chang’E data, the authors here identify more than 109,000 previously unrecognized lunar craters and date almost 19,000 craters based on transfer learning with deep neural networks. A new lunar crater database is derived and distributed to the planetary community.

Classification techniques applied to hyperspectral images are very useful for lithologic discrimination and geological mapping. Classifiers are often applied either to all spectral channels or only to absorption spectral channels. However, it is difficult to obtain different lithology information using specific absorption regions from the narrow bandwidth and contiguous spectral channels due to spectral variability among rocks. In this article, we propose a band selection (BS) method for hyperspectral lithologic discrimination, in which the lithological superpixels are first gathered. A spectral bands selection criterion is learned by measuring the homogeneity and the variation of the lithological superpixels, and lithologic discriminating bands are identified by an efficient clustering algorithm based on affinity propagation. In this article, two geologic test sites, i.e., the Airborne Visible/Infrared Imaging Spectrometer data of the Cuprite, Nevada, USA, including 11 lithologic units (9 types of rocks) and the Hyperion data of Junggar, China, with 5 lithologic units, are chosen for validation. The performance of the proposed BS method is compared with those of using all the bands, specific absorption spectral channels, and two literature BS techniques. Experimental results show that the proposed method improves mapping accuracy by selecting fewer bands with higher lithologic discrimination capability than the other considered methods.

Attribute profiles are well-acknowledged as one of the most significant techniques to characterize spectral-spatial properties of a hyperspectral image. The spectral-spatial content of an attribute profile is influenced by the threshold values considered during its construction. In this article, we propose a robust method to detect the threshold values automatically by overcoming the limitations of the existing techniques. The proposed method employs a tree structure representing the connected components of the image and evaluates attribute values at each node. Then, a total characteristic function (TCF) is defined that represents these attribute values in a nondecreasing order. The defined TCF is analyzed using a novel technique to detect a few informative thresholds for the construction of a low-dimensional attribute profile representing substantial spectral-spatial information. The proposed threshold detection method is computationally efficient. To assess the effectiveness of the proposed technique experiments are conducted on three real hyperspectral datasets using six different attributes and the results are compared to the recent state-of-the-art method. The results demonstrate that the proposed method has several advantages over the existing state-of-the-art method.

Phase delay caused by atmospheric effects due to spatial and temporal variations of pressure, temperature, and water vapor content is one of the major error sources in estimation of ground deformation by interferometric synthetic aperture radar (InSAR>). Therefore, accuracy of ground deformation measurement is highly contingent on the robustness of the atmospheric correction techniques. These techniques rely either on auxiliary data such as numerical weather models (NWMs) or on the analysis of the interferometric phase itself. The accuracy in phase delays estimation of mixing effects of turbulent delay in atmosphere and stratified delay in lower troposphere is a key factor in determination of performance of each technique. Hence, the performance evaluation of the techniques is required in order to assess their potentials, robustness, and limitations. This article analyzes and evaluates the performance of four NWMs (i.e., ERA-Interim, ERA5, MERRA2, and WRF) and two phase-based techniques (i.e., linear and power law) to estimate phase delay using Sentinel-1A/B data over the Corvara landslide located in the Alps. The GPS data and GACOS product were used to validate the results. We generally found that ERA5 outperformed among other weather models with a phase standard deviation reduction of 77.7% (with respect to the InSAR phase), a correlation coefficient of 0.86 (between InSAR phase and estimated tropospheric delay) and a less significant error in the velocity estimation of the landslide.

This article presents an approach to improve the estimation of the glacier mass balance (GMB) of six selected alpine glaciers in the European Alps. This is achieved by combining three complementary data sources: hydroclimatological model, remote sensing (RS) data, and ground measurements. The hydroclimatological model provides spatially distributed mass balances. RS supplies spatially distributed surface characteristics. The ground point measurements provide the mass balance at the local scale. The combination of these data sources allows us to improve the spatial resolution of the model output and its GMB estimates. We used the alpine multiscale numerical distributed simulation engine model (AMUNDSEN), which considers the processes of accumulation and ablation of snow and ice for the area of the entire glacier (with a given spatial and temporal resolution). In the proposed integration approach, we first compute the deviations between the GMB simulation (afforded by the hydroclimatological model) and the ground measurements. Then, the RS data are used to define a feature space (which objectively characterizes the glacier surface properties). The method estimates the adjustment required to the model, for each unlabeled sample, leveraging on its neighboring labeled samples in the feature space. This allows us to apply similar adjustment to samples sharing similar glacier surface conditions. Experimental results show that the proposed integration approach achieves an average root-mean-square error of 460 mm (compared to 732 and 661 mm obtained by the hydroclimatological model and the standard regression models, typically used for parameters estimation).

In the presence of abrupt change events, multitemporal synthetic aperture radar (SAR) data represent a precious supporting tool for quantifying changes, in particular in urban areas. A large amount of SAR data also exists at very high resolution (VHR). Over urban areas, the introduction of the VHR imagery moves the analysis down to the single building scale. However, VHR imagery is also characterized by a large heterogeneity and a more complex representation of the building. In this work, we propose a geometrical model for describing partially destroyed buildings and derive the corresponding multitemporal backscattering signature by applying the ray-tracing method. The model is integrated into an unsupervised automatic approach for the detection of both fully and partially destroyed buildings. The strategy considers a hierarchical structure of the changes. Experimental results conducted on two multitemporal VHR SAR datasets show a large robustness of the approach and good accuracy in the detection of the classes for damaged buildings with different severity levels.

Data acquisition in planetary remote sensing missions is influenced by complex environmental resource- and instrument-specific constraints. This impedes to perform observations at any given time during the mission and with any of the instruments composing the scientific payload. This article presents an approach to automatic scheduling of acquisition operations of a remote sensing instrument composing the scientific payload of a mission. The methodology first subdivides the long available observation time intervals into shorter segments and then performs a selection of them, producing an acquisition schedule, optimized with respect to scientific requirements, instrument characteristics, and mission constraints. The scheduling problem is modeled as a multiobjective optimization problem and solved by using Genetic Algorithms (GAs). GAs are able to efficiently explore the solution space by considering different competing objective functions, reaching high-quality solutions. These solutions represent different optimized tradeoffs among the considered instrument-specific quality metrics. The approach is demonstrated on the operations of Radar for Icy Moons Exploration (RIME), a radar sounder onboard Jupiter Icy Moons Explorer. The obtained results show a high potential of the proposed methodology.

With the continuing improvement of remote-sensing (RS) sensors, it is crucial to monitor Earth surface changes at fine scale and in great detail. Thus, semantic change detection (SCD), which is capable of locating and identifying “from-to” change information simultaneously, is gaining growing attention in RS community. However, due to the limitation of large-scale SCD datasets, most existing SCD methods are focused on scene-level changes, where semantic change maps are generated with only coarse boundary or scarce category information. To address this issue, we propose a novel convolutional network for large-scale SCD (SCDNet). It is based on a Siamese UNet architecture, which consists of two encoders and two decoders with shared weights. First, multi-temporal images are given as input to the encoders to extract multi-scale deep representations. A multi-scale atrous convolution (MAC) unit is inserted at the end of the encoders to enlarge the receptive field as well as capturing multi-scale information. Then, difference feature maps are generated for each scale, which are combined with feature maps from the encoders to serve as inputs for the decoders. Attention mechanism and deep supervision strategy are further introduced to improve network performance. Finally, we utilize softmax layer to produce a semantic change map for each time image. Extensive experiments are carried out on two large-scale high-resolution SCD datasets, which demonstrates the effectiveness and superiority of the proposed method.

In the study of hyperspectral image classification based on machine learning theory and techniques, the problems related to the high dimensionality of the images and the scarcity of training samples are widely discussed as two main issues that limit the performance of the data-driven classifiers. These two issues are closely interrelated, but are usually addressed separately. In our study, we try to kill two birds with one stone by constructing an ensemble of lightweight base models embedded with spectral feature refining modules. The spectral feature refining module is a technique based on the mechanism of channel attention. This technique can not only perform dimensionality reduction, but also provide diversity within the ensemble. The proposed ensemble can provide state-of-the-art performance when the training samples are quite limited. Specifically, using only a total of 200 samples from each of the four popular benchmark data sets (Indian Pines, Salinas, Pavia University and Kennedy Space Center), we achieved overall accuracies of 89.34%, 95.75%, 93.58%, and 98.14%, respectively.

The empirical line (EL) calibration method is commonly used for atmospheric correction of remotely sensed spectral images and recovery of surface reflectance. The current EL-based methods are applicable to calibrate only single images. Therefore, the use of the EL calibration is impractical for imaging campaigns, where many (partially overlapped) images are acquired to cover a large area. In addition, the EL results are unconstrained and an undesired reflectance with negative values or larger than 100% can be obtained. In this paper, we use the standard EL model to formulate a new generalized empirical line (GEL) model. Based on the GEL, we present a novel method for simultaneous and constrained calibration of multiple images. This new method allows for calibration through multiple image constrained empirical line (MIcEL) and three additional calibration modes. Given a set of images, we use the available ground targets and automatically extracted tie points between overlapping images to calibrate all the images in the set simultaneously. Quantitative and visual assessments of the proposed method were carried out relatively to the off-the-shelf method quick atmospheric correction (QUAC), using real hyperspectral images and field measurements. The results clearly show the superiority of MIcEL with respect to the minimization of the difference between the reflectance values of the same object in different overlapping images. An assessment of the absolute accuracy, with respect to 11 field measurement points, shows that the accuracy of MIcEL, with an average mean absolute error (MAE) of ~11%, is comparable with respect to the QUAC.

Radar sounders, used for the geophysical exploration of celestial objects in the solar system, possess striking similarities to bat sonars. Here, the authors adapt and implement the bat clutter mitigation mechanism to radar geophysical exploration of planetary bodies.

Different methods are available for retrieving chlorophyll-a (Chl-a) in inland waters from optical imagery, but there is still a need for an inter-comparison among the products. Such analysis can provide insights into the method selection, integration of products, and algorithm development. This work aims at inter-comparison and consistency analyses among the Chl-a products derived from publicly available methods consisting of Case-2 Regional/Coast Colour (C2RCC), Water Color Simulator (WASI), and OC3 (3-band Ocean Color algorithm). C2RCC and WASI are physics-based processors enabling the retrieval of not only Chl-a but also total suspended matter (TSM) and colored dissolved organic matter (CDOM), whereas OC3 is a broadly used semi-empirical approach for Chl-a estimation. To pursue the inter-comparison analysis, we demonstrate the application of Sentinel-2 imagery in the context of multitemporal retrieval of constituents in some Italian lakes. The analysis is performed for different bio-optical conditions including subalpine lakes in Northern Italy (Garda, Idro, and Ledro) and a turbid lake in Central Italy (Lake Trasimeno). The Chl-a retrievals are assessed versus in situ matchups that indicate the better performance of WASI. Moreover, relative consistency analyses are performed among the products (Chl-a, TSM, and CDOM) derived from different methods. In the subalpine lakes, the results indicate a high consistency between C2RCC and WASI when aCDOM(440) < 0.5 m−1, whereas the retrieval of constituents, particularly Chl-a, is problematic based on C2RCC for high-CDOM cases. In the turbid Lake Trasimeno, the extreme neural network of C2RCC provided more consistent products with WASI than the normal network. OC3 overestimates the Chl-a concentration. The flexibility of WASI in the parametrization of inversion allows for the adaptation of the method for different optical conditions. The implementation of WASI requires more experience, and processing is time demanding for large lakes. This study elaborates on the pros and cons of each method, providing guidelines and criteria on their use.

Forest aboveground biomass (AGB) is a prime forest parameter that requires global level estimates to study the global carbon cycle. Light detection and ranging (LiDAR) is the state-of-the-art technology for AGB prediction but it is expensive, and its coverage is restricted to small areas. On the contrary, spaceborne Earth observation data are effective and economical information sources to estimate and monitor AGB at a large scale. In this paper, we present a study on the use of different spaceborne multispectral remote sensing data for the prediction of forest AGB. The objective is to evaluate the effects of temporal, spectral, and spatial capacities of multispectral satellite data for AGB prediction. The study was performed on multispectral data acquired by Sentinel-2, RapidEye, and Dove satellites which are characterized by different spatial resolutions, temporal availability, and number of spectral bands. A systematic process of least absolute shrinkage and selection operator (lasso) variable selection generalized linear modeling, leave-one-out cross-validation, and analysis was accomplished on each satellite dataset for AGB prediction. Results point out that the multitemporal data based AGB models were more effective in prediction than the single-time models. In addition, red-edge and short wave infrared (SWIR) channel dependent variables showed significant improvement in the modeling results and contributed to more than 50% of the selected variables. Results also suggest that high spatial resolution plays a smaller role than spectral and temporal information in the prediction of AGB. The overall analysis emphasizes a good potential of spaceborne multispectral data for developing sophisticated methods for AGB prediction especially with specific spectral channels and temporal information.

A new era of spaceborne hyperspectral imaging has just begun with the recent availability of data from PRISMA (PRecursore IperSpettrale della Missione Applicativa) launched by the Italian space agency (ASI). There has been pre-launch optimism that the wealth of spectral information offered by PRISMA can contribute to a variety of aquatic science and management applications. Here, we examine the potential of PRISMA level 2D images in retrieving standard water quality parameters, including total suspended matter (TSM), chlorophyll-a (Chl-a), and colored dissolved organic matter (CDOM) in a turbid lake (Lake Trasimeno, Italy). We perform consistency analyses among the aquatic products (remote sensing reflectance (Rrs) and constituents) derived from PRISMA and those from Sentinel-2. The consistency analyses are expanded to synthesized Sentinel-2 data as well. By spectral downsampling of the PRISMA images, we better isolate the impact of spectral resolution in retrieving the constituents. The retrieval of constituents from both PRISMA and Sentinel-2 images is built upon inverting the radiative transfer model implemented in the Water Color Simulator (WASI) processor. The inversion involves a parameter (gdd) to compensate for atmospheric and sun-glint artifacts. A strong agreement is indicated for the cross-sensor comparison of Rrs products at different wavelengths (average R ≈ 0.87). However, the Rrs of PRISMA at shorter wavelengths (gdd through the inversion that suggests an underestimated atmospheric path radiance of PRISMA level 2D products compared to the atmospherically corrected Sentinel-2 data. The results indicate the high potential of PRISMA level 2D imagery in mapping water quality parameters in Lake Trasimeno. The PRISMA-based retrievals agree well with those of Sentinel-2, particularly for TSM.

Automatic ship detection in complicated maritime background is a challenging task in the field of optical remote sensing image interpretation and analysis. In this paper, we propose a novel and reliable ship detection framework based on a visual saliency model, which can efficiently detect multiple targets of different scales in complex scenes with sea clutter, clouds, wake and islands interferences. Firstly, we present a reliable background prior extraction method adaptive for the random locations of targets by computing boundary probability and then generate a saliency map based on the background prior. Secondly, we compute the prior probability of salient foreground regions and propose a weighting function to constrain false foreground clutter, gaining the foreground-based prediction map. Thirdly, we integrate the two prediction maps and improve the details of the integrated map by a guided filter function and a wake adjustment function, obtaining the fine selection of candidate regions. Afterwards, a classification is further performed to reduce false alarms and produce the final ship detection results. Qualitative and quantitative evaluations on two public available datasets demonstrate the robustness and efficiency of the proposed method against four advanced baseline methods.

The recent PlanetScope constellation (130+ satellites currently in orbit) has shifted the high spatial resolution imaging into a new era by capturing the Earth’s landmass including inland waters on a daily basis. However, studies on the aquatic-oriented applications of PlanetScope imagery are very sparse, and extensive research is still required to unlock the potentials of this new source of data. As a first fully physics-based investigation, we aim to assess the feasibility of retrieving bathymetric and water quality information from the PlanetScope imagery. The analyses are performed based on Water Color Simulator (WASI) processor in the context of a multitemporal analysis. The WASI-based radiative transfer inversion is adapted to process the PlanetScope imagery dealing with the low spectral resolution and atmospheric artifacts. The bathymetry and total suspended matter (TSM) are mapped in the relatively complex environment of Venice lagoon during two benchmark events: The coronavirus disease 2019 (COVID-19) lockdown and an extreme flood occurred in November 2019. The retrievals of TSM imply a remarkable reduction of the turbidity during the lockdown, due to the COVID-19 pandemic and capture the high values of TSM during the flood condition. The results suggest that sizable atmospheric and sun-glint artifacts should be mitigated through the physics-based inversion using the surface reflectance products of PlanetScope imagery. The physics-based inversion demonstrated high potentials in retrieving both bathymetry and TSM using the PlanetScope imagery.

The semantic segmentation of remote sensing images (RSIs) is important in a variety of applications. Conventional encoder-decoder-based convolutional neural networks (CNNs) use cascade pooling operations to aggregate the semantic information, which results in a loss of localization accuracy and in the preservation of spatial details. To overcome these limitations, we introduce the use of the high-resolution network (HRNet) to produce high-resolution features without the decoding stage. Moreover, we enhance the low-to-high features extracted from different branches separately to strengthen the embedding of scale-related contextual information. The low-resolution features contain more semantic information and have a small spatial size; thus, they are utilized to model the long-term spatial correlations. The high-resolution branches are enhanced by introducing an adaptive spatial pooling (ASP) module to aggregate more local contexts. By combining these context aggregation designs across different levels, the resulting architecture is capable of exploiting spatial context at both global and local levels. The experimental results obtained on two RSI datasets show that our approach significantly improves the accuracy with respect to the commonly used CNNs and achieves state-of-the-art performance.

Presents corrections to the above named paper.

Interests in synthetic aperture radar (SAR) data analysis is driven by the constantly increased spatial resolutions of the acquired images, where the geometries of scene objects can be better defined than in lower resolution data. This paper addresses the problem of the built-up areas extraction in high-resolution (HR) SAR images, which can provide a wealth of information to characterize urban environments. Strong backscattering behavior is one of the distinct characteristics of built-up areas in a SAR image. However, in practical applications, only a small portion of pixels characterizing the built-up areas appears bright. Thus, specific texture measures should be considered for identifying these areas. This paper presents a novel texture measure by combining the proposed labeled co-occurrence matrix technique with the specific spatial variability structure of the considered land-cover type in the fuzzy set theory. The spatial variability is analyzed by means of variogram, which reflects the spatial correlation or non-similarity associated with a particular terrain surface. The derived parameters from the variograms are used to establish fuzzy functions to characterize the built-up class and non built-up class, separately. The proposed technique was tested on TerraSAR-X images acquired of Nanjing (China) and Barcelona (Spain), and on a COSMO-SkyMed image acquired of Hangzhou (China). The obtained classification accuracies point out the effectiveness of the proposed technique in identifying and detecting built-up areas.

This work is aimed at showing the present capabilities and future potentialities of an imaging radar system that can be mounted onboard flexible aerial platforms, such as helicopters or small airplanes, and may operate in the UHF and VHF frequency bands as Sounder and as Synthetic Aperture Radar (SAR). More specifically, the Sounder operates at 165 MHz, whereas the SAR may operate either at 450 MHz or at 860 MHz. In the work, we present the first results relevant to a set of Sounder and SAR data collected by the radar during a helicopter-borne campaign conducted in 2018 over a desert area in Erfoud, Morocco, just after the conclusion of a system upgrading procedure. In particular, a first analysis of the focusing capabilities of the Sounder mode and of the polarimetric and interferometric capabilities of the SAR mode is conducted. The overall system, originally developed by CO.RI.S.T.A. according to a ASI funding set up in 2010, has been upgraded in the frame of a contract signed in 2015 between ASI and different private and public Italian Research Institutes and Universities, namely CO.RI.S.T.A., IREA-CNR, Politecnico di Milano and University of Trento.

This paper presents a new concept to derive the snow water equivalent (SWE) based on the joint use of snow model (AMUNDSEN) simulation, ground data, and auxiliary products derived from remote sensing. The main objective is to characterize the spatial-temporal distribution of the model-derived SWE deviation with respect to the real SWE values derived from ground measurements. This deviation is due to the intrinsic uncertainty of any theoretical model, related to the approximations in the analytical formulation. The method, based on the k-NN algorithm, computes the deviation for some labeled samples, i.e., samples for which ground measurements are available, in order to characterize and model the deviations associated to unlabeled samples (no ground measurements available), by assuming that the deviations of samples vary depending on the location within the feature space. Obtained results indicate an improved performance with respect to AMUNDSEN model, by decreasing the RMSE and the MAE with ground data, on average, from 154 to 75 mm and from 99 to 45 mm, respectively. Furthermore, the slope of regression line between estimated SWE and ground reference samples reaches 0.9 from 0.6 of AMUNDSEN simulations, by reducing the data spread and the number of outliers.

The Copernicus Sentinel-1 mission provides synthetic aperture radar (SAR) acquisitions over large areas with high temporal and spatial resolution. This new generation of satellites providing open-data products has enhanced the capabilities for continuously studying Earth surface changes. Over the past two decades, several studies have demonstrated the potential of differential synthetic aperture radar interferometry (DInSAR) for detecting and quantifying land surface deformation. DInSAR limitations and challenges are linked to the SAR properties and the field conditions (especially in mountainous environments) leading to spatial and temporal decorrelation of the SAR signal. High temporal decorrelation can be caused by changes in vegetation (particularly in nonurban areas), atmospheric conditions, or high ground surface velocity. In this study, the kinematics of the complex and vegetated Corvara landslide, situated in Val Badia (South Tyrol, Italy), are monitored by a network of three permanent and 13 monthly measured benchmark points measured with the differential global navigation satellite system (DGNSS) technique. The slope displacement rates are found to be highly unsteady and reach several meters a year. This paper focuses firstly on evaluating the performance of DInSAR changing unwrapping and coherence parameters with Sentinel-1 imagery, and secondly, on applying DInSAR with DGNSS measurements to monitor an active and complex landslide. To this end, 41 particular SAR images, coherence thresholds, and 2D and 3D unwrapping processes give various results in terms of reliability and accuracy, supporting the understanding of the landslide velocity field. Evolutions of phase changes are analysed according to the coherence, the changing field conditions, and the monitored ground-based displacements.

This paper presents an analysis of the integration between interferometric and intensity-offset tracking-based SAR remote sensing for landslide hazard mitigation in the Italian Alps. Despite the advantages of Synthetic Aperture Radar Interferometry (InSAR) methods for quantifying landslide deformation, some limitations remain. The temporal decorrelation, the 1-D Line Of Sight (LOS) observation restriction, the high velocity rate and the multi-directional movement properties make it difficult to monitor accurately complex landslides in areas covered by vegetation. Therefore, complementary and integrated approaches, such as offset tracking-based techniques, are needed to overcome these InSAR limitations for monitoring ground surface deformations. As sub-pixel offset tracking is highly sensitive to data spatial resolution, the latest generations of SAR sensors, such as TerraSAR-X and COSMO-SkyMed, open interesting perspective for a more accurate hazard assessment. In this paper, we consider high-resolution X-band data acquired by the COSMO-SkyMed (CSK) constellation for Permanent Scatterers Interferometry (PSI), Multi-Aperture Interferometry (MAI) and offset tracking processing. We analyze the offset tracking techniques considering area and feature-based matching algorithms to evaluate their applicability to CSK data by improving sub-pixel offset estimations. To this end, PSI and MAI are used for extracting LOS and azimuthal displacement components. Then, four well-known area-based and five feature-based matching algorithms (taken from computer vision) are applied to 16 X-band corner reflectors. Results show that offset estimation accuracy can be considerably improved up to less than 3% of the pixel size using the combination of the different feature-based detectors and descriptors. A sensitivity analysis of these techniques applied to CSK data to monitor complex landslides in the Italian Alps provides indications on advantages and disadvantages of each of them.