High-resolution satellite images can provide abundant, detailed spatial information for land cover classification, which is particularly important for studying the complicated built environment. However, due to the complex land cover patterns, the costly training sample collections, and the severe distribution shifts of satellite imageries caused by, e.g., geographical differences or acquisition conditions, few studies have applied high-resolution images to land cover mapping in detailed categories at large scale. To fill this gap, we present a large-scale land cover dataset, . It contains more than labeled pixels of 150 high-resolution Gaofen-2 (4 m) satellite images, annotated in a 24-category system covering , , and classes. In addition, we propose a deep-learning-based unsupervised domain adaptation approach that can transfer classification models trained on labeled dataset (referred to as the ) to unlabeled data (referred to as the ) for large-scale land cover mapping. Specifically, we introduce an end-to-end Siamese network employing dynamic pseudo-label assignment and class balancing strategy to perform adaptive domain joint learning. To validate the generalizability of our dataset and the proposed approach across different sensors and different geographical regions, we carry out land cover mapping on five megacities in China and six cities in other five Asian countries severally using: PlanetScope (3 m), Gaofen-1 (8 m), and Sentinel-2 (10 m) satellite images. Over a total study area of 60,000 km, the experiments show promising results even though the input images are entirely unlabeled. The proposed approach, trained with the dataset, enables high-quality and detailed land cover mapping across the whole country of China and some other Asian countries at meter-resolution.

Remote sensing (RS) image scene classification has obtained increasing attention for its broad application prospects. Conventional fully-supervised approaches usually require a large amount of manually-labeled data. As more and more RS images becoming available, how to make full use of these unlabeled data is becoming an urgent topic. Semi-supervised learning, which uses a few labeled data to guide the self-training of numerous unlabeled data, is an intuitive strategy. However, it is hard to apply it to cross-dataset (i.e., cross-domain) scene classification due to the significant domain shift among different datasets. To this end, semi-supervised domain adaptation (SSDA), which can reduce the domain shift and further transfer knowledge from a fully-labeled RS scene dataset (source domain) to a limited-labeled RS scene dataset (target domain), would be a feasible solution. In this paper, we propose an SSDA method termed bidirectional sample-class alignment (BSCA) for RS cross-domain scene classification. BSCA consists of two alignment strategies, unsupervised alignment (UA) and supervised alignment (SA), both of which can contribute to decreasing domain shift. UA concentrates on reducing the distance of maximum mean discrepancy across domains, with no demand for class labels. In contrast, SA aims to achieve the distribution alignment both from source samples to the associate target class centers and from target samples to the associate source class centers, with awareness of their classes. To validate the effectiveness of the proposed method, extensive ablation, comparison, and visualization experiments are conducted on an RS-SSDA benchmark built upon four widely-used RS scene classification datasets. Experimental results indicate that in comparison with some state-of-the-art methods, our BSCA achieves the superior cross-domain classification performance with compact feature representation and low-entropy classification boundary. Our code will be available at https://github.com/hw2hwei/BSCA.

Spaceborne imaging spectroscopy is a highly promising data source for all agricultural management and research disciplines that require spatio-temporal information on crop properties. Recently launched science-driven missions, such as the Environmental Mapping and Analysis Program (EnMAP), deliver unprecedented data from the Earth's surface. This new kind of data should be explored to develop robust retrieval schemes for deriving crucial variables from future routine missions. Therefore, we present a workflow for inferring crop carbon content ( ), and aboveground dry and wet biomass ( , ) from EnMAP data. To achieve this, a hybrid workflow was generated, combining radiative transfer modeling (RTM) with machine learning regression algorithms. The key concept involves: (1) coupling the RTMs PROSPECT-PRO and 4SAIL for simulation of a wide range of vegetation states, (2) using dimensionality reduction to deal with collinearity, (3) applying a semi-supervised active learning technique against a 4-years campaign dataset, followed by (4) training of a Gaussian process regression (GPR) machine learning algorithm and (5) validation with an independent dataset acquired during the ESA Hypersense experiment campaign at a German test site. Internal validation of the GPR- and GPR-AGB models achieved coefficients of determination ( ) of 0.80 for and 0.80, 0.71 for and , respectively. The mapping capability of these models was successfully demonstrated using airborne AVIRIS-NG hyperspectral imagery, which was spectrally resampled to EnMAP spectral properties. Plausible estimates were achieved over both bare and green fields after adding bare soil spectra to the training data. Validation over green winter wheat fields generated reliable estimates as suggested by low associated model uncertainties (< 40%). These results suggest a high degree of model reliability for cultivated areas during active growth phases at canopy closure. Overall, our proposed carbon and biomass models based on EnMAP spectral sampling demonstrate a promising path toward the inference of these crucial variables over cultivated areas from future spaceborne operational hyperspectral missions.

The recently launched and upcoming hyperspectral satellite missions, featuring contiguous visible-to-shortwave infrared spectral information, are opening unprecedented opportunities for the retrieval of a broad set of vegetation traits with enhanced accuracy through novel retrieval schemes. In this framework, we exploited hyperspectral data cubes collected by the new-generation PRecursore IperSpettrale della Missione Applicativa (PRISMA) satellite of the Italian Space Agency to develop and test a hybrid retrieval workflow for crop trait mapping. Crop traits were mapped over an agricultural area in north-east Italy (Jolanda di Savoia, FE) using PRISMA images collected during the 2020 and 2021 vegetative seasons. Leaf chlorophyll content, leaf nitrogen content, leaf water content and the corresponding canopy level traits scaled through leaf area index were estimated using a hybrid retrieval scheme based on PROSAIL-PRO radiative transfer simulations coupled with a Gaussian processes regression algorithm. Active learning algorithms were used to optimise the initial set of simulated data by extracting only the most informative samples. The accuracy of the proposed retrieval scheme was evaluated against a broad ground dataset collected in 2020 in correspondence of three PRISMA overpasses. The results obtained were positive for all the investigated variables. At the leaf level, the highest accuracy was obtained for leaf nitrogen content (LNC: r=0.87, nRMSE=7.5%), while slightly worse results were achieved for leaf chlorophyll content (LCC: r=0.67, nRMSE=11.7%) and leaf water content (LWC: r=0.63, nRMSE=17.1%). At the canopy level, a significantly higher accuracy was observed for nitrogen content (CNC: r=0.92, nRMSE=5.5%) and chlorophyll content (CCC: r=0.82, nRMSE=10.2%), whereas comparable results were obtained for water content (CWC: r=0.61, nRMSE=16%). The developed models were additionally tested against an independent dataset collected in 2021 to evaluate their robustness and exportability. The results obtained (i. LCC: r=0.62, nRMSE=27.9%; LNC: r=0.35, nRMSE=28.4%; LWC: r=0.74, nRMSE=20.4%; LAI: r=0.84, nRMSE=14.5%; CCC: r=0.79, nRMSE=18.5%; CNC: r=0.62, nRMSE=23.7%; CWC: r=0.92, nRMSE=16.6%) evidence the transferability of the hybrid approach optimised through active learning for most of the investigated traits. The developed models were then used to map the spatial and temporal variability of the crop traits from the PRISMA images. The high accuracy and consistency of the results demonstrates the potential of spaceborne imaging spectroscopy for crop monitoring, paving the path towards routine retrievals of multiple crop traits over large areas that could drive more effective and sustainable agricultural practices worldwide.

Satellite imaging spectroscopy for terrestrial applications is reaching maturity with recently launched and upcoming science-driven missions, e.g. PRecursore IperSpettrale della Missione Applicativa (PRISMA) and Environmental Mapping and Analysis Program (EnMAP), respectively. Moreover, the high-priority mission candidate Copernicus Hyperspectral Imaging Mission for the Environment (CHIME) is expected to globally provide routine hyperspectral observations to support new and enhanced services for, among others, sustainable agricultural and biodiversity management. Thanks to the provision of contiguous visible-to-shortwave infrared spectral data, hyperspectral missions open enhanced opportunities for the development of new-generation retrieval models of multiple vegetation traits. Among these, canopy nitrogen content (CNC) is one of the most promising variables given its importance for agricultural monitoring applications. This work presents the first hybrid CNC retrieval model for the operational delivery from spaceborne imaging spectroscopy data. To achieve this, physically-based models were combined with machine learning regression algorithms and active learning (AL). The key concepts involve: (1) coupling the radiative transfer models PROSPECT-PRO and SAIL for the generation of a wide range of vegetation states as training data, (2) using dimensionality reduction to deal with collinearity, (3) applying an AL technique in combination with Gaussian process regression (GPR) for fine-tuning the training dataset on in field collected data, and (4) adding non-vegetated spectra to enable the model to deal with spectral heterogeneity in the image. The final CNC model was successfully validated against field data achieving a low root mean square error (RMSE) of 3.4 and coefficient of determination ( ) of 0.7. The model was applied to a PRISMA image acquired over agricultural areas in the North of Munich, Germany. Mapping aboveground CNC yielded reliable estimates over the whole landscape and meaningful associated uncertainties. These promising results demonstrate the feasibility of routinely quantifying CNC from space, such as in an operational context as part of the future CHIME mission.

As remote sensing (RS) data obtained from different sensors become available largely and openly, multimodal data processing and analysis techniques have been garnering increasing interest in the RS and geoscience community. However, due to the gap between different modalities in terms of imaging sensors, resolutions, and contents, embedding their complementary information into a consistent, compact, accurate, and discriminative representation, to a great extent, remains challenging. To this end, we propose a shared and specific feature learning (S2FL) model. S2FL is capable of decomposing multimodal RS data into modality-shared and modality-specific components, enabling the information blending of multi-modalities more effectively, particularly for heterogeneous data sources. Moreover, to better assess multimodal baselines and the newly-proposed S2FL model, three multimodal RS benchmark datasets, i.e., - hyperspectral and multispectral data, - hyperspectral and synthetic aperture radar (SAR) data, - hyperspectral, SAR, and digital surface model (DSM) data, are released and used for land cover classification. Extensive experiments conducted on the three datasets demonstrate the superiority and advancement of our S2FL model in the task of land cover classification in comparison with previously-proposed state-of-the-art baselines. Furthermore, the baseline codes and datasets used in this paper will be made available freely at https://github.com/danfenghong/ISPRS_S2FL.

Aerial scene recognition is a fundamental visual task and has attracted an increasing research interest in the last few years. Most of current researches mainly deploy efforts to categorize an aerial image into one scene-level label, while in real-world scenarios, there often exist multiple scenes in a single image. Therefore, in this paper, we propose to take a step forward to a more practical and challenging task, namely multi-scene recognition in single images. Moreover, we note that manually yielding annotations for such a task is extraordinarily time- and labor-consuming. To address this, we propose a prototype-based memory network to recognize multiple scenes in a single image by leveraging massive well-annotated single-scene images. The proposed network consists of three key components: 1) a prototype learning module, 2) a prototype-inhabiting external memory, and 3) a multi-head attention-based memory retrieval module. To be more specific, we first learn the prototype representation of each aerial scene from single-scene aerial image datasets and store it in an external memory. Afterwards, a multi-head attention-based memory retrieval module is devised to retrieve scene prototypes relevant to query multi-scene images for final predictions. Notably, only a limited number of annotated multi-scene images are needed in the training phase. To facilitate the progress of aerial scene recognition, we produce a new multi-scene aerial image (MAI) dataset. Experimental results on variant dataset configurations demonstrate the effectiveness of our network. Our dataset and codes are publicly available.

Existing techniques of 3-D reconstruction of buildings from SAR images are mostly based on multibaseline SAR interferometry, such as PSI and SAR tomography (TomoSAR). However, these techniques require tens of images for a reliable reconstruction, which limits the application in various scenarios, such as emergency response. Therefore, alternatives that use a single SAR image and the building footprints from GIS data show their great potential in 3-D reconstruction. The combination of GIS data and SAR images requires a precise registration, which is challenging due to the unknown terrain height, and the difficulty in finding and extracting the correspondence. In this paper, we propose a framework to automatically register GIS building footprints to a SAR image by exploiting the features representing the intersection of ground and visible building facades, specifically the near-range boundaries in the building polygons, and the double bounce lines in the SAR image. Based on those features, the two data sets are registered progressively in multiple resolutions, allowing the algorithm to cope with variations in the local terrain. The proposed framework was tested in Berlin using one TerraSAR-X High Resolution SpotLight image and GIS building footprints of the area. Comparing to the ground truth, the proposed algorithm reduced the average distance error from 5.91 m before the registration to -0.08 m, and the standard deviation from 2.77 m to 1.12 m. Such accuracy, better than half of the typical urban floor height (3 m), is significant for precise building height reconstruction on a large scale. The proposed registration framework has great potential in assisting SAR image interpretation in typical urban areas and building model reconstruction from SAR images.

Physically-based methods in remote sensing provide benefits over statistical approaches in monitoring biophysical characteristics of vegetation. However, physically-based models still demand large computational resources and often require rather detailed informative priors on various aspects of vegetation and atmospheric status. Spectral invariants and photon recollision probability theories provide a solid theoretical framework for developing relatively simple models of forest canopy reflectance. Empirical validation of these theories is, however, scarce. Here we present results of a first empirical validation of a model based on photon recollision probability at the level of individual trees. Multiangular spectra of pine, spruce, and oak tree seedlings (height = 0.38-0.7 m) were measured using a goniometer, and tree hemispherical reflectance was derived from those measurements. We evaluated the agreement between modeled and measured tree reflectance. The model predicted the spectral signatures of the tree seedlings in the wavelength range between 400 and 2300 nm well, with wavelength-specific bias between -0.048 and 0.034 in reflectance units. In relative terms, the model errors were the smallest in the near-infrared (relative RMSE up to 4%, 7%, and 4% for pine, spruce, and oak seedlings, respectively) and the largest in the visible wavelength region (relative RMSE up to 34%, 20%, and 60%). The errors in the visible region could be partly attributed to wavelength-dependent directional scattering properties of the leaves. Including woody parts of tree seedlings in the model improved the results by reducing the relative RMSE by up to 10% depending on species and wavelength. Spectrally invariant model parameters, i.e. total and directional escape probabilities, depended on spherically averaged silhouette to total area ratio (STAR) of the tree seedlings. Overall, the modeled and measured tree reflectance mainly agreed within measurement uncertainties, but the results indicate that the assumption of isotropic scattering by the leaves can result in large errors in the visible wavelength region for some tree species. Our results help increasing the confidence when using photon recollision probability and spectral invariants -based models to interpret satellite images, but they also lead to an improved understanding of the assumptions and limitations of these theories.

This paper addresses the problem of semi-supervised transfer learning with limited cross-modality data in remote sensing. A large amount of multi-modal earth observation images, such as multispectral imagery (MSI) or synthetic aperture radar (SAR) data, are openly available on a global scale, enabling parsing global urban scenes through remote sensing imagery. However, their ability in identifying materials (pixel-wise classification) remains limited, due to the noisy collection environment and poor discriminative information as well as limited number of well-annotated training images. To this end, we propose a novel cross-modal deep-learning framework, called X-ModalNet, with three well-designed modules: self-adversarial module, interactive learning module, and label propagation module, by learning to transfer more discriminative information from a small-scale hyperspectral image (HSI) into the classification task using a large-scale MSI or SAR data. Significantly, X-ModalNet generalizes well, owing to propagating labels on an updatable graph constructed by high-level features on the top of the network, yielding semi-supervised cross-modality learning. We evaluate X-ModalNet on two multi-modal remote sensing datasets (HSI-MSI and HSI-SAR) and achieve a significant improvement in comparison with several state-of-the-art methods.

Retrieval of vegetation properties from satellite and airborne optical data usually takes place after atmospheric correction, yet it is also possible to develop retrieval algorithms directly from top-of-atmosphere (TOA) radiance data. One of the key vegetation variables that can be retrieved from at-sensor TOA radiance data is leaf area index (LAI) if algorithms account for variability in atmosphere. We demonstrate the feasibility of LAI retrieval from Sentinel-2 (S2) TOA radiance data (L1C product) in a hybrid machine learning framework. To achieve this, the coupled leaf-canopy-atmosphere radiative transfer models PROSAIL-6SV were used to simulate a look-up table (LUT) of TOA radiance data and associated input variables. This LUT was then used to train the Bayesian machine learning algorithms Gaussian processes regression (GPR) and variational heteroscedastic GPR (VHGPR). PROSAIL simulations were also used to train GPR and VHGPR models for LAI retrieval from S2 images at bottom-of-atmosphere (BOA) level (L2A product) for comparison purposes. The BOA and TOA LAI products were consistently validated against a field dataset with GPR ( of 0.78) and with VHGPR ( of 0.80) and for both cases a slightly lower RMSE for the TOA LAI product (about 10% reduction). Because of delivering superior accuracies and lower uncertainties, the VHGPR models were further applied for LAI mapping using S2 acquisitions over the agricultural sites Marchfeld (Austria) and Barrax (Spain). The models led to consistent LAI maps at BOA and TOA scale. The LAI maps were also compared against LAI maps as generated by the SNAP toolbox, which is based on a neural network (NN). Maps were again consistent, however the SNAP NN model tends to overestimate over dense vegetation cover. Overall, this study demonstrated that hybrid LAI retrieval algorithms can be developed from TOA radiance data given a cloud-free sky, thus without the need of atmospheric correction. To the benefit of the community, the development of such hybrid models for the retrieval vegetation properties from BOA or TOA images has been streamlined in the freely downloadable ALG-ARTMO software framework.

Parameter retrieval and model inversion are key problems in remote sensing and Earth observation. Currently, different approximations exist: a direct, yet costly, inversion of radiative transfer models (RTMs); the statistical inversion with data that often results in problems with extrapolation outside the study area; and the most widely adopted hybrid modeling by which statistical models, mostly nonlinear and non-parametric machine learning algorithms, are applied to invert RTM simulations. We will focus on the latter. Among the different existing algorithms, in the last decade kernel based methods, and Gaussian Processes (GPs) in particular, have provided useful and informative solutions to such RTM inversion problems. This is in large part due to the confidence intervals they provide, and their predictive accuracy. However, RTMs are very complex, highly nonlinear, and typically hierarchical models, so that very often a single (shallow) GP model cannot capture complex feature relations for inversion. This motivates the use of deeper hierarchical architectures, while still preserving the desirable properties of GPs. This paper introduces the use of deep Gaussian Processes (DGPs) for bio-geo-physical model inversion. Unlike shallow GP models, DGPs account for complicated (modular, hierarchical) processes, provide an efficient solution that scales well to big datasets, and improve prediction accuracy over their single layer counterpart. In the experimental section, we provide empirical evidence of performance for the estimation of surface temperature and dew point temperature from infrared sounding data, as well as for the prediction of chlorophyll content, inorganic suspended matter, and coloured dissolved matter from multispectral data acquired by the Sentinel-3 OLCI sensor. The presented methodology allows for more expressive forms of GPs in big remote sensing model inversion problems.

Optical remote sensing imagery is at the core of many Earth observation activities. The regular, consistent and global-scale nature of the satellite data is exploited in many applications, such as cropland monitoring, climate change assessment, land-cover and land-use classification, and disaster assessment. However, one main problem severely affects the temporal and spatial availability of surface observations, namely cloud cover. The task of removing clouds from optical images has been subject of studies since decades. The advent of the Big Data era in satellite remote sensing opens new possibilities for tackling the problem using powerful data-driven deep learning methods. In this paper, a deep residual neural network architecture is designed to remove clouds from multispectral Sentinel-2 imagery. SAR-optical data fusion is used to exploit the synergistic properties of the two imaging systems to guide the image reconstruction. Additionally, a novel cloud-adaptive loss is proposed to maximize the retainment of original information. The network is trained and tested on a globally sampled dataset comprising real cloudy and cloud-free images. The proposed setup allows to remove even optically thick clouds by reconstructing an optical representation of the underlying land surface structure.

Human settlement extent (HSE) information is a valuable indicator of world-wide urbanization as well as the resulting human pressure on the natural environment. Therefore, mapping HSE is critical for various environmental issues at local, regional, and even global scales. This paper presents a deep-learning-based framework to automatically map HSE from multi-spectral Sentinel-2 data using regionally available geo-products as training labels. A straightforward, simple, yet effective fully convolutional network-based architecture, Sen2HSE, is implemented as an example for semantic segmentation within the framework. The framework is validated against both manually labelled checking points distributed evenly over the test areas, and the OpenStreetMap building layer. The HSE mapping results were extensively compared to several baseline products in order to thoroughly evaluate the effectiveness of the proposed HSE mapping framework. The HSE mapping power is consistently demonstrated over 10 representative areas across the world. We also present one regional-scale and one country-wide HSE mapping example from our framework to show the potential for upscaling. The results of this study contribute to the generalization of the applicability of CNN-based approaches for large-scale urban mapping to cases where no up-to-date and accurate ground truth is available, as well as the subsequent monitor of global urbanization.

Coastal wetlands, composed of coastal vegetation and non-vegetated tidal flats, play critical roles in biodiversity conservation, food production, and the global economy. Coastal wetlands in China are changing quickly due to land reclamation from sea, aquaculture, industrialization, and urbanization. However, accurate and updated maps of coastal wetlands (including vegetation and tidal flats) in China are unavailable, and the detailed spatial distribution of coastal wetlands are unknown. Here, we developed a new pixel- and phenology-based algorithm to identify and map coastal wetlands in China for 2018 using time series Landsat imagery (2,798 ETM+/OLI images) and the Google Earth Engine (GEE). The resultant map had a very high overall accuracy (98%). There were 7,474.6 km of coastal wetlands in China in 2018, which included 5,379.8 km of tidal flats, 1,856.4 km of deciduous wetlands, and 238.3 km of evergreen wetlands. Jiangsu Province had the largest area of coastal wetlands in China, followed by Shandong, Fujian, and Zhejiang Provinces. Our study demonstrates the high potential of time series Landsat images, pixel- and phenology-based algorithm, and GEE for mapping coastal wetlands at large scales. The resultant coastal wetland maps at 30-m spatial resolution serve as the most current dataset for sustainable management, ecological assessments, and conservation of coastal wetlands in China.

The 2016 National Land Cover Database (NLCD) product suite (available on www.mrlc.gov), includes Landsat-based, 30 m resolution products over the conterminous (CONUS) United States (U.S.) for land cover, urban imperviousness, and tree, shrub, herbaceous and bare ground fractional percentages. The release of NLCD 2016 provides important new information on land change patterns across CONUS from 2001 to 2016. For land cover, seven epochs were concurrently generated for years 2001, 2004, 2006, 2008, 2011, 2013, and 2016. Products reveal that land cover change is significant across most land cover classes and time periods. The land cover product was validated using existing reference data from the legacy NLCD 2011 accuracy assessment, applied to the 2011 epoch of the NLCD 2016 product line. The legacy and new NLCD 2011 overall accuracies were 82% and 83%, respectively, (standard error (SE) was 0.5%), demonstrating a small but significant increase in overall accuracy. Between 2001 and 2016, the CONUS landscape experienced significant change, with almost 8% of the landscape having experienced a land cover change at least once during this period. Nearly 50% of that change involves forest, driven by change agents of harvest, fire, disease and pests that resulted in an overall forest decline, including increasing fragmentation and loss of interior forest. Agricultural change represented 15.9% of the change, with total agricultural spatial extent showing only a slight increase of 4778 km, however there was a substantial decline (7.94%) in pasture/hay during this time, transitioning mostly to cultivated crop. Water and wetland change comprised 15.2% of change and represent highly dynamic land cover classes from epoch to epoch, heavily influenced by precipitation. Grass and shrub change comprise 14.5% of the total change, with most change resulting from fire. Developed change was the most persistent and permanent land change increase adding almost 29,000 km over 15 years (5.6% of total CONUS change), with southern states exhibiting expansion much faster than most of the northern states. Temporal rates of developed change increased in 2001-2006 at twice the rate of 2011-2016, reflecting a slowdown in CONUS economic activity. Future NLCD plans include increasing monitoring frequency, reducing latency time between satellite imaging and product delivery, improving accuracy and expanding the variety of products available in an integrated database.

Green fractional vegetation cover ( ) is an important phenotypic factor in the fields of agriculture, forestry, and ecology. Spatially explicit monitoring of via relative vegetation abundance (RA) algorithms, especially those based on scaled maximum/minimum vegetation index (VI) values, has been widely investigated in remote sensing research. Although many studies have explored the effectiveness of RA algorithms over the past 30 years, a literature review summarizing the corresponding theoretical background, issues, current state-of-the-art techniques, challenges, and prospects has not yet been published. The overall objective of the present study was to accomplish a comprehensive and systematic review of RA algorithms considering these factors based on the scientific papers published from January 1990 to November 2019. This review revealed that the key issues related to RA algorithms is the determination of the appropriate normalized difference vegetation index (NDVI) values of the full vegetation cover and bare soil (denoted hereafter by NDVI and NDVI, respectively). The existing methods used to correct for these issues were investigated, and their advantages and disadvantages are discussed in depth. In literature trends, we found that the number of reported studies in which RA algorithms were used has increased consistently over time, and that most authors tend to utilize the linear NDVI model, rather than other models in the RA algorithm family. We also found that RA algorithms have been utilized to analyze the images with spatial resolutions ranging from the sub-meter to kilometer, most commonly, using images of 30-m spatial resolution. Finally, current challenges and forward-looking insights in remote estimation of using RA algorithms are discussed to guide future research and directions.

Automatic building extraction from optical imagery remains a challenge due to, for example, the complexity of building shapes. Semantic segmentation is an efficient approach for this task. The latest development in deep convolutional neural networks (DCNNs) has made accurate pixel-level classification tasks possible. Yet one central issue remains: the precise delineation of boundaries. Deep architectures generally fail to produce fine-grained segmentation with accurate boundaries due to their progressive down-sampling. Hence, we introduce a generic framework to overcome the issue, integrating the graph convolutional network (GCN) and deep structured feature embedding (DSFE) into an end-to-end workflow. Furthermore, instead of using a classic graph convolutional neural network, we propose a gated graph convolutional network, which enables the refinement of weak and coarse semantic predictions to generate sharp borders and fine-grained pixel-level classification. Taking the semantic segmentation of building footprints as a practical example, we compared different feature embedding architectures and graph neural networks. Our proposed framework with the new GCN architecture outperforms state-of-the-art approaches. Although our main task in this work is building footprint extraction, the proposed method can be generally applied to other binary or multi-label segmentation tasks.

Hyperspectral dimensionality reduction (HDR), an important preprocessing step prior to high-level data analysis, has been garnering growing attention in the remote sensing community. Although a variety of methods, both unsupervised and supervised models, have been proposed for this task, yet the discriminative ability in feature representation still remains limited due to the lack of a powerful tool that effectively exploits the labeled and unlabeled data in the HDR process. A semi-supervised HDR approach, called iterative multitask regression (IMR), is proposed in this paper to address this need. IMR aims at learning a low-dimensional subspace by jointly considering the labeled and unlabeled data, and also bridging the learned subspace with two regression tasks: labels and pseudo-labels initialized by a given classifier. More significantly, IMR dynamically propagates the labels on a learnable graph and progressively refines pseudo-labels, yielding a well-conditioned feedback system. Experiments conducted on three widely-used hyperspectral image datasets demonstrate that the dimension-reduced features learned by the proposed IMR framework with respect to classification or recognition accuracy are superior to those of related state-of-the-art HDR approaches.

The local climate zone (LCZ) scheme was originally proposed to provide an interdisciplinary taxonomy for urban heat island (UHI) studies. In recent years, the scheme has also become a starting point for the development of higher-level products, as the LCZ classes can help provide a generalized understanding of urban structures and land uses. LCZ mapping can therefore theoretically aid in fostering a better understanding of spatio-temporal dynamics of cities on a global scale. However, reliable LCZ maps are not yet available globally. As a first step toward automatic LCZ mapping, this work focuses on LCZ-derived land cover classification, using multi-seasonal Sentinel-2 images. We propose a recurrent residual network (Re-ResNet) architecture that is capable of learning a joint spectral-spatial-temporal feature representation within a unitized framework. To this end, a residual convolutional neural network (ResNet) and a recurrent neural network (RNN) are combined into one end-to-end architecture. The ResNet is able to learn rich spectral-spatial feature representations from single-seasonal imagery, while the RNN can effectively analyze temporal dependencies of multi-seasonal imagery. Cross validations were carried out on a diverse dataset covering seven distinct European cities, and a quantitative analysis of the experimental results revealed that the combined use of the multi-temporal information and Re-ResNet results in an improvement of approximately 7 percent points in overall accuracy. The proposed framework has the potential to produce consistent-quality urban land cover and LCZ maps on a large scale, to support scientific progress in fields such as urban geography and urban climatology.

The relative motion of moving objects is an essential research topic in geographical information science (GIScience), which supports the innovation of geodatabases, spatial indexing, and geospatial services. This analysis is very popular in the domains of urban governance, transportation engineering, logistics and geospatial information services for individuals or industrials. Importantly, data models of moving objects are one of the most crucial approaches to support the analysis for dynamic relative motion between moving objects, even in the age of big data and cloud computing. Traditional geographic information systems (GIS) usually organize moving objects as point objects in absolute coordinated space. The derivation of relative motions among moving objects is not efficient because of the additional geo-computation of transformation between absolute space and relative space. Therefore, current GISs require an innovative approach to directly store, analyze and interpret the relative relationships of moving objects to support their efficient analysis. This paper proposes a relative space-based GIS data model of moving objects (RSMO) to construct, operate and analyze moving objects' relationships and introduces two algorithms (relationship querying and relative relationship dynamic pattern matching) to derive and analyze the dynamic relationships of moving objects. Three scenarios (epidemic spreading, tracker finding, and motion-trend derivation of nearby crowds) are implemented to demonstrate the feasibility of the proposed model. The experimental results indicates the execution times of the proposed model are approximately 5-50% those of the absolute GIS method for the same function of these three scenarios. It's better computational performance of the proposed model when analyzing the relative relationships of moving objects than the absolute methods in a famous commercial GIS software based on this experimental results. The proposed approach fills the gap of traditional GIS and shows promise for relative space-based geo-computation, analysis and service.