Regional disaster resilience quantifies the changing nature of physical risks to inform policy instruments ranging from local immediate recovery to international sustainable development. While many existing state-of-practice methods have greatly advanced the dynamic mapping of exposure and hazard, our understanding of large-scale physical vulnerability has remained static, costly, limited, region-specific, coarse-grained, overly aggregated, and inadequately calibrated. With the significant growth in the availability of time-series satellite imagery and derived products for exposure and hazard, we focus our work on the equally important yet challenging element of the risk equation: physical vulnerability. Given this unique problem, we leverage machine learning methods that flexibly capture spatial contextual relationships, limited temporal observations, and uncertainty in a unified probabilistic spatiotemporal inference framework. We therefore introduce Graph Variational State-Space Model (GraphVSSM), a novel modular spatiotemporal approach that uniquely integrates graph deep learning, state-space modeling, and variational inference using time-series data and prior expert belief systems in a weakly supervised or coarse-to-fine-grained manner. We present three major results: a city-wide demonstration in Quezon City, Philippines; an investigation of sudden changes in the cyclone-impacted coastal Khurushkul community (Bangladesh) and the mudslide-affected Freetown (Sierra Leone); and an open geospatial dataset, METEOR 2.5D, that spatiotemporally enhances the existing global static dataset for 46 UN-recognized Least Developed Countries (as of 2020). Beyond advancing the practice of regional disaster resilience assessment and improving our understanding of global progress in disaster risk reduction, our method also offers a probabilistic deep learning approach, contributing to broader urban studies that require compositional data analysis in weakly supervised settings.

Geolocated text data are a promising data source for spatial analyses in many fields, from disease surveillance to the spatial humanities. This study investigates the relationship between texts’ thematic categories and their likelihood of containing usable geolocation information by quantifying and modelling this relationship across seven diverse English text datasets of different types, including web forums, microblogs, news, and magazines. We find that the likelihood of geoinformation is highly variant, being high for the category ‘Travel, Tourism & Migration’ and low for ‘Private Life, Family & Relationships’. The rank-correlation of this likelihood between datasets is moderate to strong. These findings indicate that the topic plays a significant role in determining the frequency of geospatial references within the text, and that the effect is not entirely dataset-specific. This contributes to the empirical study of the concept of spatiality and provides valuable insights for bias mitigation in the increasing use of text as data for spatial analyses.

The building footprint extraction (BFE) from aerial images is important for the creation and continuous monitoring of building inventories useful for urban planning, among others. Existing methods frequently extract roofs of buildings from aerial images assuming that they overlap with the footprint. This assumption does not hold in the case of off-nadir images. This letter proposes a novel multilabel learning of oblique building features—footprint, roof, and shape—with a Vision Transformer (ViT) for accurate BFE from off-nadir aerial images. A shape calculation algorithm is developed to derive shape polygons from the existing footprint and roof polygons. The method is compared with several convolutional neural networks (CNNs) and ViTs, and a postprocessing algorithm is further devised to achieve regular building footprint polygons. The proposed method outperforms existing scores of BFE on the BONAI dataset (0.727 versus 0.643F1), and our shape calculation algorithm provides labels as accurate as Segment Anything 2 without the need for a GPU. The results conclude that models trained with shapes in addition to the footprint and roof provide consecutively higher scores (F1 score: 0.747 w/ shape versus 0.727 w/o shape versus 0.682 w/ only footprint) and substantially improve the BFE on off-nadir images. The codes and datasets are available at:https://github.com/bipulneupane/Multilabel-BONAI/.

Semantic segmentation of remote sensing imagery plays an important role in applications such as environmental monitoring and disaster response. However, challenges such as complex spatial patterns of variable target objects, significant scale variations, and high inter-class similarity challenge accurate segmentation. Most existing methods based on convolutional neural networks (CNNs) and Transformers face limitations in modeling multi-scale global-local dependencies or often incur high computational costs. Therefore, we propose a multi-scale convolution and mamba fusion model (MF-Mamba) that integrates a CNN encoder with a Mamba-based decoder. The decoder incorporates a Global-Local State Space (GLSS) module with eight-directional selective scanning mechanisms and multi-kernel parallel convolutions to capture the rich global-local context. To enhance multi-scale feature representation, we developed a channel-spatial attention and dense multi-scale feature fusion (CSDF) module, which combines channel-spatial attention and atrous convolutions for multi-scale feature fusion. Additionally, a multi-scale lateral connection is developed to align encoder features for efficient integration. Experiments on the data sets of ISPRS Vaihingen, ISPRS Potsdam, and the Wuhan Dense Labeling Dataset (WHDLD) demonstrate the superior performance of MF-Mamba compared to existing state-of-the-art methods. It achieves Mean F1 scores of 86.71%, 90.70%, and 77.07%, respectively. The code is available at https://github.com/Mango-Mars/MF-Mamba.

Current research and development efforts at DLR`s Center for Satellite Based Crisis Information (ZKI) focus on deploying automated image analysis methods as part of rapid mapping processing routines. The use of machine learning methods enables processing of large amounts of heterogeneous satellite, aerial and drone images at varying spatial scales and temporal frequencies. In this work, we introduce an automated and scalable image processing chain for rapid building damage assessment, optimize it for inference on different hardware and provide application examples from recent natural disasters. We show the scalability of the method from high-frequency live-mapping with drones on a laptop to large-scale processing of satellite and aerial images on a high-performance computing cluster.

Small water bodies and their temporal changes are, especially in urban areas, closely related to the urban climate, people's daily life, among others. Mapping of small water bodies with optical remote sensing images in complex urban landscapes is challenging: that is to establish a balance between reducing incorrect water detection and increasing the integrity of water extraction. In this work we propose a spatial information-integrated small water bodies mapping (SWM) method to achieve a complete and accurate extraction and temporal change monitoring of small water bodies. The spatial contextual information is exploited by the proposed water index roughness feature to compensate for the indistinguishability of small water bodies in spectral information. Results using Landsat and Sentinel-2 data show that the proposed algorithm achieves better water extraction performance, i.e. higher completeness and less incorrect extractions. It proves the ability to observe the changes of surface water.

We present a classification postprocessing (CPP) technique based on fully convolutional neural networks (CNNs) for semantic remote sensing image segmentation. Conventional CPP techniques aim to enhance the classification accuracy by imposing smoothness priors in the image domain. Contrary to that, here, a relearning strategy is proposed where the initial classification outcome of a CNN model is provided to a subsequent CNN model via an extended input space to guide the learning of discriminative feature representations in an end-to-end fashion. This deep relearning CNN (DRCNN) explicitly accounts for the geospatial domain by taking the spatial alignment of preliminary class labels into account. Hereby, we evaluate to learn the DRCNN in a cumulative and noncumulative way, i.e., extending the input space based on all previous or solely preceding model outputs, respectively, during an iterative procedure. Besides, the DRCNN can also be conveniently coupled with alternative CPP techniques such as object-based voting (OBV). The experimental results obtained from two test sites of WorldView-II imagery underline the beneficial performance properties of the DRCNN models. They can increase the accuracies of the initial CNN models on average from 72.64% to 76.01% and from 92.43% to 94.52% in terms of$\kappa $statistic. An additional increase of 1.65 and 2.84 percentage points can be achieved when combining the DRCNN models with an OBV strategy. From an epistemological point of view, our results underline that CNNs can benefit from the consideration of preliminary model outcomes and that conventional CPP techniques can profit from an upstream relearning strategy.

Previous applications of machine learning in remote sensing for the identification of damaged buildings in the aftermath of a large-scale disaster have been successful. However, standard methods do not consider the complexity and costs of compiling a training data set after a large-scale disaster. In this article, we study disaster events in which the intensity can be modeled via numerical simulation and/or instrumentation. For such cases, two fully automatic procedures for the detection of severely damaged buildings are introduced. The fundamental assumption is that samples that are located in areas with low disaster intensity mainly represent nondamaged buildings. Furthermore, areas with moderate to strong disaster intensities likely contain damaged and nondamaged buildings. Under this assumption, a procedure that is based on the automatic selection of training samples for learning and calibrating the standard support vector machine classifier is utilized. The second procedure is based on the use of two regularization parameters to define the support vectors. These frameworks avoid the collection of labeled building samples via field surveys and/or visual inspection of optical images, which requires a significant amount of time. The performance of the proposed method is evaluated via application to three real cases: the 2011 Tohoku-Oki earthquake–tsunami, the 2016 Kumamoto earthquake, and the 2018 Okayama floods. The resulted accuracy ranges between 0.85 and 0.89, and thus, it shows that the result can be used for the rapid allocation of affected buildings.

Agricultural activities cause rapid changes in vegetation development at local and regional scales. Those modifications affect the small-scale behavior of animals, like the foraging ground usage of breeding white storks. Only recently, a novel approach, that enables to quantify the relationship between mowing and harvesting activities and a prolonged foraging time of storks by combining remote sensing time series with GPS telemetry, has been proposed. This study examines the stability of this approach. We investigate two potential influencing factors: different vegetation indices and time lags over which vegetation dynamics were retrieved. Mostly independent from the vegetation index and time lag, we observed that storks spent large proportions of foraging time in areas characterized by a recent drop in vegetation indices, indicative for a preferred usage after harvesting and mowing events. This suggest that the proposed approach is relatively stable and hence, provides a reasonable basis to investigate the effects of anthropogenic vegetation alterations on animal behavior at small spatiotemporal scales.

Polycentricity refers to urban regions with more than one center. These additional (sub-) centers, e.g. spatial concentrations of jobs, are characteristic for the transformation of monocentric towards polycentric urban patterns. Frequently assessed with socioeconomic data, the phenomenon is also reflected in the built morphology of urban landscapes. Only recently, a methodology for large-scale morphological characterization of built-up structures in urban areas relying on TanDEM-X and Sentinel-2 data has been introduced. Thus, a new way to investigate morphologic polycentricity in and among cities is provided. Relying on this approach, we derive the distribution of urban mass concentrations in four city regions. We identify high urban mass concentrations - proxies for (sub-) centers - using a threshold approach. A comparison between the studied regions reveals that only one city tends to have a polycentric urban structure. Our study highlights a new and promising possibility to study the urban morphologic development at global scales.

The TanDEM-X mission (TDM) is a spaceborne radar interferometer which delivers a global digital surface model (DSM) with a spatial resolution of 0.4 arcsec. In this letter, we propose an automatic workflow for digital terrain model (DTM) generation from TDM DSM data through additional consideration of Sentinel-2 imagery and open-source geospatial vector data. The method includes the automatic and robust compilation of training samples by imposing dedicated criteria on the multisource geodata for subsequent learning of a classification model. The model is capable of supporting the accurate distinction of elevated objects (OBJ) and bare earth (BE) measurements in the TDM DSM. Finally, a DTM is interpolated from identified BE measurements. Experimental results obtained from a test site which covers a complex and heterogeneous built environment of Santiago de Chile, Chile, underline the usefulness of the proposed workflow, since it allows for substantially increased accuracies compared to a morphological filter-based method.

We propose a novel cost-sensitive multitask active learning (CSMTAL) approach. Cost-sensitive active learning (CSAL) methods were recently introduced to specifically minimize labeling efforts emerging from ground surveys. Here, we build upon a CSAL method but compile a set of unlabeled samples from a learning set which can be considered relevant with respect to multiple target variables. To this purpose, a multitask meta-protocol based on alternating selection is implemented. It comprises a so-called one-sided selection (i.e., single-task AL selection for a reference target variable with simultaneous labeling of the residual target variables) with a changing leading variable in an iterative selection process. Experimental results are obtained for the city of Cologne, Germany. The target variables to be predicted, using features from remote sensing and a support vector machine framework, are “building type” and “roof type.” Comparative model accuracy evaluations underline the capability of the CSMTAL method to provide beneficial solutions with respect to a random sampling strategy and noncost-sensitive multitask active sampling.

In this letter, we establish two sampling schemes to select training and test sets for supervised classification. We do this in order to investigate whether estimated generalization capabilities of learned models can be positively biased from the use of spatial features. Numerous spatial features impose homogeneity constraints on the image data, whereby a spatially connected set of image elements is attributed identical feature values. In addition to a frequent occurrence of intrinsic spatial autocorrelation, this leads to extrinsic spatial autocorrelation with respect to the image data. The first sampling scheme follows a spatially random partitioning into training and test sets. In contrast to that, the second strategy implements a spatially disjoint partitioning, which considers in particular topological constraints that arise from the deployment of spatial features. Experimental results are obtained from multi- and hyperspectral acquisitions over urban environments. They underline that a large share of the differences between estimated generalization capabilities obtained with the spatially disjoint and non-disjoint sampling strategies can be attributed to the use of spatial features, whereby differences increase with an increasing size of the spatial neighborhood considered for computing a spatial feature. This stresses the necessity of a proper spatial sampling scheme for model evaluation to avoid overoptimistic model assessments.

Morphological operators (MOs) and their enhancements such as morphological profiles (MPs) are subject to a lively scientific contemplation since they are found to be beneficial for, for example, classification of very high spatial resolution panchromatic, multi-, and hyperspectral imagery. They account for spatial structures with differing magnitudes and, thus, provide a comprehensive multilevel description of an image. In this paper, we introduce the concept of object-based MPs (OMPs) to also encode shape-related, topological, and hierarchical properties of image objects in an exhaustive way. Thereby, we seek to benefit from the so-called object-based image analysis framework by partitioning the original image into objects with a segmentation algorithm on multiple scales. The obtained spatial entities (i.e., objects) are used to aggregate multiple sequences obtained with MOs according to statistical measures of central tendency. This strategy is followed to simultaneously preserve and characterize shape properties of objects and enable both the topological and hierarchical decompositions of an image with respect to the progressive application of MOs. Subsequently, supervised classification models are learned by considering this additionally encoded information. Experimental results are obtained with a random forest classifier with heuristically tuned hyperparameters and a wrapper-based feature selection scheme. We evaluated the results for two test sites of panchromatic WorldView-II imagery, which was acquired over an urban environment. In this setting, the proposed OMPs allow for significant improvements with respect to classification accuracy compared to standard MPs (i.e., obtained by paired sequences of erosion, dilation, opening, closing, opening by top-hat, and closing by top-hat operations).

In this letter, we present an object-based postclassification relearning approach for enhanced supervised remote sensing image classification. Conventional postclassification processing techniques aim to enhance the classification accuracy by imposing smoothness priors in the image domain (based on, for example, majority filtering or Markov random fields). In contrast to that, here, a supervised classification model is learned for the second time, with additional information generated from the initial classification outcome to enhance the discriminative properties of relearned decision functions. This idea is followed within an object-based image analysis framework. Therefore, we model spatial-hierarchical context relations with the preliminary classification outcome by computing class-related features using a triplet of hierarchical segmentation levels. Those features are used to enlarge the initial feature space and impose spatial regularization in the relearned model. We evaluate the relevance of the method in the context of classifying of a high-resolution multispectral image, which was acquired over an urban environment. The experimental results show an enhanced classification accuracy using this method compared to both per-pixel-based approach and outcomes obtained with a conventional object-based postclassification processing technique (i.e., object-based voting).

In this letter, a new approach is proposed to classify tsunami-induced building damage into multiple classes using pre- and post-event high-resolution radar (TerraSAR-X) data. Buildings affected by the 2011 Tohoku earthquake and tsunami were the focus in developing this method. In synthetic aperture radar (SAR) data, buildings exhibit high backscattering caused by double-bounce reflection and layover. However, if the buildings are completely washed away or structurally destroyed by the tsunami, then this high backscattering might be reduced, and the post-event SAR data will show a lower sigma nought value than the pre-event SAR data. To exploit these relationships, a rapid method for classifying tsunami-induced building damage into multiple classes was developed by analyzing the statistical relationship between the change ratios in areas with high backscattering and in areas with building damage. The method was developed for the affected city of Sendai, Japan, based on the decision tree application of a machine learning algorithm. The results provided an overall accuracy of 67.4% and a kappa statistic of 0.47. To validate its transferability, the method was applied to the town of Watari, and an overall accuracy of 58.7% and a kappa statistic of 0.38 were obtained.

The TanDEM-X mission (TDM) is a spaceborne radar interferometer which delivers a global digital surface model (DSM) with an unprecedented spatial resolution. This allows resolving objects above ground such as buildings. Extracting and characterizing those objects in an automated manner represents a challenging problem but opens simultaneously a broad range of large-area applications. In this paper, we discuss and evaluate the suitability of morphological filters (MFs) for the derivation of normalized DSMs from the TDM in complex urban environments and introduce a novel region-growing-based progressive MF procedure. This approach is jointly proposed and can be combined with a postclassification processing scheme to specifically allow for a viable reconstruction of urban morphology in a challenging terrain. The filter approach comprises a multistep procedure using concepts of morphological image filtering, region growing, and interpolation techniques. Therefore, it extends the idea of progressive MFs. The latter aim to identify nonground pixels in the DSM by gradually increasing the size of a structuring element and applying iteratively an elevation difference threshold. After the identification of initial nonground pixels, here, potential nonground pixels are identified within each iteration, and their similarity with respect to neighboring nonground pixels is assessed. Pixels are finally labeled as nonground if a constraint is fulfilled. The postclassification processing scheme adapts techniques of object-based image analyses to further refine regions of classified nonground pixels. Digital terrain models are subsequently generated by interpolating between identified ground pixels. Experimental results are obtained for settlement areas that cover large parts of the cities of Izmir (Turkey) and Wuppertal (Germany). They confirm the capability of the proposed approaches for a reduction of omission errors compared to basic MF-based methods when classifying ground pixels, which is favorable in a mountainous terrain with steep slopes.

In this study, a semi-automated method to estimate building damage in a tsunami affected area is developed using pre- and post-event high-resolution synthetic aperture radar (TerraSAR-X) data. For development, some coastal areas affected by the 2011 Tohoku earthquake tsunami were focused. The method for estimating building damage consists of three steps, 1) To detect flooded areas by the tsunami, 2) To detect built-up areas, 3) To estimate building damage inside the flooded built-up areas. The previously proposed methods using high-resolution SAR data needs building footprint data for estimating building damage[1]. However, this problem was improved by developing a new method which does not need building footprint data to estimate building damage caused by the tsunami. The developed method was validated on the other test sites and the estimated results showed good consistency with the ground truth data.