Conference Agenda

Urban security also refers to urban ground subsidence as well as to exposure and vulnerability to urbanized climate impacts. To this end, a two-fold approach has been promoted in the course of Dragon 5:

Urban ground subsidence. Regarding the complex evolution problem of urban ground subsidence in Beijing under the combined influence of climate change and human activities, a machine learning model was developed that combines physical mechanisms and data-driven approaches to reveal the underlying causes and mechanisms of the coupling between groundwater level dynamics and differential ground subsidence evolution in typical areas of Beijing. The model integrates InSAR remote sensing technology with actual measurement datasets from national field scientific observation stations, as well as shallow surface static and dynamic load stress fields, regional groundwater seepage fields, etc.

The model was applied for the:

(1) Simulation and prediction of nonlinear evolution of long-term surface subsidence in the typical area of Beijing Plain. Technical methods such as machine learning, spatial analysis, and statistics were integrated and combined with multi-source data including stratified compressible layer thickness and groundwater levels, to determine the optimal hyperparameter dataset, and construct an extreme random tree-Monte Carlo surface subsidence simulation model for regional land subsidence simulation.

(2) In response to the complex and discontinuous characteristics of the surface subsidence problem, an integrated, multi-scale, variable-resolution surface subsidence modeling method based on near-field dynamics theory is proposed. Taking the complex land subsidence system in Tongzhou District, Beijing as the research area, the multi-body theory of complex systems was applied to create a three-dimensional simplified model of the hydrogeological body and perform a near-field dynamics simulation of the surface subsidence evolution in Tongzhou District from 2021 to 2035.

(3) The impact of natural and human-caused factors on surface subsidence was quantified by combining spatial data mining and machine learning methods. The results indicate that in the Beijing area, the compressible layer thickness contributes the most to the evolution of land subsidence, accounting for over 35%, followed by changes in groundwater levels, with a total contribution rate exceeding 70%.

Urban Climate. A comprehensive set of climate indicators for urban areas was developed, aiming to describe the complex interconnections between climate change and urban environments, with a particular focus on the current and future thermal conditions. Through a suite of climate indicators and high-resolution EO-based products, the framework facilitates the extraction of urban and thermal parameters. The ultimate objective is to transition from generic urban adaptation plans to localized interventions, thereby enhancing the overall quality of life.

(1) New tools were considered to study the relationship between urban form and the state of the urban thermal environment. To this respect, the initial focus is on the association between Land Surface Temperature (LST), multiple climate indices, and the Local Climate Zones (LCZ) in Athens, Greece, and Beijing, China. Trends of the selected indicators are also assessed in terms of their dependence on LCZ classes.

(2) In terms of conducting a local-scale evaluation of the thermal environment, satellite-derived LST with a fine spatial scale and a high temporal frequency are required. To achieve this a statistical downscaling procedure was developed. LST was sharpened based on its statistical relation with surface features, which were used as independent variables (predictors). The provision of such indicators can bridge the gap between climate information and the practical end-users’ needs for better management of heat-related risks and adaptation to climate change.

The pace of urbanization has been unprecedented. Rapid urbanization poses significant social and environmental challenges, including sprawling informal settlements, increased pollution and urban heat island effects, loss of biodiversity and ecosystem services, and increased vulnerability to disasters. Therefore, timely and accurate information on urban change patterns is crucial to support sustainable and resilient urban planning and monitoring of the UN 2030 SDGs. Leveraging Earth observation (EO) big data and AI, this project aims to develop innovative, robust and globally applicable methods, for urban land cover mapping and urbanization monitoring to address.

In recent years, deep learning algorithms have shown promise in urban mapping and change detection, especially when integrating various Earth observation data sources. Despite advancements, challenges persist, including cloud interference and the need for labelled data.

To address these challenges, the EO-AI4Urban team has developed varous deep learning-based methods for urban mapping and change detection. For urban mapping, a novel Domain Adaptation (DA) approach using semi-supervised learning has been developed for urban extraction. The DA approach jointly exploits Sentinel-1 SAR and Sentinel-2 MSI data to improve across-region generalization for built-up area mapping [1]. Furthermore, we developed a multi-modal urban mapping method that utilizes a reconstruction network to approximate the features of the optical modality when only SAR data is available [2].

For urban change detection, several novel methods have been developed including a dual-stream U-Net [3], a Siamese Difference Dual-Task network with Multi-Modal Consistency Regularization [4], a high-resolution feature difference attention network (HDANet) using the Siamese network structure [5]. Another novel procedure was designed to search for built-up changing patterns with the joint use of temporal and spatial properties, using high-frequency SAR time series [6] [7]. To comprehensively capture the scene-level changes between the bi-temporal VHR images, a novel DAN-CPFS framework that integrates differential aggregation network and class probability-based fusion strategy was proposed [8]. The Double U-Net (W-Net) method, a semi-supervised dual-head approach, was introduced to tackle challenges in change detection such as incomplete buildings and blurred edges. By integrating UNet++ and a superpixel module, W-Net improves spatio-spectral characteristics and corrects edge discrepancies, while employing a self-training method to enhance change detection data volume and reduce labelling costs [9]. Finally, to identify similar urban areas quickly and to reduce the cost of manually labeled data, a multisource data reconstruction-based deep unsupervised hashing method was proposed for unisource remote sensing image retrieval, called MrHash, which consists of a label generation network and a deep hashing network [10]. Furthermore, to address the increasing frequency of extreme heat events due to climate change and urbanization, we proposed a heat health risk assessment framework for Beijing, focusing on hazard, exposure, and vulnerability. It utilizes remote sensing data to analyze the spatial pattern of green infrastructure and its impact on heat health risk.

Using ESA Sentinel-1 SAR, Sentinel-2 MSI, the results showed that the proposed DA approach achieves strong improvements upon fully supervised learning and offers great potential to be adapted to produce easily updateable human settlements maps at a global scale [1] [2]. Using the OSCD dataset, the results showed that the dual-stream U-Net outperformed other U-Net-based approaches with feature level fusion of SAR and optical data [3]. Using bi-temporal SAR and MSI image pairs as input, the Siamese Difference Dual-Task network with Multi-Modal Consistency achieved higher F1 score than that of several supervised models when applied to the sites located outside of the source domain [4]. Using several public building change detection datasets, the experimental results showed that the HDANet can achieve a high building change detection accuracy, compared with the current mainstream methods, with public building change detection datasets [5]. Using Hanyang Scene Change Detection (HY-SCD) dataset, the results show that the proposed DAN-CPFS change detection method outperformed some other state-of-the-art methods [8]. Experimenting on two public datasets and the Shanghai JD large-scene dataset, the W-Net exhibits remarkable performance across datasets. Conducting experiments on Sentinel-2 and GF-1 satellite images, the results showed that MrHash yielded the best performance among all methods [10]. The urban heat study confirms that areas with abundant green infrastructure exhibited a low likelihood of becoming high-risk areas, underscoring the importance of expanding green spaces and water bodies to mitigate heat health risks and offering insights for enhancing urban thermal resilience through nature-based climate adaptation.

References

[1] Hafner, S., Ban, Y. and Nascetti, A., 2022. Unsupervised domain adaptation for global urban extraction using Sentinel-1 SAR and Sentinel-2 MSI data. Remote Sensing of Environment, 280, p.113192.

[2] Hafner, S. and Ban, Y., 2023, July. Multi-Modal Deep Learning for Multi-Temporal Urban Mapping with a Partly Missing Optical Modality. In IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium (pp. 6843-6846).

[3] Hafner, S., Nascetti, A., Azizpour, H. and Ban, Y., 2021. Sentinel-1 and Sentinel-2 data fusion for urban change detection using a dual stream U-Net. IEEE Geoscience and Remote Sensing Letters, 19, pp.1-5.

[4] Hafner, S., Ban, Y. and Nascetti, A., 2023. Semi-Supervised Urban Change Detection Using Multi-Modal Sentinel-1 SAR and Sentinel-2 MSI Data. Remote Sensing, 15(21), p.5135.

[5] Wang, X., P. Du, et al., 2022. A high-resolution feature difference attention network for the application of building change detection, International Journal of Applied Earth Observation and Geoinformation, Volume 112, 102950.

[6] M. Che, A. Vizziello and P. Gamba, 2022. Spatio-temporal Urban Change Mapping with Time-Series SAR data, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

[7] Che, M., A. Vizziello, P. Gamba. 2021. Spatio-temporal Change Mapping with Coherence Time-Series. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

[8] H. Fang, et al., " Scene Change Detection by Differential Aggregation Network and Class Probability-Based Fusion Strategy,"IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1-18, 2023, Art no. 5406918,

[9] Tan, K. et al., Dual U-Net (W-Net): A Semi-Supervised Dual-Head Change Detection Network for Urban Macro-Scene Monitoring (under review).

[10] Y. Sun, Y. Ye, et al. 2022 Multisource Data Reconstruction-based Deep Unsupervised Hashing for Unisource Remote Sensing Image Retrieval. IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-16.