Editorial: Building resilient cities with remote sensing and AI

Cities are interconnected systems in which physical landscapes, functional activities, environmental processes, and governance structures interact. Building resilience in a warming and rapidly urbanizing world therefore requires understanding these dimensions together, rather than treating the city as a single mapped surface. Remote sensing has traditionally been strongest at the physical layer, while multi-source data and artificial intelligence (AI) are extending urban observation toward function, process, and decision-making. This Research Topic suggests a coherent value chain - from observation and representation, through understanding and diagnosis, to intervention - toward the urban intelligence needed for resilient cities (Figure 1). Their broader message is that the contribution of remote sensing and AI to urban resilience lies not simply in producing better maps, but in connecting urban structures and functions with environmental processes and action.

That progression begins with a foundational question of how well different observing systems represent heterogeneous urban surfaces. A cross-sensor comparison of Landsat, Sentinel-2, and PRISMA imagery over the same urban setting finds that no single system captures that heterogeneity uniformly. Hyperspectral information improves the separation of some transitional urban-density classes, but these gains are class-specific and remain conditioned by spatial resolution, acquisition timing, preprocessing, and classifier behavior. Sensor selection is therefore not a search for a universally superior platform, but a choice among complementary observational strengths. As shown in Figure 1, this contribution primarily illustrates L1 (multi-source sensing) and the observation-to-harmonization connection with L2.

Yet spectral, spatial, and temporal information alone describes what a city looks like more readily than what it does. A second contribution integrates high-resolution imagery with nighttime lights, building information, points of interest, and areas of interest to identify urban functional zones across Beijing, Chengdu, and Shanghai. It links data fusion (L2) and mapping (L3) to the functional-system node of urban-system intelligence (L4) (Figure 1) (Wang et al.). Overall accuracies above 90% demonstrate the value of combining physical and socioeconomic representations, but cross-city transfer performance declines markedly, indicating how strongly functional signatures remain tied to local urban structure. This pattern is consistent with wider work linking remote sensing to social sensing and semantic information for urban-function mapping (Zhang et al., 2024). Together, these two contributions trace a path from spectral representation to multi-source functional intelligence, while also exposing the persistent challenge of transferability.

Representation becomes more useful for resilience when it helps explain environmental processes associated with urban form and function. Seasonal analysis of land surface temperature across four functional zones on Xiamen Island showed that urban heat is not governed by a single city-wide relationship, an example of the environmental-system node of urban-system intelligence in Figure 1 L4 (Xu). The relative influence of geographic position, building density, population density, floor-area ratio, and landscape configuration varies by land use and season. This context dependence echoes broader evidence that the cooling effects of green infrastructure, and the drivers of urban heat more generally, shift with morphology, season, and place (Qasim et al., 2026; Hong et al., 2023). Resilience planning therefore requires spatially differentiated and temporally adaptive analysis rather than universal endorsements.

That analysis only matters if it changes what gets built or governed, which is where the Research Topic turns from analysis to intervention. A quasi-natural-experiment evaluation of China’s AI innovation-pilot-zone policy finds that AI-oriented policy can contribute to lower urban carbon emissions through pathways including green technological innovation and environmental regulation, with effects conditioned by enabling factors such as digital infrastructure and financial development (Wan and Li). These findings link the governance-system node of L4 to the low-carbon-development outcome in L5 (Figure 1). It also broadens the role of AI beyond an analytical technique by showing that technological capability can contribute to resilience-related outcomes when embedded in institutions, infrastructure, and policy.

Three lessons emerge from reading the Research Topic together. First, urban resilience requires multiple representations: no single sensor, dataset, or model captures a city’s physical, functional, environmental, and socioeconomic complexity at once. Second, context and transferability matter at every stage: sensor performance, functional-zone mapping performance, thermal drivers, and policy effectiveness vary with place, season, urban form, and institutional capacity, and models developed in one setting may not transfer cleanly to another. Third, resilience depends on connecting analytical levels. Observation alone is insufficient; value is created when representation supports understanding, understanding enables diagnosis, and diagnosis informs intervention.

These lessons point to where the field goes next. The gaps the Research Topic Collection originally set out to address (multi-sensor integration, model reliability, transferability, temporally dynamic monitoring, and usability for planners) remain open, but the tools available to address them have advanced since the call was written. Therefore, the conceptual framework linking multi-source urban sensing and data combines the demonstrated and prospective elements of the Research Topic Collection. For instance, none of the four contributions implements a geospatial foundation model or a digital twin. Foundation models and digital twins are included in L3 (Figure 1) to indicate forward-looking opportunities from the wider literature rather than achievements of the Research Topic. Recent work on geospatial foundation models is shifting attention from bespoke, single-city classifiers toward reusable representations learned from large and increasingly multimodal geospatial archives (Janowicz et al., 2025). In parallel, urban digital twins offer environments in which heterogeneous observations, models, simulation, visualization, and feedback can be organized around planning and operational decisions (Theodorakopoulos and Theodoropoulou, 2026). These developments could help bridge the stages represented in Figure 1, but greater technical capability does not remove the need for interpretability, uncertainty communication, and institutional accountability. Explainable GeoAI remains particularly important as model outputs move closer to decisions with material consequences (Papadakis et al., 2022). The next generation of urban GeoAI should therefore be assessed not only by classification accuracy or predictive performance, but also by whether it generalizes across cities, represents uncertainty, integrates physical and social dimensions, and delivers information at the scales and times at which urban decisions are made. The transition from observing cities to building usable urban intelligence is where remote sensing and AI can make their most consequential contribution to resilient urban futures.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited and reviewed by: Robert J. Frouin, University of California, San Diego, United States