Conference Agenda

Unplanned urban expansion poses significant challenges for rapidly growing regions in the Global South, where weak planning systems and fragmented governance structures affects the sustainable city development. This paper demonstrates how digital data infrastructures can shift urban management from reactive regulation to proactive territorial governance, using the City Regional Development Observatory (CRDO) in Gujarat, as an empirical case. The CRDO integrates multi-source geospatial intelligence, such as high-resolution satellite imagery, land use data, infrastructure networks, and urban growth indicators, into a single and unified spatial decision platform.

Unlike the traditional land-use planning, the observatory enables continuous surveillance of peripheral development, early identification of the areas where land is being developed illegally, and clear understanding of how peri-urban areas are changing over time. The observatory supports institutional coordination between Urban Local Bodies, Development authorities, and various state agencies, providing evidence-based decision making, infrastructure alignment and environmental risk mitigation. The study reveals that data-driven planning not only improves spatial transparency, but also reconfigures institutional power, reducing discretionary decision making and enhance the regulatory accountability. However, it also faces challenges such as standardised data, limited technical skills, and resistance to data sharing. This paper argue that digital observatories represent a paradigm shift in governing city-regional expansion in Global South, embedding predictive spatial intelligence into planning systems and offering a replicable model for managing urban transitions.

Urban vegetation plays a critical role in maintaining sustainable, low-carbon cities by regulating microclimates, supporting biodiversity, and providing essential ecosystem services. However, accurately mapping urban vegetation remains challenging due to spatial heterogeneity, mixed land use, and seasonal dynamics. This study introduces a novel approach that combines high-resolution (3 m) Planet Scope imagery with monthly time series analysis to enhance urban vegetation classification, including the differentiation of dominant tree species—a task rarely studied in urban remote sensing. The research was conducted in Modi’in, Israel, using Planet Scope imagery acquired between 2021 and 2023. We generated monthly composites with cloud cover below 5%, grouping the data into four analytical categories: (1) yearly composite, (2) seasonal composites, (3) individual seasons, and (4) individual months. For classification, we combined spectral bands with four vegetation indices (Normalized difference in vegetation index (NDVI), Enhanced vegetation index (EVI), Green normalized difference in vegetation index (GNDVI), Modified soil adjusted vegetation index (MSAVI) and Near infrared vegetation index (NIRv) to capture phenological patterns across the annual cycle. A Random Forest classifier was applied at the pixel level for both major vegetation types and tree species. Our results demonstrate that time series analysis at high spatial resolution significantly improves urban vegetation mapping. The yearly composite achieved the highest accuracy (Overall Accuracy = 94.33%, Kappa = 0.92), outperforming seasonal composites (OA = 93.49%, Kappa = 0.91) and individual months. Summer yielded the best seasonal result (OA = 91.41%, Kappa = 0.88), while monthly accuracies varied, with July performing well (OA = 91.21%, Kappa = 0.88) and January the lowest (OA = 75.88%). Tree species classification achieved consistent accuracy across both yearly and seasonal composites, with F-scores ranging from 0.67–0.90, despite the challenges of structural complexity and spectral overlap in urban environments. These findings highlight the value of integrating high spatial and temporal resolution imagery with phenological analysis for urban ecosystem monitoring. By capturing subtle seasonal variations, this method supports improved biodiversity monitoring, carbon assessment, green infrastructure planning, disaster risk reduction, and urban climate adaptation. The approach demonstrates the potential of remote sensing time series for advancing fine-scale, species-level vegetation mapping in complex urban landscapes.

Rapid urbanization and climate change are intensifying urban heat islands (UHI), with highly variable impacts across land-use types. Conventional heat studies often rely on land surface temperature (LST) or land use/land cover (LULC) data to extrapolate limited air temperature measurements. However, LST alone does not effectively capture the impact of land use on air temperature and LULC lacks sufficient granularity. There is a research in the existing models which provide static snapshots rather than actionable simulations to guide urban heat mitigation. This study presents a hybrid modeling framework that integrates Graph Neural Networks (GNNs) and machine learning methods, particularly Random Forests, to predict spatial–temporal variations in urban air temperature. In Ahmedabad, traverse-based heat mapping was conducted across 17 routes representing 13 land-use typologies, selected through an Analytic Hierarchy Process (AHP). GNNs trained on LST and LULC achieved high predictive accuracy (R² = 0.94). A Random Forest model, using predictors such as NDVI, NDBI, LST, and local vegetation density, further quantified the impacts of land-use change scenarios (R²= 0.64). Simulations showed that converting 183 km² of barren and sparse vegetation to dense vegetation reduced average air temperature by 0.16 °C, whereas expanding impervious surfaces increased it by 1.63 °C. The results reveal that applying GNN on a combination of LST and LULC generates a more realistic Air Temperature layer that can be used to aid policymaking. Secondly, from the Random Forest prediction model, it becomes clear that regulating the growth of impervious areas within in a city may have a larger impact on than solely growing trees. A combination of these two models, therefore, provides information about heat hotspots in a more effective way for urban policy.

Cities are at the forntiers of convergence research, facing the dual challenge of supporting urban wellbing of residents while reducing carbon emissions. This study aims to identify urban fabrics suitable for implementing net-zero carbon city planning strategies. Three fabric types are included—walking-oriented, transit-oriented, and car-dependent—defined according to the theory of urban fabrics. A reproducible geospatial workflow has been conceptualized, leveraging the best available open data sources, including functional Points of Interest (POIs), building footprints, and road networks. The methodology has been piloted in the case study city of Dresden, with potential applicability to global urban contexts. To classify land-use activities, a natural language processing (NLP) algorithm (e.g., Doc2Vec) is used for getting POIs vector embedding which helps to cluster the semantic patterns. Each cluster is annotated later by urban fabric type which receives scores at the grid-cell level by analyzing thematic relationships among POIs, building density characteristics, and road network configuration. Sensitivity analysis of grid sizes—200, 500, and 1000 m²—was conducted to investigate sensitivity of spatial resolution. The resulting dataset can be combined with high-quality local urban data to generate actionable spatial insights and identify intervention hotspots. These support context-specific planning tools, such as walkability improvements, public transport enhancement, and localized deployment of public e-charging infrastructure. By blending urban structure, land use, and infrastructure, this GeoNLP-based approach enables place-based comparisons and assessments that contribute to climate action—toward more livable, low-carbon cities.