Accurate air temperature (T_a)  forecasting in urban areas is crucial for various socio-economic aspects, including risk warning and optimization of electricity systems. However, forecasting within urban environments faces substantial challenges due to the coarse spatial resolution and inadequate urban representation in numerical weather prediction (NWP) models. In this study, we present a novel multimodal deep learning framework that learns local dynamics from ground-level weather stations while effectively informing large-scale weather patterns for short-range (1- 24  hour lead time) T_a forecasting. The framework first employs graph neural networks (GNNs) to model intra-city spatiotemporal dynamics across 35 weather stations in Hong Kong, achieving over 12% forecast improvement compared to modeling individual time series, primarily through mean state regularization. We further develop an end-to-end multimodal framework by integrating the GNN with synoptic weather patterns, achieving an additional 23% improvement, with particular expertise in winter and capturing cold spell events. Our study demonstrates the effectiveness of incorporating multi-scale information from diverse data sources and reveals that weather patterns within approximately 2000 km are critical for local city-scale forecasting. This framework can be readily adapted to other urban areas and will benefit significantly from the increasing deployment of smart IoT sensors to effectively address intra-city temperature heterogeneity.

How to cite: Yang, J. and Wang, H.: Skillful urban air temperature forecast system with multimodal deep learning Framework, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-15, https://doi.org/10.5194/icuc12-15, 2025.

Under the dual challenges of global warming and rapid urbanization, Wuhan faces escalating thermal risks exacerbated by frequent heatwaves and intensified urban heat island effects. The city's distinctive land use land cover configuration with hundreds of lakes in a basin creates a self-reinforcing thermal trap: extensive water networks elevate humidity through evaporation, while vertical urban canyons restrict ventilation and prolong heat and moisture retention. This study develops an innovative hybrid modeling framework to unravel Wuhan's complex heat-moisture dynamics and generate hourly meter-scale heat index mapping. To overcome existing methodological constraints, where conventional physics-based models face expensive computational cost bottlenecks and data-driven methods face limitations in observational data quality, we present a three-phase framework synergizing physical modeling and deep learning. Phase one enhances urban climate simulations through WRF-UCM optimized with 1km three-dimensional urban canopy parameters and updated land use data, achieving significantly improved temperature accuracy across 183 urban weather stations. Phase two employs a hybrid CNN-Transformer architecture that fuses multi-source data streams, including calibrated WRF outputs, IoT sensor networks, and sub-meter remote sensing layers to predict hyper-resolution HI through spatiotemporal feature fusion. Phase three reveals critical diurnal thermal patterns through SHAP-enhanced interpretability analysis, quantifying water bodies' substantial moisture contribution and identifying high-risk zones in compact urban cores with pronounced HI diurnal fluctuations. Our framework demonstrates superior computational efficiency with high spatial accuracy in HI prediction, establishing the first operational hourly meter-scale heat monitoring system for Wuhan. The methodology advances urban climate modeling through physics-AI hybridization while providing urban planners with three-dimensional heat mitigation insights from blue infrastructure optimization to urban morphology planning, supporting climate-resilient and healthy city design.

How to cite: Song, J. and Zhang, Q.: A Hybrid Modeling Framework for High-Resolution Humid Heat Mapping in Metropolitan Wuhan: Integrating Physical Simulations and Deep Learning, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-201, https://doi.org/10.5194/icuc12-201, 2025.

Urban heat is a significant contemporary challenge caused by the combined effect of urban development and global climate change. There has been substantial research investigating urban heat and assessing the effectiveness of heat mitigation strategies for different cities. Much of this research uses satellite-based Land Surface Temperature (LST) to assess urban heat through bird's-eye view surface temperatures, whereas canopy urban heat, measured by air temperature (Ta), is more directly relevant for public health and citizen thermal comfort. However, the sparse spatial coverage of Ta measurements fails to capture the significant temporal and spatial variability of air temperature in urban areas. Therefore, there is a need to produce gridded air temperature maps that represent these variations at scales relevant to people. Using crowdsourced air temperature measurements and machine learning techniques, we developed an innovative approach for estimating gridded air temperature for Sydney, Australia. We achieved this by using Landsat LST data as well as incorporating urban datasets characterizing urban form, fabric, and geography. A Convolutional Neural Network (CNN) was employed to infill gaps in the crowdsourced sensor data, achieving high performance and enhancing the spatial resolution of the temperature data. This study presents an effective approach for generating high-resolution, city-scale air temperature maps for cities aiming to enhance their temperature mapping and improve urban climate resilience.

How to cite: Naserikia, M., Hart, M., Asadi Shamsabadi, E., Bechtel, B., and Nazarian, N.: Advancing urban air temperature mapping with machine learning, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-243, https://doi.org/10.5194/icuc12-243, 2025.

Representative climate days are specific days that reflect typical weather conditions for a given location. Representative climate days are essential for reducing the computational demands of urban building energy and microclimate models, which typically require annual simulations. Traditionally, to identify representative climate days from typical meteorological year (TMY) weather files, unsupervised Machine Learning (ML) methods are employed. Traditional ML methods, though effective for high-dimensional data, struggle with bimodal weather distributions, where two distinct climatic regimes occur within the same period. This study introduces a novel hybrid ML framework for accurately clustering bimodal weather data through a multiphase unsupervised clustering process.

The framework starts by applying principal component analysis (PCA) to TMY data for reducing dimensionality. Next, misclustered days—identified using the silhouette score—undergo iterative re-clustering using k-Means and Gaussian Mixture Models (GMM) until fewer than 30 remain. Finally, representative climate days are determined from the properly clustered groups using a medoid-based weighted sampling approach. The potential of this hybrid framework is demonstrated using TMY data of the Tel Aviv, comprising of bimodal distribution. The k-Means, GMM, and hierarchical agglomerative clustering achieved higher silhouette scores through multiphase clustering over traditional approach. While k-Means-based multiphase clustering achieved the highest silhouette score, GMM demonstrated superior clustering performance by preserving month-to-month continuity, which is crucial for capturing seasonal variations. By maintaining seasonal continuity in representative days, this approach enhances the reliability of climate-based urban performance simulations, supporting more accurate and computationally efficient modelling.

How to cite: Manapragada, N. V. S. K. and Natanian, J.: A Hybrid Machine Learning Framework for Identifying Representative Climate Days in Bimodal Weather Distributions, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-169, https://doi.org/10.5194/icuc12-169, 2025.

Urban overheating has become a global issue, exacerbated by climate change and that may lead to severe effects on both public health as well as urban sustainability. This study is intended to permit the prediction of the longevity and severity of future urban overheating events by integrating field measurements and machine learning models, focusing on the impact of urban greening under different global warming (GW) scenarios. Field measurements have been conducted during summer 2024 in an office campus at Ottawa, a city located in cold climate zone. Microclimate data were measured at four locations within the campus, the four locations have different types and coverage levels of urban greenings – large lawn area without trees (Lawn), parking lot without any greening (Parking), greenery area with sparsely distributed trees (Tree) and an area with 100% coverage of trees (Forest).  Models, such as Artificial Neural Networks (ANN), and Recurrent Neural Networks (RNN), and Long Short-Term Memory network models (LSTM) were trained on local microclimate data, with LSTM chosen for its superior performance predictions. Four Global Warming (GW) scenarios were considered to represent different Shared Socioeconomic Pathways (SSP) by 2050 and 2090. The results show that the UTCI at the “Parking” location increased from around 27 °C under GW1.0 to 31 °C under GW3.5. Besides, low health risk (UTCI > 26 °C) will be increased in all locations due to climate change impacts, regardless of urban greening conditions. However, the tree area like 'Tree' and 'Forest' are effective in eliminating the occurrence of extremely high-risk heat conditions (UTCI > 38.9 °C). The findings demonstrate that urban greening plays a crucial role in reducing severe thermal stress, thereby enhancing thermal comfort under future climate scenarios.

How to cite: Zou, J., Wang, L., Yang, S., Lacasse, M., and Wang, L.: Evaluating the Impacts of Natural Based Soluations on Long-term Urban Overheating through Machine Learning and Field Measurements, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-310, https://doi.org/10.5194/icuc12-310, 2025.

Heat stress significantly impacts public health and urban sustainability, necessitating accurate and efficient assessment methods. Numerical models are commonly used to simulate key variables such as temperature, wind speed, and humidity for calculating heat stress indices. However, their accuracy is often hindered by uncertainties of various sources. This study developed a novel, high accuracy "WRF+AI" framework that integrates the Weather Research and Forecasting (WRF) model with Artificial Intelligence (AI) to enhance heat stress estimation. Five heat stress indices derived from air temperature, wind speed, and relative humidity were evaluated within this framework, addressing three critical questions: (1) Which AI method provides the best balance of accuracy and ease of use in this framework? (2) Is it more effective to estimate heat stress directly or indirectly via related basic variables? (3) How can the number of features in the “WRF+AI” framework be reduced to create a lightweight model? The results show that Automated Machine Learning method achieves high accuracy without the need for hyperparameter tuning. Direct heat stress estimation using the “WRF+AI” framework significantly reduces RMSE by 67.3%–82.6% and MAE by 70.0%–81.6% compared to traditional WRF simulations, outperforming indirect estimation based on basic variables produced by the “WRF+AI” framework. Additionally, SHAP (SHapley Additive exPlanations model)-based feature selection method proved effective in minimizing the number of features while maintaining model performance. This framework notably improves the accuracy of heat stress estimations, particularly in capturing diurnal peak variations, providing a reliable tool for heat stress risk assessment and urban heat management.

How to cite: Gao, M. and Li, H.: A "WRF+AI" framework for enhanced heat estimation in urban environments, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-325, https://doi.org/10.5194/icuc12-325, 2025.

This study develops a Physics-Informed Neural Network (PINN) model using a multi-layer perceptron (MLP) to estimate vegetation cooling effects, specifically spatially averaged air temperature reduction (ΔT) and vegetation-surrounding UTCI reduction (ΔUTCI), across heterogeneous urban contexts. The model incorporates building, vegetation, and climate features and is trained on ENVI-met simulations from 22 representative sites in Singapore, covering monsoon and inter-monsoon scenarios. These sites, selected through context-based mapping, capture diverse building and greenery configurations. By embedding physics constraints on sensible and latent heat fluxes from the Surface Energy Balance into the loss function, the model captures the overall influence of vegetation at the pedestrian level (2 m), with the physics loss enhancing accuracy and generalizability. A sensitivity analysis and an explanatory study using SHapley Additive exPlanations (SHAP) are conducted to assess the local and global impacts of features. The contributions of building, vegetation, and climate features to vegetation cooling effects are quantified across day/night and monsoon/inter-monsoon seasons. Results indicate that during daytime, ΔT is largely influenced by vegetation features (40%), whereas ΔUTCI is primarily driven by background climate features (47%), with seasonal variations affecting individual feature importance. These insights inform scenario-based urban design guidelines for optimizing vegetation cooling potential. The trained model is applied to built-up areas in Singapore, generating a vegetation cooling distribution map that reveals a mean ΔT of 0.5 °C and a mean ΔUTCI of 2 °C across all sites around noon. Targeted interventions are proposed for areas with suboptimal vegetation cooling performance, providing an urban design solution for leveraging greenery as a passive cooling strategy to mitigate urban heat in high-density cities.

How to cite: Zhang, L. and Yuan, C.: Physics-Informed Machine Learning for Mapping the Heat Mitigation Potential of Vegetation in Singapore, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-357, https://doi.org/10.5194/icuc12-357, 2025.

In a warming climate, Furthermore, with an increasing world population over the course of this century, more and more people will be at risk. Additionally, urbanization is expected to continue increasing, and due to the urban heat island (UHI) effect, heat will be intensified in the most densely populated areas.  In this project, we investigate the severity of hot spells with the help of health-relevant heat indices, namely wet bulb temperature, the universal thermal climate index (UTCI) and the amount of tropical nights, among others. We process reanalysis data from ERA5-Land from 1950 onwards and simulation data from the downscaled EURO-CORDEX simulations for various climate scenarios until 2100 to generate these indicators for Europe at a resolution of 0.1°. We train a shallow machine learning model (XGBoost) to downscale reanalysis and simulation data to the city level at a resolution of 100m for select European cities. For model validation, temperature series from 12 European cities’ urban measurement networks of are used. The indicators will be applied to four pilot cities (Oslo, Bern, Lyon and Naples), as part of the EU project healthRiskADAPT under the framework of Horizon Europe. In a subsequent phase, weather forecast data may be downscaled with the same algorithm to predict urban heatwaves. Furthermore, advanced modeling techniques such as Weather Research and Forecasting (WRF) models or computational fluid dynamics (CFD) may be applied to better understand the compound effects of heat and pollution in cities.

How to cite: Pierce, C.: Health-relevant Heat Indices for Urban Areas: A Machine Learning Approach with Downscaled Climate Data and City Measurement Networks, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-212, https://doi.org/10.5194/icuc12-212, 2025.

Machine learning is a subfield of artificial intelligence, rooted into statistics. It is a flexible and interdisciplinary tool that can be used for solving several types of problems. This work focused on the use of multi-layer perceptron (MLP) to solve a regression type problem: downscale land surface temperature (LST) from the SEVIRI sensor onboard Meteosat Second Generation series of satellites to a 750 m spatial grid. The choice of an MLP in preference of other machine learning algorithms was motivated by the intention of developing the simplest algorithm that provides acceptable results. Data from VIIRS onboard Suomi National Polar-Orbiting Partnership was used to compute the target LST. The resulting model was trained on 2019-2022 data and its performance accessed on 2023 data. To compare downscaled LST at different hours than the ones retrieved by SNPP, data from three additional sensors were used: VIIRS onboard both Joint Polar Satellite System 1 and 2 and the SLSTR onboard Sentinel 3A.

The comparison between the target and the models’ results are optimistic given that its performance on new data is similar to its training performance, i.e. no overfitting. The distribution of model temperature values follows the one of target temperature values.

The new downscaled dataset is then used to analyze the diurnal behavior of the surface urban heat island with an improved spatial resolution than the one a geostationary is able to provide, for the city of Madrid during cold/heat waves.

How to cite: Hurduc, A., Ermida, S., and DaCamara, C.: A multi-layer perceptron approach to downscaling geostationary land surface temperature in urban areas, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-447, https://doi.org/10.5194/icuc12-447, 2025.

Accurate high-resolution urban microclimate modeling, including wind and temperature prediction, is essential for urban planning, occupant comfort analysis, and building energy efficiency optimization. However, traditional computational fluid dynamics (CFD) methods are computationally expensive and time-intensive for applications requiring rapid urban microclimate estimation. This work presents a novel deep learning framework that directly downscales kilometer-scale Weather Research and Forecasting (WRF) model outputs to a 10-meter level resolution 3D urban microclimate for a given geographical setting. By incorporating building geometries as model inputs, our approach captures fine-scale building-induced effects in urban wind and temperature fields, which are absent in WRF's coarse-resolution outputs.

The deep learning model is trained and evaluated using urban microclimate data simulated with PALM for a realistic geographical setting in Berlin, Germany, where one week's worth of low-resolution WRF outputs serve as boundary conditions. Our proposed approach follows a two-stage training process. First, a conditional neural field (CNF) encodes the coarse WRF boundary conditions and generates a smooth, building-agnostic 3D flow field at PALM resolution. Next, a geometry-aware Fourier neural operator (FNO) refines this field by incorporating high-resolution building geometries, accurately capturing the complex interactions between airflow and urban structures. To effectively represent complex building geometries, we introduce a multi-directional distance feature (MDDF) that captures long-range spatial relationships between buildings. 

By producing building-resolved microclimate data from WRF outputs in near-real-time, our approach facilitates applications that are otherwise impractical with conventional CFD solvers. Despite being trained on a limited set of WRF boundary conditions, our model generalizes effectively to unseen conditions, underscoring its potential as a powerful and flexible tool for rapid urban microclimate forecasting and analysis.

How to cite: Qin, S., Geng, D., Vogel, J., Afshari, A., and Wang, L. (.: Deep Learning for Urban Microclimate Downscaling: From Coarse WRF Data to Building-Resolved PALM Simulations, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-568, https://doi.org/10.5194/icuc12-568, 2025.

The formation of Urban Heat Islands (UHI) creates substantial effects on building energy usage alongside human comfort standards throughout the world's urban areas. Current methods for mapping urban temperatures struggle to create a balance between detailed spatial coverage and accurate time-specific data. In this research, we designed a reference station-based method to create high-resolution temperature maps of urban areas at low cost, which is implemented in Shenzhen, China as the case study. A combination of Local Climate Zone classification with satellite data and machine learning algorithms generates spatiotemporally continuous temperature field results. The XGBoost-based mapping framework can achieve an MAE of 0.56°C with an R² value of 0.980. Building simulation together with thermal comfort analysis can benefit substantially from this methodology as it allows users to develop representative high-resolution microclimates through Typical Meteorological Year (TMY) weather data. The created model enables architects and engineers and urban planners to support their decisions in building design, climate change adaptation, and energy management practices. The developed approach delivers advanced air temperature mapping at affordable costs and requires easy implementation. The proposed data collection method offers detailed temperature information with high spatial resolution and temporal precision which makes it possible to improve urban planning and forecast building as well as renewable energy system performance in urban areas.

How to cite: Shen, P.: A novel reference station-based methodology for high-resolution urban temperature mapping utilizing machine learning techniques, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-596, https://doi.org/10.5194/icuc12-596, 2025.

Accurate near-surface air temperature data at high spatiotemporal resolution is crucial for understanding the urban thermal environment, especially given rapid urbanization and global climate change. However, existing research often focuses on daily air temperature metrics, neglecting the importance of sub-daily variations. This study addresses this gap by developing a Random Forest model to estimate sub-daily air temperature in major Chinese cities. Leveraging MODIS-derived land surface temperature (LST) from 2013 to 2023, the model incorporates 18 auxiliary variables encompassing time-related (e.g., atmospheric indices) and space-related (e.g., elevation, land cover) factors. To account for diurnal and seasonal variations, the model was trained and evaluated separately under four distinct conditions: daytime/nighttime and warm/cold. Cross-validation was employed to assess model performance. Results indicate optimal performance during warm nighttime conditions, achieving a low root mean square error. Analysis of variable importance revealed LST as the most influential predictor across all conditions, followed by humidity-related variables. Furthermore, the study found that the relative importance of auxiliary variables shifts with time of day and season. Time-related variables exert greater influence during warm conditions and daytime, while space-related variables become more important in cold seasons and nighttime. This highlights the importance of including diverse auxiliary variables for accurate sub-daily air temperature estimation. Developed using open-source data and cloud computing platforms like Google Earth Engine, the model offers a readily accessible and adaptable tool for urban climate research. This research not only provides valuable high spatiotemporal resolution air temperature data for Chinese cities but also presents a transferable methodology applicable to urban climate studies globally. The resulting data can contribute significantly to a deeper understanding of urban thermal dynamics and the development of effective urban heat island mitigation strategies.

How to cite: Guo, Y., Unger, J., and Gál, T.: Model development for estimating urban air temperature in China integrating satellite-based LST and auxiliary variables with machine learning, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-608, https://doi.org/10.5194/icuc12-608, 2025.

Detailed long-term climate scenarios over cities are needed for reliable future urban risk assessment and for designing climate resilient cities. For many cities, only coarse model information is however available. A small portion of mainly large cities in the global North possesses (sub-)kilometer long-term climate information, often acquired in the context of dedicated modelling case studies. Producing intra-urban climate projections for the multitude of medium to small cities is impossible using traditional downscaling techniques due to the high computational cost.

Therefore, we explored the potential of machine learning to simulate high-resolution near-surface air temperature over European cities. The European Random Forest Urban Climate Emulator (Eu-RaFUCE) framework was built and tested on hectometer-scale 2 m-air temperature of the urban climate model UrbClim for which data for 100 European cities is available (Lauwaet et al., 2024). Feature importance analysis showed that both temporal, such as surface net solar radiation, and spatial, such as land cover, inputs are important for the machine learning model to capture the urban heat island (UHI) effect. By applying Eu-RaFUCE, hourly ERA5 reanalysis data can be directly downscaled to (sub-)kilometer 2 m-temperature. Eu-RaFUCE captures the spatial and temporal UHI characteristics of multiple test cities well, resulting in high model accuracy over cities and years that were not included during the training. Out of all test cities Madrid has the largest root mean square error (RMSE), amounting 1.82 K, while the lowest RMSE of 0.85 K was found for Tallin. This proof of concept paves the way for the application of Eu-RaFUCE in downscaling 2 m-temperature projections to hectometer resolution over many cities, showing the potential of machine learning for urban climate studies.

Reference: Lauwaet et al. (2024). High resolution modelling of the urban heat island of 100 European cities. Urban Climate, 54, 101850.

How to cite: Top, S., Blancke, J., Delbare, K., Covaci, A., Caluwaerts, S., Hamdi, R., and De Cruz, L.: Hectometer-scale 2 m-temperature climate data over every city?!, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-647, https://doi.org/10.5194/icuc12-647, 2025.

Pedestrian-level wind plays a critical role in shaping the urban microclimate and is
significantly influenced by urban form and geometry. The most common method for
determining spatial wind speed patterns in cities relies on numerical computational
fluid dynamics (CFD) simulations, which resolve Navier-Stokes equations around
buildings. While effective, these simulations are computationally intensive and require
specialised expertise, limiting their broader applicability. To address these limitations,
this study proposes a more cost-effective alternative while achieving 90% performance
in capturing the mean and maintaining spatial wind patterns captured by CFD. We
developed a machine learning (ML) approach with U-net architecture to predict time
mean wind speed patterns from prevailing wind directions and three-dimensional
urban morphology, which are increasingly available for global cities. The model
is trained and tested using a comprehensive dataset of 512 numerical simulations
of urban neighbourhoods, representing diverse morphological configurations in cities
worldwide. We find that the ML algorithm accurately predicts complex wind patterns,
achieving a normalised mean absolute error of less than 10%, which is comparable
to wind anemometer measurement in a low wind speed environment. In predicting
wind statistics, the ML model also surpasses that of regression models based solely
on statistical representations of urban morphology. The R2 values measuring grid-
level agreement between ML and CFD range from 0.94-0.99 and 0.65-0.95 for the
idealised and whole datasets, respectively. However, we find that grid-based R2 is not
an effective metric for evaluating the 2D model performance due to localised biases
arising from faster wind speed grid regions, which is revealed by the wind probability
density function. These findings demonstrate that complex pedestrian wind patterns
can be effectively predicted using an image-based ML approach, offering the potential
to emulate physics-based LES models, which are computationally expensive, thereby
significantly reducing computing costs.

How to cite: Lu, J., Li, W., Hobeichi, S., Azad, S., and Nazarian, N.: Machine Learning Predicts Pedestrian Wind Flowfrom Urban Morphology and Prevailing WindDirection, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-101, https://doi.org/10.5194/icuc12-101, 2025.

Urban areas are increasingly vulnerable to climate change impacts, particularly heatwaves, due to their unique characteristics. However, the influence of urban form and land cover on future outdoor thermal comfort remains insufficiently quantified in existing climate models. In this study, we present UHTC-NN, a novel deep learning model designed to predict human thermal comfort (UTCI) in complex urban environments at an unprecedented 1-meter spatial resolution. UHTC-NN provides rapid, high-resolution predictions of pedestrian-level UTCI fields, enabling systematic and quantitative analysis of urban heat stress.

We demonstrate the capabilities of UHTC-NN by downscaling a CMIP5 regional climate model ensemble to 1-meter resolution for a 5.0 km x 2.5 km area in Freiburg, Germany. The results reveal significant increases in heat stress hours under the RCP4.5 and RCP8.5 scenarios, with the climate signal emerging as the dominant driver. Our analysis highlights substantial intra-urban variability in heat stress hours for both the reference period (1990–2019) and projected future changes (2070–2099), emphasizing the critical need for high-resolution prediction models like UHTC-NN.

Additionally, our findings reveal distinct day-night patterns: future daytime heat stress increases more uniformly across the city, whereas nighttime heat stress exhibits greater spatial heterogeneity, driven by factors such as shading, building density, and land cover. The high-resolution UTCI predictions generated by UHTC-NN represent a significant advancement in data-driven heat stress modeling, offering a holistic understanding of climate change impacts, complex urban structures, and diurnal variations.

How to cite: Briegel, F., Schrodie, S., Sulzer, M., Brox, T., Pinto, J. G., and Christen, A.: High Resolution City-Scale Climate Projections of Urban Heat Stress based on an Deep Learning Approach, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-699, https://doi.org/10.5194/icuc12-699, 2025.

Intra-urban induced heating (IUIH) in hot desert cities exhibits distinct patterns and complex diurnal interactions with built environment features, differing significantly from those in temperate areas and remains not fully understood. Understanding how various built environment features contribute to intra-urban thermal variability is an essential first step in developing sub-diurnal targeted heat mitigation strategies. This study presents a data-driven examination of IUIH dynamics in Kuwait’s desert metropolis. It employs the framework of urban induced heating (UIH) to disconnect from the urban-to-nature comparative fundamental to the urban heat island (UHI) definition. This approach facilitates a methodology that specifically excludes non-urban systems and highlights intra-urban thermal variability, proving more relevant for assessing the effectiveness of urban heat mitigation interventions. Near-surface air temperature observations were collected using high-resolution loop-type traverses at selected hours during a representative summer day to determine IUIH variability. The diurnal impacts of built environment features were modeled using an ensemble learning approach and interpreted with SHapley Additive exPlanations. Among several candidate machine learning regressors evaluated, Random Forest demonstrated strong predictive power (R² = 0.954) with acceptable error (RMSE =0.096, MAPE =0.001) and least bias (MBE =0.008). The study’s significance lies in its assessment framework that emphasizes explainability of sub-diurnal dynamics, offering detailed insights that challenge traditional assumptions and inform both immediate local climate interventions and strategic urban planning. The findings reveal that simple day-evening comparatives might overlook nuanced sub-diurnal dynamics, such as potential irrigation-induced warming by shrubs observed at mid-day and the complex trade-offs between radiative and transpirative processes by trees in the afternoon and evening. Additionally, the study identifies cooling effects associated with natural land cover, presenting a critical optimization challenge between compact and open urban forms to effectively modulate near-surface air temperatures.

How to cite: AlKhaled, S. and Ramadan, A.: Intra-urban induced heating assessment in Kuwait’s desert metropolis using explainable machine learning, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-317, https://doi.org/10.5194/icuc12-317, 2025.

High summer air temperatures lead to increased morbidity and mortality in urban areas worldwide.　To mitigate the adverse effects of extreme heat, policymakers and urban planners have developed high-temperature adaptation strategies and public health management plans. The effective implementation of these measures requires a precise spatial understanding of heat generation patterns within urban environments. With the advancement of IoT technology, the use of third-party sensors, which can be deployed and operated at relatively low cost, has become increasingly common for monitoring urban temperatures as a method of fixed-point measurement. Some studies have suggested that data from these sensors can complement information from existing primary sensors. However, the extent to which these sensors enhance the understanding of the urban thermal environment remains an area of ongoing research.

Therefore, This study aims to quantify the contribution of high-density sensors operated by local governments to improving the performance and reliability of AI-based models for estimating high-resolution heat-related information in urban areas. Specifically, this study (1) constructs a baseline model using a commonly adopted set of auxiliary variables including land surface temperature (LST) derived from Landsat 8 satellite imagery and (2) develops an alternative model incorporating observational data from high-density sensors operated by the Seoul Metropolitan Government in South Korea as additional auxiliary variables. By quantitatively assessing the role of high-density sensors deployed by metropolitan governments, this study seeks to provide valuable insights for enhancing the accuracy of thermal hazard information. The findings will support metropolitan and local government decision-makers and urban planners in developing more effective strategies for adapting to extreme heat in the future.

How to cite: Aida, A. and Park, C.: Enhancing Urban Heat Monitoring: Assessing the Contribution of High-Density Environmental Sensors in AI-Based Models, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-996, https://doi.org/10.5194/icuc12-996, 2025.

In this study, we aimed to evaluate the accuracy of wrist-mounted wearable sensors in measuring and predicting individuals’ thermal comfort sensations in transitional and outdoor environments. To achieve this, we combined mobile measurements, wearable devices, and surveys to generate a reliable dataset from outdoor settings. We assessed the universal thermal climate index (UTCI) and wrist-mounted wearable data in relation to thermal comfort votes (TCV) and thermal sensation votes (TSV). Our findings revealed that UTCI strongly correlates with individuals’ thermal comfort sensations and serves as a reliable indicator of outdoor thermal comfort, particularly in Sydney's outdoor environments. We observed that wrist air temperature demonstrates a correlation pattern with TCV similar to that of UTCI and exhibits an even stronger correlation with TSV. This finding suggests that wrist air temperature can serve as an effective indicator of thermal comfort sensations in the absence of UTCI. Using a random forest machine learning algorithm, we developed a prediction model for UTCI based on wrist-mounted sensor data. The results demonstrated the potential of wrist-mounted sensors to accurately predict UTCI, further validating their effectiveness in assessing outdoor thermal comfort. Furthermore, we utilized wearable data to directly develop prediction models for TCV and TSV using the same machine-learning approach. Feature importance analysis revealed that mean radiant temperature, wind speed, and wrist air temperature significantly influence the prediction models, emphasizing that thermal conditions play a critical role in these predictions.

How to cite: Parchami, M., Nazarian, N., Hart, M. A., Liu, S., and Martilli, A.: Machine-learning approach for predicting individual's thermal comfort and thermal sensation in outdoor environments, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-246, https://doi.org/10.5194/icuc12-246, 2025.

It is crucial to know the airflows within the urban district for the improvement of the urban atmospheric environment such as air pollution, thermal comfort, wind load, etc. However, it is not easy to know their spatial distribution in observation. And numerical simulation still requires a high computational cost. Instead, this study attempts to reproduce the local velocity distribution within the urban canopy layer by means of a supervised machine learning model. In order to learn the relationship between urban structures and turbulent velocity distributions, the model is trained on large datasets of turbulent flow simulation data and detailed urban geometries of the Tokyo districts. The simulation cases of urban turbulent airflows have been conducted using lattice Boltzmann equation model, which was selected thanks to its high-computation performance on a massive GPU cluster.

The developed model was applied onto several cities other than Tokyo. Irrespective of the building geometries and arrangement, some general characteristics of the urban airflows have been confirmed; such as, an acceleration of the flows in streets parallel to the main flow, and a down-draft and divergent flows in front of tall buildings. The results were also compared with observation datasets taken directly from the urban districts examined.

How to cite: Inagaki, A., Takamatsu, R., and Kanda, M.: Reproduction of local turbulent flow characteristics within urban district using supervised machine learning model, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-650, https://doi.org/10.5194/icuc12-650, 2025.

Microclimates play a crucial role in shaping occupant health, thermal comfort, and building energy performance. Urbanized landscapes are characterized by great variation in microclimates, often over very short distances that can be related to the character of the built landscape (including the layout and dimensions of buildings, the properties of construction materials and the natural landscaping).  These microclimates can enhance background climate and weather in cities, affecting the indoor and outdoor environment. The impact of these urban effects are greatest for those living in historically marginalized and low-income communities, where buildings and neighbourhoods are poorly constructed and designed and residents face disproportionate exposure to extreme heat, with limited access to adaptive resources. Addressing these challenges requires an integrated, data-driven approach to urban and building-level heat mitigation.

This study combines real-world weather data collection, Computational Fluid Dynamics (CFD) simulations, and Machine Learning (ML) models to evaluate the efficacy of targeted mitigation strategies in a historically redlined neighborhood in a hot and humid Texas city. At the urban scale, strategies such as increased tree canopy coverage, reflective/cool pavements, high-albedo surfaces, and optimized street layouts are analyzed for their potential to reduce localized heat accumulation. At the building scale, interventions include enhanced insulation, cool roofs, natural ventilation optimization, and external shading solutions. Machine learning algorithms are employed to identify patterns in urban heat distribution, predict temperature fluctuations under different scenarios, and optimize mitigation measures based on building typologies and urban configurations.

The results provide actionable insights for planners and policymakers by quantifying the relative effectiveness of different interventions in reducing urban heat exposure and improving outdoor/indoor thermal comfort. By integrating CFD modeling with ML-driven analysis, this study presents a scalable framework for urban heat mitigation, with implications for equitable and climate-resilient urban planning.

How to cite: Hashemi, F., Mills, G., Tran, Q. V., and Najafian, P.: Data-Driven Urban Heat Mitigation: Integrating CFD and Machine Learning for Adaptive Cooling Strategies, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-985, https://doi.org/10.5194/icuc12-985, 2025.

Urban extreme precipitation presents a considerable challenge owing to its escalating societal repercussions and the constraints associated with opportune forecasting. This study presents an artificial intelligence (AI) model aimed at improving daily precipitation forecasts for complex urban environments, driven by a fully urbanized Weather Research and Forecasting (uWRF) model. Our use case is New York City, which has historically experienced significant social, infrastructural, and economic consequences from such events. Using New York state mesonet station observations, it was found that there were around 63 extreme precipitation (>39 mm/day) cases during the 2018–2023 summer season. For all these extreme precipitation cases, the uWRF model adequately predicts precipitation; however, biases exist in both the spatial and temporal occurrence of maximum precipitation. To address these issues, a well-tested AI model, Attention U-Net, has been applied to improve precipitation forecasts. In this case, uWRF hourly precipitation data at 1 km spatial resolution serves as the input, while the Multi-Radar/Multi-Sensor System (MRMS) daily accumulated precipitation data is used as the target variable. Future research will focus on evaluating the performance of the Attention U-Net model for several out-of-sample extreme precipitation events and further addressing spatial biases using MRMS data, with the goal of improving both the accuracy and reliability of forecasts for urban extreme precipitation. We will also explore transferability to other locations. 

How to cite: Swain, M., Montoya-Rincon, J. P., Schmude, J., and González-Cruz, J. E.: AI-Driven Correction of Precipitation Forecasts in Dense Urban Environments, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-972, https://doi.org/10.5194/icuc12-972, 2025.

Urban areas cover 3 to 5% of total land area, however they contain nearly 2/3 of global population. It has been shown that cities also experience increased effects of climate change. Therefore for a sustainable future it is vital to be able to model not only the large-scale climate change, but also the small-scale changes of weather patterns on an intra-urban scale.

Modern methods of climate modelling do not provide sufficient resolutions to reflect on the 100 meter scale of the city processes. Downscaling methods are used to increase the spatial resolution of climate data. Dynamical downscaling is the use of mesoscale weather models. Statistical downscaling includes machine learning and deep learning methods.

This work employs both methods to create high-resolution air temperature, wind speed and thermal comfort fields. Mesoscale models are used to model the target vartiables with high resolution. Then the data is used to train a U-Net neural network. The network takes ERA5 low-resolution fields and high resolution fields of land surface data. 

The network has shown a decrease in errors in comparison wint ERA5 which does not reproduce urban microclimate. For key urban stations the mean error was reduced by several times. Moreover, there is a significant gain of speed. Computing 1 year with a weather model takes 17 days, while it takes the NN 5 minutes to do the same. This creates opportunities for high-res climate modelling

How to cite: Kospanov, A., Varentsov, M., Krinitskiy, M., and Stepanenko, V.: The use of a U-Net neural network for high-resolutionc urban climate modelling, 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-1022, https://doi.org/10.5194/icuc12-1022, 2025.

The urban drag coefficient (C_d​) is a key parameter in urban climate modeling, influencing airflow, pollutant dispersion, and energy exchange. This study aims to map the spatial distribution of Cd​ in the Guangdong-Hong Kong-Macau Greater Bay Area (GBA) by combining computational fluid dynamics (CFD) simulations and machine learning techniques. We employ the Reynolds-Averaged Navier-Stokes (RANS) turbulence model to simulate airflow over idealized urban surfaces with varying building density, average height, and height variability. Two groups of urban configurations are designed—one with uniform building heights and another with heterogeneous heights—considering multiple plan area fractions (λ_p​) and standard deviation of building heights (σ_H​). Machine learning models, particularly Random Forest, are trained on CFD-derived Cd​ values to predict spatially distributed drag coefficients at a 1-km resolution. SHapley Additive exPlanations (SHAP) analysis reveals that λ_pand σ_H​ are the dominant factors influencing C_d​. The resulting drag coefficient map captures the spatial variability of urban aerodynamic resistance, providing a refined parameterization scheme for mesoscale climate models. This study enhances the representation of urban canopy processes and can be extended to other metropolitan regions, improving the accuracy of urban climate simulations.

How to cite: Mo, Z. and Chen, J.: Mapping the urban canopy drag coefficient using CFD and machine learning , 12th International Conference on Urban Climate, Rotterdam, The Netherlands, 7–11 Jul 2025, ICUC12-312, https://doi.org/10.5194/icuc12-312, 2025.