Amplified warming in tropical and subtropical cities at 2 °C climate change

Amplified warming in tropical and subtropical cities at 2 °C climate change | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Physical Sciences - Article Amplified warming in tropical and subtropical cities at 2 °C climate change Sarah Berk, Manoj Joshi, Clare Goodess, Peer Nowack This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4623186/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Feb, 2026 Read the published version in Proceedings of the National Academy of Sciences → Version 1 posted You are reading this latest preprint version Abstract Cities are often warmer than rural surroundings due to a phenomenon known as the urban heat island, which can be influenced by various factors, such as regional climate. Under climate change, cities face not only the challenge of increasing temperatures in their surrounding hinterland, but also the challenge of potential changes in their heat islands. Making projections of future climate at the city scale is difficult given limitations of Earth System Model (ESMs), which has limited studies to a small number of urban areas – mostly megacities. Here, we address these limitations by applying a novel process-based machine learning model to ESM outputs, to provide projections of changes in land surface temperature (LST) for 104 medium-sized cities (population 300K to 1M) in the subtropics and tropics. Under a 2°C global warming scenario, annual mean LST in 81% of these cities is projected to increase faster than the surrounding area. In 16% of these cities, mostly in India and China, mean LST is projected to increase by an additional 50–112% above ESM projections of the surrounding area. These findings suggest that the potential impacts of climate change are underestimated at present for millions of people in cities. Earth and environmental sciences/Climate sciences/Climate change/Projection and prediction Earth and environmental sciences/Climate sciences/Climate change/Climate-change impacts/Environmental health Earth and environmental sciences/Environmental sciences/Environmental impact urban heat island surface urban heat island machine-learning land surface temperature climate change Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction The urban heat island (UHI) is a phenomenon whereby the temperature in a city differs from the surrounding rural area, typically being warmer. This leads to increased heat-related health risks for urban inhabitants in comparison to their rural counterparts 1 . In 2018, it was estimated that over half the world’s population resided in cities and this proportion is projected to increase to 68% by 2050 2 . Climate change results in rising global temperatures and increased frequency of extreme heat events 3 , which can have severe human health impacts including increased mortality 4–6 . UHIs are influenced by both climate and city attributes (e.g., city area, rural aridity) 7 , all of which can change over time. The constructed model is based on these processes and therefore can assess the consequence of changes in attribute properties on the UHI. A deeper understanding of such shifts in UHI intensities will inform city planners as they design cities aiming to optimize human comfort and health, as well informing adaptation to the impacts of climate change. Modelling and projecting changes in UHI remains a challenge 8 . Global climate model outputs have spatial resolutions larger than the scale of most cities due to limitations in computational power. Regional climate models have higher resolutions but are also constrained by computational expense, limiting their ability to model many cities simultaneously. The above limitations mean that projections of the impacts of climate change on the UHI are also limited to either the largest cities 9 , or to smaller cities in certain geographical regions, at a lower resolution 10 . Indeed, much of the current research focus of the UHI is on megacities, which represent just 12% of the urban population 11,12 . Furthermore, several regions of the world are under-represented in the UHI literature, e.g., Africa and South and Central America 12,13 . Typically, as cities expand the intensity of their UHI also grows. However, it is observed that saturation of the UHI with city size occurs in very large cities (e.g., London) 14 . Cities where saturation of the UHI has occurred may respond differently to climate change than those medium-sized cities where this point has not yet been reached. A complete picture of UHI behaviour under climate change can therefore only be gained by addition of the examination of such medium-sized cities. In this paper we use a process-based machine learning (ML) model which uses earth system model (ESM) projections to make future surface UHI (SUHI) projections under 2 °C global warming for selected medium-sized cities in the tropics and subtropics. ML approaches have been used before in SUHI studies to explore relationships between predictor and target variables 15 , but this work takes the approach further, using the fitted model to predict SUHI and its changes in the future. Such projected changes will be in addition to both temperature change projected in the surrounding area by ESMs, and the currently experienced SUHI. We use the SUHI- the urban-rural difference in land surface temperature (LST)- as a proxy for the UHI due to limited availability of air temperature measurements. Weather station data is irregular and sparse in coverage, and air temperature sensor networks in cities are rare and have limited temporal coverage 16 . Satellite data, on the other hand, has a high resolution and global coverage. Although the additional effect of advection complicates the coupling between the two, a significant positive correlation exists between urban LST and air temperature 17,18 because LST controls air temperature in the lower layers of the atmosphere. A novel method for SUHI projections Here we present our results for the day-time SUHI at 13:30 hours. Results for night-time, when changes in SUHI with climate change are generally much smaller, can be found in the extended material, alongside the observed values of current SUHIs. First, an overview of the selected cities is given. Next, we describe our ML model and evaluate its performance. We then describe the projections made by combining the ML model with ESM outputs. From a dataset of global urban areas 2 , we impose city selection criteria to return a subset containing medium-sized cities with additional restrictions to remove non-climatic influences. For example, coastal cities or those in mountainous regions are not included. Selection criteria are listed in the methods section. The locations of the 104 selected cities are shown in Figure 1. Figure 2a outlines the procedure used to generate the changes in urban LSTs. The ML model, Regression Enhanced Random Forest (RERF) 19 , is set up to predict SUHI magnitudes from factors such as urban-rural vegetation differences and relative humidity. These factors are acquired from satellite and reanalysis data for the model development. Projected changes in these drivers under a global 2 °C warming, obtained from CMIP6 ESM projections, are then used to project changes in the SUHI (see Methods). Our ML model performs well for present day climates across the selected cities, successfully predicting SUHI magnitudes for a range of observed values, as shown in Figure 2b. Across all cities, the test data have overall performance statistics of R-squared 0.87 and RMSE 0.86 °C, giving confidence in the ability of the model to make projections of the SUHI on unseen data. We consider various validation scenarios to ensure a robust model for the required application (see Methods). SUHIs increase with 2 °C warming For most of the 104 cities, the current SUHI is projected to become more positive. This is apparent in Figure 1. Under 2 °C warming, 81 % of 13:30 SUHIs are projected to increase in their annual mean. This change is in addition to the regional background warming projected by ESMs also shown in Figure 1. The overall mean of this amplification is 0.4 °C, increasing the overall change in city temperature (urban ∆LST) from 2.2 °C (the ESM regional ∆LST) to 2.6 °C. Cities in the Middle East, India and China all undergo large additional annual warming, as shown in Figure 1a. Figure 3 zooms in on the annual changes shown in Figure 1. In the Middle East increases in SUHI are a particular cause for concern as these regions are already very hot, and also face a considerable increase in ESM-based regional LST. In these areas, the current SUHI is negative (an urban cool island), due to greater vegetation and irrigation in the urban area in contrast to the rural. The projected increase in the urban LST in these regions indicates the SUHI becoming less negative, and in some cases, positive. Increases in SUHI magnitude are especially likely to impact on human health during the warmest months of the year. To investigate this, the data was split into four calendar quarters and the warmest season defined as that with the highest mean 2m air temperature for each region. This warm season projected change can be seen in Figure 1b (with a magnified version in Extended Data Figure 3), which shows that 75% of the SUHIs increase, with an overall mean change of 0.3 °C (with ESM ∆LST being 2.4 °C and urban ∆LST 2.7 °C). Warm season increases in SUHI magnitude are particularly noticeable for cities in Northeastern China. Major shifts in highly populated regions For many highly populated countries, such as India and China, projected changes in the SUHI are shown to be particularly pronounced in comparison to background levels of warming (Figure 3). For all the studied cities in India, mean LST is projected to increase by an additional 45% above ESM projections of the surrounding area, and in China by an additional 40%. A major reason for this is the influence of vegetation, which is associated with increased cooling due to evapotranspiration. Predictor importances, determined using accumulated local effects 20 , find vegetation to be a strong influencer on the SUHI magnitude (Extended Data Figure 4). ESM projections of large-scale changes to vegetation or moisture availability, which have a cooling influence on rural areas, do not typically affect cities to the same extent, as they are made up of artificial impervious surfaces and drainage systems that carry away surface water 21 . In the areas where there are increases in regional vegetation (ESM projections can be seen in Extended Data Figure 5), the SUHI becomes more positive. Here, these changes in vegetation, which lead to an increased magnitude in urban-rural vegetation difference, are responsible for the largest changes in the SUHI. In parts of Brazil the opposite effect is seen (Figure 3), and the SUHI becomes smaller. Figure 4 summarises how the inclusion of city specific projections can have a substantial influence on the overall ∆LST as a function of ESM ∆LST. Whilst only 3 city regions experience an increase above 3 °C based on ESM LST, 26 cities experience increases in median urban modelled LST above 3 °C. For two of the cities, Patiala, India and Kasur, Pakistan, the additional change in SUHI results in the city ∆LST being twice that of the ESM projection. Prediction intervals, based on the ML model are shown on Figure 4. The cities which have small projected changes in SUHI tend to have the largest prediction intervals, indicating these SUHIs are less influenced by the input climatic variables in the ML model. When changes in SUHI are considered on top of the changes in regional LST, it is clear that almost all of the cities studied undergo larger LST increases than their rural hinterlands. The overall influence of including city-specific projections, rather than simply examining the ESM grid cell, skews the probability distribution of ∆LST towards larger magnitudes for both the annual and warm season mean values (Extended Data Figure 6). Discussion We have investigated the effects of climate change on the daytime SUHI of 104 medium-sized cities in the tropics and subtropics, which are currently home to over 50 million inhabitants. City temperatures are already amplified due to the UHI in all but the most arid regions, and globally all areas face increases in temperature due to climate change. On top of these known factors, we have demonstrated the potential for urban warming to be amplified in many cities, i.e. city LSTs increasing faster than ESM projections suggest. We note that such a trend has already been observed over the last twenty years 22 . Our results are of immediate relevance to policymakers who will need to account for the increased hazards many urban citizens will face over the coming decades. The cities studied here are located in the warmer parts of the world, which makes this increase even more impactful for human health and the urban environment 23,24 . More generally, medium-sized cities represent a large proportion of global cities with more than 2.5 times as many cities in this category than cities with over 1 million population 2 . Our novel method, which combines state-of-the-art climate change projections with process-based ML models, enables more informed planning for these future risks. The projected SUHI increases are particularly noticeable in the highly populated regions of north India and northeastern China. This is concerning as both these areas are projected to experience more frequent and intense heatwaves 25,26 . In hot temperatures, outdoor workers are subject to numerous negative impacts of heat exposure 27 and economic impacts should they forgo a day’s work. India is projected to require large cooling demands in the future, which is problematic as the infrastructure may not be able to cope with this increased load, and the costs are prohibitive for many 28 . Increased energy usage also brings consequences for climate change mitigation. The need for UHI mitigation and heat adaption in these regions is therefore even more pressing. A caveat of this study is that city expansion has not been considered. Including future urban expansion would likely lead to an even greater warming compared to the above projected changes 29 . Relaxing the city selection criteria to include more cities would make the results more generalisable, although the addition of variables into the ML model would increase complexity. In particular, including coastal cities and those near large waterbodies, where humidity is more important, and an important aspect of human health and comfort, is a source for future research. Methods City Selection The included cities had a population between 300,000 and one million and a latitude of less than 40°. Additionally, the surrounding features (lakes, hills, oceans) of cities were considered in order to control other variables and isolate the impact of climate (this could be relaxed in future work). An overview of the criteria and datasets used is provided below. Population: 300,000 to 1,000,000 24 Location: 42 km from any other city with >300,000 population 24 Coastal distance: > 100 km from shoreline 25 Water proximity: > 50 km from lakes >50 km wide or > 22 km from lakes >1 km wide 26 Topography: 5 km 2 city area in 2002 28 Generating the SUHI and predictor variables The dataset used to develop the ML model was generated using satellite and reanalysis data. The workflow diagram for the model build and use can be seen in Figure 2a. For satellite data, cloud contamination thresholds were set so at least 70% of the overall (rural + urban) area and 50% of the urban have usable pixels. For images deemed acceptable, any remaining poor-quality pixels (pre-defined in the datasets via quality flags) were masked in the analysis to promote accuracy. SUHI was quantified as the difference between the monthly mean city LST and the surrounding rural reference area; where LST urban , n urban , LST rural , n rural represents the LST and number of urban and rural pixels, respectively. The rural reference area was defined as a rectangular box surrounding the city, where the city takes up the 10% of the area in the centre. The city and all other urban pixels in the area are masked out. The satellite data, including spatial and temporal characteristics utilised for SUHI quantification are outlined below. Terra LST 8-Day Global (MOD11A2) 30 (1km resolution, available every 8 days from 2002-2020) ESA Land Cover Climate Change Initiative: Global Land Cover Maps, Version 2.0.7 31 (300m resolution, each year between 1992 and 2020) The predictor variables used (selected based on maximising R-squared) are monthly means of the following. Relative humidity (RH) Total precipitation (TP) Urban enhanced vegetation index (EVI_U) Urban – rural enhanced vegetation index difference (EVI_D) Log 10 of city area (LOG_AREA) Urban – rural white sky albedo difference (WSA_D) Urban - rural elevation difference (ELEVATION_D) Standard deviation of urban elevation (STD_ELEVATION_U) These variables are known to influence UHI magnitudes, and are tested through model evaluation techniques such as accumulated local effects 20 (Extended Data Figure 4) to ensure the ML model accurately reflects the physical processes that cause the SUHI to vary. City area is calculated using the ESA landcover data 31 and elevation variables using the topography 27 . The additional satellite and reanalysis datasets used are as follows. RH, TP from ERA-5 32 (9km resolution, monthly from 1981 to present) EVI_U, EVI_D from MODIS MYD13A2, MOD13A2 33,34 (1000m resolution, every 16 days from 2002 to 2020) WSA_D from MODIS MCD43A3, MCD43A2 35,36 (500m resolution, daily from 2002 to 2020) The urban – rural difference variables use the same rural and urban areas and equation 1 as SUHI quantification. Developing the ML model The chosen ML model, RERF 19 , is a hybrid of Ridge Regression and Random Forest Regression, using a ‘base’ model (Ridge Regression in this case) to make a prediction and adjusting this using a Random Forest Regression prediction of the residuals. The model therefore capitalises on the benefits of both fitting a linear relationship between variables via the Ridge base model, which is more robust in a changing climate system than other ML techniques 37 , and the flexibility of a non-parametric Random Forest. Data was split into test and training data and RERF hyperparameters were tuned using 5-fold cross validation on the training data only. The split was done based on odd and even years, but validation was also undertaken using various splits (e.g., early and later years) and performance remained similar. Before using the RERF to make projections, it was refitted on the entire dataset using the hyperparameters from cross validation. This gives the best constrained model on the largest possible training data range, still with the objectively best hyperparameter settings for the REFR fit. Merging the ML model with Earth System Model Projections Changes in the future SUHI are investigated by combining the RERF functions learned from observations with climate model projections for future regional changes (i.e., areas surrounding the cities considered) for the predictor variables from the most recent phase of the Coupled Model Intercomparison Project (CMIP)- CMIP6. ESM projections are used to quantify potential future changes in vegetation and climate, so they can then be added into the dataset of predictor variables. A key challenge is that ESM climate projections show different rates of warming due to the forcings, feedbacks and parameterisations used 38 . An alternative approach is to analyse climate at a 2 °C global mean temperature rise from pre-industrial 39 . This is additionally relevant to policymakers as it is easier for those without expertise in climate modelling to understand. Here we use the SSP3-7.0 pathway 40 . To use ESM projections in the RERF, variables are converted according to the following pre-processing steps: Calculate a pre-industrial mean global temperature for each ESM, defined as the mean global temperature from 01/01/1850 to 01/01/1900. Find the 20-year period where the mean global temperature is 2 °C higher than preindustrial baseline. This will be known as future period. Re-grid the ESMs so they are all on the same grid. This is the coarsest grid, CanESM5, which has spatial resolution 2.8 ° latitude x 2.8 ° longitude. Use the ESM outputs from the pre-industrial period to get a baseline for the climate (surface RH and TP) and vegetation (LAI) variable outputs. Use the ESM outputs from the future period (2 °C global mean warming from pre-industrial) to get a projection for the future climate and vegetation variables. Calculate the change in the climate and vegetation variables using these two ESM outputs. This gives a change in LAI, RH and TP for each ESM (5 total). By looking at the difference between pre-industrial and 2 °C warming in the model rather than the absolute prediction for each predictor variable, some bias adjustment is implicitly performed on the ESMs. The changes in the predictor variables are then added to observations. This means the resolution of the variables will remain that of the observations, and not those of the ESMs. This technique is known as the delta change method 41 . An implicit assumption of the approach is that climate forcing will be constant throughout an ESM grid box. Well studied and validated ESMs, with the required variables and scenario available from CMIP6 were chosen. The ESMs are as follows; CanESM5 42 , CNRM-CM6-1 43 , ACCESS-ESM1-5 44 , IPSL-CM6A-LR 45 and UKESM1-1-LL 46 . Declarations Acknowledgments This work was supported by the Natural Environment Research Council and the ARIES Doctoral Training Partnership [grant number NE/S007334/1]. This work used JASMIN, the UK collaborative data analysis facility. Author Contributions All authors contributed to the design of the study. S.B. conducted the analysis and produced the figures. All authors contributed to the interpretation and presentation of the results. S.B. wrote the manuscript. M.J. and P.N. edited the manuscript which was also reviewed by C.M.G. References Heaviside, C., Macintyre, H. & Vardoulakis, S. The Urban Heat Island: Implications for Health in a Changing Environment. Current environmental health reports vol. 4 296–305 (2017). United Nations, Department of Economic and Social Affairs, P. D. World Urbanization Prospects: The 2018 Revision (ST/ESA/SER.A/420) . New York: United Nations https://population.un.org/wup/Publications/Files/WUP2018-Report.pdf (2019) doi: 10.4054/DemRes.2005.12.9 . Horton, R. M., Mankin, J. S., Lesk, C., Coffel, E. & Raymond, C. A Review of Recent Advances in Research on Extreme Heat Events. Current Climate Change Reports vol. 2 242–259 (2016). Vicedo-Cabrera, A. M. et al. The burden of heat-related mortality attributable to recent human-induced climate change. Nat. Clim. Chang. 11, 492–500 (2021). Dang, T. N., Van, D. Q., Kusaka, H., Seposo, X. T. & Honda, Y. Green Space and Deaths Attributable to the Urban Heat Island Effect in Ho Chi Minh City. 108, 137–143 (2018). Huang, H., Deng, X., Yang, H., Zhou, X. & Jia, Q. Spatio-Temporal Mechanism Underlying the Effect of Urban Heat Island on Cardiovascular Diseases. 49, 1455–1466 (2020). Manoli, G. et al. Magnitude of urban heat islands largely explained by climate and population. Nature 573, 55–60 (2019). Goodess, C. et al. Climate change projections for sustainable and healthy cities. 2, 812 (2021). Andrade, C., Fonseca, A. & Santos, J. A. Climate Change Trends for the Urban Heat Island Intensities in Two Major Portuguese Cities. Sustain. 15, (2023). McCarthy, M. P., Harpham, C., Goodess, C. M. & Jones, P. D. Simulating climate change in UK cities using a regional climate model, HadRM3. Int. J. Climatol. 32, 1875–1888 (2011). Lamb, W. F., Creutzig, F., Callaghan, M. W. & Minx, J. C. Learning about urban climate solutions from case studies. Nat. Clim. Chang. 9, 279–287 (2019). Zhou, D. et al. Satellite remote sensing of surface urban heat islands: Progress, challenges, and perspectives. Remote Sens. 11, 1–36 (2019). Bai, X. Six research priorities for cities and climate change. Nature 555, 23–25 (2018). Jones, P. D. & Lister, D. H. The urban heat island in central London and urban-related warming trends in central London since 1900. Weather 64, 323–327 (2009). Ma, L. et al. Changing Effect of Urban Form on the Seasonal and Diurnal Variations of Surface Urban Heat Island Intensities (SUHIIs) in More Than 3000 Cities in China. Sustain. 2021, Vol. 13, Page 2877 13, 2877 (2021). Muller, C. L., Chapman, L., Grimmond, C. S. B., Young, D. T. & Cai, X. Sensors and the city: A review of urban meteorological networks. Int. J. Climatol. 33, 1585–1600 (2013). Cao, J., Zhou, W., Zheng, Z., Ren, T. & Wang, W. Within-city spatial and temporal heterogeneity of air temperature and its relationship with land surface temperature. Landsc. Urban Plan. 206, 103979 (2021). Amani-Beni, M., Chen, Y., Vasileva, M., Zhang, B. & Xie, G. di. Quantitative-spatial relationships between air and surface temperature, a proxy for microclimate studies in fine-scale intra-urban areas? Sustain. Cities Soc. 77, 103584 (2022). Zhang, H., Nettleton, D. & Zhu, Z. Regression-Enhanced Random Forests. JSM Proc. (2019) doi: https://doi.org/10.48550/arXiv.1904.10416 . Apley, D. W. & Zhu, J. Visualizing the Effects of Predictor Variables in Black Box Supervised Learning Models. (2016). Mohajerani, A., Bakaric, J. & Jeffrey-Bailey, T. The urban heat island effect, its causes, and mitigation, with reference to the thermal properties of asphalt concrete. Journal of Environmental Management vol. 197 (2017). Liu, Z. et al. Surface warming in global cities is substantially more rapid than in rural background areas. Commun. Earth Environ. 3, (2022). Liu, J. et al. Heat exposure and cardiovascular health outcomes: a systematic review and meta-analysis. Lancet Planet. Heal. 6, e484–e495 (2022). Harlan, S. L. et al. Heat-related deaths in hot cities: Estimates of human tolerance to high temperature thresholds. Int. J. Environ. Res. Public Health 11, 3304–3326 (2014). Rohini, P., Rajeevan, M. & Mukhopadhay, P. Future projections of heat waves over India from CMIP5 models. Clim. Dyn. 53, 975–988 (2019). Domeisen, D. I. V. et al. Prediction and projection of heatwaves. Nat. Rev. Earth Environ. 4, 36–50 (2023). Moda, H. M., Filho, W. L. & Minhas, A. Impacts of climate change on outdoor workers and their safety: Some research priorities. Int. J. Environ. Res. Public Health 16, (2019). Sherman, P., Lin, H. & McElroy, M. Projected global demand for air conditioning associated with extreme heat and implications for electricity grids in poorer countries. Energy Build. 268, 112198 (2022). Huang, K., Li, X., Liu, X. & Seto, K. C. Projecting global urban land expansion and heat island intensification through 2050. Environ. Res. Lett. 14, (2019). Wan, Z., Hook, S. & Hulley, G. MYD11A2 MODIS/Aqua Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V006 [Data set]. NASA EOSDIS Land Processes DAAC (2015) doi: https://doi.org/10.5067/MODIS/MYD11A2.006 . ESA Land Cover CCI project team; Defourny, P. ESA Land Cover Climate Change Initiative (Land_Cover_cci): Global Land Cover Maps, Version 2.0.7. Centre for Environmental Data Analysis (2019). Muñoz Sabater, J. ERA5-Land monthly averaged data from 1981 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS) (2019) doi: 10.24381/cds.68d2bb30 . National Aeronautics and Space Administration. MOD13A2 - MODIS/Terra Vegetation Indices 16-Day L3 Global 1km SIN Grid. (2021). National Aeronautics and Space Administration. MYD13A2 - MODIS/Aqua Vegetation Indices 16-Day L3 Global 1km SIN Grid. (2021). National Aeronautics and Space Administration. MCD43A3 - MODIS/Terra + Aqua BRDF/Albedo Daily L3 Global – 500m. (2021). National Aeronautics and Space Administration. MCD43A2 - MODIS/Terra + Aqua BRDF/Albedo Quality Daily L3 Global – 500m. (2021). Nowack, P. & Watson-Parris, D. Opinion: Why all emergent constraints are wrong but some are useful - a machine learning perspective. Egusph. [preprint] 1–28 (2024) doi: https://doi.org/10.5194/egusphere-2024-1636 . Eyring, V. et al. Taking climate model evaluation to the next level. Nat. Clim. Chang. 9, 102–110 (2019). Joshi, M., Hawkins, E., Sutton, R., Lowe, J. & Frame, D. Projections of when temperature change will exceed 2°C above pre-industrial levels. Nat. Clim. Chang. 1, 407–412 (2011). Lee, J.-Y. et al. Future Global Climate: Scenario-based Projections and Near-term Information. Clim. Chang. 2021 Phys. Sci. Basis. Contrib. Work. Gr. I to Sixth Assess. Rep. Intergov. Panel Clim. Chang. [Masson-Delmotte , V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger. N. Caud, Y. Chen , 553–672 (2021) doi: 10.1017/9781009157896.006 . Navarro-Racines, C., Tarapues, J., Thornton, P., Jarvis, A. & Ramirez-Villegas, J. High-resolution and bias-corrected CMIP5 projections for climate change impact assessments. Sci. Data 7, 1–14 (2020). Swart, N. C. et al. The Canadian Earth System Model version 5 (CanESM5.0.3). 5, 4823–4873 (2019). Voldoire, A. et al. Evaluation of CMIP6 DECK Experiments Journal of Advances in Modeling Earth Systems. J. ofAdvances Model. Earth Syst. 11, 2177–2213 (2019). Ziehn, T. et al. The Australian Earth System Model: ACCESS-ESM1.5. J. ofSouthern Hemisph. Earth Syst. Sci. 70, 193–214 (2020). Boucher, O. et al. Presentation and Evaluation of the IPSL-CM6A‐LR Climate Model. J. ofAdvances Model. Earth Syst. 12, 1–52 (2020). Mulcahy, J. P. et al. UKESM1.1: Development and evaluation of an updated configuration of the UK Earth System Model. Geosci. Model Dev (2022) doi: https://doi.org/10.5194/gmd-2022-113 . Additional Declarations There is NO Competing Interest. Supplementary Files ExtendedData.docx Extended Data Cite Share Download PDF Status: Published Journal Publication published 02 Feb, 2026 Read the published version in Proceedings of the National Academy of Sciences → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4623186","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Physical Sciences - Article","associatedPublications":[],"authors":[{"id":319348792,"identity":"9f664991-a775-4df4-abeb-f98f7cee4210","order_by":0,"name":"Sarah Berk","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAoUlEQVRIiWNgGAWjYDACCQYGxoYKOJtoLWdI1tLYRooW+dnNxz7OnHc4j7+B+eBtHmK0GNw5ljxz47bDxRIH2JKtidMikWPM+HDb4cQNDDxm0kRpkZ+R/5nx4RyQFv5vxGlhuJHDzLixAWwLG3FagH4xZpxxLD1xxmE2Y8s5RDlsdvNjxp4a68T+9uaHN94Q5TA4YCZN+SgYBaNgFIwCfAAA+dIw3MsPpkwAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-3958-3827","institution":"University of East Anglia","correspondingAuthor":true,"prefix":"","firstName":"Sarah","middleName":"","lastName":"Berk","suffix":""},{"id":319348793,"identity":"3f111647-6984-427a-8d2c-66894efbf7f0","order_by":1,"name":"Manoj Joshi","email":"","orcid":"https://orcid.org/0000-0002-2948-2811","institution":"University of East Anglia","correspondingAuthor":false,"prefix":"","firstName":"Manoj","middleName":"","lastName":"Joshi","suffix":""},{"id":319348794,"identity":"59481b02-829b-46f8-8bde-f4c9ccdf1a31","order_by":2,"name":"Clare Goodess","email":"","orcid":"","institution":"University of East Anglia","correspondingAuthor":false,"prefix":"","firstName":"Clare","middleName":"","lastName":"Goodess","suffix":""},{"id":319348795,"identity":"b52ceff4-276a-4543-83bd-d3c67908011e","order_by":3,"name":"Peer Nowack","email":"","orcid":"https://orcid.org/0000-0003-4588-7832","institution":"Karlsruhe Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Peer","middleName":"","lastName":"Nowack","suffix":""}],"badges":[],"createdAt":"2024-06-22 21:20:08","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-4623186/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4623186/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1073/pnas.2502873123","type":"published","date":"2026-02-03T00:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":59992783,"identity":"de94cbc4-fb24-4cae-82fc-c6144e3bd58d","added_by":"auto","created_at":"2024-07-10 08:52:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":331525,"visible":true,"origin":"","legend":"\u003cp\u003eLocations of selected cities, and projected LST changes for the background regional area and the additional SUHI driven changes with 2 °C warming. Maps show regional changes in mean LST projected by the ESMs, with the additional LST changes in the city projected by the ML model for a) annual values and b) the warm season.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/5ffef5a237980fd5e4ba4bbd.png"},{"id":59993360,"identity":"4e9b1db0-9e2e-4e35-b14f-1d3b411e81db","added_by":"auto","created_at":"2024-07-10 09:00:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":158500,"visible":true,"origin":"","legend":"\u003cp\u003eModelling overview and ML model performance. a) Schematic showing the process to generate projections of changes in SUHI and regional and city LSTs b) Scatterplot of ML SUHI predictions (horizontal axis) versus observations (vertical axis) for the test data. Data was split by alternate years, with test data odd years. Each point represents the monthly mean SUHI observation and prediction for all 104 cities.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/d38457175736cab4d19bf986.png"},{"id":59992780,"identity":"3250a451-a7c9-4e8a-8679-8560085b7258","added_by":"auto","created_at":"2024-07-10 08:52:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":300616,"visible":true,"origin":"","legend":"\u003cp\u003eA closer view of the projected annual LST changes for the background regional area and the additional SUHI driven changes. The map in Figure 1a is split into regions showing a) North and South America, b) Africa and Middle East, c) South Asia, d) China and Southeast Asia. The same figure for the warm season can be seen in Extended Data Figure 3.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/533fa0660a1a2bec5f5678a1.png"},{"id":59992784,"identity":"66ae3059-ed74-4fda-8b8a-ef327d9f2b44","added_by":"auto","created_at":"2024-07-10 08:52:08","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":191639,"visible":true,"origin":"","legend":"\u003cp\u003eProjected LST changes with and without SUHI changes under a 2 °C warming for the 104 individual cities. The red dots show the median projected LST changes for the overall ESM grid cell region. Orange dots show the median ML projected changes in the city LST in additional to this, driven by the changes in the ESM climate variables. A 68% prediction interval for the ML model is shown in black, representing the likelihood that the true value of each projection lies within this range.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/71fb9bc8ae36a9a3bd41efbd.png"},{"id":102048692,"identity":"5d2b2fe1-51bc-4276-9fb3-97aa9c539e57","added_by":"auto","created_at":"2026-02-06 14:30:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1446906,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/be6633d4-0cea-401f-a780-920907ae37df.pdf"},{"id":59992782,"identity":"8a3e7f2e-783c-42a8-bd21-ed074d8e4a1b","added_by":"auto","created_at":"2024-07-10 08:52:07","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3198891,"visible":true,"origin":"","legend":"Extended Data","description":"","filename":"ExtendedData.docx","url":"https://assets-eu.researchsquare.com/files/rs-4623186/v1/bf0f5db150528f8b644cb752.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Amplified warming in tropical and subtropical cities at 2 °C climate change","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe urban heat island (UHI) is a phenomenon whereby the temperature in a city differs from the surrounding rural area, typically being warmer. This leads to increased heat-related health risks for urban inhabitants in comparison to their rural counterparts\u003csup\u003e1\u003c/sup\u003e. In 2018, it was estimated that over half the world\u0026rsquo;s population resided in cities and this proportion is projected to increase to 68% by 2050\u003csup\u003e2\u003c/sup\u003e.\u0026nbsp;Climate change results in rising global temperatures and increased frequency of extreme heat events\u003csup\u003e3\u003c/sup\u003e, which can have severe human health impacts including increased mortality\u003csup\u003e4\u0026ndash;6\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUHIs are influenced by both climate and city attributes (e.g., city area, rural aridity)\u003csup\u003e7\u003c/sup\u003e, all of which can change over time. The constructed model is based on these processes and therefore can assess the consequence of changes in attribute properties on the UHI. A deeper understanding of such shifts in UHI intensities will inform city planners as they design cities aiming to optimize human comfort and health, as well informing adaptation to the impacts of climate change.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eModelling and projecting changes in UHI remains a challenge\u003csup\u003e8\u003c/sup\u003e. Global climate model outputs have spatial resolutions larger than the scale of most cities due to limitations in computational power. Regional climate models have higher resolutions but are also constrained by computational expense, limiting their ability to model many cities simultaneously.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe above limitations mean that projections of the impacts of climate change on the UHI are also limited to either the largest cities\u003csup\u003e9\u003c/sup\u003e, or to smaller cities in certain geographical regions, at a lower resolution\u003csup\u003e10\u003c/sup\u003e. Indeed, much of the current research focus of the UHI is on megacities, which represent just 12% of the urban population\u003csup\u003e11,12\u003c/sup\u003e. Furthermore, several regions of the world are under-represented in the UHI literature, e.g., Africa and South and Central America\u003csup\u003e12,13\u003c/sup\u003e. Typically, as cities expand the intensity of their UHI also grows. However, it is observed that saturation of the UHI with city size occurs in very large cities (e.g., London)\u003csup\u003e14\u003c/sup\u003e. Cities where saturation of the UHI has occurred may respond differently to climate change than those medium-sized cities where this point has not yet been reached. A complete picture of UHI behaviour under climate change can therefore only be gained by addition of the examination of such medium-sized cities.\u003c/p\u003e\n\u003cp\u003eIn this paper we use a process-based machine learning (ML) model which uses earth system model (ESM) projections to make future surface UHI (SUHI) projections under 2 \u0026deg;C global warming for selected medium-sized cities in the tropics and subtropics. ML approaches have been used before in SUHI studies to explore relationships between predictor and target variables\u003csup\u003e15\u003c/sup\u003e, but this work takes the approach further, using the fitted model to predict SUHI and its changes in the future. Such projected changes will be in addition to both temperature change projected in the surrounding area by ESMs, and the currently experienced SUHI.\u003c/p\u003e\n\u003cp\u003eWe use the SUHI- the urban-rural difference in land surface temperature (LST)- as a proxy for the UHI due to limited availability of air temperature measurements. Weather station data is irregular and sparse in coverage, and air temperature sensor networks in cities are rare and have limited temporal coverage\u003csup\u003e16\u003c/sup\u003e. Satellite data, on the other hand, has a high resolution and global coverage. Although the additional effect of advection complicates the coupling between the two, a significant positive correlation exists between urban LST and air temperature\u003csup\u003e17,18\u003c/sup\u003e because LST controls air temperature in the lower layers of the atmosphere.\u003c/p\u003e\n\u003ch3\u003eA novel method for SUHI projections\u003c/h3\u003e\n\u003cp\u003eHere we present our results for the day-time SUHI at 13:30 hours. Results for night-time, when changes in SUHI with climate change are generally much smaller, can be found in the extended material, alongside the observed values of current SUHIs. First, an overview of the selected cities is given. Next, we describe our ML model and evaluate its performance. We then describe the projections made by combining the ML model with ESM outputs.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFrom a dataset of global\u0026nbsp;urban areas\u003csup\u003e2\u003c/sup\u003e, we impose city selection criteria to return a subset containing medium-sized cities with additional restrictions to remove non-climatic influences. For example, coastal cities or those in mountainous regions are not included. Selection criteria are listed in the methods section. The locations of the 104 selected cities are shown in Figure 1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 2a outlines the procedure used to generate the changes in urban LSTs. The ML model, Regression Enhanced Random Forest (RERF)\u003csup\u003e19\u003c/sup\u003e, \u0026nbsp;is set up to predict SUHI magnitudes from factors such as urban-rural vegetation differences and relative humidity. These factors are acquired from satellite and reanalysis data for the model development. Projected changes in these drivers under a global 2 \u0026deg;C warming, obtained from CMIP6 ESM projections, are then used to project changes in the SUHI (see Methods).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur ML model performs well for present day climates across the selected cities, successfully predicting SUHI magnitudes for a range of observed values, as shown in Figure 2b. Across all cities, the test data have overall performance statistics of R-squared 0.87 and RMSE 0.86 \u0026deg;C, giving confidence in the ability of the model to make projections of the SUHI on unseen data. We consider various validation scenarios to ensure a robust model for the required application (see Methods).\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eSUHIs increase with 2 \u0026deg;C warming\u003c/h2\u003e\n\u003cp\u003eFor most of the 104 cities, the current SUHI is projected to become more positive. This is apparent in Figure 1. Under 2 \u0026deg;C warming, 81 % of 13:30 SUHIs are projected to increase in their annual mean. This change is in addition to the regional background warming projected by ESMs also shown in Figure 1. The overall mean of this amplification is 0.4 \u0026deg;C, increasing the overall change in city temperature (urban ∆LST) from 2.2 \u0026deg;C (the ESM regional ∆LST) to 2.6 \u0026deg;C. Cities in the Middle East, India and China all undergo large additional annual warming, as shown in Figure 1a. Figure 3 zooms in on the annual changes shown in Figure 1. In the Middle East increases in SUHI are a particular cause for concern as these regions are already very hot, and also face a considerable increase in ESM-based regional LST. In these areas, the current SUHI is negative (an urban cool island), due to greater vegetation and irrigation in the urban area in contrast to the rural. The projected increase in the urban LST in these regions indicates the SUHI becoming less negative, and in some cases, positive.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIncreases in SUHI magnitude are especially likely to impact on human health during the warmest months of the year. To investigate this, the data was split into four calendar quarters and the warmest season defined as that with the highest mean 2m air temperature for each region. This warm season projected change can be seen in Figure 1b (with a magnified version in Extended Data Figure 3), which shows that 75% of the SUHIs increase, with an overall mean change of 0.3 \u0026deg;C (with ESM ∆LST being 2.4 \u0026deg;C and urban ∆LST 2.7 \u0026deg;C). Warm season increases in SUHI magnitude are particularly noticeable for cities in Northeastern China.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eMajor shifts in highly populated regions\u003c/h2\u003e\n\u003cp\u003eFor many highly populated countries, such as India and China, projected changes in the SUHI are shown to be particularly pronounced in comparison to background levels of warming (Figure 3). For all the studied cities in India, mean LST is projected to increase by an additional 45% above ESM projections of the surrounding area, and in China by an additional 40%. A major reason for this is the influence of vegetation, which is associated with increased cooling due to evapotranspiration. Predictor importances, determined using accumulated local effects\u003csup\u003e20\u003c/sup\u003e, find vegetation to be a strong influencer on the SUHI magnitude (Extended Data Figure 4). ESM projections of large-scale changes to vegetation or moisture availability, which have a cooling influence on rural areas, do not typically affect cities to the same extent, as they are made up of artificial impervious surfaces and drainage systems that carry away surface water\u003csup\u003e21\u003c/sup\u003e. In the areas where there are increases in regional vegetation (ESM projections can be seen in Extended Data Figure 5), the SUHI becomes more positive. Here, these changes in vegetation, which lead to an increased magnitude in urban-rural vegetation difference, are responsible for the largest changes in the SUHI. In parts of Brazil the opposite effect is seen (Figure 3), and the SUHI becomes smaller.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 4\u0026nbsp;summarises how the inclusion of city specific projections can have a substantial influence on the overall ∆LST as a function of ESM ∆LST. Whilst only 3 city regions experience an increase above 3 \u0026deg;C based on ESM LST, 26 cities experience increases in median urban modelled LST above 3 \u0026deg;C. For two of the cities, Patiala, India and Kasur, Pakistan, the additional change in SUHI results in the city ∆LST being twice that of the ESM projection.\u003c/p\u003e\n\u003cp\u003ePrediction intervals, based on the ML model are shown on Figure 4. The cities which have small projected changes in SUHI tend to have the largest prediction intervals, indicating these SUHIs are less influenced by the input climatic variables in the ML model.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWhen changes in SUHI are considered on top of the changes in regional LST, it is clear that almost all of the cities studied undergo larger LST increases than their rural hinterlands. The overall influence of including city-specific projections, rather than simply examining the ESM grid cell, skews the probability distribution of ∆LST towards larger magnitudes for both the annual and warm season mean values (Extended Data Figure 6).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe have investigated the effects of climate change on the daytime SUHI of 104 medium-sized cities in the tropics and subtropics, which are currently home to over 50 million inhabitants. City temperatures are already amplified due to the UHI in all but the most arid regions, and globally all areas face increases in temperature due to climate change. On top of these known factors, we have demonstrated the potential for urban warming to be amplified in many cities, i.e. city LSTs increasing faster than ESM projections suggest. We note that such a trend has already been observed over the last twenty years\u003csup\u003e22\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur results are of immediate relevance to policymakers who will need to account for the increased hazards many urban citizens will face over the coming decades. The cities studied here are located in the warmer parts of the world, which makes this increase even more impactful for human health and the urban environment\u003csup\u003e23,24\u003c/sup\u003e. More generally, medium-sized cities represent a large proportion of global cities with more than 2.5 times as many cities in this category than cities with over 1 million population\u003csup\u003e2\u003c/sup\u003e. Our novel method, which combines state-of-the-art climate change projections with process-based ML models, enables more informed planning for these future risks.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe projected SUHI increases are particularly noticeable in the highly populated regions of north India and northeastern China. This is concerning as both these areas are projected to experience more frequent and intense heatwaves\u003csup\u003e25,26\u003c/sup\u003e. In hot temperatures, outdoor workers are subject to numerous negative impacts of heat exposure\u003csup\u003e27\u003c/sup\u003e and economic impacts should they forgo a day\u0026rsquo;s work. India is projected to require large cooling demands in the future, which is problematic as the infrastructure may not be able to cope with this increased load, and the costs are prohibitive for many\u003csup\u003e28\u003c/sup\u003e. Increased energy usage also brings consequences for climate change mitigation. The need for UHI mitigation and heat adaption in these regions is therefore even more pressing.\u003c/p\u003e\n\u003cp\u003eA caveat of this study is that city expansion has not been considered. Including future urban expansion would likely lead to an even greater warming compared to the above projected changes\u003csup\u003e29\u003c/sup\u003e. Relaxing the city selection criteria to include more cities would make the results more generalisable, although the addition of variables into the ML model would increase complexity. In particular, including coastal cities and those near large waterbodies, where humidity is more important, and an important aspect of human health and comfort, is a source for future research. \u0026nbsp;\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eCity Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe included cities had a population between 300,000 and one million and a latitude of less than 40\u0026deg;. Additionally, the surrounding features (lakes, hills, oceans) of cities were considered in order to control other variables and isolate the impact of climate (this could be relaxed in future work). \u0026nbsp;An overview of the criteria and datasets used is provided below.\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003ePopulation:\u0026nbsp;300,000 to 1,000,000\u003csup\u003e24\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eLocation: \u0026lt; 40 \u0026deg; Latitude\u003csup\u003e24\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eCity distance: \u0026gt; 42 km from any other city with \u0026gt;300,000 population\u003csup\u003e24\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eCoastal distance: \u0026gt; 100 km from shoreline\u003csup\u003e25\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eWater proximity: \u0026gt; 50 km from lakes \u0026gt;50 km wide or \u0026gt; 22 km from lakes \u0026gt;1 km wide\u003csup\u003e26\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eTopography: \u0026lt; \u0026plusmn;150 m (standard deviation) in elevation within 55 km\u003csup\u003e2\u003c/sup\u003e of surrounding area\u003csup\u003e27\u003c/sup\u003e\u003c/li\u003e\n \u003cli\u003eCity area: \u0026gt; 5 km\u003csup\u003e2\u003c/sup\u003e city area in 2002\u003csup\u003e28\u003c/sup\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eGenerating the SUHI and predictor variables\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset used to develop the ML model was generated using satellite and reanalysis data. The workflow diagram for the model build and use can be seen in\u0026nbsp;Figure 2a. For satellite data, cloud contamination thresholds were set so at least 70% of the overall (rural + urban) area and 50% of the urban have usable pixels. For images deemed acceptable, any remaining poor-quality pixels (pre-defined in the datasets via quality flags) were masked in the analysis to promote accuracy.\u003c/p\u003e\n\u003cp\u003eSUHI was quantified as the difference between the monthly mean city LST and the surrounding rural reference area;\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"411\" height=\"61\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere LST\u003csub\u003eurban\u003c/sub\u003e, n\u003csub\u003eurban\u003c/sub\u003e, LST\u003csub\u003erural\u003c/sub\u003e, n\u003csub\u003erural\u003c/sub\u003e represents the LST and number of urban and rural pixels, respectively.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe rural reference area was defined as a rectangular box surrounding the city, where the city takes up the 10% of the area in the centre. The city and all other urban pixels in the area are masked out. The satellite data, including spatial and temporal characteristics utilised for SUHI quantification are outlined below.\u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eTerra LST 8-Day Global (MOD11A2)\u003csup\u003e30\u003c/sup\u003e (1km resolution,\u0026nbsp;available every 8 days from 2002-2020)\u003c/li\u003e\n \u003cli\u003eESA Land Cover Climate Change Initiative: Global Land Cover Maps, Version 2.0.7\u003csup\u003e31\u003c/sup\u003e (300m resolution, each year between 1992 and 2020)\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe predictor variables used (selected based on maximising R-squared) are monthly means of the following.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eRelative humidity (RH)\u003c/li\u003e\n \u003cli\u003eTotal precipitation (TP)\u003c/li\u003e\n \u003cli\u003eUrban enhanced vegetation index (EVI_U)\u003c/li\u003e\n \u003cli\u003eUrban \u0026ndash; rural enhanced vegetation index difference (EVI_D)\u003c/li\u003e\n \u003cli\u003eLog\u003csub\u003e10\u003c/sub\u003e of city area (LOG_AREA)\u003c/li\u003e\n \u003cli\u003eUrban \u0026ndash; rural white sky albedo difference (WSA_D)\u003c/li\u003e\n \u003cli\u003eUrban - rural elevation difference (ELEVATION_D)\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eStandard deviation of urban elevation (STD_ELEVATION_U)\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThese variables are known to influence UHI magnitudes, and are tested through model evaluation techniques such as accumulated local effects\u003csup\u003e20\u003c/sup\u003e (Extended Data Figure 4) to ensure the ML model accurately reflects the physical processes that cause the SUHI to vary. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCity area is calculated using the ESA landcover data\u003csup\u003e31\u003c/sup\u003e\u0026nbsp; and elevation variables using the topography\u003csup\u003e27\u003c/sup\u003e. The additional satellite and reanalysis datasets used are as follows.\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eRH, TP from\u0026nbsp;ERA-5\u003csup\u003e32\u003c/sup\u003e (9km resolution, monthly from 1981 to present)\u003c/li\u003e\n \u003cli\u003eEVI_U, EVI_D from MODIS MYD13A2, MOD13A2\u003csup\u003e33,34\u003c/sup\u003e (1000m resolution, every 16 days from 2002 to 2020)\u003c/li\u003e\n \u003cli\u003eWSA_D from MODIS MCD43A3, MCD43A2\u003csup\u003e35,36\u003c/sup\u003e (500m resolution, daily from 2002\u0026nbsp;\u003cbr\u003eto 2020)\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe urban \u0026ndash; rural difference variables use the same rural and urban areas and equation 1 as SUHI quantification.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeveloping the ML model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe chosen ML model, RERF\u003csup\u003e19\u003c/sup\u003e, is a hybrid of Ridge Regression and Random Forest Regression, using a \u0026lsquo;base\u0026rsquo; model (Ridge Regression in this case) to make a prediction and adjusting this using a Random Forest Regression prediction of the residuals. The model therefore capitalises on the benefits of both fitting a linear relationship between variables via the Ridge base model, which is more robust in a changing climate system than other ML techniques\u003csup\u003e37\u003c/sup\u003e, and the flexibility of a non-parametric Random Forest.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData was split into test and training data and RERF hyperparameters were tuned using 5-fold cross validation on the training data only. The split was done based on odd and even years, but validation was also undertaken using various splits (e.g., early and later years) and performance remained similar. Before using the RERF to make projections, it was refitted on the entire dataset using the hyperparameters from cross validation. This gives the best constrained model on the largest possible training data range, still with the objectively best hyperparameter settings for the REFR fit.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMerging the ML model with Earth System Model Projections\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eChanges in the future SUHI are investigated by combining the RERF functions learned from observations with climate model projections for future regional changes (i.e., areas surrounding the cities considered) for the predictor variables from the most recent phase of the Coupled Model Intercomparison Project (CMIP)- CMIP6. ESM projections are used to quantify potential future changes in vegetation and climate, so they can then be added into the dataset of predictor variables.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA key challenge is that ESM climate projections show different rates of warming due to the forcings, feedbacks and parameterisations used\u003csup\u003e38\u003c/sup\u003e. An alternative approach is to analyse climate at a 2 \u0026deg;C global mean temperature rise from pre-industrial\u003csup\u003e39\u003c/sup\u003e. This is additionally relevant to policymakers as it is easier for those without expertise in climate modelling to understand. Here we use the SSP3-7.0 pathway\u003csup\u003e40\u003c/sup\u003e. To use ESM projections in the RERF, variables are converted according to the following pre-processing steps:\u0026nbsp;\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eCalculate a pre-industrial mean global temperature for each ESM, defined as the mean global temperature from 01/01/1850 to 01/01/1900.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eFind the 20-year period where the mean global temperature is 2 \u0026deg;C higher than preindustrial baseline. This will be known as future period.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eRe-grid the ESMs so they are all on the same grid. This is the coarsest grid, CanESM5, which has spatial resolution 2.8 \u0026deg; latitude x 2.8 \u0026deg; longitude.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eUse the ESM outputs from the pre-industrial period to get a baseline for the climate (surface RH and TP) and vegetation (LAI) variable outputs. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eUse the ESM outputs from the future period (2 \u0026deg;C global mean warming from pre-industrial) to get a projection for the future climate and vegetation variables.\u003c/li\u003e\n \u003cli\u003eCalculate the change in the climate and vegetation variables using these two ESM outputs. This gives a change in LAI, RH and TP for each ESM (5 total). By looking at the difference between pre-industrial and 2 \u0026deg;C warming in the model rather than the absolute prediction for each predictor variable, some bias adjustment is implicitly performed on the ESMs.\u003c/li\u003e\n \u003cli\u003eThe changes in the predictor variables are then added to observations. This means the resolution of the variables will remain that of the observations, and not those of the ESMs. This technique is known as the delta change method\u003csup\u003e41\u003c/sup\u003e.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eAn implicit assumption of the approach is that climate forcing will be constant throughout an ESM grid box.\u003c/p\u003e\n\u003cp\u003eWell studied and validated ESMs, with the required variables and scenario available from CMIP6 were chosen. The ESMs are as follows; CanESM5\u003csup\u003e42\u003c/sup\u003e, CNRM-CM6-1\u003csup\u003e43\u003c/sup\u003e, ACCESS-ESM1-5\u003csup\u003e44\u003c/sup\u003e, IPSL-CM6A-LR\u003csup\u003e45\u003c/sup\u003e and UKESM1-1-LL\u003csup\u003e46\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Natural Environment Research Council and the ARIES Doctoral Training Partnership [grant number NE/S007334/1]. This work used JASMIN, the UK collaborative data analysis facility. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors contributed to the design of the study. S.B. conducted the analysis and produced the figures. All authors contributed to the interpretation and presentation of the results. S.B. wrote the manuscript. M.J. and P.N. edited the manuscript which was also reviewed by C.M.G. \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHeaviside, C., Macintyre, H. \u0026amp; Vardoulakis, S. The Urban Heat Island: Implications for Health in a Changing Environment. Current environmental health reports vol. 4 296\u0026ndash;305 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUnited Nations, Department of Economic and Social Affairs, P. D. \u003cem\u003eWorld Urbanization Prospects: The 2018 Revision (ST/ESA/SER.A/420)\u003c/em\u003e. \u003cem\u003eNew York: United Nations\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://population.un.org/wup/Publications/Files/WUP2018-Report.pdf\u003c/span\u003e\u003cspan address=\"https://population.un.org/wup/Publications/Files/WUP2018-Report.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4054/DemRes.2005.12.9\u003c/span\u003e\u003cspan address=\"10.4054/DemRes.2005.12.9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHorton, R. M., Mankin, J. S., Lesk, C., Coffel, E. \u0026amp; Raymond, C. A Review of Recent Advances in Research on Extreme Heat Events. Current Climate Change Reports vol. 2 242\u0026ndash;259 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVicedo-Cabrera, A. M. \u003cem\u003eet al.\u003c/em\u003e The burden of heat-related mortality attributable to recent human-induced climate change. Nat. Clim. Chang. 11, 492\u0026ndash;500 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDang, T. N., Van, D. Q., Kusaka, H., Seposo, X. T. \u0026amp; Honda, Y. Green Space and Deaths Attributable to the Urban Heat Island Effect in Ho Chi Minh City. 108, 137\u0026ndash;143 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, H., Deng, X., Yang, H., Zhou, X. \u0026amp; Jia, Q. Spatio-Temporal Mechanism Underlying the Effect of Urban Heat Island on Cardiovascular Diseases. 49, 1455\u0026ndash;1466 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eManoli, G. \u003cem\u003eet al.\u003c/em\u003e Magnitude of urban heat islands largely explained by climate and population. Nature 573, 55\u0026ndash;60 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoodess, C. \u003cem\u003eet al.\u003c/em\u003e Climate change projections for sustainable and healthy cities. 2, 812 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndrade, C., Fonseca, A. \u0026amp; Santos, J. A. Climate Change Trends for the Urban Heat Island Intensities in Two Major Portuguese Cities. Sustain. 15, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcCarthy, M. P., Harpham, C., Goodess, C. M. \u0026amp; Jones, P. D. Simulating climate change in UK cities using a regional climate model, HadRM3. Int. J. Climatol. 32, 1875\u0026ndash;1888 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLamb, W. F., Creutzig, F., Callaghan, M. W. \u0026amp; Minx, J. C. Learning about urban climate solutions from case studies. Nat. Clim. Chang. 9, 279\u0026ndash;287 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, D. \u003cem\u003eet al.\u003c/em\u003e Satellite remote sensing of surface urban heat islands: Progress, challenges, and perspectives. Remote Sens. 11, 1\u0026ndash;36 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai, X. Six research priorities for cities and climate change. Nature 555, 23\u0026ndash;25 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJones, P. D. \u0026amp; Lister, D. H. The urban heat island in central London and urban-related warming trends in central London since 1900. Weather 64, 323\u0026ndash;327 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa, L. \u003cem\u003eet al.\u003c/em\u003e Changing Effect of Urban Form on the Seasonal and Diurnal Variations of Surface Urban Heat Island Intensities (SUHIIs) in More Than 3000 Cities in China. Sustain. 2021, \u003cem\u003eVol. 13, Page 2877\u003c/em\u003e 13, 2877 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuller, C. L., Chapman, L., Grimmond, C. S. B., Young, D. T. \u0026amp; Cai, X. Sensors and the city: A review of urban meteorological networks. Int. J. Climatol. 33, 1585\u0026ndash;1600 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao, J., Zhou, W., Zheng, Z., Ren, T. \u0026amp; Wang, W. Within-city spatial and temporal heterogeneity of air temperature and its relationship with land surface temperature. Landsc. Urban Plan. 206, 103979 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmani-Beni, M., Chen, Y., Vasileva, M., Zhang, B. \u0026amp; Xie, G. di. Quantitative-spatial relationships between air and surface temperature, a proxy for microclimate studies in fine-scale intra-urban areas? Sustain. Cities Soc. 77, 103584 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, H., Nettleton, D. \u0026amp; Zhu, Z. Regression-Enhanced Random Forests. \u003cem\u003eJSM Proc.\u003c/em\u003e (2019) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.1904.10416\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1904.10416\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eApley, D. W. \u0026amp; Zhu, J. Visualizing the Effects of Predictor Variables in Black Box Supervised Learning Models. (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohajerani, A., Bakaric, J. \u0026amp; Jeffrey-Bailey, T. The urban heat island effect, its causes, and mitigation, with reference to the thermal properties of asphalt concrete. Journal of Environmental Management vol. 197 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Z. \u003cem\u003eet al.\u003c/em\u003e Surface warming in global cities is substantially more rapid than in rural background areas. Commun. Earth Environ. 3, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, J. \u003cem\u003eet al.\u003c/em\u003e Heat exposure and cardiovascular health outcomes: a systematic review and meta-analysis. Lancet Planet. Heal. 6, e484\u0026ndash;e495 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarlan, S. L. \u003cem\u003eet al.\u003c/em\u003e Heat-related deaths in hot cities: Estimates of human tolerance to high temperature thresholds. Int. J. Environ. Res. Public Health 11, 3304\u0026ndash;3326 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRohini, P., Rajeevan, M. \u0026amp; Mukhopadhay, P. Future projections of heat waves over India from CMIP5 models. Clim. Dyn. 53, 975\u0026ndash;988 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDomeisen, D. I. V. \u003cem\u003eet al.\u003c/em\u003e Prediction and projection of heatwaves. Nat. Rev. Earth Environ. 4, 36\u0026ndash;50 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eModa, H. M., Filho, W. L. \u0026amp; Minhas, A. Impacts of climate change on outdoor workers and their safety: Some research priorities. Int. J. Environ. Res. Public Health 16, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSherman, P., Lin, H. \u0026amp; McElroy, M. Projected global demand for air conditioning associated with extreme heat and implications for electricity grids in poorer countries. Energy Build. 268, 112198 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, K., Li, X., Liu, X. \u0026amp; Seto, K. C. Projecting global urban land expansion and heat island intensification through 2050. Environ. Res. Lett. 14, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWan, Z., Hook, S. \u0026amp; Hulley, G. MYD11A2 MODIS/Aqua Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V006 [Data set]. \u003cem\u003eNASA EOSDIS Land Processes DAAC\u003c/em\u003e (2015) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5067/MODIS/MYD11A2.006\u003c/span\u003e\u003cspan address=\"10.5067/MODIS/MYD11A2.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eESA Land Cover CCI project team; Defourny, P. ESA Land Cover Climate Change Initiative (Land_Cover_cci): Global Land Cover Maps, Version 2.0.7. Centre for Environmental Data Analysis (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMu\u0026ntilde;oz Sabater, J. ERA5-Land monthly averaged data from 1981 to present. \u003cem\u003eCopernicus Climate Change Service (C3S) Climate Data Store (CDS)\u003c/em\u003e (2019) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.24381/cds.68d2bb30\u003c/span\u003e\u003cspan address=\"10.24381/cds.68d2bb30\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Aeronautics and Space Administration. MOD13A2 - MODIS/Terra Vegetation Indices 16-Day L3 Global 1km SIN Grid. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Aeronautics and Space Administration. MYD13A2 - MODIS/Aqua Vegetation Indices 16-Day L3 Global 1km SIN Grid. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Aeronautics and Space Administration. MCD43A3 - MODIS/Terra\u0026thinsp;+\u0026thinsp;Aqua BRDF/Albedo Daily L3 Global \u0026ndash;\u0026thinsp;500m. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Aeronautics and Space Administration. MCD43A2 - MODIS/Terra\u0026thinsp;+\u0026thinsp;Aqua BRDF/Albedo Quality Daily L3 Global \u0026ndash;\u0026thinsp;500m. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNowack, P. \u0026amp; Watson-Parris, D. Opinion: Why all emergent constraints are wrong but some are useful - a machine learning perspective. Egusph. [preprint] 1\u0026ndash;28 (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5194/egusphere-2024-1636\u003c/span\u003e\u003cspan address=\"10.5194/egusphere-2024-1636\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEyring, V. \u003cem\u003eet al.\u003c/em\u003e Taking climate model evaluation to the next level. Nat. Clim. Chang. 9, 102\u0026ndash;110 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoshi, M., Hawkins, E., Sutton, R., Lowe, J. \u0026amp; Frame, D. Projections of when temperature change will exceed 2\u0026deg;C above pre-industrial levels. Nat. Clim. Chang. 1, 407\u0026ndash;412 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, J.-Y. \u003cem\u003eet al.\u003c/em\u003e Future Global Climate: Scenario-based Projections and Near-term Information. \u003cem\u003eClim. Chang. 2021 Phys. Sci. Basis. Contrib. Work. Gr. I to Sixth Assess. Rep. Intergov. Panel Clim. Chang. [Masson-Delmotte\u003c/em\u003e, V., P. Zhai, A. Pirani, S.L. Connors, C. P\u0026eacute;an, \u003cem\u003eS. Berger. N. Caud, Y. Chen\u003c/em\u003e, 553\u0026ndash;672 (2021) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/9781009157896.006\u003c/span\u003e\u003cspan address=\"10.1017/9781009157896.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNavarro-Racines, C., Tarapues, J., Thornton, P., Jarvis, A. \u0026amp; Ramirez-Villegas, J. High-resolution and bias-corrected CMIP5 projections for climate change impact assessments. Sci. Data 7, 1\u0026ndash;14 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSwart, N. C. \u003cem\u003eet al.\u003c/em\u003e The Canadian Earth System Model version 5 (CanESM5.0.3). 5, 4823\u0026ndash;4873 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVoldoire, A. \u003cem\u003eet al.\u003c/em\u003e Evaluation of CMIP6 DECK Experiments Journal of Advances in Modeling Earth Systems. J. ofAdvances Model. Earth Syst. 11, 2177\u0026ndash;2213 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZiehn, T. \u003cem\u003eet al.\u003c/em\u003e The Australian Earth System Model: ACCESS-ESM1.5. J. ofSouthern Hemisph. Earth Syst. Sci. 70, 193\u0026ndash;214 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoucher, O. \u003cem\u003eet al.\u003c/em\u003e Presentation and Evaluation of the IPSL-CM6A‐LR Climate Model. J. ofAdvances Model. Earth Syst. 12, 1\u0026ndash;52 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMulcahy, J. P. \u003cem\u003eet al.\u003c/em\u003e UKESM1.1: Development and evaluation of an updated configuration of the UK Earth System Model. Geosci. Model Dev (2022) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5194/gmd-2022-113\u003c/span\u003e\u003cspan address=\"10.5194/gmd-2022-113\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"urban heat island, surface urban heat island, machine-learning, land surface temperature, climate change","lastPublishedDoi":"10.21203/rs.3.rs-4623186/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4623186/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCities are often warmer than rural surroundings due to a phenomenon known as the urban heat island, which can be influenced by various factors, such as regional climate. Under climate change, cities face not only the challenge of increasing temperatures in their surrounding hinterland, but also the challenge of potential changes in their heat islands. Making projections of future climate at the city scale is difficult given limitations of Earth System Model (ESMs), which has limited studies to a small number of urban areas \u0026ndash; mostly megacities. Here, we address these limitations by applying a novel process-based machine learning model to ESM outputs, to provide projections of changes in land surface temperature (LST) for 104 medium-sized cities (population 300K to 1M) in the subtropics and tropics. Under a 2\u0026deg;C global warming scenario, annual mean LST in 81% of these cities is projected to increase faster than the surrounding area. In 16% of these cities, mostly in India and China, mean LST is projected to increase by an additional 50\u0026ndash;112% above ESM projections of the surrounding area. These findings suggest that the potential impacts of climate change are underestimated at present for millions of people in cities.\u003c/p\u003e","manuscriptTitle":"Amplified warming in tropical and subtropical cities at 2 °C climate change","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-10 08:52:03","doi":"10.21203/rs.3.rs-4623186/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2ba0f38a-50ba-4caa-a725-bab6fce5285c","owner":[],"postedDate":"July 10th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":33765947,"name":"Earth and environmental sciences/Climate sciences/Climate change/Projection and prediction"},{"id":33765948,"name":"Earth and environmental sciences/Climate sciences/Climate change/Climate-change impacts/Environmental health"},{"id":33765949,"name":"Earth and environmental sciences/Environmental sciences/Environmental impact"}],"tags":[],"updatedAt":"2026-02-06T14:30:11+00:00","versionOfRecord":{"articleIdentity":"rs-4623186","link":"https://doi.org/10.1073/pnas.2502873123","journal":{"identity":"proceedings-of-the-national-academy-of-sciences","isVorOnly":true,"title":"Proceedings of the National Academy of Sciences"},"publishedOn":"2026-02-03 00:00:00","publishedOnDateReadable":"February 3rd, 2026"},"versionCreatedAt":"2024-07-10 08:52:03","video":"","vorDoi":"10.1073/pnas.2502873123","vorDoiUrl":"https://doi.org/10.1073/pnas.2502873123","workflowStages":[]},"version":"v1","identity":"rs-4623186","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4623186","identity":"rs-4623186","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}