Insights Into Urban Heat Island and Heat Waves Synergies Revealed by a Land‐Surface‐Physics‐Based Downscaling Method

Researchers have recently focused on the interplay of the urban heat island (UHI) effect and heat waves (HWs). However, the synergies of these two phenomena remains inconclusive at present. To address this gap, this study investigated UHIs and HWs synergies during the last 30 years in the Tokyo metropolitan area, through a unique and novel approach named Land‐Surface‐Physics‐Based Downscaling (LSP‐DS). LSP‐DS integrates the widely used Noah‐Multiparameterization (Noah‐MP) land‐surface model coupled with urban canopy‐process physics, aiming to conduct high‐resolution, long‐term urban‐specific simulations with much less computational resources. Our comprehensive analysis combining observation data and numerous LSP‐DS simulations confirms exacerbated UHIs during HWs. Specifically, HWs amplify the temperature differences between urban and rural environments, which is quantified by UHI intensity (UHII). During HWs, UHII increased more at night in inland areas and more during daytime in coastal areas. HWs present especially a heightened threat to coastal regions where daytime UHII increased by approximately 1°C during HWs. The Bowen ratio can explain the increase in the daytime UHII, and the daytime accumulated storage heat increase during HWs can explain the increase in nighttime UHII. Based on future projections of the increasing frequency of high temperatures, our findings highlight the impending heat‐related health challenges faced by urban residents.


Introduction
Cities are centers of economic activity characterized by more advanced infrastructures and public services, such as roads, transport, health care, education, and cultural facilities.More than half of the world's population (approximately 4.2 billion) currently live in cities, and the urban population is expected to reach two-thirds of the total population by 2050 (Dodman et al., 2022).The growing urban population, especially in unplanned and informal settlements of Asia and Africa, faces heightened climate-related risks, including sea-level rise, thermal stress, tropical cyclones, storm surges, heavy rainfall, and coastal flooding (Q.-V.Doan et al., 2022;Dodman et al., 2022).
Of these, extreme heat stands out as one of the most important and serious urban meteorological hazard (D.N. Tran et al., 2020;Qian et al., 2022).Cities make their own climate, and a well-documented example is the urban heat island (UHI) effect, which is characterized by warmer urban centers than the surrounding rural areas (Santamouris, 2020).This effect is primarily attributed to the replacement of substantial vegetation with impermeable surfaces characterized by low albedo, high heat capacity, and high solar absorption (Y.Li et al., 2020;Santamouris, 2020;Yun et al., 2020).Significant reduction in vegetation reduces evapotranspiration and thus latent heat, leading to higher urban temperatures (D.Li et al., 2019;Manoli et al., 2019;Zhao et al., 2014).Simultaneously, urban structures serve as heat reservoirs, and the complex urban morphology traps radiation and reduces ventilation (Q.V. Doan et al., 2016;Y. Li et al., 2020;Santamouris, 2020;Wang et al., 2023).In addition, the release of anthropogenic heat from human activities cannot be ignored, as emphasized by X. Xu et al. (2018) and V. Q. Doan et al. (2019).
UHI effect has multiple impacts on urban dwellers's well-being and socio-economical activities.For instance, to deal with higher temperatures, humans consume more energy for cooling down consequently releasing more carbon dioxide into the atmosphere (Santamouris et al., 2001(Santamouris et al., , 2018)).In addition, there is an increase in heatrelated mortality and morbidity associated with the UHI effect (Baccini et al., 2008;N. D. Tran et al., 2018; V. T. Nguyen et al., 2022;Vicedo-Cabrera et al., 2021).
Recent research focuses on investigating the synergies between regional-scale extreme heat phenomena like heat waves (HWs) and the local-scale UHI effect (Founda et al., 2015; H. S. Khan et al., 2020;Richard et al., 2021).Many studies reported that UHI effects could be enhanced during HWs (Ao et al., 2019;D. Li et al., 2016;Ramamurthy, Li, & Bou-Zeid, 2017).An et al. (2020) have shown that the differences between urban and rural thermal conditions throughout almost the entire boundary layer were enhanced during HWs, and the temperature difference between urban and rural areas at nighttime reached up to 8°C at a central urban station in Beijing.Similarly, D. Li and Bou-Zeid (2013) revealed that not only do HWs increase the ambient temperatures, but they also intensify the difference between urban and rural temperatures.Despite a growing number of studies, research on the synergies of UHI and HWs remains relatively limited, particularly in terms of geographical heterogeneity, as highlighted by Kong et al. (2021).
To date, several mechanisms have been proposed to explain how UHI effects change during HWs.Most of them are based on land surface energy balance theory.For instance, studies have shown that due to clear skies and reduced water vapor conditions during HWs, the net all-wave radiation typically increases, being roughly 1.5 times higher compared to regular days (An et al., 2020;Black et al., 2004;Jiang et al., 2019;D. Li et al., 2015).The increased urban-rural difference in latent heat flux is also reported to be a key contributor to the increase of UHI during HWs (Z.Gao et al., 2019;Ramamurthy, Li, & Bou-Zeid, 2017).Another potential factor contributing to the synergy between HWs and UHI could be the enhanced anthropogenic heat release during HWs (An et al., 2020;Ao et al., 2019;Z. Gao et al., 2019;X. He et al., 2020).However, the lack of sufficient observation data, including surface heat fluxes, wind profile, etc., hinders the comprehensive understanding of the mechanisms underlying the UHI-HW synergies.Therefore, most recent studies framed the problem using a numerical modeling approach, such as using a coupled regional climate model like the Weather Research and Forecasting Model (WRF) (Chen et al., 2014;Chew et al., 2021;Giannaros et al., 2018).For instance, research by Chew et al. (2021) reveals that HWs don't amplify the UHI intensity, which consistently peaks around 2.5°C during both HWs and regular periods.While coupled Regional Climate Models (RCMs) have proven to be valuable tools for enhancing our understanding of the issue, the intricate processes depicted by RCMs and the inherent model biases (such as in solar radiation and precipitation) interpret the results as challenging and potentially inaccurate.
Therefore, we developed a novel urban downscaling approach named "Land-Surface-Physics-Based Downscaling (LSP-DS)," which provides more accurate simulations of urban land surface energy budgets by using observed atmospheric forcing conditions instead of potentially biased couple models simulated weather variables.Our LSP-DS closely aligns with the recent statistical dynamical downscaling approach developed by DuchÊne et al. (2020).Unlike conventional direct downscaling, which employs coupled numerical weather prediction systems to represent atmospheric motion and associated thermodynamic effects, our tool is closer to, though not a fully diagnostic approach, operating without dynamic feedback to the atmosphere.The overarching science question to be addressed is: how did the HWs in the last 30 years affect UHIs in the Tokyo metropolitan area, Japan?It places exclusive emphasis on the role of land surface heat balance.LSP-DS consists of the utilization of the Noah with Multi-parameterization (Noah-MP) land-surface model (LSM, Niu et al., 2011) coupled to a singlelayer urban canopy model (SLUCM) (Kusaka et al., 2001) within the NCAR High-Resolution Land Data Assimilation System (HRLDAS) framework (Chen et al., 2007;C. He et al., 2023).The offline HRLDAS/Noah-MP will be driven by atmospheric forcing from ERA5 reanalysis, followed by using remote sensing data and observational air temperature data to validate the LSP-DS.Once the model's accuracy and reliability are confirmed, we will further explore the UHIs during HWs.
The paper is organized as follows: the "Method and Data" section offers a comprehensive explanation of the LSP-DS approach and the data sources used; the "Results and Discussions" section presents the analysis of observational data, along with the results of LSP-DS validated against observations, an explanation of the causal relationship between the heat budget and the synergies of UHI and HWs, and an in-depth exploration of the findings.The study concludes with a "Summary" in Section 4.

Land-Surface-Physics-Based Downscaling (LSP-DS)
LSMs are useful tools to simulate the interactions between land surface and atmosphere.By simulating various processes that occur at the land-atmosphere interface, including energy balance, water cycle, carbon cycle, heat and momentum exchange, soil processes, vegetation dynamics, etc., LSMs resolve terrestrial responses to and interactions with the atmosphere, ocean, glacier, and sea ice in the earth system.In the last few decades, LSMs have undergone significant development, and the open-source community Noah-MP is one of the most widely used state-of-the-art LSMs in the world (C.He et al., 2023;Niu et al., 2011).Noah-MP uses multiple physics options for key land-atmosphere interaction processes, which allows the multi-physics model ensemble experiments for uncertainty assessment and testing competing hypotheses.It has been used in many important applications including tackling HWs (Imran et al., 2018;Patel et al., 2022), high-resolution climate modeling (Y.Gao et al., 2017;C. Liu et al., 2017;Rasmussen et al., 2023), crop and agricultural management (X.Liu et al., 2016;D.-M. Nguyen et al., 2023;Warrach-Sagi et al., 2022), urbanization and heat island (Patel et al., 2022;Salamanca et al., 2018;X. Xu et al., 2018).
Urban areas have unique characteristics due to the concentration of buildings, roads, vegetation, and other infrastructure, complex 3-D geometry and human activities, which can significantly impact the interactions between urban and the atmosphere above (Q.-V.Doan et al., 2020Doan et al., , 2023;;A. Khan et al., 2023;Wang et al., 2021Wang et al., , 2023)).Therefore, specific dynamical and thermal parameterizations should be implemented to resolve these complex interactions.In the last 20 years, urban parameterization has developed and can generally be categorized into three main types: slab model, SLUCM (Kusaka et al., 2001;Masson, 2000) and multi-layer UCM (V.Q. Doan & Kusaka, 2019;Hamdi & Masson, 2008;Martilli et al., 2002;Qian et al., 2022).The slab model essentially modifies parameters in the soil-vegetation-atmosphere transfer models based on observations to simulate the dynamic effects and thermal effects of urban environments.This model is very simple and cannot solve the 3-D geometry of urban.SLUCM represents the city with a simplified urban canyon geometry which is represented by the ratio of the height of the building and the width of the road, the thermal properties of the surface materials of buildings and roads including albedo, emissivity, heat capacity, and thermal conductivity are also considered to simulate the thermal effects.It can capture most of the 3-D physical processes, and recent studies suggested that a simple single-layer model is adequate for UHI modeling at a regional scale (Best & Grimmond, 2015;Daniel et al., 2019;Jänicke et al., 2017;Kusaka et al., 2012;Trusilova et al., 2016).
LSP-DS integrated the Noah-MP LSM coupled with both SLUCM and multilayer UCM, through the offline HRLDAS framework, which is illustrated in Figure 1.Noah-MP implemented with SLUCM resolves the response of terrain to the atmosphere, so the input comprises two parts: static land-use land-cover types, soil texture, and terrain features and time-varying atmospheric forcings.Terrain status can be easily generated by using the WRF pre-processing system (WPS), by setting the domain of the study area in the namelist, and WPS will generate a geo_em file including the domain, land-use, terrain soil texture, monthly leaf area index (LAI) and green fraction (GREENFRAC), etc. Monthly LAI and GREENFRAC, valid on the 15th of each month, are linearly interpolated to the daily values.HRLDAS relies on two main components for driving model simulations: set-up files (i.e., initial conditions) and atmospheric forcings.The set-up files, define the initial soil and snow conditions for the simulation, encompassing parameters such as skin temperature, snow depth, soil temperature across four layers, and volumetric soil moisture across the same layers.On the other hand, atmospheric forcings encompass crucial meteorological variables, including specific humidity, temperature, wind components, radiation (surface solar radiation downwards and surface thermal radiation downwards), pressure, and precipitation (surface pressure and total precipitation).HRLDAS creates forcings for offline Noah-MP by interpolating the atmospheric forcings according to the geo_em file both spatially and temporally.Specifically, due to the difference in resolution between forcings and geo_em file, temperature adjustment involves first adapting it to mean sea level based on the geopotential height of forcings.Subsequently, the temperature is further adjusted to a designated height above the ground level based on the topography height.The outputs of HRLDAS/Noah-MP are surface energy budgets, water budgets, soil temperature, soil moisture, surface temperature, air temperature, etc.After using HRLDAS to resolve the response of terrain to the atmosphere, we can get refined, downscaled land surface states.Because of its offline nature, one distinct advantage of LSP-DS is its computationally efficient and a prototype of LSP-DS was used to investigate the UHI and mitigation strategies (M.Gao et al., 2019Gao et al., , 2020;;Monaghan et al., 2014).

Method of Validation and Application
To evaluate the performance of LSP-DS in simulating the UHI effect, validating temperature is essential, as the UHI effect is defined as the temperature differences between urban and rural areas (i.e., the former minus the latter).As outlined in the flow chart (Figure 2), we conducted spin-up simulations to ascertain the appropriate length of spin-up time for the urban areas within our study.To determine this duration, first the air temperatures after the 1-month and 12-month spin-up were compared and found to be identical, second the air temperature after the 1-month spin-up and no spin-up were compared and finally the exact duration of the spin-up was determined to be 15 days.Once the spin-up time was established, we proceeded to verify the reliability and accuracy of LSP-DS.LSP-DS from 1 July to 31 August for the years 1991-2020, excluding the first 15 days of spin-up time, the analysis period spans from 16 July to 31 August for each year.This timeframe corresponds to the midsummer period in Japan, characterized by consistently sunny weather and high temperatures (Japan Meteorological Agency, n.d.-a).Then the verification involves comparing surface temperatures after downscaling with remote sensing data and comparing air temperatures after downscaling with observational data.After ensuring the model's accuracy and reliability, we will delve deeper into studying the UHI Intensity (UHII) (see below for definition).and GLDAS (top right, orange box).Outputs include land surface states such as temperature and energy budget (bottom, green box).The chosen LSM is the high-resolution land data assimilation system (central, pink box), which incorporates the single-layer urban canopy model.

Definition of Heat Waves (HWs) and Extreme Heat
There is not a single universal definition of a heat wave (HW) due to the varying climatic conditions across regions.Generally, the characterization of HWs can vary along three dimensions: temperature metric, intensity (or temperature threshold), and duration (Z.Xu et al., 2016).
Given that our study spans 30 years and focuses only on the second half of summer each year, a HW is characterized here when the surface air temperature surpasses a percentile-based threshold for a minimum of 3 consecutive days.In this study, three specific types of HWs have been defined: daytime HW, nighttime HW, and compound HW: 1. Daytime HW: when the daily maximum temperature exceeds the 75th percentile of the analysis period of that year for at least 3 consecutive days.2. Nighttime HW: when the daily minimum temperature exceeds the 75th percentile of the analysis period of that year for at least 3 consecutive days.3. Compound HW: when the conditions of both daytime and nighttime HW thresholds are met concurrently for at least 3 consecutive days.Thus a HW event occurs when the daily maximum (minimum or both) temperature exceeds the 75th percentile of the analysis period of that year for at least 3 consecutive days.Therefore, an event with more than 3 consecutive days exceeding this threshold is classified as one HW event.
Extreme heat is identified when the air temperature exceeds the 90th percentile threshold, based on the data of the whole analysis period of 30 years, for at least 3 consecutive days.This categorization also has three types, analogous to the HWs definition.
Different criteria were chosen for extreme heat to ensure convenience for further analysis.The HWs are defined by the 75th percentile thresholds of the annual data to avoid the concentration of the detected events in specific warm years and to ensure that the number of events is meaningfully large enough for analysis (a large annual "local" threshold might result in a few or no detected events).For very extreme hot events, however, the 90th percentile threshold for the entire 30-year analysis period is used.In such a case, a significantly large number of detected events is ensured though they are not necessary not occur every year.

Definition of Urban Heat Island Intensity (UHII) and Added Heat Load (AHL)
The UHI effect arises from cities typically exhibiting higher air temperatures than their surrounding rural areas (An et al., 2020;Richard et al., 2021).Consequently, the UHII is defined as the difference in air temperature between urban and rural sites, as shown in Equation 1 (Founda et al., 2015;Rasilla et al., 2019;Ward et al., 2016): where T urban is the air temperature of the urban station and T rural is the air temperature of the rural stations.
The Added Heat Load (AHL), representing the additional heat load caused by HWs, is defined as the difference in UHII between HW and no-HW days, as shown in Equation 2 (Ward et al., 2016): where UHII HW is the UHII during HWs and UHII no HW is the UHII during non-HW days.A positive AHL indicates an increase in UHII during HWs, while a negative AHL signifies a decrease in UHII during HWs.Therefore, AHL can reflect the trend and magnitude of UHII changes during HWs.

Land-Cover, Land-Use, and Study Area
Two land-use data sets are compared for our study area: Moderate Resolution Imaging Spectroradiometer (MODIS) (default in WPS with a 1 km resolution) and the High-resolution Land Use and Land Cover Map Products from the Japan Aerospace Exploration Agency (JAXA) with a 100 m resolution.The MODIS data set presents a significantly larger urban area in Kanto than the JAXA data set (Figure 3).However, upon comparison with remote sensing maps, JAXA's representation appears to be more accurate to the actual landscape.
Therefore, we calibrate the MODIS Kanto land-use data ( 2001) based on the JAXA data set (2018-2020, version 21.11).Because the urban area in the land-use of MODIS fully covers the urban area in the land-use of JAXA, the calibration adheres to the following criteria (Table 1): 1. Urban grids are determined by JAXA's land-use data.
2. For grids categorized as urban in MODIS but not in JAXA, the land-use assignment is derived from JAXA due to their differing category distinctions.3.For grids marked as urban in MODIS but lacking data in JAXA, we retain the original MODIS land-use classification.

Urban Parameters
The thermal effects of shadowing and radiative trapping in urban canyons are primarily associated with the geometry of the canyon which is described by the building height, road width, roof width, and sky view factor in SLUCM (Kusaka et al., 2001).The thermal effect of long-wave radiation emitted from buildings and roads is very sensitive to thermal properties like heat capacity and thermal conductivity.Therefore, these urban parameters are crucial in the simulation of the thermal environment of urban areas using SLUCM and different cities should have distinct parameters.After carefully examining the map, referring to papers which also did research in the Kanto area and conducting tuning, the urban parameters used in SLUCM are as follows: the urban fraction is 0.7, the building height is 9.0 m, roof width is 9 m, road width is 16.5 m, sky view factor is 0.59, heat capacity for roof, wall and road are all 3.2E6 J m 3 K 1 and thermal conductivity is 1.1 Jm 1 s 1 K 1 (Kusaka et al., 2014).Anthropogenic heat emissions are also a crucial factor that demands consideration.Research conducted by Kikegawa et al. (2014) reveals that the highest anthropogenic heat emissions in commercial areas of Tokyo can reach up to 220 W/m 2 , while in residential areas, they typically range from 10 to 20 W/m 2 (0.5-1 km).After  referring to the anthropogenic heat data set AH4GUC, the maximum daily anthropogenic heat is set at 150 W/m 2 (1 km) (Varquez et al., 2021).

Reanalysis Data
ECMWF Reanalysis v5 (ERA5) is the latest climate reanalysis, providing hourly data on many atmospheric, landsurface, and sea-state parameters together with estimates of uncertainty, it is one of the most widely used reanalysis data in the world (Hersbach et al., 2020).ERA5 provides long-time high-resolution hourly data (0.1°× 0.1°) and 137 model-level data which is necessary for UCM.Considering the mean height of buildings in Tokyo, we have employed forcing data at model level 136, which corresponds to approximately 30 m above the ground.We recognize that ERA5 utilizes the σ coordinate system, upon comprehensive examination of temporal and spatial variability at the chosen model level, we observed these variations to be relatively stable within a 0.15m range.Moreover, such minor variations could lead to a temperature impact of approximately 0.0065*0.15°C,which is less than 0.001°C.Given the inherent uncertainties and data precision limitations, such a degree of temperature change is deemed negligible and insufficient to substantially affect our study's outcomes.The other atmospheric data which is necessary for HRLDAS such as radiation, surface pressure and total precipitation are downloaded from the ERA5-Land Data set.In the ERA5 data set, radiation and precipitation are accumulated over 24 hr, de-accumulate to hourly data is necessary before using HRLDAS to create forcing for offline Noah-MP.ERA5-Land Data set also provides the initial condition including soil temperature, soil moisture, snow depth, and skin temperature for HRLDAS.

Observation Data
Automated Meteorological Data Acquisition System (AMeDAS) is a collection of Automatic Weather Stations (AWSs) run by the JMA for automatic observation which is the most widely used observation data in Japan.
AMeDAS began operating on 1 November 1974, AWSs are installed in schoolyards, government yards, agricultural experiment stations, fire stations, water treatment plants, etc. in general.Currently, there are about 1,300 precipitation monitoring stations (at intervals of about 17 km) throughout Japan.Of these, about 840 stations (at intervals of about 21 km) observe wind direction and speed, temperature, humidity, and precipitation, while about 330 stations in regions with heavy snowfall also measure the depth of snow cover.The temperature is automatically collected hourly with an accuracy of 0.1°C.
After considering factors such as data quality, data completeness, altitude, terrain, land-use, environment, and extreme weather, 13 observation stations were selected in our study area (Figure 4, Table 2).Urban stations are in relatively dense urban areas with more than 70% impervious surfaces within 1 km 2 of the station.Rural stations are in a relatively plain area surrounded by natural properties and have minimum impervious surfaces.All these 13 stations are in the plain area excluding the impact of mountains.Further, we divided these stations into two groups: inland stations and coastal stations, including 7 inland stations (5 urban stations, 2 rural stations) and 6 coastal stations (4 urban stations and 2 rural stations).Of these, the Nerima and Tokyo sites were relocated, and we treat the before and after relocation as two different sites.The criterion for classification is that a coastal station is within 1 km of the coast (Table 3).

Remote Sensing Data
Land Surface Temperature (LST) serves as a critical indicator of land surface energy balances.The advent of remote sensing technology has revolutionized the ability to measure LST remotely, supplying invaluable data for a broad spectrum of scientific, environmental, and practical uses.Accurately estimating LST through remote sensing data enhances our comprehension of land surface phenomena and the environmental repercussions of human interventions.
The Terra MODIS LST/Emissivity Daily (MOD11A1, Version 6.1) product offers daily, per-pixel LST and Emissivity (LST&E) at a 1 km spatial resolution within a 1,200 by 1,200 km grid (Wan et al., 2021).This pixel temperature data originates from the MOD11_L2 swath product.
On the other hand, Himawari 8, the 8th in the line of Himawari geostationary weather satellites, is managed by the JMA.Stationed above the equator at 140.7°E longitude, its chief observation scope encompasses East/Southeast  Asia and Oceania.Currently, Himawari-8 offers updates with sub-daily latency, disseminating data on an hourly basis.The temperature in central urban areas is highest, contrasting sharply with the lower temperatures in the western mountain regions.This pattern aligns perfectly with the ERA5 data set and observational data.During the downscaling process, terrain adjustments resulted in a decrease in mountainous temperatures from the initial 22-18°C.At the same time, the application of SLUCM in urban areas has led to an increase in temperatures, rising from 28 to 29°C in the metropolitan center.The metropolitan center, which is less affected by the sea, has the highest temperatures.The high-temperature zone extends from the metropolitan center to the valleys in the

Google map
Note.Inland urban stations require an urban fraction exceeding 0.7, whereas inland rural stations necessitate an urban fraction below 0.3.Coastal stations, regardless of urban or rural classification, must be situated within 1 km of the coastline.northwest.However, extending from the metropolitan center toward the coast, the temperature decreases due to prevailing sea breezes.This behavior is highly consistent with the pattern identified in the ERA5 data set.
Compared to observational data, ERA5 exhibits an underestimation with a mean error of 0.99°C, and LSP-DS shows improved performance with a mean error of 0.4°C suggesting that the results of LSP-DS are closer to the observational values.The root mean square error for ERA5 is 1.75 and 1.32°C for LSP-DS indicates a lower overall level of error after LSP-DS.The correlation coefficient between ERA5 and observations is 0.93, and LSP-DS exhibits a stronger linear relationship with observations with a correlation coefficient of 0.94.This suggests that LSP-DS performs exceptionally well in simulating the observed data trends.
By comparing the surface temperature after LSP-DS with that from remote sensing on 20 August 2020, both the distribution and magnitude are similar (Figure 6).Various land-use and terrain features are clearly visible, especially the high temperatures in urban areas.The surface temperatures after LSP-DS in urban areas are higher than remotely sensing data, while surface temperatures in non-urban areas are mostly lower than remotely sensing data (Figure 7). Figure 8 indicates that at 02:00 local time, the surface temperatures after LSP-DS in urban areas are about 7°C higher than remote sensing data, but surface temperatures after LSP-DS in rural areas are about 2°C lower than remote sensing data, and the results at 14:00 local time are similar except for rural areas is about 5°C lower than remote sensing data.Since the surface temperature is highly sensitive to thermal properties such as heat capacity and thermal conductivity, these temperature differences are mainly due to model parameter uncertainties.
Figure 9 indicates the standard deviation of the daily maximum air temperature and daily minimum air temperature of each station and the correlation coefficient with observation data in a Taylor diagram.In order to make a cross-site comparison of differences and to present the figures clearly and easily understandable, we normalized the standard deviation ahead.The standard deviation of observation data of each station was set to be 1.0, and the normalized standard deviation of LSP-DS results is the fraction of the standard deviation with observation data.The standardized standard deviation for these 13 stations is around 0.85 which is very close to 1.0.The correlation coefficient of the daily maximum air temperature is around 0.95, slightly higher than the daily minimum air temperature which is around 0.9.The correlation coefficient for the daily maximum temperatures of station Kamogawa and station Tsujidou is slightly lower compared to the others, approximately 0.85.These two stations are coastal rural and coastal urban stations, respectively.
Figure 10 compares the LSP-DS results, which represent the average daily maximum and minimum temperatures during each summer for the selected 13 stations, with the observation data and ERA5 data.The results for the 30 years show that LSP-DS effectively captures the air temperature trend during the summer of each year, both for daily maximum and minimum temperatures.The average daily maximum temperatures during summer are about 2°C lower than the observation data for both urban and rural areas.Comparing LSP-DS, ERA5, and observation data, the LSP-DS enhances the ERA5 data and reduces differences with observation data for both daily maximum and minimum temperatures.The only exception is the daily maximum temperature in rural areas which is lower than both observation data and ERA5 data.

Events and Duration of Heat Waves (HWs) and Extreme Heat
Figure 11 presents the statistics of HW events for 9 urban stations over 30 years.Due to the strict definition of compound HWs, both observations and LSP-DS indicate that compound HWs events occur least frequently, at about half the frequency of the other two types of HW events.For all stations, observational data indicate that nighttime HW events are more frequent than daytime events.However, this trend is reversed in the LSP-DS, where daytime HWs occur slightly more frequently than nighttime ones.In general, both data sets exhibit similar numbers of identified HW events: compound HWs are the lowest at about 22 occurrences, while daytime and nighttime HWs are around 50 occurrences each.
Figure 12 presents the total HW days for 9 urban stations over 30 years.Regardless of the type of HW, the total HW days identified from the LSP-DS data is consistently 20-30 days longer than those identified from observational data, or approximately one additional day per year.Consistent with the results of the event statistics, the total HW days of compound HWs remain the shortest, about half that of the other two types, averaging about 3 days per year.In both data sets, the total HW days for nighttime HWs are slightly higher than those of daytime HWs.In summary, both data sets consistently depict total HW days: compound HWs average 3 days per year, while daytime and nighttime HWs average around 7 days per year.
The overall characteristics of extreme heat are similar to HW.However, daytime and nighttime HW occur on average once every 2 years, lasting about four to 5 days each time; compound heatwaves occur approximately once every 5 years, lasting about 3 days each time (Figure not shown).

Urban Heat Island Intensity (UHII)
Figure 13 indicates the diurnal profile of UHII during regular days and during HWs.The UHII of observation data reach its lowest at 09:00 local time then gradually increases until sunset, remains stable at nighttime, and decreases in the early morning.The UHII reach its highest at nighttime indicating the UHI is mainly a nighttime phenomenon.The UHII of LSP-DS has a similar pattern, but reaches the lowest at 06:00 local time and does not keep stable at nighttime but decreases immediately after sunset.As illustrated in Figure 14, according to the LSP-DS results, the 2 m air temperature at urban stations decreases more rapidly than at rural stations which leads to a decrease in the UHII during nighttime.To investigate this further, additional simulations without anthropogenic heat are run.As Figure 15 illustrates, anthropogenic heat is not the reason for the faster decrease of UHII at nighttime.Moreover, we compared the temperatures of individual tiles (vegetation, bare land, and urban) within each grid (Figure 16).It can be found that the main factor contributing to the decrease of UHII at night is the difference in the rate of temperature decrease of non-urban tiles, which is higher in urban areas than in rural areas.
The reason for this difference is mainly due to the difference in forcings between urban and rural grids and also the difference in vegetation type and soil type.Compared with the observational data, the temperature decrease rate of non-urban tile in urban grids is overestimated, especially for vegetation tile, while the temperature decrease rate in rural grids is underestimated, especially for bare land tile.On the whole, the UHII estimated from LSP-DS is higher than that from observations, except for the nighttime.To find how UHII changes during HWs for both daytime and nighttime, daytime UHII and nighttime UHII are defined by 14:00 and 22:00 (local time), respectively.During HWs, the UHII increases from 0.43 to 0.72°C for daytime and increases from 1.2 to 1.75°C for nighttime according to observation data.However, the increase in UHII according to LSP-DS is consistent, with a rise of 0.3°C in both daytime and nighttime.

Added Heat Load (AHL)
AHL is defined as the increase of UHII during HWs, Figure 17 shows the AHL (daytime and nighttime) of 22 pairs of urban stations and rural stations during 6 scenarios of HWs.Except for a few pairs where daytime AHL is negative, all the others are positive, indicating an increase in UHII during HWs.Overall, there is no significant difference in AHL for the same pair under 6 scenarios of HWs.LSP-DS got the same result with observation data.However, there are significant differences between daytime and nighttime, and these vary between pairs.The main indication is that for inland areas, nighttime AHL is higher than daytime AHL, whereas in coastal areas, the opposite is observed with higher daytime AHL compared to nighttime AHL.During the day, the AHL for inland areas is relatively low, around 0.3°C, while it's slightly higher at night, approximately 0.6°C.For coastal areas, the daytime AHL is quite high, around 1°C, and it drops slightly at night to about 0.8°C.During the extreme heat, the daytime AHL reaches up to 2°C.This indicates that during HWs, the UHII increase is most pronounced in coastal areas in the daytime, followed by coastal areas at nighttime, and finally, inland areas during the nighttime and daytime.AHL reflects the different reactions between urban and rural areas, with a positive AHL indicating a greater increase in air temperature in urban areas than in rural areas during HWs and vice versa.Therefore, after a comprehensive comparison of inland and coastal, urban and rural areas, it can be concluded that the order of air temperature increase during HWs is: In further detail, during the daytime, the temperature in inland rural areas is larger than that in coastal urban areas, but at night the reverse is true.The inland urban areas have the largest temperature increase, while coastal rural areas still have the smallest temperature increase for both daytime and nighttime.

Energy Budgets Analysis
Surface energy balance can be expressed as: where Q* is net all-wave radiation.Q E and Q H are latent and sensible heat fluxes, and Q S is ground heat flux, that is, heat fluxed stored in ground, vegetation or urban canopies.
Overall, LSP-DS can capture the spatial variability in the surface heat budget though it uses the same urban parameters, say urban morphology, thermal inertia, etc., over urban grid cells (not shown).These spatial variabilities in heat budgets could be attributed to the difference in forcing data (from ERA5) and the impacts of local urban conditions such as soil type, vegetation type, etc.Note that the LSP-DS approach divided each urban grid into three tiles: vegetation tile, bare tile, and urban tile, and the output represents a weighted average of these three components.Given that the urban parameters remain constant across urban grid cells, three-tile approach explains the variations in surface energy budgets among different sites.
We utilize the LSP-DS simulated energy budget to explain AHL and its variations.We explain the heat budget for two distinct areas, inland and coastal.
In inland areas, LSP-DS consistently simulates lower Q* in urban sites than in rural sites.This pattern is partially attributed to stronger upward longwave radiation due to high surface temperatures in the urban sites compared to the rural ones.However, during HWs, the Q* appears to increase more in urban sites compared to rural sites.For instance, the peak urban Q* (in the daytime) shows an increase of 130 W/m 2 compared to a rise of 108 W/m 2 in rural sites.This result aligns with findings by previous studies conducted in Shanghai (Ao et al., 2019) and Singapore (Chew et al., 2021), which reported Q* increases of 128 and 170 W/m 2 , respectively.This higher Q* increase in urban sites during HWs is possibly explained by the increased incoming shortwave radiation.This phenomenon is likely linked to the lower albedo in urban environments compared to rural ones.Cities typically consist of concentrated asphalt and tar rooftops, which exhibit lower reflectivity compared to the water, grass, and trees that are more prevalent in rural settings (Trlica et al., 2017).Consequently, with the same surplus of incoming radiation during cloud-free HWs, urban areas absorb a greater portion than rural ones, leading to a larger increase in Q*.
To gain insight into more details, we separate heat fluxes into various components, for example, sensible heat flux (Q H ), latent heat flux (Q E ), and ground heat flux (Q S ).We reveal that the evapotranspiration capacity might play a crucial role in AHL, particularly when looking at how surplus Q* transforms into sensible and latent heat.Specifically, in inland urban sites, Q E exhibits a negligible increase of only 1.2 W/m 2 during HWs, compared to a significantly higher increase of 56 W/m 2 in rural sites (Figure 18).This substantial discrepancy can be attributed to the extensive presence of impermeable surfaces in urban environments, which limits their ability to convert the surplus radiation into latent heat.In contrast, rural sites with abundant vegetation and soil moisture still retain the capacity for this conversion.Therefore, a larger portion of the surplus Q* is transformed into Q H in urban sites compared to rural areas, consequently leading to AHL.
From a different aspect, the Bowen ratio indicator, β, which is defined as the ratio of sensible heating to latent heating (Q H /Q E ) is used to express the above phenomenon.A high Bowen ration, that is, β > 1, indicates that more heat is being transferred through sensible heat (warming or cooling the air), while a low Bowen ration, that is, β < 1, suggests that more heat is being transferred through latent heat (due to evaporation or condensation).In inland rural sites during non-HWs, the Bowen ratio is 0.33, meaning latent heating is three times larger than sensible heating.This ratio remains stable during HWs, indicating that even under increased energy input, the vegetation and soil retain their capacity for latent heat conversion.Conversely, in inland urban sites, sensible heat flux predominates due to the absence of vegetation and soil.Here, the Bowen ratio increases from 1.94 during non-HWs to 2.44 during HWs.This signifies a rise in Q H that exacerbates UHI effects during HWs.
Because buildings and roads cover a large portion of urban areas, the urban areas have a higher peak value of storage heat flux, Q S , and more than three times that of the rural areas (Figure 19).When HWs occur, the increase in Q S in urban areas is also more significant.Q S accumulates in the daytime and begins to be released into the atmosphere after around 15:00 local time.In conclusion, the daytime AHL in the inland area is primarily determined by the difference between urban-rural β, while nighttime AHL is mainly influenced by the difference in the increase of accumulated daytime Q S .These results align with what was demonstrated by Ao et al. (2019) and shown by Kong et al. (2021).
Challenges arise when examining coastal areas.While AHL is also observed there, the explanation for it is not straightforward with LSP-DS.Some findings in coastal areas contradict those confirmed in inland areas.One notable contradiction is that the Q* increase in coastal urban sites is lower than that in coastal rural sites (95 vs. 108 W/m 2 ), opposite the results for inland sites.Moreover, a greater decrease in latent heating and a higher increase in sensible heating is seen in rural sites compared to urban ones.These pose difficulty in explaining AHL straightforwardly.
One speculation for the aforementioned contradiction is the influence of dynamical processes.Since air temperature is directly determined by sensible heating transmitted from the surface due to the temperature disparity between the surface and the air, the heat exchange coefficient regulates this relationship.In other words, air temperature is not solely determined by the additional sensible heating amount, it also depends on turbulencecaused exchangeability.These exchange coefficients highly depend on wind speed.Variations in wind speed distribution in coastal sites, along with the presence of sea-land breeze, which is not resolved in LSP-DS, could cause difficulty in explaining AHL over the coastal areas.However, as this study is limited to a specific number of urban and rural sites, it is also challenging to justify the abovementioned claims.Understanding the synergies between HWs and UHIs, along with other dynamic factors such as sea-land breeze, is not straightforward.It requires more comprehensive dynamic numerical simulations, which go beyond the scope of LSP-DS, and necessitates a finer scale and a larger number of observation sites.Evaluation of surface energy budgets is crucial, but unfortunately, direct observations are currently difficult to obtain.We believe it is worthwhile to investigate this issue in future research.

Summary
This study investigated the synergies between UHIs and heat waves (HWs) in the Tokyo metropolitan area of Japan from 1991 to 2020 using a combination of observation analysis and a land-surface-physics-based downscaling (LSP-DS) approach analysis.The UHI effect is represented by its intensity, or UHII, which is quantified as the difference in temperature between urban and rural.To make the quantitative results more reliable, a total of 22 urban-rural station pairs were selected in inland and coastal areas.Heatwaves are defined by continuous 3 days when the temperature is greater than the 75% percentile.Similar to UHII, multiple definitions of heatwaves, that is, based on daily maximum, minimum or compound two, are employed to obtain a robust conclusion.In this study, we define the AHL as an index to measure the increased degree of UHII during HWs compared with non-HWs.
In order to investigate the mechanism behind the AHL, we used a novel approach so-named LSP-DS.LSP-DS is driven by reanalysis data set ERA5 to estimate 1-km gridded land surface meteorological data, and verified against in situ observations before its results on heat fluxes are used to explain the AHL.
The observational analysis for the last 30 years of summer shows that the air temperature observed at stations around Tokyo shows asymmetric warming during the attack of HWs.The highest warming is observed in the inland urban stations, followed by inland rural and coastal urban stations, meanwhile, the warming in coastal rural stations observed the least temperature increase.
This leads to the AHL, that is, the amplification of UHII during HWs, but such amplification is not symmetric between the coastal and inland areas.The AHL becomes pronounced in coastal areas in the daytime (1°C), followed by coastal areas at nighttime (0.8°C), and finally, inland areas during the nighttime (0.6°C) and daytime (0.3°C).
The analysis of energy budgets estimated by LSP-DS reveals that the daytime AHL is primarily determined by the limited capacity for evapotranspiration in urban settings.This limitation restricts latent heating, resulting in a higher proportion of sensible heating being released to the air from the urban surface during heat waves.At the same time, nighttime AHL is mainly influenced by the difference in the increase of accumulated daytime Q S .
The LSP-DS approach exclusively addresses surface physics, omitting considerations of atmospheric dynamics, microphysics, and precipitation processes.Consequently, the reliability of LSP-DS outcomes hinges heavily on the accuracy of forcing data and the precise representation of local conditions, including land use, soil type, and vegetation type, along with the appropriateness of related parameters.Specifically for urban areas, crucial parameters pertain to urban canopy modeling, encompassing aspects like urban fraction, building height, road width, thermal properties, and anthropogenic heat.It is crucial to recognize that land surface models do not consider momentum and energy exchange between neighboring grid points and with the upper atmosphere.This limitation leads to the representation of phenomena such as land-sea breeze, and foehn strongly depends on the quality of the forcing data.Also, due to the coarse resolution of the forcing data, spatial heterogeneity at the subgrid scale in these data is not accounted for.For instance, if in this study, both urban and rural points belong to one ERA5 grid, then they are forced by the same data even though there might exist a difference in this such as due to the UHI effect.Furthermore, uncertainties in urban model schemes, such as the absence of sophisticated treatments for urban physics (e.g., multi-layer vertical profiles of heat exchange, urban park, urban hydrology), contribute to the overall limitations.For urban areas, because the SLUCM uses a tiles average approach (impervious tile and vegetation tile), especially when the only hydrological process considered for impervious tile is evaporation after precipitation using an empirical approach, is not accurate enough to represent the Q E .Also, the tiles average approach did not capture the nighttime UHII well due to the different diurnal profiles of impervious tile and non-urban tile.
This research showcased the effectiveness of utilizing LSP-DS in simulating the detailed interplay between UHI and HWs, with the demonstration for the Tokyo metropolitan area.LSP-DS, by its simplification with a focus on only land surface physics, offers an effective tool to elucidate the phenomena which strongly related to surface heat balance theory.Also, note that this study solely concentrates on LSP-DS, and a comparison between LSP-DS and conventional dynamical downscaling using RCMs like WRF is not within the scope of this paper.These comparative aspects will be addressed in forthcoming studies.

Figure 1 .
Figure1.Land-Surface-Physics-Based Downscaling (LSP-DS) flow chart depicts the input, output, and land-surface model (LSM) selected.Inputs comprise the terrain status from the Weather Research and Forecasting Model Pre-processing System (WPS)-generated geo_em file (top left, purple box) and atmospheric forcings derived from reanalysis data sets like ERA5 and GLDAS (top right, orange box).Outputs include land surface states such as temperature and energy budget (bottom, green box).The chosen LSM is the high-resolution land data assimilation system (central, pink box), which incorporates the single-layer urban canopy model.

Figure 2 .
Figure 2. Research flowchart illustrating the methodology employed in this paper, progressing from left to right along the central axis, following the directional arrows.The central box outlines the spin-up experiments and the validation process for thermal variables output by Land-Surface-Physics-Based Downscaling (LSP-DS).Subsequently, validated thermal variables from LSP-DS will be utilized in the calculation of Urban Heat Island Intensity.

Figure 3 .
Figure 3. Study area and its land-use consist mainly of urban (red), rural (yellow), forest (green), and water (blue).The color bar on the bottom shows the Moderate Resolution Imaging Spectroradiometer (MODIS) category of land-cover and landuse.Panel (a) show the urban land-use (red) in this region from MODIS and panel (b) is from JAXA.The urban areas in MODIS are overestimated, compared to the more accurate JAXA.

Figure 4 .
Figure 4. Terrain and selected observation stations of the study area.The right-hand color bar indicates terrain altitude.The xaxis represents longitudes, while the y-axis depicts latitudes.As indicated in the legend, urban stations are denoted by triangles (△), and rural stations by crosses (×).Red symbols identify inland stations, while yellow symbols represent coastal stations.The corresponding station names are also marked in the figure.

Figure 5
Figure5illustrates the spatial distribution of the mean 2-m air temperature for August 2020 from ERA5 (left) and from LSP-DS (middle), and the spatial distribution of observational data is also shown (right).There is a pronounced enhancement in the delineation of key geographic features including urban centers, croplands, mountains, and coastal areas.Notably, the temperature of water areas (sea, lakes, and rivers), not calculated by Noah-MP, is masked out in the figure.

Figure 5 .
Figure 5.The mean 2-m air temperature (units: °C) (a) before Land-Surface-Physics-Based Downscaling (LSP-DS) (11 km), (b) after LSP-DS (1 km), and (c) observational data in August 2020.The white part in the figure after LSP-DS is the water (rivers, lakes, and sea) which is not included in the LSP-DS.Urban areas have the highest 2-m air temperatures, while mountainous regions covered by forests have the lowest 2-m temperatures.

Figure 6 .
Figure 6.Comparison of surface temperatures (units: °C) obtained after applying Land-Surface-Physics-Based Downscaling (left, 1 km) and surface temperatures derived from remote sensing data (right) on 20 August 2020.The first row is the surface temperature at 20 August 2020 02:00 local time compared with the remote sensing data set of Himawari (2 km); The second row is the surface temperature at 20 August 2020 14:00 local time compared with the remote sensing data set of Himawari (2 km); the last row is the mean surface temperature of 20 August 2020 compared with the remote sensing data set of Moderate Resolution Imaging Spectroradiometer (1 km).Because of the clouds, some parts of the figure of Himawari are missing (white parts on land).

Figure 7 .
Figure 7. Differences between surface temperature obtained after applying Land-Surface-Physics-Based Downscaling and surface temperature from remote sensing data (Himawari, 2 km) for urban areas (left) and rural areas (right) at 20 August 2020 02:00 local time (upper) and 14:00 local time (lower).The surface temperature of Himawari has been interpolated into 1 × 1 km resolution.

Figure 8 .
Figure 8. Differences between surface temperature obtained after applying Land-Surface-Physics-Based Downscaling (LSP-DS) and surface temperature from remote sensing data (Himawari, 2 km) at 20 August 2020 02:00 local time (upper) and 14:00 local time (lower).The left panels display scatter plots of surface temperature for both urban and rural grids, with contour lines representing the probability density for 5 levels.The right panels display box plots of surface temperature for both urban and rural grids, with the x-axis indicating the differences between after applying LSP-DS and Himawari (LSP-DS minus Himawari).

Figure 9 .
Figure 9.Taylor diagram of mean daily maximum 2-m air temperature and mean daily minimum 2-m air temperature at 13 selected observation stations after applying Land-Surface-Physics-Based Downscaling (LSP-DS) from 1991 to 2020.The standard deviation of the observed air temperature for each station is set to 1.0 (orange line), and the standard deviation of the air temperature after LSP-DS is normalized accordingly.The radial axes represent normalized standard deviations, while the arc-axis represents the Pearson correlation coefficient with the observed data.The orange represents the mean daily maximum 2-m air temperature for each station, and the blue represents the mean daily minimum 2-m air temperature for each station.The legend shows the stations indicated by the symbols.

Figure 10 .
Figure 10.Time series showing (a, b) the annual mean of daily maximum 2-m air temperatures (units: °C) and (c, d) the annual mean of daily minimum 2-m air temperatures (units: °C) for each summer (16 July to 31 August, 1991 to 2020) for urban and rural areas separately.Figures for urban areas are averaged across 9 selected urban observation stations and figures for rural areas are averaged across 4 selected rural observation stations.The data is presented for both Land-Surface-Physics-Based Downscaling (LSP-DS) (orange solid lines), observational data (purple dashed lines), and ERA5 (green dashed lines) values.Ninety-five percent confidence intervals error bands of LSP-DS are shown by the shaded regions.

Figure 11 .
Figure 11.Histograms displaying events statistics for three types of heat waves (HWs) (green: compound HW, orange: daytime HW, and blue: nighttime HW) at nine selected urban observation stations from 16 July to 31 August for the years 1991 to 2020.The comparison includes both (a) observed and (b) downscaled data.The x-axis represents the station names, and the y-axis shows the total number of events for each type of HW.The corresponding data labels are provided for each column.

Figure 12 .Figure 13 .
Figure12.The same as Figure11but for total heat wave days.

Figure 14 .
Figure 14.Time series illustrates the diurnal profile of 2 m air temperature of urban stations and rural stations for both observed and downscaled data.The x-axis represents the hour of the day, spanning from 05:00 to 04:00 local time of the following day; the y-axis represents the 2 m air temperature (units: °C).The scatters represent the 2 m air temperature of observational data, and the lines represent the 2 m air temperature after Land-Surface-Physics-Based Downscaling (LSP-DS).The orange segment of each line corresponds to daytime (from 05:00 to 19:00 local time), while the purple segment corresponds to nighttime (from 19:00 to 04:00 local time of the following day).Dashed vertical lines highlight the 2 m air temperature at 14:00 and at 22:00 local time.Ninety-five percent confidence intervals error bands of LSP-DS are shown by the shaded regions.

Figure 15 .
Figure 15.The same as Figure 13; however, subplot (a) includes anthropogenic heat while subplot (b) does not.

Figure 17 .
Figure17.Heat maps illustrate the added heat load (AHL) (units: °C) for 22 pairs of urban-rural stations during 3 types of heat waves (HWs) for both daytime (left) and nighttime (right).Each row corresponds to a pair which is annotated on the left, the urban station name followed by the rural station name.Above the black dashed line are inland station pairs, while below are coastal station pairs.Columns represent AHL during compound HWs, daytime HWs, and nighttime HWs for observed data, as well as during compound HWs, daytime HWs, and nighttime HWs for Land-Surface-Physics-Based Downscaling data.Red denotes a positive AHL (urban heat island intensity (UHII) increase during HWs), while blue denotes a negative AHL (UHII decrease during HWs), with darker colors indicating greater changes.The corresponding values are also provided.

Figure 18 .
Figure 18.Time series illustrate the latent heat flux (Q E ) for four stations: (a) an inland urban station (Nerima (2)), (b) an inland rural station (Ushiku), (c) a coastal urban station (Chiba), and (d) a coastal rural station (Miura) during heat waves (HWs) (solid line) and non-HW (dashed line) using Land-Surface-Physics-Based Downscaling data.The x-axis represents the hour of the day, spanning from 05:00 to 04:00 local time of the following day; the y-axis represents the Q E (units: W/m 2 ).The orange segment of each line corresponds to daytime (from 05:00 to 19:00 local time), while the purple segment corresponds to nighttime (from 19:00 to 04:00 local time of the following day).Dashed lines highlight 14:00 local time, indicative of daytime Q E , and 22:00 local time, representing nighttime Q E .Error bands are displayed.For inland stations, the Q E increased during HWs which is around (a) 1.2 W/m 2 for the urban station and (b) 56 W/m 2 for the rural station.Conversely, coastal stations exhibit the opposite trend.During HWs, the Q E decreases in coastal stations, with the coastal rural station (d) experiencing a greater decrease of 45 W/m 2 compared to the coastal urban station (c) with a decrease of 22 W/m 2 .

Figure 19 .
Figure 19.Time series illustrates the storage heat flux (Q S ) for four stations: (a) an inland urban station (Nerima (2)), (b) an inland rural station (Ushiku), (c) a coastal urban station (Chiba), and (d) a coastal rural station (Miura) during heat waves (HWs) (solid line) and non-HW (dashed line) using Land-Surface-Physics-Based Downscaling data.The x-axis represents the hour of the day, spanning from 05:00 to 04:00 local time of the following day; the y-axis represents the Q S (units: W/m 2 ).The orange segment of each line corresponds to daytime (from 05:00 to 19:00 local time), while the purple segment corresponds to nighttime (from 19:00 to 04:00 local time of the following day).Dashed lines highlight 14:00 local time, indicative of daytime Q S , and 22:00 local time, representing nighttime Q S .Error bands are displayed.In inland areas, during HWs, the daytime accumulated Q S shows a substantial increase of 505 Wh/m 2 (a), and the rural station, which increases by only 98 Wh/m 2 (b).However, in coastal areas, during HWs, the accumulated Q S in the daytime increased by 440 Wh/m 2 in the urban station (c) and increased by 193 Wh/m 2 in the rural station (d).The differences in Q S between inland urban-rural stations are much more significant than those observed in coastal urban-rural stations.

Table 1
The Left Column Is the Categories of JAXA and the Right Column Is the Corresponding Land-Use Categories in Moderate Note.The category water (1) in JAXA only covers the water on land which corresponds to the lake (21) in MODIS.No solar panels (12 in JAXA) exist in the study area.

Table 2
Selected Observation Stations Details Including Location, Type, Station ID, Station Name, Latitude, Longitude, Height, Land  Use, etc

Table 3
Principles for the Selection of the Four Types of Observation Stations, Inland Urban, Inland Rural, Coastal Urban and Coastal Rural, and Their Google Maps Satellite Photographs Within 1 km