1 Introduction

The global urban areas with more than half of the world's population are highly vulnerable to climate change. About 55% of the world's population now lives in urban areas, which is expected to rise to 70% by 2050. India is projected to lead this urban transition along with China and Nigeria (United Nations, 2018). With the increasing urban population, more natural surfaces will be converted into built-up areas, potentially modifying the surface and atmospheric conditions. With the absorption of the incoming solar radiation and lower albedo over the buildings, roofs, roads, and urban canyons, the temperature over the urban areas is relatively warmer than the surrounding rural or suburban areas (Oke et al., 2017; Grimmond et al., 2016; Grimmond, 2007). High concentrations of greenhouse gases (mainly CO2) and aerosols emitted in urban areas from numerous sources enhances this warming (Grimmond et al., 2016). Though covering less than 3% of the global land area, they are a major driving force behind anthropogenic climate change through their extreme demand for energy-intensive anthropogenic activities (Grimmond et al., 2016). It is known that temperature is rising globally due to anthropogenic climate change. When coupled with urbanization, this leads to enhanced warming in urban centres. Past studies have suggested that the contribution of urbanization to large-scale warming is negligible (Jones et al., 1990; Parker, 2006; Peterson et al., 1999). However, the recent Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) report suggests that future urban development will amplify the temperature projections over cities, with their overall impacts being as significant as global warming.

Attempts have been made to separate the various factors contributing to the temperature changes in urban areas. One common technique to estimate the urbanization effect is comparing the temperature trends in urban and surrounding rural areas, using in-situ or satellite measurements. The basic assumption in this analysis is that rural areas are free from the effects of urbanization. However, this assumption may not be valid in all cases (Wang & Yan, 2016). Nevertheless, such estimates are still widely used as an indicator of urbanization.

Over Europe, of the 0.18 °C warming (per decade), a study showed that urbanization leads to 0.0026 °C. In this study, the increase in temperature considering all stations was compared with the stations with very low levels of urbanization (Chrysanthou et al., 2014). Then the difference in trends was attributed to urbanization. A recent study using model simulations to estimate urban-induced surface and climate warming over U.S. cities has shown an overall increase of 0.15 °C due to urbanization. The effect amplifies to 0.24 °C in cities built in vegetated areas and is suppressed by -0.25 °C in arid regions. The obtained trend is comparable with the 0.13 °C/decade of global warming (Bounoua et al., 2021). The overall urban-induced warming was found to be much more prominent in developing countries undergoing rapid growth. Using an optimal fingerprinting technique, (Sun et al., 2016) suggested that out of the 1.44 °C warming observed over China from 1961 to 2013, 0.49 °C, almost one-third is due to urbanization, indicating that urbanization effects regionally may be substantial. Another study using the Empirical Orthogonal function analysis found that 0.77 °C in the observed warming of 1.37 °C over Korea is due to the urban effect, and the greenhouse effect causes the remaining 0.60 °C (Kim & Kim, 2011). In India, several studies related to UHI and its effect on temperature were conducted over multiple cities (Butsch et al., 2017; Mohan et al., 2013, 2012, 2011; Sethi et al., 2022; Sussman et al., 2019; Swain et al., 2017, 2020). However, systematically segregating and quantifying the impact of urbanization and the regional effect (related to climate change) is still unexplored over the Indian region. One widely used method to quantify the land use and land cover (LULC) and urbanization effect on temperature is the observation minus reanalysis (OMR) methodology introduced by Kalnay & Cai, 2003. In India, few studies used the OMR technique to discern the effect of LULC changes on observed temperature trends over different regions (Nayak & Mandal, 2012, 2019; Prijith et al., 2021; Nayak et al., 2021; Nayak, 2021; Nayak et al., 2022). Most of the above studies focused on the attribution of regional temperature changes due to LULC, mainly the conversion from shrubland, forest, or barren land to agriculture. The conversion from natural surfaces to artificial impervious urban surfaces was not considered. However, a similar study conducted over the Eastern state of Odisha showed that out of the 0.3 °C of the warming observed over the region, ~ 25 to 50% is attributed to LULC changes (Gogoi et al., 2019). The LULC-led changes in temperature are amplified over cities by up to 50% (for Bhubaneswar). This study indicated urbanization's effect on surface temperature. However, their spatial pattern (including hotspots), transition, and diurnal differences were unexplored due to a lack of high spatial resolution long-term temperature observations at city scales. In addition, a recent review conducted on urban heat island studies in India concluded that numerical modelling studies to understand UHI and their transition are largely lacking in India (Veena et al., 2020). Therefore, in this study, the Weather Research and Forecasting (WRF) model is used to discern, segregate and quantify the effect of urbanization and regional climate change on surface/air temperature observed over the fastest growing tier-2 city (designated as a smart city for future development), Bhubaneswar in India.

2 Study region

Bhubaneswar, one of India's fastest-growing tier-two (with a population between 50,000 and 1,00,000 as per the 2001 census) cities and the capital of Odisha state, is situated on the eastern coast (see Fig. 1 a and b). Recent studies have shown that during 2000–2014, the urban area over Bhubaneswar has undergone an increase of ~ 83% and a simultaneous decrease in dense vegetation and croplands (Swain et al., 2017) with a SUHI intensity of 1.75 °C. From 2003 to 2017, the land surface temperature (LST) over the urban region increased at the rate of 0.25 °C/year (Anasuya et al., 2019). Sethi et al., 2022, using 20 years of high-resolution MODIS LST data, showed that LST changed by 0.34 °C per decade during daytime and 0.18 °C per decade during nighttime. Besides these, Bhubaneswar also experiences heatwave conditions during summer regularly. In addition, Swain et al., 2020 showed that urbanization-linked extreme rainfall events and their intensity are increasing over Bhubaneswar.

Fig. 1
figure 1

Study location and the LULC changes over Bhubaneswar. a WRF model domain, domain 1(d01), domain 2(d02), and domain 3(d03) at the resolution of 36 km, 12 km, and 4 km, respectively. b The state of Odisha, Bhubaneswar, is marked as a blue star. The default (c) USGS LULC data, (d) MODIS LULC in the model and AWiFS LULC over Bhubaneswar during (e) 2004 and (f) 2015 at a resolution of 30 s

Full size image

Figure 1 (e & f) shows the LULC changes over Bhubaneswar between 2004 and 2015 from the Advanced Wide Field Sensor (AWiFS) LULC dataset. A significant increase can be seen in the urban built-up areas of the city (shown in red) (Fig. 1 e & f). The expansion is mainly towards the city's south, southwest, and northern parts, consistent with the results obtained by Swain et al., 2017. A significant land cover change observable (Fig. 1 e & f) is the conversion of the fallow land into either cropland or urban areas. The deciduous forest cover (over Chandaka forest) has increased over the years. It also appears that there are many scattered water bodies identified in the AWiFs dataset (especially in 2015). These are not artefacts but parts of river Mahanadi and its tributaries that are not seen in 2004 LULC dataset. This may be due to the difference in the acquisition time of imageries (between rainy and non-rainy periods) used for classification. Albeit such minor issues, built-up areas are well classified. In the present study, the model spatial resolution is 4.0 km2. Therefore these little differences are not expected to create significant discrepancies in the results discussed later.

Though LULC from United States Geological Survey (USGS) and Moderate Resolution Imaging Spectroradiometer (MODIS) are shown here (see Fig. 1 c & d), the present study uses an India-specific LULC developed by the National Remote Sensing Centre for model studies. It is clear that both USGS and MODIS show much smaller urban built-up areas (used as default in the WRF modelling system) due to their earlier acquisition dates (more details in Data and Methodology section).

3 Data and methodology

Model simulations were performed using WRF version 4.0. using a nested domain with the outermost domain covering the Indian region, the inner domain covering the eastern part of India, and the innermost domain covering Bhubaneswar with a resolution of 36, 12, and 4 km, respectively (Fig. 1a). The initial meteorological boundary conditions for the model were obtained from the National Centre for Environmental Prediction-Final Analysis (NCEP-FNL) with a temporal resolution of 6 h and a spatial resolution of 1°x1°. The model's default LULC was replaced with Indian Remote Sensing Satellite-P6 (IRS-P6) AWiFS data with a resolution of 30 s. The 24-category default USGS global land cover characterization LULC dataset was derived from April 1992 Advanced Very High Resolution Radiometer (AVHRR) 1 km data, and the more recent MODIS LULC dataset at 1 km resolution in the model is based on the year 2001. Indian Space Research Organisation (ISRO) has created WRF model compatible LULC with 24 categories, at 30-s resolution for each year from 2004–2018, based on AWiFS data. This LULC is an India-specific dataset developed for studies in this region. It is expected to have better quality control and hence confidence over this region, which is clear from the figure (Fig. 1 c to f). Past studies have shown that replacing the default United States Geological Survey (USGS) LULC dataset with the AWiFS LULC dataset leads to better results for simulations (Badarinath et al., 2012; Unnikrishnan et al., 2016; Boyaj et al., 2020; Sahoo et al., 2020).

Since detailed information on the urban morphological parameters of the city is unavailable, the simple bulk parameterization scheme, which is a function of LULC, is used (Taha, 1999). Previous studies have shown that even though the complex Urban Canopy Models (UCM) have potential, they do not outperform the simple slab model (Jänicke et al., 2017). Since the study analyses the general temperature changes over the city, the simple bulk scheme is sufficient, as suggested by Salamanca et al., 2011. The NOAH multi-parametrization land surface model is used since it is found to be better than NOAH for estimating the daily evolution of temperature (Salamanca et al., 2018). The other physics parameterization schemes used are the WRF single-moment 6-class for microphysics (Hong et al., 2005), Dudhia for short wave radiation (Dudhia, 1989), Rapid radiative transfer model for longwave radiation (Mlawer et al., 1997), the Boulac scheme for the planetary boundary layer (Bougeault and Lacarrere, 1989), and the Kain-Fritsch scheme for cumulus parameterization in the first and second domain (Kain & Fritsch, 1990). These physics parameterization schemes were selected based on previous UHI studies conducted over India using the WRF model and have obtained good results (Kedia et al., 2021; Mohan & Bhati, 2011; Sultana & Satyanarayana, 2019; Vinayak et al., 2022). The model was run for 90 days from December to February. The first month is avoided in the further analysis as it is taken as a spin-up period for 2004 and 2015.

A recent study by Sethi et al., 2022, estimated SUHI for Bhubaneswar using 20 years of daily satellite-based LST measurements during different seasons and found that nighttime SUHI is highest during the winter. In addition, rainfall occurrence is lowest during the dry winter period, and hence, cooling or warming effects due to both surface and atmospheric moisture could be assumed to be minimal during this period. Several studies have indicated that the UHI effect is discernible during nighttime (Masson et al., 2020; Mathew et al., 2018; Sethi et al., 2022). Based on the above considerations, winter and nighttime temperatures are primarily considered for the simulation and further analysis.

To discern the effect of LULC changes, the following simulations were carried out,

  1. 1.

    Control simulations for the years 2015 (1st December 2014 to 1st March 2015) and 2004 (1st December 2003 to 1st March 2004) (refered to as CNTL2015 and CNTL2004 hereon) with the initial and lateral boundary (LBC) conditions (including LULC) of 2015 and 2004, respectively. The two periods (2015 and 2004) were chosen only based on the availability of LULC datasets from NRSC.

  2. 2.

    Two additional experiments were carried out, i. EXP1 with the Initial and Boundary conditions of 2015, but LULC replaced for 2004 and ii. EXP2 kept the initial and boundary conditions of 2004, but LULC was replaced by 2015.

These simulations and their time were carefully chosen to minimize other compounding factors that may influence surface temperatures, such as moisture, clouds and rainfall. The three hourly meteorological observations (2-m air temperature) made by the Indian Meteorological Department (IMD) were used for model validation. The MODIS Land Surface Temperature (LST) products MOD11A2 during the months of January and February for 2004 and 2015 were used to compare the spatial pattern of LST simulated by the model with the satellite observations.

4 Results and discussions

4.1 Validation of the model

Figure 2 compares model simulations and observations from the IMD Automatic Weather Station (AWS) data at 20.15°N latitude and 85.50°E longitude. The model-simulated values of 2-m air temperature agree well with observations with a correlation coefficient of 0.93 (0.94) and a mean absolute error of 2.52 °C (2.48 °C) for 2004 (2015). The root mean square error (RMSE) was 3.16 °C (3.00 °C), and the index of agreement was 0.88 (0.88) for 2004 (2015).

Fig. 2
figure 2

Comparing the model simulated temperature values with observations. a Scatter plot showing the relationship between model-simulated temperature and the observations for 2004 and 2015. b Diurnal temperature variation over the city from model and observations for 2004 and 2015

Full size image

Considering the diurnal variation in temperature over the city, the model overestimates the temperature values with a mean bias of 2.21 °C for 2004 and 2.22 °C for 2015. Note the similarity in the overall bias for both years. The simulated values are closer to observations at higher temperatures, the minimum bias is obtained at 11.30 Indian Standard Time (IST), and the bias is higher at lower temperatures; the maximum values were obtained at 5.30 IST. The model can capture the variability albeit the bias, especially at night. Overall, the model can reasonably simulate the 2-m air temperature.

Figure 3 shows the spatial pattern of LST simulated by the model. The model well simulates the spatial pattern of temperature with high temperatures over the urban region (Bhubaneswar city (Fig. 1e) and low temperatures over the forest region (Chandaka (Fig. 1 e)). The new hotspots (Sethi et al., 2022) outside the city observed in 2015 are well simulated and are seen in observations. It is observed that the model is showing both an enhanced LST and UHI effect compared to satellite-based estimates (shown as contour lines), indicating a positive bias that is similar in both years. The city's overall spatial patterns, directional growth, and their effect on LST are reasonable. Minor differences may arise due to the instantaneous snapshot nature of satellite imageries (10:30 pm, IST) compared to the three hourly average (11:30 pm, IST) outputs of the model used in this analysis. The major objective in showing Fig. 3 is not to discern the absolute values of the UHI, SUHI, or LST but to highlight that the model simulates the spatial pattern of SUHI reasonably well. It may be noted that all the results discussed in the following sections rely on the differences between simulations. Therefore, the biases observed in Figs. 2 and 3 are expected to not change the conclusions. In the next section, the temperature and its changes are further discussed.

Fig. 3
figure 3

Comparing model simulated surface skin temperature with MODIS Land surface temperature. (a) 2004 (b) 2015. The color bar shows model simulated values, and the contour shows the MODIS (Terra)-LST values

Full size image

4.2 Temperature changes over Bhubaneswar

This section discusses the surface and 2-m air temperature simulated by the model and its changes between 2004 and 2015. The land surface temperature (LST) represents the surface radiant temperature (using which the SUHI is estimated), which mainly depends on the surface properties. In comparison, the air temperature (using which the UHI is estimated) represents the temperature at 2 m height controlled by the surface sensible heat flux and air advection from surrounding regions.

In Fig. 4, an overall higher nighttime air (a & b) and surface temperature (d & e) can be seen over Bhubaneswar city (urban domain in Fig. 1 e & f). The temperature is gradually decreasing as we move away from the urban centre, highlighting a clear discernible SUHI/UHI for both 2004 and 2015.

Fig. 4
figure 4

The model simulated temperature over the city and surroundings. Night time temperature at 2 m for (a) 2004 (b) 2015 and (c) Difference (2015–2004 (b-a)), Night time land surface temperature for (d) 2004 (e) 2015 and (f) Difference (2015–2004 (e-d)). The black dots represent statistical significance at a 90% confidence level

Full size image

On small spatial scales, the incoming solar radiation can be assumed to be almost similar in both urban and surrounding rural areas. The higher heat capacities and the lower evaporation over the dry impervious urban surfaces lead to the formation of UHI/SUHI, with a minor contribution from the urban–rural albedo differences (Masson et al., 2020). In the model, these properties of urban surfaces are incorporated by reducing the albedo of the urban surfaces to 0.15. The volumetric heat capacity and soil thermal conductivity are 3.0 × 106 Jm3K−1 and 3.24 Wm−1 K−1, respectively. This is slightly higher than concrete and asphalt materials' volumetric heat capacity and soil thermal conductivity.

Over the rural areas (grids other than urban from Fig. 1), the air temperature values exceed the LST values, and the LST exceeds the air temperature over the urban areas. Similar results have been obtained for Birmingham, a city in U.K. using observations and for the city of Berlin, Germany using WRF model simulations. A study conducted over Birmingham using observations suggests that over the urban areas, air temperature is still higher than the LST with significantly less difference compared to rural areas.. Whereas, the model simulations over Berlin suggest that the air temperature and LST are close to each other over urban centres, and LST exceeds air temperature at some points (Azevedo et al., 2016; Li et al., 2018). The contrasting LST and air temperature patterns over urban and rural areas are attributed to the increased thermal capacity over urban surfaces (Azevedo et al., 2016).

To the west of the city, there is a cool island due to the presence of the Chandaka forest. The cool island is more discernible in LST than in air temperature because the spatially continuous LST is more closely related to the biophysical changes in the surface (Winckler et al., 2019). Though the forests have lower albedo and absorb more shortwave radiation during the day, similar to urban areas, the warming is counteracted by the more significant latent heat loss due to higher evapotranspiration, which is absent over the urban dry islands (Li et al., 2015).

In addition to showing a higher air and surface temperature over the urban areas, there is also a marked increase in the area of higher temperatures (shown in red) between 2004 (Fig. 4 a & d) and 2015 (Fig. 4 b & e). This indicates that the simulations show not just the UHI or SUHI but also its spatial expansion with time. In 2015, there are a couple of additional urban hotspots in the southern part of the city but still within the planned development area of Bhubaneswar (Fig. 1). There is also more warming in 2015 in comparison to 2004, which is visible in Fig. 4 (c & f). Overall, the difference between 2015 and 2004 shows enhanced warming over newly developed urban areas and spatially homogeneous warming in the existing urban areas and the surrounding regions.

Over the urban areas, the nighttime air temperature has increased by 0.58 °C (21.58–22.16 °C), and in the surrounding rural areas, the increase is around 0.72 °C (18.95–19.67 °C). Considering the LST changes, in urban areas, the growth is around 0.88 °C (22.4–23.28 °C), and in rural areas, it is almost 1 °C (17.15–18.14 °C). The rate of warming in the rural surroundings appears higher than the rate in the urban areas. The urbanization effect over a city is generally quantified using the UHI intensity, the temperature difference between the urban and the rural areas. In this case, the UHI and SUHI values for 2004 and 2015 decreased from 2.62 in 2004 to 2.49 °C in 2015 and 5.24 (2004) to 5.13 °C (2015), respectively.

The spatial plot shows the highest warming over the newly urbanized regions. Still, this spatial signature is lost while averaging for all the urban grids, leading to a misconception that the UHI effect is decreasing even though the urban area has expanded. There is warming over the whole domain and enhanced warming over the newly urbanized areas. Overall, we encounter warming due to three possibilities.

  1. i)

    Urbanization over the newly developed urban areas,

  2. ii)

    Due to large-scale changes, including climate change which we call regional/decadal change explicitly to avoid confusion with the term climate change considering the short time difference in the simulations (2004 and 2015)

  3. iii)

    The potential effect of urban areas on the temperatures of the surrounding rural areas.

Hereafter, we use the term local/LULC and regional to refer to these changes.

4.3 Quantifying regional and LULC effects on temperature

Generally, the effect of urbanization is discerned using the difference between urban and rural temperatures. While using in-situ observations, it is difficult to fully segregate urban and rural stations (Sun et al., 2016). While using satellite observations, it is often difficult to define the urban boundary because most Indian cities border on peri-urban areas (Martin-Vide et al., 2015). Modelling is an effective tool to separate these local and regional effects. Previously, a modelling study conducted over North-West India used WRF model simulations to quantify the temperature changes observed over the region due to the land-use changes, mainly the conversion of barren land to open scrubland and scrubland to cropland (Prijith et al., 2021). WRF simulations were carried out by Lal et al., 2021, over India, for the period 2009–2015 with the LULC of 2002 and LULC of 2015, and the temperature difference between these two simulations was attributed to LULC changes. In the above studies, the conversion of natural land cover to urban areas is not considered. Bounoua et al., 2021 used model simulations to separate and quantify the effect of urban-induced land cover changes and climate change on surface temperature over multiple U.S. cities but didn't consider the spatial pattern of these changes. Here, the study is conducted over a much smaller region, over a particular city. The regional effect and local/LULC effect on temperature are quantified, and the spatial pattern of regional and local effects are analyzed. The details are provided in Table 1.

Table 1 Details of the methodology used for segregating into regional and LULC effects
Full size table

The regional effect refers to resultant change in temperature as a consequence of all factors whereas, LULC effect refers to changes induced by LULC alone. Regional effect is obtained as the difference between simulations run with same LULC and LULC effect is obtained as the difference between simulations with same LBC (Fig. 5).

The combined effect of climate change and LULC changes shows an increase in temperature (land surface and 2-m air) over the whole domain, with the highest increase over the city's periphery (Fig. 4 c & f). While considering only the regional effect, the temperature increase is more or less homogeneous over the entire domain irrespective of the LULC (Fig. 5 (a&c) & (e & g)). The obtained pattern is similar in the case of both experiments and for both air temperature and land surface temperature, with ~ 0.5 °C temperature increase. The newly formed urban regions show the highest increase in temperature, where the LULC effect (Fig. 5 ((b&d) & (f&h))  amplifies the temperature rise caused by the regional effects of climate change. Over the city centre, there is not much change due to LULC because the city region is already developed.

Fig. 5
figure 5

The LULC and regional effects on temperature. Night-time temperature changes, a Regional effect(EXP1-CNTL2004), b Local/LULC effect(CNTL2015-EXP1), c Regional effect (CNTL2015-EXP2), d Local/LULC effect (EXP2-CNTL2004). e-f,  same as (a-d) but for air temperature

Full size image

Previously a study conducted over Jakarta, the capital city of Indonesia showed that the global effect is more or less similar over all land use classes (Darmanto et al., 2019). Considering the land use and land cover effects, the highest increase in temperature is observed over the periphery of the city and the newly developed urban centers towards the south of the existing city (Fig. 5 (b&d) & (f&h)). The local effects over the newly urbanized areas seem more substantial than the global warming signal making the region more vulnerable to climate change (Darmanto et al., 2019).

To quantify the effect of LULC and regional change, the temperature values over the region inside the box in Fig. 5 (d) were averaged. There was an overall increase of ~ 1.40 ± 0.16 °C (1.9 ± 0.37 °C) in the city's nighttime air temperature (LST) from 2004 to 2015 (Fig. 6). The regional and LULC effect is discerned, and the relative contribution is quantified from both experiments. Out of the 1.40 °C (1.9 °C) increase in temperature between 2004 and 2015, Method1 shows that the regional effect causes a 0.56 ± 0.03 °C (0.52 ± 0.03) rise and LULC effect causes a 0.84 ± 0.16 °C (1.39 ± 0.37 °C) rise. Method2 suggests that the regional effect causes an increase of 0.44 ± 0.02 °C (0.35 ± 0.07 °C), and the LULC effect causes an increase of 0.97 ± 0.16 °C (1.56 ± 0.4 °C). Thus during nighttime, the local effects seem to be dominant with a relative contribution of almost 60–70% in the rise in air temperature and almost 70–80% rise in the surface temperature compared to regional effects. The interesting point here is that even for a tier-2 city, the local LULC changes dominate (up to 70 to 80%) in terms of its effect on both surface and air temperature. These numbers are similar to past observational studies that indicated up to 50% additional warming (air temperature) due to urbanization during the period 2000–2010 (Gogoi et al., 2019) and up to 62% (LST) using differences in urban–rural trends from satellite measurements (Sethi et al., 2022, Personal Communication). Thus, the model results discussed are consistent with other studies that used different observational methodologies.

Fig. 6
figure 6

Quantifying the temperature changes. (a) The difference in air temperature and surface temperature in degree Celcius (b) difference in temperature in terms of percentage

Full size image

Overall, the present study shows compelling evidence that local/urbanization-linked effect overwhelms (60–70% & 70–80%) the total changes in temperature (air and land-surface) over the city of Bhubaneswar. In the era of anthropogenic climate change, the dominant role of local effects also points to the potential for local mitigation/intervention efforts to reduce urban-induced warming.

5 Summary and conclusions

The model simulation reasonably replicates the UHI effect and its changes during 2004 and 2015. The region experiences a nighttime UHI of ~ 2.5 °C and SUHI of ~ 5 °C. The nighttime air temperature increased by almost 1.4 °C over the urban area between 2004 and 2015. Though the warming signatures mainly concentrated over the city, the changes appear to be dominated in the periphery of the urban domain surrounding the city, where recent growth has occurred. Our analysis reveals that ~ 60–70% of the overall increase in air temperature and ~ 70–80% of the overall increase in surface temperature between 2004 and 2015 is due to local LULC changes (urbanization). The remaining ~ 30–40% increase in air temperature and ~ 20–30% increase in surface temperature are due to other regional climate effects (loosely may be attributed to climate change).

The study implies that the local LULC effects have a vital role in rising temperature over the cities; local LULC modification can control more than half of the increase in temperature. In addition, the small size of cities also means that mitigation efforts are achievable on shorter time scales, with potentially cooler, comfortable cities.