2.1. Study area

Kuala Lumpur (Figure 1), the capital city of Malaysia, was selected as the pilot study area due to its rapid urbanisation, dense built environment, and high susceptibility to UHI effects and extreme heat events (Miniandi et al., 2024). Located in the Klang Valley (3°08′24″N, 101°41′24″E), the city experiences a tropical rainforest climate with persistently high temperatures, humidity and bimodal rainfall driven by the monsoon system. In recent years, Kuala Lumpur has also experienced extreme rainfall, frequent flash floods and steadily rising temperatures, further intensifying climate-related risks (Zawawi et al., 2024). The metropolitan region contains a heterogeneous urban fabric, including high-rise commercial districts, medium-density residential zones, and peri-urban green and water bodies, making it well-suited for LCZ classification and UHI assessment. Due to their dense populations, cities like Kuala Lumpur are not only major contributors to global greenhouse gas (GHG) emissions. However, they are also extremely vulnerable to the impacts of climate change, including heatwaves, floods, severe storms and droughts. These overlapping risks make Kuala Lumpur an ideal case study for investigating the interactions between urban morphology, UHI dynamics and extreme heat under current and projected climate conditions.

2.2. LCZ classification

The LCZ classification was conducted using the LCZ Generator web platform (Demuzere et al., 2021), with training areas derived from building footprint and LULC datasets. This web tool has been widely used in previous studies, including those by Abougendia (2023), Senyel Kurkcuoglu et al. (2025) and Rahmani and Sharifi (2025), to assess UHI intensity across different LCZ classes. Building footprint data provided urban morphological characteristics such as density, coverage and height, which were critical for distinguishing compact and open built-up LCZ types. In contrast, LULC data represented natural surfaces including vegetation, bare soil and water. To ensure balanced training samples, large homogeneous polygons (>1.5 km²) were reduced to a radius of ∼350 m following recommended WUDAPT protocols. The training areas, together with a suite of Earth observation predictors (Landsat, Sentinel-1/2, terrain and canopy height features), were used as input to a random forest classifier implemented in Google Earth Engine. The classification produced LCZ maps at 100 m spatial resolution, along with a Gaussian-filtered version and class probability layer.

The LCZ classification consists of 17 standardised classes representing urban and natural land-cover types based on surface cover, building structure and land use. Urban classes include compact high-rise (LCZ 1), compact mid-rise (LCZ 2), compact low-rise (LCZ 3), open high-rise (LCZ 4), open mid-rise (LCZ 5), open low-rise (LCZ 6), lightweight low-rise (LCZ 7), large low-rise (LCZ 8), sparsely built (LCZ 9) and heavy industry (LCZ 10). Natural land-cover classes comprise dense trees (LCZ A), scattered trees (LCZ B), bush or scrub (LCZ C), low plants (LCZ D), bare rock or paved surfaces (LCZ E), bare soil or sand (LCZ F) and water (LCZ G) (Stewart & Oke, 2012).

2.3. UHI derivation

The UHI intensity was derived using both remote sensing and numerical modelling approaches. In the first approach, MODIS land surface temperature (LST) data from Terra and Aqua satellites were processed in Google Earth Engine (GEE) to extract daytime and nighttime LST for the period 2000–2024. Urban and rural reference areas were identified based on the LULC classification. Built-up areas were considered urban references, while vegetated land cover classes, such as forest and agricultural areas, were used as rural references. The rural reference areas were selected outside the urban core to represent background surface temperature conditions. UHI intensity was calculated as the difference between urban and rural LST, where positive values indicate warmer urban surfaces relative to surrounding rural areas. In the second approach, the WRF model was employed to simulate near-surface skin temperature (TSK). The urban canopy model (UCM) was coupled with the WRF software version 4.5 for the simulation of the urban climate of Kuala Lumpur city. The study domain adopted a three-level nested configuration, with d01 (9 km × 9 km) covering the Malay Peninsula, d02 (3 km × 3 km) and d03 (1 km × 1 km) focused on Kuala Lumpur (Figure 2). This nested domain configuration enables large-scale atmospheric processes to be captured while providing high-resolution simulations over the study area. April 2024 was chosen as the case study period due to several occurrences of heatwaves during that month in the city (NADMA, 2024). The physics setting used for the WRF is the tropical suite (UCAR, 2025).

Three urban parameter configurations were tested to represent the LCZ classification within WRF: (i) Single Layer Urban Canopy Model (SLUCM), (ii) WRF with default LCZ parameters, and (iii) WRF with localised LCZ parameters (LCZ_local) tailored to the study area’s urban morphology, derived from the CGLC-MODIS-LCZ_100m dataset. These simulations were used to estimate UHI intensity during a documented heatwave event in April 2024 and to assess the sensitivity of future UHI conditions to a +1 °C atmospheric temperature perturbation. The +1 °C perturbation was adopted as a sensitivity experiment to evaluate the response of urban thermal conditions to incremental warming rather than to simulate future climate under specific emissions pathways. Model performance was evaluated by comparing simulated TSK with MODIS-derived LST as the reference dataset. No ground-based observations or meteorological station data were used for model validation. The comparison was conducted using statistical metrics including mean absolute error (MAE), root mean square error (RMSE), bias, correlation coefficient (r), coefficient of determination (R²), and Willmott’s index of agreement (d), calculated using Python software.

2.4. Heatwave analysis

A heatwave is defined as a period of unusually high temperatures that poses significant risks to human health, infrastructure and ecosystems, characterised by sustained excess heat accumulation over multiple days. In this study, heatwaves were evaluated using the Excess Heat Factor (EHF), a metric designed to monitor and forecast heatwave intensity and severity with a focus on human health impacts (Nairn & Fawcett, 2015). The EHF was calculated from daily mean temperature (DMT) values derived from the hourly ERA5-Land reanalysis dataset (Hersbach, 2020). ERA5-Land provides hourly air temperature at 2 m spatially interpolated to a ∼9 km grid, offering a consistent long-term climate record suitable for heatwave assessment. Previous ecological and climatological studies conducted across Southeast Asia have extensively validated and applied ERA5-Land (Linarka et al., 2025), confirming its reliability as a proxy for interannual climate variability. The DMT, denoted as T_i, which is the average of the daily maximum and minimum temperatures for day i, reflecting the physiological impact of a hot night following a hot day. The EHF was calculated using two excess heat indices over a three-day period (TDP). The first index, the significance index, is given by:

where T₉₅ is the local 95^th percentile of DMT, calculated independently for each grid cell using the ERA5 hourly reanalysis dataset over the climatological reference period 1971–2000. This index assesses absolute heat relative to the annual climate. A positive EHF_sig, indicates the TDP is unusually warm, a prerequisite for a heatwave, typically excluding winter half-year events. The second index, the acclimatisation index, is:

This measures the deviation of the three-day DMT from the average over the prior 30 days, accounting for short-term human and system adaptation. The final EHF was computed as:

$$EHF = EHI_{\mathrm{sig}} \times \max\left(1,\, EHI_{\mathrm{accl}}\right)$$

yielding positive values only when conditions are unusually hot, with units in °C² for intensity quantification.

This enables severity classification: low-intensity for 0 < EHF < EHI_as (where EHI_as is the 85^th percentile of positive EHF values), severe for, EHI_as < EHF < 3 × EHI_as, and extreme for EHF ≥ 3 × EHF_as allowing normalised comparisons across diverse climates.

Compared with traditional definitions, which rely on sequences of daily maximum temperatures that exceed fixed thresholds (e.g., three days above the 90th percentile), EHF offers distinct advantages. It incorporates minimum temperatures to capture overnight heat retention and cumulative thermal stress, critical for health impacts, rather than focusing solely on daytime peaks. By including the acclimatisation index, EHF reduces false positives due to gradual warming, better aligning with human vulnerability. Additionally, its continuous intensity scale and severity classification provide a more nuanced impact assessment than binary traditional methods, enhancing forecasting accuracy and supporting tailored mitigation strategies (Nairn & Fawcett, 2015).

For the assessment of heatwaves using the EHF method, we evaluated four key characteristics:

Heatwave intensity (HWI), quantified as the annual mean EHF, measured the severity of heatwave events by combining temperature exceedance and acclimatisation effects. This annual mean provided a standardised metric for comparing severity across regions and years.

Heatwave duration (HWD) is the total number of days across all heatwave events in a year, where each event requires at least three consecutive days with EHF > 0. It is calculated as:

$$HWD = \sum_{j=1}^{m} d_j$$

where m is the number of heatwave events, and d_j is the number of consecutive days in the jth event. This metric quantifies cumulative heat exposure, crucial for assessing impacts on health, agriculture and infrastructure (Perkins-Kirkpatrick & Lewis, 2020).

Heatwave frequency (HWF) is the number of distinct heatwave events in a year, each comprising at least three consecutive days with EHF > 0, separated by at least one day with EHF ≤ 0. It is expressed as:

$$HWF = m.$$

Heatwave peak temperature (HWP) is the yearly averaged maximum DMT during heatwave events, calculated as:

$$HWP = \frac{1}{m} \sum_{j=1}^{m} T_{\max,j}$$

where T_max,f is the maximum DMT during the j^th heatwave event.

Overall methodology is shown in Figure 3.

3.1. UHI trends from 2000 to 2024

The analysis of UHI intensity from 2000 to 2024 reveals clear spatial and temporal variability, with statistically significant differences between LCZ classes (p < 0.05) (Figures 4–6). The following subsections present the spatial, LCZ-specific and diurnal trends in detail.

3.2. Performance assessment of WRF_Simulated temperatures with SLUCM, LCZ and LCZ_local land use inputs

Daytime and nighttime LST were analysed to evaluate the performance of WRF-simulated models (SLUCM, LCZ and LCZ_local) against reference data derived from MODIS (via GEE). For daytime conditions, both LCZ and LCZ_local simulations produced mean values closest to the MODIS reference, indicating relatively better agreement (Figures 7 and 8). SLUCM, however, simulated the lowest extremes, with minimum and maximum values of 29 °C and 37 °C, respectively, suggesting an underestimation of daytime variability (Figure 8). Statistical evaluation in Table 1 confirmed these patterns: LCZ_local achieved the lowest errors (MAE ≈ 2.2 °C; RMSE ≈ 2.5 °C), while all models exhibited a negative bias of approximately −2.0 °C to −2.7 °C. Correlation was moderate (r ≈ 0.3–0.5), reflecting an ability to capture overall patterns but with limited representation of variability. The explained variance remained modest (R² ≈ 0.1–0.2).

Meanwhile, at nighttime, SLUCM, LCZ and LCZ_local yielded similar mean values, but all were consistently higher by about +6 °C than the MODIS reference (Figures 7 and 8). This positive bias was also reflected in the minimum and maximum values, with all three models overestimating nighttime LST. Performance statistics (Table 1) indicated large errors (MAE ≈ 6.4–6.6 °C; RMSE ≈ 6.5–6.7 °C), negligible correlation (r = 0.0)and no explained variance (R² = 0.0). Willmott’s index of agreement was also very low (d ≈ 0.1–0.2), highlighting poor overall agreement between modelled and observed nighttime LST.

Although ground-based meteorological observations are commonly used to evaluate WRF simulations of near-surface air temperature (T2), this study focused on land surface temperature, the TSK, which is directly comparable with MODIS LST. The use of MODIS enabled spatially explicit validation of the simulated surface temperature and UHI patterns across Kuala Lumpur. Nevertheless, the inclusion of ground-based observations would provide complementary validation and could further improve confidence in the model performance.

Overall, the LCZ_local configuration consistently outperformed SLUCM and LCZ, particularly under daytime conditions and can therefore be considered the most reliable model setup for reproducing observed temperature patterns in this study.

3.3. UHI intensity across LCZ during heatwave periods

Since LCZ_local demonstrated the best overall performance, it was selected to calculate UHI intensity during the heatwave period in Kuala Lumpur (April 2024). The estimated UHI intensity ranged from −3.5 to 4.6 °C during the daytime and from −3.1 to 2.7 °C at nighttime (Figure 9). The negative UHI values indicate locations where surface temperatures were lower than the selected rural reference, representing localised urban cool island conditions or the influence of heterogeneous land-cover characteristics within the urban–rural reference framework. These areas are primarily associated with vegetated and open landscapes, where shading, evapotranspiration, proximity to water bodies and local topographic effects can reduce surface temperatures despite the occurrence of a regional heatwave. Results of the ANOVA indicate that UHI intensity varied significantly across LCZ classes, with a p-value < 0.05 confirming statistically significant differences at the 5% level (Table 2). Among the classes, LCZ 1–3 consistently exhibited the highest UHI intensity compared to other LCZ types, reflecting the influence of dense building morphology and high impervious surface cover (Figure 10).

 

Notably, several LCZ classes (LCZ 7, 10, A, B, D, E, F and G) were absent in the WRF outputs. This is attributable to the heterogeneous and fragmented distribution of these classes in Kuala Lumpur, which are too small in spatial extent to be resolved by the WRF’s dominant land-use representation. Their omission is not expected to substantially affect the overall UHI patterns, given their limited areal coverage within the study domain. Collectively, these results confirm that UHI intensity in Kuala Lumpur is strongly dependent on LCZ type, with the most pronounced heat stress concentrated in compact urban areas, underscoring the importance of targeted heat mitigation strategies.

3.4. UHI trends during heatwave periods

The daily UHI intensity ranged between 3.9 °C and 5.8 °C, with the peak recorded on 11 April (Figure 11(a)). The overall mean UHI was 4.7 °C and variability was relatively low (standard deviation < 2.5 °C), indicating stable conditions with no extreme fluctuations.

The diurnal cycle showed a peak UHI intensity of approximately 6.9 °C occurring during the early morning (06:00–07:00), while the lowest intensity (∼2.0 °C) was observed around midnight to the early nighttime hours (Figure 10(b)). The early-morning peak is physically plausible, as it corresponds to the pre-sunrise period, when the contrast between urban and rural thermal responses is typically strongest. During nighttime, rural areas generally experience more rapid cooling due to efficient longwave radiation loss and lower thermal storage capacity, whereas urban surfaces retain heat accumulated during the daytime due to higher heat capacity, reduced sky-view factor and limited nocturnal radiative cooling (Cimolai & Aguilar, 2025; Pron & Zyane, 2026). Consequently, the urban–rural temperature difference increases and reaches its maximum before sunrise. This diurnal behaviour is consistent with previous studies based on both remote sensing observations and in situ measurements, which commonly report stronger UHI intensity during nighttime and early morning periods, followed by a reduction during daytime as solar heating decreases the temperature contrast between urban and non-urban areas (Mathew et al., 2018; Memon et al., 2009). However, the exact timing and magnitude of the peak UHI can vary among cities, depending on local characteristics such as urban morphology, vegetation cover, surface properties and prevailing weather conditions. Following sunrise, increasing solar radiation warms both urban and rural surfaces, reducing the relative temperature contrast and weakening UHI intensity. The diurnal variation was approximately 5 °C between the minimum and maximum values. The relatively low variability (standard deviation < 1.3 °C) throughout the cycle indicates a consistent and stable diurnal UHI pattern during the heatwave period.

3.5. Projected impacts of +1 °C temperature increase on UHI across LCZ

Given that LCZ_local demonstrated the highest overall performance, it was applied to estimate the projected UHI intensity during the heatwave period in Kuala Lumpur (April 2024) under the +1 °C perturbation scenario. The simulated UHI intensity ranged from −3.7 °C to 4.5 °C during the daytime and from −3.2 °C to 2.6 °C at nighttime (Figure 12). Spatially, the strongest UHI effects were concentrated in compact and dense built-up areas (LCZ 1–3), whereas vegetated and open zones exhibited lower UHI values, indicating localised cooling. The results of the ANOVA confirmed that UHI intensity varied significantly by LCZ (p < 0.05) (Table 3), with post-hoc analysis further showing that LCZ 1–3 were significantly different from the more open and natural LCZ classes (Figure 13).

Temporally, both daytime and nighttime UHI displayed variability, with hotspots distributed across both mid- and high-rise built-up areas. The +1 °C perturbation amplified UHI intensity across all LCZ classes, with the largest increases occurring in compact urban forms, underscoring the disproportionate heat stress in these areas. Overall, LCZ_local not only reproduced observed spatial–temporal patterns more accurately than SLUCM and standard LCZ but also highlighted the heightened vulnerability of dense urban LCZs during extreme heat events (Figure 12).

3.6. Heatwave characteristics

Heatwave characteristics were analysed at both regional and city scales to distinguish the influence of large-scale climate variability from localised urban thermal effects. The analysis across Peninsular Malaysia provides the broader climatic context of changing heat extremes. At the same time, the Kuala Lumpur-specific assessment highlights how these regional changes translate into increasing thermal stress within a highly urbanised environment. Although heatwaves are not directly caused by UHI, the interaction between background warming, prolonged heat events and urban land surface characteristics can amplify heat exposure in cities. Therefore, evaluating heatwave intensity, duration, frequency and peak temperature provides important context for interpreting the magnitude and persistence of UHI effects under extreme heat conditions.

Figure 14 reveals pronounced spatial and temporal changes in the HWI across Peninsular Malaysia over the past four decades. Between 1980 and 1999, the average HWI was 0.41 °C², with values ranging from 0.17 °C² to 0.83 °C². Higher intensities were concentrated in the northern and central regions, while southern parts, including Johor and the coastal lowlands, exhibited relatively weaker heatwave signals. In contrast, during 2000–2024, the mean HWI increased markedly to 0.52 °C², with local maxima reaching 1.13 °C², particularly in northern inland areas such as Kedah and Perak. This spatial shift suggests a northward and inland intensification of heatwave conditions, possibly reflecting both changes in background climate and land surface modifications. The anomaly map, showing differences between the two periods, indicates a widespread positive signal, with anomalies ranging from 0.1 °C² to 0.5 °C² across most of the peninsula. The strongest anomalies are observed in the northwestern and northeastern regions, while minor or near-neutral changes occur in a few southern coastal zones. This pattern implies that the increase in heatwave intensity has been both spatially extensive and geographically differentiated, with interior and northern regions experiencing the greatest amplification.

The time series for Kuala Lumpur provides further temporal detail by demonstrating how regional-scale heatwave intensification manifests in one of Southeast Asia’s most rapidly urbanising metropolitan areas. This city-scale assessment complements the UHI analysis by identifying periods when enhanced background heat conditions may increase the vulnerability of urban populations to additional warming from built environments. From 1980 to the early 1990s, HWI values fluctuated between 0.2 °C² and 0.5 °C², followed by a brief decline around the mid-1990s. After 2000, however, a clear upward trend emerged, with several pronounced peaks—most notably around 2010–2015, when HWI values exceeded 0.6 °C², the highest in the record. Although interannual variability remains substantial, the overall mean during 2000–2024 is consistently higher than during 1980–1999.

Taken together, these results highlight a significant intensification of heatwaves across Peninsular Malaysia, with a regional mean increase of about 0.1 °C² and local increases exceeding 0.3–0.4 °C². The stronger signal over northern and highly urbanised regions, including Kuala Lumpur, indicates the potential compounding effects of regional climate warming and urban land surface modification on heat exposure. While HWI primarily reflects background heatwave characteristics, its upward trend provides important climatic context for understanding how extreme heat events may interact with UHI processes in urban areas. This intensification not only underscores the growing risks of heat stress in densely populated and agricultural regions but also signals the need for adaptive planning to mitigate the health and environmental impacts of rising heat extremes.

Between 1980 and 1999, the average heatwave duration across the peninsula was 15.2 days, with spatial variations ranging from 12.5 to 17.3 days (Figure 15). Heatwave events were relatively short and evenly distributed, with slightly longer durations observed in the northern inland and east-coastal areas. This pattern suggests that heatwave conditions were relatively moderate and localised during this earlier baseline period. In contrast, the 2000–2024 period exhibits a substantial intensification in both the magnitude and spatial extent of heatwaves. The mean HWD nearly doubled to 29.6 days, with durations exceeding 40 days in some southern and central regions, including Pahang, Negeri Sembilan and Johor. This spatial shift toward longer, more persistent heatwaves indicates a significant increase in the frequency and duration of hot conditions across the entire peninsula. The anomaly map further highlights this escalation, showing widespread positive anomalies between +4 and +28 days, with the strongest increases occurring in southern and interior zones, while even traditionally cooler northern areas experienced notable lengthening of heatwave events.

The time series for Kuala Lumpur reflects this broader regional pattern. Between 1980 and 1999, the city experienced short, sporadic heatwave periods, typically lasting fewer than 20 days per year, with occasional peaks around 1985 and 1995, reaching 60–70 days. After 2000, however, a marked shift occurred: HWD values increased sharply, with frequent episodes exceeding 50 days and the most extreme peaks surpassing 100 days around 2010–2015. This trend underscores a transition from intermittent heatwaves to sustained, prolonged events in recent decades.

Figure 16 shows clear spatial and temporal increases in HWF across Peninsular Malaysia between 1980–1999 and 2000–2024. During the earlier period, heatwaves were relatively rare, averaging 2.68 events per year (range 1.7–4.15), with most areas experiencing only two to four events annually. By contrast, from 2000–2024, the mean frequency more than doubled to 5.40 events, with several southern and central regions recording up to 7–8 events per year. The anomaly map highlights widespread increases of 0.4–5.4 events, most pronounced in southern and inland zones.

Simulation results for Kuala Lumpur reflect a similar pattern. From 1980 to 1999, heatwave events remained below four per year, interrupted by occasional spikes. After 2000, frequencies rose sharply, with repeated peaks exceeding 10–14 events, especially during 2005–2017. The results indicate a substantial and geographically consistent rise in heatwave occurrences, driven by regional warming and intensified by urbanisation. This increase in both magnitude and recurrence underscores growing thermal stress in urban areas, emphasising the need for improved heat management and adaptation planning.

Regarding HWP, spatial patterns indicate only a slight increase in the maximum temperature reached during heatwaves over the past four decades (Figure 17). The mean HWP rose marginally from 27.3 °C in 1980–1999 to 27.4 °C in 2000–2024, with maxima around 29.6–29.7 °C and minima near 23 °C for both periods. The anomaly map confirms this stability, showing generally weak positive differences—mostly within ±0.3 °C and locally up to 0.26 °C—suggesting that while heatwaves have become more frequent and longer (as seen in other indicators), their peak intensity has changed very little. The time series for Kuala Lumpur supports this finding. From 1980 to 2024, HWP fluctuates between 27 °C and 29 °C, with no pronounced upward trend. Occasional peaks around 1990 and 2005–2010 reflect episodic extreme years rather than a consistent rise. Despite this relative stability in peak temperature, these values still coincide with notable increases in heatwave duration and frequency, indicating that heatwaves are lasting longer and occurring more often, even if their maximum temperatures remain similar. The result suggests that the severity of heatwaves in terms of peak temperature has remained nearly constant, while other dimensions—frequency, duration and intensity—have intensified. This pattern implies that the region’s growing heat risk stems less from higher absolute temperatures and more from the persistence and recurrence of extreme heat conditions.