Enhanced nighttime heatwaves over African urban clusters

Despite the threat that nighttime heatwave poses to public health and the environment in the developing world, it remains relatively understudied in Africa, especially in its rapidly expanding urban areas with large populations. Using meteorological observations, climate reanalysis, remote sensing datasets, and integrated methodology, we found that during 1981–2020, nighttime heatwaves dramatically increased with early onset dates over Africa. Large scale circulation induced dry conditions over land which explains the major heatwaves across all climate zones. Meanwhile, the increasing trend of nighttime heatwaves in urban areas than rural areas in both tropical and temperate climates is largely due to urbanization, which amplifies heatwaves with urban heat island (UHI) effects. The strongest contribution of urbanization to nighttime heatwaves was observed in temperate climate, leading to a 46% increase in the heatwave duration. In arid climate, urban expansion has a negative impact on nighttime heatwaves, due to the cool island effect of urban greens and weak urbanization. The major contribution of urbanization to the urban-rural contrast of nighttime heatwave trends in temperate climate can be attributed to stronger UHI intensity resulting from lower windspeed and less evapotranspiration. Without appropriate adaptation efforts to reduce heatwave exposure, the risks posed in Africa’s urban areas will continue to increase under future warming.


Introduction
Under a warming climate, heatwaves have been increasing globally, posing serious threats to ecosystem services, energy and water security, and human health (Mora et al 2017, Kornhuber et al 2020. Heatwaves are considered as a period of extremely hot weather during which temperatures exceed a specified threshold (Perkins and Alexander 2013). Due to their devastating impacts, it is becoming imperative to understand the spatial-temporal changes of heatwaves, particularly in Africa which has been considered as one of the continents vulnerable to heatwave events due to its large population, lack of cooling infrastructure, and background climate change (Russo et al 2016). With a surge in urban populations in Africa, several major urban clusters with large economic activities have emerged in the last several decades including Lagos, Nairobi, Johannesburg, Kinshasa etc (Güneralp et al 2018). Though Africa's urbanization is increasing at a faster rate than other urbanizing regions such as Asia, the actual processes shaping urbanization are different as Africa's urban expansion is unplanned and unregulated. The impacts of heatwaves may be enormous considering the region's low adaptive capacity.
In recent decades, significant increase in heatwave frequency and duration were observed over Africa (Harrington andOtto 2020, Perkins-Kirkpatrick andLewis 2020). Climate projections expect further increases in temperatures, as well as hot extremes over Africa (Nangombe et al 2018). These studies focused on analyzing daytime heatwave patterns. Due to differences in climate regimes and urbanization rate, heat-induced impacts could likely vary across different regions, particularly at nighttime. The impacts of nighttime heatwaves are greater than daytime, because the body cannot recover from daytime heat exposure, thus increasing and the risk of heat-related cardiovascular diseases and mortalities in urban areas (Obradovich et al 2017. Improved prediction of heatwave impacts will benefit from understanding the evolution of nighttime heatwave events at local scales (Sarangi et al 2021, Xie et al 2022. Also, understanding the changing risk of heatwaves associated with rapid urbanization at local scale is critical for improving climate change adaptation measures (Zschenderlein et al 2019, Tuholske et al 2021. Due to increased impervious surfaces from urbanization, the impacts of nighttime heatwaves in urban areas could be aggravated by the urban heat island (UHI) effect. Compared to daytime temperatures, urban expansion has a strong impact on nighttime temperatures in urban areas because, impervious surfaces store heat during the daytime and release it at nighttime hence, increasing the UHI effect (Yang et al 2017, Sarangi et al 2021. As reported, the UHI effect intensified more at nighttime compared to daytime in urbanized areas during heatwave events (Huang et al 2021). Also, the synergistic effect between nighttime heatwaves and UHI were stronger in some major urban agglomerations in China (Jiang et al 2019), the United States (Sarangi et al 2021), and Greece (Founda and Santamouris 2017). However, whether and how urbanization magnifies nighttime heatwaves across various climate regimes over large domain are poorly understood. The scope of this study was therefore framed to provide answers to this question over Africa.
The combined impacts of nighttime heatwaves and UHI would likely be enormous in African urban clusters considering rapid urban expansion and limited infrastructures to deal with such intense nighttime heat. Here, we incorporate climate records and satellite data products to examine changing patterns of nighttime heatwaves and the contribution of urbanization over various climate zones in Africa. The objectives of this study are to (a) examine the decadal change in the regional patterns of nighttime heatwaves, (b) assess the evolution of heatwaves in urban areas, and (c) evaluate the impact of urbanization on nighttime heatwaves over large domain.

Data
To investigate nighttime heatwave events (1981-2020), we used daily minimum temperature and evapotranspiration datasets from the land version of the fifth-generation reanalysis product of the European Centre for Medium Range Weather Forecasts (ECMWF), i.e. ERA5-land at 0.1 • × 0.1 • spatial resolution (Hersbach et al 2020) and qualityassured daily HadISD (Hadley Integrated surface dataset, version 3.2.0.202109p) station data with minimum temperature and wind speed. Large-scale mechanism were assessed using sea surface temperature (SST), 850 hPa winds, 500 hPa geopotential height pressure (HGT) and precipitation datasets from ERA5 reanalysis at 0.25 • × 0.25 • spatial resolution.
To better understand the spatial heterogeneity of nighttime heatwaves across various climate zones, we used the Köppen-Geiger climate classification of Africa into three major climate zones (Beck et al 2018) including the arid, tropical and temperate climates (figure S1). Urbanization rate was quantified using impervious surface area. The impervious surfaces are artificial structures used to represent human settlement and development. The impervious surface area over Africa was obtained from the global artificial impervious area dataset  at 1 km spatial resolution (Gong et al 2020). Impervious surface areas increased over large parts of Africa during 1985-2018 (figures 1(a) and (b)) with major intensification in the tropical and temperate climates. To examine the changes in vegetation cover, we used MODIS normalized difference vegetation index (NDVI) data (MOD13C1, collection 6) from 2001 to 2020, with 0.05 • and 16 day spatial and temporal resolutions respectively (Didan 2015).

Evaluation of nighttime heatwave events and large-scale mechanism
Here, a nighttime heatwave is defined as a period of at least three consecutive days during which daily minimum temperature exceeds the 90th percentile of long-term daily threshold (Perkins and Alexander 2013). Over a 40 year period (1981-2020), the 90th percentile threshold was calculated by ranking 15 day minimum temperature data surrounding a calendar day (i.e. seven days before and after the calendar day). This threshold was estimated for each station of HadISD and grid point of ERA5.
We assessed four nighttime heatwave metrics: (a) heatwave intensity (HWA, units in • C), (b) heatwave duration (HWD, unit in day), (c) heatwave number (HWN, units in event) and (d) heatwave onset (HWO, units in day of year). The HWA is the annual minimum temperature anomaly for heatwaves above 90th percentile threshold; the HWD is the duration for the longest heatwave event; the HWN is the number of annual heatwave events and the HWO is the onset date for the first heatwave event in a year. To account for heatwave events across all seasons and sub-regions, each heatwave characteristic is estimated over the course of a longer summer season (April-September in the Northern Hemisphere (NH), October-March in the Southern Hemisphere (SH)).
We calculated linear trends of nighttime heatwave metrics over 1981-2020 and estimated their statistical significance using the Mann-Kendall test. The long-term comparison between the minimum temperatures of HadISD stations and the nearest ERA5 grid cell shows good consistency within a correlation coefficient of 0.7, root mean square error of 1 • C and standard deviation of 1.25 • C (figure S2). We examined the evolution of nighttime heatwaves in urban and rural areas using analysis results based on HadISD dataset. Rural areas in this study are taken as background without urbanization impact to highlight their contrasts with urban nighttime heatwaves.
To analyze the pattern of possible large-scale mechanisms associated with nighttime heatwaves, we utilize composite anomalies of SST, 850 hPa winds, 500 hPa geopotential height pressure and precipitation. Composite anomalies are produced for all variables over all nighttime heatwave days for the longer summer months in the NH (April-September), and SH (October-March) respectively. Statistical significance of composites was then assessed using student's t-test.

Classifying urban and rural stations
Using varying annual impervious surface area, HadISD stations were dynamically classified into urban and rural stations. To identify an urban or rural station, we first evaluate the urbanization rate around each station within a buffer region. We evaluated various circular buffers with radius ranging from 1 to 10 km around each station, and then computed the correlation coefficient between the nighttime heatwave characteristics and urbanization rate within different radius around stations during 1985-2018. We then chose the radius at which a stable correlation coefficient between nighttime heatwave characteristics and urbanization rate was observed (figures S3(a)-(d)). As a result, all nighttime heatwaves characteristics in urban and rural areas were estimated based on the 8 km radius. If the proportion of impervious surface area within 8 km radius of a station is less than the annual mean fraction of all stations, then it is considered as a rural station. Otherwise, it is classified as an urban station. The mean impervious surface fraction surrounding HadISD stations was about 5.4% in 1985, however, it rose to 54.2% in 2018 (figure S4(a)). Urban and rural stations were updated annually based on the impervious surface area fractions within the buffer (figures 1(c) and (d); figures S4 (b) and (c)).
We classified stations pre-1985, based on the impervious surface area of 1985. This was done because though Africa is experiencing rapid urbanization, this pattern became noticeable only in the early 20th century (Potts 2012, Güneralp et al 2018. So, the assumption of a stable urbanization process before early 2000s was adopted in this study. Also, based on slow economic development and urbanization after 2018 over Africa (World Bank 2019, OECD 2020, Kelsall et al 2021), we assumed stable urban growth rate and classified stations after 2018 based on the impervious surface areas of 2018. This method of assuming stable urban growth over certain periods due to slow economic development and urbanization, and data availability is consistent with previous studies (Liao et al 2017, Yang et al 2017, Shi et al 2021.

Evaluating nighttime heatwave contrast of urban and rural stations
We divided the study area into longitude-latitude grids to examine the urban-rural contrast of nighttime heatwave characteristics (figures S5(a) and (b)). This was done with the assumption that for each 5 • × 5 • grid cell, natural forcing and climate variability are similar across all stations. Considering the heterogeneous distribution of HadISD stations across Africa, we test different grid sizes including 2.5 • × 2.5 • , 5 • × 5 • , 7.5 • × 7.5 • , and 10 • × 10 • to determine which would have more grid cells with minimum one urban and one rural station for the evaluation of nighttime heatwave events. Urban and rural stations were well covered and well represented using the 5 • × 5 • grid cells (figures S5(a) and (b)).
Throughout the study period, a total of 55 5 • × 5 • grid cells with a minimum of one urban and one rural station are available (see shaded grid cells in figure S5(b)). The nighttime heatwave characteristics (HWA, HWD, HWN, and HWO) for the urban and rural stations were estimated respectively. The urban and rural heatwave characteristics within each 5 • × 5 • grid cell are averaged to represent the urban and rural nighttime heatwave characteristics at the grid level. To eliminate topographical effect on the estimated nighttime heatwave characteristics, rural and urban stations were removed if their elevations are 500 m higher than the lowest elevation within the grid cell. As a result, a total of 307 HadISD stations were used to examine nighttime heatwave events.
To understand the effects of urbanization on nighttime heatwave characteristics, we evaluated the difference in nighttime heatwave trend between urban (T urban ) and rural (T rural ) stations for each grid cell. The relative contribution of urbanization (Uc in %) to heatwave trends (Ren and Zhou 2014, Luo and Lau 2018) is expressed as: Uc > 0 and ⩽100% indicates positive contribution of urbanization; Uc ⩽ 0% indicates negative impact of urbanization. Simple linear regression was used to estimate the trends, and a nonparametric student's t-test (Zimmerman and Zumbo 1993) was used to assess the difference in trends between urban and rural nighttime heatwave characteristics. We first postulate that urban growth rate is responsible for the observed urban-rural nighttime heatwave characteristic. To test this hypothesis, the relationship between the urban-rural contrasting nighttime heatwave trend and urban expansion based on the 8 km buffered area around each station was evaluated.
In theory, a greater shift in the UHI intensity indicates higher impact of urban expansion on urban heatwaves (Schär et al 2004, Liao et al 2018. Hence, we compared the UHI intensity based on the mean annual minimum temperature between urban and rural grids in all climate zones using the probability density function and cumulative distribution function (CDF). The difference in CDF curves was assessed using a two-sample Kolmogorov-Smirnov test (Xiao 2017).
To explain the UHI and nighttime heatwave patterns induced by urbanization, we assessed changes in daily wind speed and evapotranspiration, for each station and each grid cell. Furthermore, to estimate changes in vegetation cover in urban and rural areas, the greenest pixel composite method (Trianni et al 2015) was first used to create new annual composite images based on yearly maximum pixel values of the NDVI. This was done to correct inconsistent coverage due to cloud cover and seasonal signals. Linear trend was then estimated based on annual maximum NDVI values for 8 km buffered region around the urban and rural stations.
We acknowledge that uneven station distribution and incomplete spatial coverage may introduce some uncertainties in our results. Hence, a comparison of urbanization and contrasting nighttime HWD was conducted by randomly selecting grid cells (20) with the same urbanization rate across climate regions. Correlations results remain significant (0.53) at 95% level after being repeated 100 times. This suggests a significant relationship between urbanization rate and nighttime heatwave trend which is not influenced by the distribution of HadISD stations.

Spatial heterogeneity of nighttime heatwave changes
Over large parts of Africa nighttime heatwaves have become more frequent, stronger and longer-lasting, as demonstrated by both ERA5 and HadISD datasets (figures 2(a)-(h)). Our analysis reveals major nighttime HWD in the north and south Africa compared to central Africa (figures 2(b) and (f)). While, nighttime HWA is minor in the central areas with the exceptions in the north-western and southern parts (figures 2(a) and (e)). The nighttime HWN increased with significant trends in the north-western and southern Africa (figures 2(c) and (g)). On the other hand, earlier nighttime HWO were observed over the continent (figures 2(d) and (h)).
Overall, the temperate climate in Africa experienced a significant increasing trend of nighttime HWA, HWD and HWN, than the arid and tropical climate, as indicated by both ERA5 and HadISD (figures S6(a)-(i)). Earlier nighttime HWO was observed across all climate zones (figures S6(j)-(l)). Significant trend was observed in the temperate climate for both ERA5 (17.5 days decade −1 ) and HadISD (20.4 days decade −1 ). Although the methods for constructing ERA5 differ from HadISD, they are consistent in regional nighttime heatwaves trends with correlation coefficient above 0.5 (p < 0.01) (figures S6(a)-(l)), indicating both datasets can represent trends of nighttime extreme heat events (Raymond et al 2020, Rogers et al 2021. For all heatwave days of the longer summer months in the NH and SH respectively, the composite anomalies show that Africa's regional SST was above-normal, with peak values up to 4 • C in Mediterranean Sea, Indian, Northern and Southern Atlantic oceans, close to regions with major heatwave events ( figure 3(a)). In both hemispheres, the circulation patterns (850 hPa wind) showed strong warm advection from major positive SST regions to land areas with major heatwave events particularly in temperate climate ( figure 3(a)). Regions of major nighttime heatwave events corresponded with high pressure systems ( figure 3(b)) and above-normal drier conditions ( figure 3(c)), implying that circulation patterns likely influenced dry conditions in both hemispheres. In contrast, wetter conditions in the tropical climate were observed where smaller or negative changes in nighttime heatwave were reported. Such patterns were reported previously (Horton et al 2015) and suggested the contribution of large-scale circulation and abnormally high SST to heatwave events. As large-scale drivers, the contribution of this large-scale atmospheric circulation pattern and ocean heat to nighttime heatwaves is evenly distributed in urban and rural areas in both hemispheres. The local scale impacts and drivers may vary across different regions especially over populated urban areas leading to spatial heterogeneity of nighttime heatwave. To understand this, we further evaluated urban-rural contrast of nighttime heatwaves over Africa.

Urban-rural contrasts in nighttime heatwaves
In the temperate and tropical climate, nighttime urban heatwaves advanced further than those in rural areas (figures 4(a)-(h)), while opposite pattern was observed in arid climate (figures 4(i)-(l)). The trends of urban-rural contrast for HWA, HWD, and HWN are 0.25 • C decade −1 , 0.51 days decade −1 and 0.2 event decade −1 respectively in the temperate climate, 0.12 • C decade −1 , 0.21 days decade −1 and 0.07 event decade −1 respectively in the tropical climate, and −0.12 • C decade −1 , −0.51 days decade −1 and −0.02 event decade −1 in the arid climate. The nighttime HWO appeared earlier in urban than rural areas by 3.55 days decade −1 and 7.22 days decade −1 in the temperate and tropical climates, respectively, while it delayed by 1.25 days decade −1 in the arid climate. Overall, major changes in the trend of urban-rural contrast for nighttime heatwave characteristics were observed in the temperate climate. The difference in heatwave trends between urban and rural areas are significant at 95% confidence level (figures 4(a)-(l)).
The contribution of urbanization to HWA, HWD, HWN, and HWO are positive in the temperate and tropical climates and negative in the arid climate (figures 4(a)-(l); figures 5(a)-(d)). Overall, urbanization contributed more to nighttime HWD than HWA, HWN and HWO in the temperate and tropical climates (table S1). Despite being in temperate climates, the contribution of urbanization to HWD (46%) in temperate Africa is greater than that  . Trends of nighttime heatwave intensity (HWA, (a) and (e)), duration (HWD, (b) and (f)), number (HWN, (c) and (g)) and onset (HWO, (d) and (h)) with ERA5 (top) and HadISD (bottom). Trends are statistically significant at 95% confidence level according to student t-test.
(24%) in eastern China (Yang et al 2017). This difference could be due to more planned and regulated urban expansion in China than Africa (Güneralp et al 2018).

Role of urbanization in nighttime heatwave changes
Over the region, significant positive correlation between nighttime heatwave and urban expansion (p < 0.05) was observed ( figure S7(a)). Overall, urban expansion intensified more in the temperate (15.42% decade −1 ) and tropical (15.21% decade −1 ) climate than in arid climate (1.78% decade −1 ) ( figure S7(b)). Urban expansion alters the surface properties and land-atmosphere interactions, including evapotranspiration, roughness and albedo, which magnifies the magnitude of urban warming , Manoli et al 2019.
To further understand the role of urbanization in heightening nighttime heatwaves across climate regimes, we examined changes in UHI intensities. As observed, the UHI intensity shifted toward greater mean values in the temperate climate than tropical climate, while a slight reduction was observed in arid climate (figures 6(a)-(d)). This suggests that stronger nighttime heatwave trends observed in temperate climate could be linked with intensified UHI. However, to test this, the difference in mean values for two periods (1981-2000 and 2001-2020) were estimated for daily windspeed and daily evapotranspiration respectively.
As documented, the UHI effect on heatwave is usually favored by reduced windspeed (Al-Obaidi et al 2021). We observed that mean difference in daily windspeed increased minimally in arid climate, and decreased more in temperate climate than tropical climate (figure 6(e)). This substantial decrease in windspeed likely contributed to the enhanced UHI effect observed in temperate climate, while increase windspeed weakened UHI effect in arid climate. Increased surface friction due to increased surface roughness in urbanized region can also reduce horizontal windspeed, and thus decrease evapotranspiration and water vapor exchange (Li et al 2015, Fenner et al 2019. The difference in surface roughness and urban-canyon properties in arid, tropical and temperate climate could have influenced the patterns of windspeed observed, however, this requires further investigation. The Increase of UHI intensity may enhance the magnitude of heatwaves by facilitating local drying conditions through decreased evapotranspiration in urban areas (Liao et al 2017, Luo and Lau 2019. Here, mean difference of daily evapotranspiration reveal a greater reduction in temperate climate than tropical climate, and a slight increase in arid climate (figure 6(f)). Reduced evapotranspiration can be attributed to increase in urban surface albedo,  however, increased vegetation cover can reduce the effect of surface warming in urban areas (Manoli et al 2019). Compared to temperate and tropical climates, the UHI intensity and evapotranspiration were weaker in the arid climate likely due to increase in vegetation cover in urban areas than rural areas (figures S8(a) and (b)). The relationship between heatwave characteristics and NDVI reveal cooling effect of increasing vegetation in urban areas than rural areas in arid climate compared to temperate and tropical climates (figures S8(c)-(f)). As noted, urbanization has little or no impact on surface warming due to urban cool island effect from urban greening and irrigation initiatives in arid climate (Duan et al 2019, Ziter et al 2019. Decrease in vegetation cover raises sensible heat flux that contributes to the nighttime UHI effects (Huang et al 2021, Wu et al 2021. However, increasing vegetation cover and low urban growth rate can reduce the UHI effect, as observed in arid climate. Vegetation cover increase due to CO 2 fertilization (Zhu et al 2016) and intense urban agriculture and irrigation (Schumacher et al 2009, Gashaye 2020 were reported in arid climate in Africa, which likely accounts for the cooling effect observed. Other than urban expansion, the interaction between nighttime heatwaves and UHI effect can be exacerbated due to anthropogenic heating (He et al 2020). Moreover, the urban clusters density can modulate UHI effect such that regions with dense urban development experience warmer temperatures than regions with sparse development (Zhou et al 2017). However, these local scale drivers have to be investigated in future studies over Africa.

Implications for adaptation
The spatio-temporal nighttime heatwaves patterns discovered are significant because people in temperate climate in Africa are not well prepared for such heatwave events. Most people in Africa still lack access to constant power and may not be able to afford cooling systems, hence nighttime heatwave impact may be enormous. As heatwaves were not a big issue in the past for people in the temperate climate, the new normal will make much stronger impacts on health and energy use.
Overall, this study provides insight into hotspots and possible mechanism of deadly nighttime heatwaves in Africa's urban areas, which have important implications for large vulnerable people in unplanned urban clusters. This is a fundamental progress in the field of climate change adaptation at the subregional scale because nighttime heatwave is more deadly and it requires different sustainable management strategies across various climate zones and sectors such as energy, water and public health. Future research should investigate the impacts of changing

Conclusion
Africa is experiencing rapid urbanization and more heatwaves, but it is not clear to what degree urbanization contributed to nighttime heatwave intensification. Our analyses reveal major intensification of nighttime heatwaves in temperate climate with earlier onset by 17.5 days decade −1 and 20.4 days decade −1 respectively according to ERA5 and HadISD. The trends of urban-rural contrast for nighttime heatwave characteristics are stronger in the temperate climate with heatwave intensity, duration, number increasing by 0.25 • C decade −1 , 0.51 days decade −1 and 0.2 event decade −1 respectively due to urbanization. Stronger nighttime heatwave trends in temperate climate are associated with large variability in UHI intensity. Increased UHI intensity amplifies the effects of nighttime heatwaves in urban clusters, while urban green increase can attenuate the UHI effect particularly in arid climate. In the future, urbanization and climate change can increase nighttime heat risks in this region. We therefore advocate for sustainable solutions through effective policies, mitigation and adaptation measures that can help reduce the impacts of nighttime heatwaves over Africa.
All data that support the findings of this study are included within the article (and any supplementary files).