Evaluation of the Contributions of Year-Specific Climate Anomaly, Nationwide Warming, and Urban Warming to Hot Summers in Japan

Abstract

Using surface observation data for the past hundred years, the contributions of year-specific climate anomaly, nationwide warming, and urban warming to hot summers in Japan were evaluated. A number of indices in temperature were defined to indicate the severity of summer heat in each year. Then, the year-to-year time series of each index was divided into a year-specific component and a temporally smoothed component, and the latter was divided into a nationwide non-urban component and an urban component. The results show that the non-urban component began to increase after the 1990s, which is approximately attributable to global warming, although there are some temperature variations on the yearly to multidecadal scales related to the Pacific Decadal Oscillation (PDO) and the Southern Oscillation (SO), whereas urban warming became apparent since the 1960s at stations in highly urbanized areas. For the recent record-breaking summer heat, the contributions of the year-specific temperature anomaly, nationwide warming, and urban warming are all evaluated to be of the order of 1 °C.

1. Introduction

In recent years, Japan has often experienced record-breaking hot summers, which have had a variety of adverse effects on society, such as health problems [1,2,3] and crop losses [4,5,6]. Factors related to hot summers are the variability of atmosphere–ocean circulation systems on the seasonal to multidecadal scales, as well as a worldwide trend of temperature increase, namely, global warming. Additionally, the urban-induced temperature increase, hereafter referred to as urban warming, is believed to amplify the heat load in cities.

However, in discussions of hot summers in Japan, the effects of urban warming tend to be addressed separately from those of large-scale fields and global warming. The influence of atmospheric and oceanic circulations causing hot summers has been examined from various viewpoints [7,8,9,10,11,12,13,14], while a large number of studies on global warming have been compiled by the Intergovernmental Panel on Climate Change (IPCC) [15]. In addition, the Japan Meteorological Agency (JMA) is monitoring climate change in Japan [16]. Some studies have used an event attribution approach based on ensemble numerical experiments, indicating that the contribution of global warming was essential to some recent hot summers over Japan [17,18,19]. Concerning the urban contribution, the IPCC states “no recent literature has emerged to alter the AR5 finding that it is unlikely that any uncorrected effects from urbanization, or from changes in land use or land cover, have raised global Land Surface Air Temperature (LSAT) trends by more than 10%” [15], as long as non-urban data are used for analysis.

On the other hand, 70% of the population of Japan, or 8.8 million people, live in “Densely Inhabited Districts” defined as agglomerations of areas with a population density of 4000 km⁻² or more [20,21]. Society is increasingly concerned about summer heat load in cities, and the urban climate has been the subject of many studies since the late 20th century [2,22,23,24,25,26], including the interaction between urban heat islands and local thermal circulations in summer [27,28,29,30,31]. Under these circumstances, it would be meaningful to evaluate the extent to which urban effects are included in the heat perceived by urban residents, together with the contribution of general circulation field anomalies and global warming.

It is not always easy to distinguish between global warming and urban-induced warming in observed temperature [2,32,33]. The IPCC adds, “larger (urban) signals have been identified in some specific regions, especially rapidly urbanizing areas such as eastern China”, after the aforementioned statement [15], as discussed in some recent studies [33,34]. For Japan, the JMA uses data from 15 stations in monitoring the long-term temperature change [16,35]. These stations are regarded as rural, but some of them are in cities with a population exceeding 100,000. Since urban temperature increases have been detected even in small cities [36,37,38], it is recommended to use data from less urbanized areas in order to separate non-urban and urban trends more reliably.

The aim of the present study is to provide an overview of the variations in summer heat in Japan using observed data for a hundred years, and their relation to global warming and urbanization. An attempt was made to separate year-to-year climate anomalies and long-term climate variability related to summer heat, and for the latter, to separate the nationwide non-urban component and the local urban warming. Then, the climatological characteristics of each factor were examined. Although climate change and the variability in the atmosphere–ocean system are inseparably linked in reality, the study attempts to find their approximate contributions from a linear perspective to give an insight into hot summers in Japan. Section 2 describes the data and methods, and Section 3 presents the results. Section 4 provides some discussion of the results, including their methodological and conceptual limitations.

2. Data and Procedure of Analysis

2.1. Data and Homogenization

Daily mean, maximum, and minimum temperature (T_ave, T_max, and T_min) data from JMA stations were used. The analysis was made for the period from 1921 to 2024, covering 68 stations where observations had begun by 1921. Figure 1 shows the distribution of these stations. Stations on the Nansei Islands (south of 30° N) were not included in the analysis because of missing data at some of them after World War II.

Observations of surface air temperature at these stations were made using tube thermometers placed in an instrument screen (Stevenson screen) until the 1960s. In the early 1970s, these thermometers were replaced by resistance thermometers covered by a tube-type radiation shield with forced ventilation, enabling automated remote observation. It is believed that this change does not affect the evaluation of long-term temperature trends seriously, although the temperature in an instrument screen may be 0.1–0.2 °C higher than that in the open air in the daytime, and slightly lower in the night-time [39]. However, changes in the observation time schedule are accompanied by a possibility of bias by up to ~0.5 °C [38]. In order to estimate and correct for this bias, hourly data of the Automated Meteorological Data Acquisition System (AMeDAS) were used for each station and month, as in previous studies [38].

Some stations were relocated during the analysis period, with a possibility of data inhomogeneity. The JMA provides a list of stations and relocation dates where temperature data are judged to be inhomogeneous [40]. Using this list, correction for the relocation bias was attempted for each of T_ave, T_max, and T_min. The procedure was based on a principal component analysis (PCA) to express the temperature distribution of the area around the target station, and a step function to express the discontinuity due to relocation. The PCA was applied to daily temperature (T_ave, T_max, or T_min) at stations within 300 km of the target station for July and August in each five-year period before and after the relocation, covering a total of ten years. Then, a least-squares criterion was applied to the daily temperature at the target station in the form

$${\displaystyle \sum _j}[T_j-\left(A_0+\sum _{k=1}^K{A}_k{f}_{jk}+B{S}_j\right){]}^2\to \mathrm{m}\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{m}\mathrm{u}\mathrm{m},$$

(1)

where T_j is the temperature at the target station on day j, f_jk is the score of the k-th component, and S_j is a step function defined as

S_j = 1 (before relocation),
0 (after relocation).

(2)

The term including f_jk indicates day-to-day changes of the regional temperature distribution, and the term including S_j gives the stepwise change due to relocation. The number of principal components, K, was set to five. They were selected in the order of highest correlation between the score and the temperature at the target station. Least-squares coefficients to be calculated are A_k (k = 0, ---, K) and B. The coefficient B was regarded as the temperature difference between the new and old sites, and was uniformly subtracted from the data before the relocation. This method is conceptually the same as that used by the JMA to correct for relocation bias in calculating climate normals [41]. The Supplementary File S1 provides a further explanation of the process of deriving Equation (1).

2.2. Evaluation of Urbanness Around Each Station

A variety of environmental factors affect the temperature at observation stations in cities. Some stations are located outside the central built-up area or in large parks, while surface air temperature can be affected by microscale factors on a kilometer scale or less [2,23,42,43,44,45,46]. However, the purpose of this study is not to explore such diversity, but rather to evaluate the urban contribution to observed temperature in hot summers on the spatial scale of the whole of Japan. For this purpose, this study uses population density as a measure of the degree of urbanness around each station. The population data in the 2015 national census were obtained on a grid of 30″ in latitude and 45″ in longitude, covering approximately 1 km × 1 km, from the “Portal Site of Official Statistics of Japan” operated by the Statistics Bureau, Ministry of Internal Affairs and Communications of Japan [20]. For each station, the population density of the surrounding area was calculated by

(3)

where P*(g) is the population in grid cell g, r_g is the distance of the grid cell from the station, and R (=3 km) is a parameter indicating the size of the area in which the population density is to be calculated. Nine stations with P < 600 km⁻² were regarded as non-urban. Previous studies have indicated a weak urban warming at stations down to P = 100 km⁻² [36,37], but the restriction on P was relaxed in the present study in order to ensure the number of non-urban stations, with the understanding that there may be a slight urban bias. Other stations were grouped into three categories satisfying 600 km⁻² ≤ P < 1000 km⁻², 1000 km⁻² ≤ P < 3000 km⁻², and P ≥ 3000 km⁻², respectively.

It is to be noted that the use of R = 1 km or R = 10 km instead of R = 3 km does not largely affect the relationship between P and the warming rate, as found in previous studies on recent temperature trends [36]. Also, there is no basic difference in results if the fraction of urbanized surface is used instead of P as a measure of the urbanness of the observation site [36].

2.3. Indices of Hotness During Summer

In most regions of Japan, the summer climate is characterized by a rainy period (Baiu) followed by a midsummer season dominated by the subtropical high centered in the north Pacific Ocean. The midsummer season begins in early to mid-July with the end of Baiu and lasts through August as the hottest period of the year. However, the time of the end of Baiu varies from year to year, and so do the weather conditions during the midsummer period, which causes year-to-year differences in the climate of the midsummer season. There are also considerable intraseasonal variations in the weather in summer, and extreme heat occasionally appears in June or September.

In the present study, the following indices for summer heat were defined for each station and each year.

• 〈T_●〉: Average of T_● for the 62 days in July and August.
• T_●⁽¹⁰⁾: Average of the top 10 T_● values among 122 days from June to September.
• T_●⁽³⁰⁾: Average of the top 30 T_● values among 122 days from June to September.
• T_●⁽⁵⁰⁾: Average of the top 50 T_● values among 122 days from June to September.

Here, T_● denotes any one of T_ave, T_max, and T_min. These indices were left missing if there were missing data on five or more days in July and August. Additionally, T_●⁽¹⁰⁾, T_●⁽³⁰⁾, and T_●⁽⁵⁰⁾ were left missing if there were missing data on 10 or more days from June to September. Although extremely high temperatures can be observed even in May and October in very rare cases, the analysis was limited to the period from June to September from the standpoint of focusing on the summer climatic conditions.

2.4. Analysis of Interannual Variation in Summer Heat Index

Each index defined in Section 2.3 was expressed as a deviation from the 1921–1940 average. Then, its year-to-year time series was separated into a year-specific component and a temporally smoothed component. The latter was obtained by applying a weighted running average with a Gaussian weight exp {−[(t − t₀)/τ]²}, where t₀ is the target year and τ = 10 years. The averaging was made for 41 years from 20 years before t₀ and 20 years after t₀, except for missing values and years outside the analysis period. The year-specific component was defined by the deviation of the original time series from the smoothed one, and will be denoted using a prefix Δ as ΔT_ave⁽³⁰⁾. Additionally, the running standard deviation (running SD) of the year-specific component was calculated for each year using the same Gaussian weight as above, as a measure of the amplitude of their interannual variation. This procedure was applied both to the time series at each station and the average time series for a group of stations.

The smoothed time series was further divided into two parts. One is the average over stations at non-urban sites (P < 600 km⁻²), and the other is the deviation from it. Hereafter, they will be denoted by “non-urban” and “urban” time series, respectively. The non-urban time series is considered to include the global climate change, namely global warming, as well as long-term variations arising from changes in the atmosphere–ocean system. The urban time series is regarded as indicating urban warming, on the assumption that the non-urban change is spatially uniform.

For the supplement, a centennial linear trend of each index over the 104 years (1921–2024) was calculated for each station and each group of stations. Additionally, the centennial standard deviation (centennial SD) was obtained from the SD of the year-specific components over the 104 years.

3. Results

3.1. Separation of Year-Specific Components and Non-Urban and Urban Changes

Figure 2 shows the time series of T_max⁽³⁰⁾ for stations at non-urban sites (P < 600 km⁻²) and in highly urbanized areas (P ≥ 3000 km⁻²). While there are long-term variations, T_max⁽³⁰⁾ has risen over the period as a whole for both non-urban and urban stations, with a higher trend in the latter than in the former. The summers of 2023 and 2024 had record-breaking heat, exceeding the previous record in 1994. The highest temperature in the first half of the period was in the summer of 1942.

Figure 3 shows the time series of year-specific components for T_ave⁽³⁰⁾, T_max⁽³⁰⁾, and T_min⁽³⁰⁾ for P ≥ 3000 km⁻² stations. The smoothed non-urban and urban time series, and the running SD of the year-specific components, are also shown. Figure 4 is for 〈T_max〉, T_max⁽¹⁰⁾, and T_max⁽⁵⁰⁾. The running SD has peaks around 1940 and from the 1980s to the 1990s in common to the panels in Figure 3 and Figure 4, indicating large interannual variations in these periods. The urban component increased after the 1960s, but has been nearly constant since the 2000s, whereas the non-urban component increased instead. In the hot summers of 2023 and 2024, the contributions of the year-specific, non-urban, the urban components were all of the order of 1 °C. In contrast, the hotness of the 1942 summer was attributed almost solely to the year-specific anomaly. In the following, the climatological characteristics of each component are overviewed.

3.2. Climatological Aspects of Each Component

3.2.1. Year-Specific Components

It can be seen from Figure 3 that the amplitude of interannual variation in ΔT_max⁽³⁰⁾ is larger than that of ΔT_min⁽³⁰⁾. The centennial SDs of ΔT_max⁽³⁰⁾ and ΔT_min⁽³⁰⁾ for the entire 68 stations are 0.68 °C and 0.48 °C, respectively. Additionally, the variation of Δ〈T_max〉 tends to be larger than that of ΔT_max⁽¹⁰⁾ and ΔT_max⁽³⁰⁾ (Figure 3 and Figure 4). The centennial SD of Δ〈T_max〉 for the 68 stations is 0.91 °C.

Figure 5 shows the relationship between P and the centennial SD of ΔT_max⁽³⁰⁾ and ΔT_min⁽³⁰⁾ at each station. There are no significant relationships between P and the centennial SDs, indicating that the amplitude of the variation in year-specific temperature anomaly is independent of urban effects. On the other hand, there is some regional difference in the magnitude of year-to-year temperature variation. Figure 6 shows the distribution of the centennial SD of ΔT_max⁽³⁰⁾. The centennial SD is larger in northern Japan than in other regions, in agreement with the existing knowledge that summer temperature variability is large in northern Japan, especially on its eastern side [47]. However, the variations are highly coherent over the analysis area. For example, regionally averaged year-to-year values of ΔT_max⁽³⁰⁾ in northern Japan and those in western Japan have a correlation coefficient of 0.91.

3.2.2. Non-Urban and Urban Trends

Figure 7 shows the relationship between P and the centennial linear trends of T_max⁽³⁰⁾ and T_min⁽³⁰⁾. Both are positively correlated with P, with a stronger dependence on P for T_min than for T_max. Figure 8 shows the centennial trend for each index for stations in each category in P. For both non-urban and urban stations, the trend for the T_min index is larger than that for T_max. It is to be noted, although not shown, that the non-urban trend of 〈T_ave〉 is 0.69 °C century⁻¹ and is close to the global warming rate, which is 0.85 °C century⁻¹ for the same period (July–August for 1921–2024) according to the global mean temperature data provided by JMA [35].

The difference in trends among indices (〈T_●〉, T_●⁽¹⁰⁾, T_●⁽³⁰⁾, and T_●⁽⁵⁰⁾) is not noticeable for non-urban stations. For urban stations, the trend of T_max⁽¹⁰⁾ is greater than that of T_max⁽³⁰⁾, and they are greater than the trend of 〈T_max〉. The trends of the differences between T_max⁽¹⁰⁾ and 〈T_max〉 and between T_max⁽³⁰⁾ and 〈T_max〉 are 0.4−0.5 °C century⁻¹, which is significant at least at the 5% level. Thus, a more extreme index for T_max tends to have a larger urban trend.

There is considerable variation in linear trends among stations in Figure 7, implying the contribution of site-specific factors other than P on temperature increase. Trends of T_max⁽³⁰⁾ are more variable than those of T_min⁽³⁰⁾, and are particularly large at Maebashi and Kumagaya (M and K in Figure 1 and Figure 7), with a rate of 3.9 °C century⁻¹ and 3.5 °C century⁻¹, respectively. These stations are to the northwest of the Tokyo metropolitan area (Figure 1), and have experienced extremely high temperatures in recent decades, with the national record of 41.1 °C at Kumagaya in 2018 [48]. Figure 9 shows the time series of ΔT_max⁽³⁰⁾ and the urban warming component at these stations. The rapid urban warming since the 1960s reached 2 °C or more by 2000, although it slowed thereafter. This warming was found to be accompanied by a decrease in daytime atmospheric pressure [49], which indicates temperature increase over a deep layer. The mechanism of the warming remains to be further explored.

4. Discussion

This study has separated the year-specific anomalies from the long-term variability in the centennial time series of temperature indices for summer heat in Japan. The latter was further separated into non-urban and urban components. The contributions of year-specific climate anomalies, nationwide non-urban warming, and urban warming in highly urbanized areas (P ≥ 3000 km⁻²) to recent record-breaking summers are all evaluated to be of the order of 1 °C. The non-urban warming can be regarded as mainly ascribed to global warming, while urban warming also contributes to hot summers in highly urbanized areas with a magnitude comparable to global warming.

As minor differences according to temperature indices, the amplitude of year-to-year variations is larger for indices in T_max than those for T_min, and the summer mean temperature shows a larger year-to-year variation than the temperature on days of extreme temperature. The latter fact may be against intuition, but can be explained in a way that a seasonally defined hot summer is not only characterized by the peak temperature, but the length of high temperature periods.

The magnitude of urban warming is larger for T_min than for T_max, consistent with previous knowledge of urban climate [2,22,23]. Nevertheless, indices for T_max show statistically significant urban warming, indicating that daytime heat is also urban-influenced. Moreover, the effect of urban warming is larger on days of extreme temperature (T_max⁽¹⁰⁾ and T_max⁽³⁰⁾) than for the average temperature in summer (〈T_max〉). This may reflect the fact that sunny days are more frequent during hot periods, which makes urban heat islands more intense. A similar feature has been reported in previous studies in terms of the synergy between heat waves and urban heat islands [50,51,52]. On the other hand, urban warming has slowed down in the recent few decades in Japan [53], while global warming is expected to further develop in the future and become a major contributor to urban temperature rise, as shown in modeling studies on climate projection in large cities [54,55,56].

It is to be noted that the temperature data from the 15 JMA stations used for climate monitoring show a warming rate of 1.12 °C century⁻¹ for the same period as the present study, namely, July and August of 1921–2024. This is substantially larger than the non-urban trend of 0.69 °C century⁻¹ for 〈T_ave〉 evaluated in the present study. Some of the 15 stations are at moderately urbanized sites of P ≥ 1000 km⁻², which means the temperature data are likely to be affected by urban warming [38].

In the following, methodological and conceptual limitations in the results of this study are discussed. A methodological source of uncertainty is the parameter choice in calculating the long-term component, while the intricate relationship between atmospheric and oceanic circulation anomalies and long-term climate variability introduces conceptual ambiguity.

4.1. Sensitivity of Results to the Procedure of Calculating the Non-Urban Component

A series of calculations for the non-urban component was made under different conditions, and the results were compared with the original one (C0). The calculation conditions were given as follows.

• C1: The smoothed time series was calculated using a weight exp {−[(t − t₀)/τ]²} that was defined by τ = 5 years instead of τ = 10 years and was applied to the 21 years from 10 years before t₀ to 10 years after t₀.
• C2: The smoothed time series was calculated from a weighted least-squares method including a linear trend term, applying a weight exp {−[(t − t₀)/τ]²} with τ = 20 years to the 81 years from 40 years before t₀ to 40 years after t₀.
• C3: The non-urban time series was defined from the data at stations with P < 1000 km⁻², instead of those with P < 600 km⁻².

Figure 10 shows year-to-year values of T_max⁽³⁰⁾ for non-urban stations, together with the smoothed time series calculated based on each condition. The time series by C1 is more winding than that by C0, while that by C2 is smoother. The time series by C3 deviates positively from that by C0 from around the 1960s. However, these deviations are not so large as to affect the overall results obtained from C0. The centennial SD of ΔT_max⁽³⁰⁾, which is defined by the deviation of each time series, is 0.60 °C, 0.58 °C, 0.61 °C, and 0.70 °C for C0, C1, C2, and C3, respectively.

Table 1. Year-specific, non-urban, and urban contributions to the 2024 anomaly for P ≥ 3000 km⁻² stations calculated from each condition.

	Tave(30)				Tmax(30)				Tmin(30)
	C0	C1	C2	C3	C0	C1	C2	C3	C0	C1	C2	C3
Year-specific (°C)	1.00	0.70	0.73	1.00	1.12	0.79	0.82	1.12	1.02	0.73	0.74	1.02
Non-urban (°C)	1.01	1.29	1.21	1.24	0.99	1.30	1.25	1.22	1.26	1.56	1.46	1.47
Urban (°C)	1.03	1.06	1.09	0.81	0.78	0.84	0.79	0.55	1.19	1.19	1.27	0.98

shows the values of the year-specific, non-urban, and urban components for T_ave⁽³⁰⁾, T_max⁽³⁰⁾, and T_min⁽³⁰⁾ in the 2024 hot summer for P ≥ 3000 km⁻² stations. The conditions C1 and C2 give smaller values of year-specific components and larger values of non-urban components than C0, while C3 gives a larger value of the non-urban component and a smaller value of the urban component. The differences from the value by C0 are approximately 0.3 °C at the largest. Thus, there can be uncertainty of about 0.3 °C in the contribution of each factor to the 2024 summer. The situation is common to other indices not shown here.

4.2. Complications Arising from Coupled Changes in the Atmosphere–Ocean System and the Climate

Year-to-year changes in the atmosphere–ocean system are found to influence summer climate in Japan in various aspects, including teleconnection [7,10,11,13,14,57,58]. Figure 11 shows the relationship between ΔT_max⁽³⁰⁾ for non-urban stations and the teleconnection indices for the Pacific Decadal Oscillation (PDO), the Southern Oscillation (SO), and the Arctic Oscillation (AO). Their monthly data were obtained from websites [59,60,61]. The PDO index has a negative correlation of −0.49 with ΔT_max⁽³⁰⁾, which is statistically significant at the 1% level, in agreement with the general understanding that a positive/negative phase of PDO tends to be accompanied by a negative/positive temperature anomaly over Japan [58,62]. Additionally, the SO index and ΔT_max⁽³⁰⁾ have a positive correlation of 0.28, and the AO index and ΔT_max⁽³⁰⁾ have a positive correlation of 0.25. These are significant at the 5% level. These indices also have multidecadal variations that can affect temperature. Figure 12 shows the time series of temporally smoothed PDO, SO, and AO indices, together with the non-urban time series, and the SD of ΔT_max⁽³⁰⁾ for non-urban stations. The PDO and SO indices have multidecadal variations in approximately opposite phase with each other, and positive PDO (negative SO) periods correspond to large SD with relatively low ΔT_max⁽³⁰⁾.

Apart from the PDO and SO, ocean-related long-term variabilities are found in temperature and its extremes on the global scale [63,64]. Some studies have indicated recent changes in circulation fields resulting in an increase in hot summers or heatwaves, and a decrease in cool summers over Japan [12,57,58,65,66]. From these aspects, year-specific anomalies and global warming are mutually related, and their linear combination presented in this study can be regarded as simply a first approximation of climate change.

5. Summary

An attempt was made to evaluate the contribution of year-to-year climate anomalies, nationwide non-urban climate change, and local urban warming to hot summers in Japan, on the basis of a number of indices for the summer heat. The urban component appeared after 1960 and has been nearly constant since the 2000s, whereas the non-urban component increased instead. As a result, the contributions of year-specific, non-urban, and urban components to the record-breaking hot summers in 2023 and 2024 were all evaluated to be on the order of 1 °C in highly urbanized areas. However, the atmospheric and oceanic processes that affect climate variations are so complicated that there can be interactions between interannual climate changes and global warming, and it is simply an approximation to view hot summers as their linear combination.