Changes in the diurnal temperature range over East Asia from 1901 to 2018 and its relationship with precipitation

Since the 1950s, the East Asian diurnal temperature range (DTR), defined as the difference between the daily maximum (Tmax) and minimum temperatures (Tmin), has gradually decreased. Precipitation changes have often been cited as a primary cause of the change. However, the East Asian DTR change before 1950 and its relationship with precipitation remain unclear. Here, we used a newly developed China Meteorological Administration-Land Surface Air Temperature dataset v1.1 to examine the climatological patterns and long-term trends of the DTR in East Asia from 1901 to 2018 and its relationship with precipitation. The mean annual DTR averaged over East Asia for 1951–2018 was approximately 10.0 °C. East Asian DTR changes during 1901–2018 show two distinct characteristics. First, the DTR decreased significantly by approximately 0.60 °C during 1901–2018, and the decrease rate in the second half of the twentieth century (by ~0.53 °C) was significantly larger than that over the rest of the Northern Hemisphere and the global land due to rapid urbanization over East Asia. Second, before the 1950s, the DTR in East Asia showed a significant non-linear increase especially in middle latitude areas, mainly due to the warming rate of Tmax is higher than that of Tmin. Additionally, we found that the spatial pattern of long-term DTR change shows a significant negative correlation with mean precipitation patterns except in arid and semi-arid areas during 1901–2018. The decreasing trend of DTR gradually became smaller from arid regions to humid regions during 1901–2018, mainly because the difference between Tmax and Tmin warming rate gradually became smaller.


Introduction
Daily maximum (T max ) and minimum (T min ) surface air temperatures have shown significant asymmetric warming since the 1950s over most land areas, with more significant increases in nighttime T min than in daytime T max , resulting in decreases in the diurnal temperature range (DTR; DTR = T max -T min ) worldwide (e.g., Karl et al. 1991;Easterling et al. 1997;Vose et al. 2005;Alexander et al. 2006; Thorne et al. 2016;Sun et al. 2019). The DTR is coupled to cloudiness, precipitation, soil moisture, and evaporation; thus, its change can provide insights into changes in the hydrological cycle (Dai et al. 1997(Dai et al. , 1999. Moreover, previous studies have also found that DTR changes significantly impact crop yields (Lobell 2007) and ecosystems (Vasseur et al. 2014). Therefore, it is important to study the characteristics and causes of DTR changes under global warming.
Previous studies have tried to determine the causes of the long-term global and regional DTR changes. It has been found that changes in the DTR are directly related to the surface solar radiation (Makowski et al. 2009;Wild 2012;Wang and Dickinson 2013). Decreases in the DTR driven by cloud cover, precipitation, and the related soil moisture after 1950 all resulted from their asymmetrical impacts on the surface solar radiation and evaporative cooling that directly affect daytime T max (Dai et al. 1997(Dai et al. , 1999Zhou et al. 2009). Among these factors, precipitation may affect daytime T max and thus the DTR through its association with clouds (which affect solar radiation and thus T max ), its impact on soil moisture (which affects surface evaporation and thus T max ), or its direct cooling effect on daytime T max (Dai et al. 1997(Dai et al. , 1999. The negative correlation between the DTR and precipitation results mainly from precipitation's association with cloudiness and soil moisture (Dai et al. 1997;Zhou et al. 2009). This significant negative correlation between precipitation and the DTR was found to be especially apparent in the warm season (Zhou et al. 2009), as dry conditions are more associated with solar radiation and less associated with evaporative cooling (Trenberth and Shea 2005). In addition, greenhouse gases, aerosols, land-use changes (e.g., urbanization), and other climate factors can also indirectly cause DTR changes by affecting clouds and solar radiation (Gallo et al. 1996;Braganza et al. 2004;Wang et al. 2012).
As a rapidly developing region, East Asia has experienced unprecedented economic growth, resulting in great changes to the natural environment. Previous studies have found that the DTR over East Asia has decreased significantly since the 1950s (e.g., Zhai and Pan 2003;You et al. 2016;Liu et al. 2016;Sun et al. 2017a). However, the causes for the decline in the DTR over East Asia since the 1950s are not consistent with those worldwide. The decrease in the DTR in East Asia could not be explained by cloud cover (Dai et al. 1999), but is also related to the changes of aerosol increases (Liu et al. 2016), land-use changes (Ren and Zhou 2014), atmospheric water vapor (Zhao 2014), and precipitation (Shen et al. 2014). Among these factors, atmospheric water vapor and precipitation can statistically explain more than 40% of the DTR decrease over China since the 1950s (Zhao 2014;Shen et al. 2014), but physically water vapor may only have a small effect on the DTR (Dai et al. 1999). However, due to the limitation of early data quality, the previous studies on East Asian DTR mainly started in the 1950s. The change in the DTR over East Asia from 1901 to 1950 and its relationship with precipitation have not yet been investigated. In this study, we use the China Meteorological Administration-Land Surface Air Temperature dataset v1.1 (CMA-LSAT v1.1) to analyze the climatology and long-term characteristics of DTR changes over East Asia for the period from 1901 to 2018 and then examine the association of T max , T min , and DTR changes on precipitation in East Asia.
The rest of the paper is organized as follows. The data and methods used in this study are introduced in Sect. 2. In Sect. 3.1, we analyze the climatological characteristics of the DTR over East Asia from 1951 to 2018. Then, the DTR trend and its dependence on precipitation over East Asia are presented in Sects. 3.2-3.3. There are some discussions in Sect. 4. Finally, a summary of the key findings is given in Sect. 5.

Temperature data
In the present study, the DTR is defined as the difference between the daily T max and T min . The observational T max and T min data are from the CMA-LSAT v1.1 dataset (see supplementary material for the detailed dataset introduction), which provides homogenized temperatures from 1901 to 2018 over global land derived from observations made at weather stations (Sun et al. 2017b;Xu et al. 2017). CMA-LSAT v1.1 data cover East Asia (80°-150°E, 0°-60°N) fairly uniformly, especially in the northwest of East Asia (Sun et al. 2017b;Xu et al. 2017). For example, 24 stations in China, which have more than 100 years of records, are added to the new dataset.
In this study, to reduce the uncertainty of DTR calculation results throughout the whole research period and ensure that all East Asian stations have long enough observation records in the base period 1961-1990, we select stations according to the following selection standards: Case 1: To ensure that the observation records of stations in the climate reference period are relatively complete, we only select stations with records of at least 15 years in the base period and at least 3 years of records for every 10 years during the base period 1961-1990. Case 2: If there are 2 months of missing values in a year, the year will be recorded as missing values. Case 3: Because there are noticeable differences between the DTR on land and islands, we only choose the observation stations on land and remove the island stations in the western Pacific, the Indian Ocean, and the South China Sea far away from the mainland. Case 4: According to the principle of retaining as many stations as possible, only stations with records longer than 30 years are selected when constructing the time series during 1901-2018. Meanwhile, cases 1 and 2 should also be met. Case 5: In the process of calculating the series and grid box trends in the periods 1901-1950and 1951, in order to reduce the impact of short record stations, only stations with more than 25 years of observation records in the periods 1901-1950 and 1951-2018 are selected to participate in the calculation. Case 6: In the process of determining the difference in the time series of stations with different record lengths (Sect. 3.2), we also select stations with more than 100 years of records to participate in the comparison.
Finally, according to cases 1-4, we select 1968 stations in East Asia that met these conditions over East Asia (Table 1). To reduce the uncertainty caused by the stations' non-uniform spatial distribution, we divided all the stations into 124 grid boxes with a resolution of 5°× 5°. Figure 1a-c shows the spatial distribution of stations in the periods 1901-1920 (a), 1920-1940 (b), and 1940-1960 (c) over East Asia. Before 1920, East Asia stations (total 178) are mainly distributed in Japan, Russia, Philippines, India, southeast China, and northeast China (Fig. 1a). From 1920 to 1940, the number of stations increased to 695 (Fig. 1b). From 1940 to 1960, the number of stations increased rapidly to 1666 (Fig. 1c), and the number of stations is close to that of all stations from 1901 to 2018 (Fig. 1d). Figure 1e shows that stations and grid boxes' coverage increased rapidly from 1920 to 1950, reaching 50% in 1933 and stabilizing at more than 80% after 1950.
Considering the differences in the DTR between the marine continental region and inland, we divided East Asia into three 20°latitude zones for sub-regional study. Before the 1940s, the grid box coverage in mid-latitudes is significantly higher than in low-latitudes. The grid box coverage of low latitudes, mid-latitudes, and high latitudes reached 50% in 1950, 1925, and 1933, respectively (Fig. 1f).
Consequently, East Asian stations before 1920 are mainly distributed in Japan, Russia, Philippines, India, southeast China, and northeast of China. From then on, the coverage of stations and grid boxes increased rapidly, reaching 50% in 1933 and stabilizing at more than 80% after 1950. Therefore, it can be considered that the DTR changes after 1933 have relatively higher credibility.
In addition, it is worth mentioning that both East Asia and low-and mid-latitude regions have experienced a decline in grid box coverage and stations in the 1930s-1940s. To explore whether the reduction in grid box coverage dominates the low DTR values in the 1930s, we will clarify this problem by comparing different time series in Sect. 3.2.

Other data
To analyze the association of the DTR with precipitation, rain gauge precipitation data from the China Meteorological Administration-Global Historic Precipitation dataset ) are also used. We only selected stations with at least 15 years (30 years) of observational records in the base period 1961-1990 (during 1901-2018). We only use the precipitation grid boxes to match the DTR grid boxes (see Supplementary material Fig. S1). The number of match grid boxes is also close to the number of all DTR grids. This suggests that the precipitation data is relatively reliable.
In addition, to further assess the uncertainty caused by urbanization, 2010 land cover data from satellite remote sensing with a resolution of 300 m from the European Space Agency (Hollmann et al. 2013) is used in this study to distinguish between urban and rural stations. The NCAR/NCEP twentieth-century reanalysis version 2 (interpolated to 5°resolution grid boxes) is used to estimate the grid box coverage error. The CRUTS4.0.3 grid data from 1901 to 2018 are used to compare with the CMA-LSAT v1.1 data. Before comparison, the data is interpolated to 5°resolution grid boxes.

Grid box coverage error and urbanization impact estimation method
There are some uncertainties in the long-term changes of land surface air temperature, especially before the 1950s (Jones 1994;Brohan et al. 2006). Even if the stations are selected strictly according to Sect. 2.1.1, there are still some uncertainties throughout the study period. The primary sources of uncertainty in the current estimation of land temperature series are station error, grid box coverage error, urbanization, etc. (Brohan et al. 2006;Li et al. 2010;Du et al. 2012).
Referring to the method used by Du et al. (2012), the grid box coverage error is estimated approximately based on the NCAR/NCEP twentieth century reanalysis data. Taking a grid box as an example, the estimation method is as follows: First, according to the observation grid boxes, obtain sampling grid boxes from complete reanalysis grid boxes by sampling; Second, we can obtain complete time series and sampling time series on the grid box scale, then the difference between two series is the difference series on grid box scale; Finally, for East Asia as a whole, we calculate the standard deviation of the difference series across all grid boxes in East Asia for a certain year, which is the grid box coverage error for the certain year. We use the isolated forest algorithm in the machine learning method invented by Zhang et al. (2021) to select urban and rural stations. Calculation steps: first, we set the United States Climate reference network (USCRN) as a standard climate network, and then use the land use data around USCRN from the European Space Agency to train the model parameters; Second, the trained parameters in the first step are substituted into the isolated forest model, and then the Asian stations in 2010 are classified into urban stations and rural stations through the model (see supplementary material for detailed method introduction). Finally, East Asian stations in 2010 can be divided into 1436 urban stations and 346 rural stations (see Supplementary material Fig. S2a,b). The urbanization land within 12 km around the rural station is less than 5%.
The main purpose of using land use data in 2010 is to ensure that stations defined as rural are not affected by urbanization until the last few years. At the same time, using the 2010 land use data can avoid missing some stations due to data updates in the past five years. However, it may still include some problems such as the relocation of stations, and the residual urbanization effect in the rural data series due to the lack of real rural stations. As a result, the impact of urbanization will be underestimated to a certain extent. Therefore, the conclusion of the urbanization impact estimation by this method is a conservative estimate and will not overestimate the urbanization impact.
The formula for the urbanization contribution rate (C u ) is as follows (Ren and Zhou 2014): where ΔT all − rural = T all -T rural , T all and T rural represent the temperature series of all stations and rural stations.

Other methods
Classification of dry/wet regions: According to annual mean precipitation in every grid box, we divided East Asia into four different dry and wet regions. An annual mean precipitation of 0-400 mm, 400-800 mm, 800-1600 mm, and more than 1600 mm are used to indicate arid and semiarid areas, semi-humid areas, humid areas, and extremely humid areas, respectively. The precipitation grid boxes used in calculation are those matching DTR. We first averaged the station T min , T max , and DTR anomalies (relative to 1961-1990) within each grid box to derive the grid box mean value, requiring at least one station with data within each box. The gridded anomaly data were regionally averaged over East Asia (overland boxes only) with the grid box area as weight. The linear trends of the anomaly series were obtained by using the least-squares fitting, and the significance of the linear trends is judged using the two-tailed Student's t test. The trend is considered statistically significant if it is significant at the 5% (p < 0.05) significance level.

Climatological characteristics of the DTR over East Asia
In this section, we uses all available observation records to calculate the climatological characteristics of the DTR in East Asia and three sub-regions during 1951-2018, as shown in Fig. 2a,b. The observed annual mean DTR over East Asia is approximately 10.0°C (2.8-18.8°C). From the spatial patterns (Fig. 2a), we can determine that the DTR gradually increased from coastal to the Asian continent, and the DTR of the marine continent (0-20°N ) and coastal grid boxes are significantly lower than those of inland grid boxes. Larger DTR is seen in northwest China and the Hindu Kush-Himalayas region, where the annual mean DTR is more than 12.5°C. On the other hand, with the increase of latitude, the proportion of land stations increases, and the zonal average DTR increases (Fig. 2b). In addition, the DTR in the northwest of East Asia is smaller than that in the surrounding area, which indicates that the DTR does not increase with latitude, and the DTR in the zonal average shows the difference in the distribution of sea and coastal grid boxes. Furthermore, the annual cycle of the DTR over East Asia is smaller in the warm season than in the cold season, and the amplitude at high latitudes is larger than that at low latitudes (Fig. 2b). Before 1950, the DTR shows non-linear increasing. The reason for the non-linear change is mainly due to the low value of DTR in the 1930s. Additionally, we compare the time series of different data sources to verify the credibility of the long-term changes in DTR. Furthermore, the primary purpose of comparing time series with different record lengths is to verify whether the change in DTR is caused by the increase of stations. Figure 3a shows the comparison of the CMA-LSAT series and CRUTS4 series. In general, both the CMA-LSAT time series and the CRUTS4 time series show the uniform characteristics of a slow increase trend in the first half of the twentieth century and a significant decrease in the second half of the twentieth century, and both of them also have a low DTR value in the 1930s. The high value of DTR in 1940s-1950s is more evident in the CMA-LSAT series than    in the CRUTS4 series. Before the 1950s and after the 1980s, the CRUTS4 series is obviously lower than the CMA-LSAT series.

Long-term changes in the DTR over East Asia
It is worth noting that the East Asian grid box coverage only reached 50% in 1933, and then increased to more than 80% after 1950, which implies that the results of DTR estimation before the 1950s, especially before the 1930s, may contain great uncertainty. Therefore, we estimate the grid box coverage error of East Asian DTR for each decade, and the results are shown in Fig. 3a (green line). It can be found that the uncertainty of the DTR remained at very high levels before the 1950s (more than 0.2°C), and then the uncertainty level decreases with the increasing of grid boxes. After the 1950s, the grid box coverage error decreased to approximately 0.1°C. Figure 3b,c shows a piecewise time series based on stations with a record length longer than 25 years before and after 1950 (Fig. 3b), and a long record series based on stations with a record length of more than 100 years (Fig. 3c). Overall, the two time series and all station series (Fig. 3a dark blue curve) have relatively consistent interdecadal fluctuations before the 1940s. The difference is that the DTR of the long record series is lower before 1920s and higher in the 1950s than the all station series, which results in an increase trend of DTR before the 1950s. In addition, all station time series, long record time series, and piecewise time series show low DTR values in the 1930s. This suggests that the reduction in grid box coverage during the 1930s to 1940s does not cause the low DTR values. Figure 3d compares the 11 years smooth of the DTR series in three latitudes during 1901-2018. The DTR of the three sub-regions decreased significantly after the 1950s. Before the 1950s, the DTR in high and middle latitudes show decrease and increase, respectively. Because there are few stations in the low latitude marine continental region before 1950, the DTR in low latitude changes always with a wide range of changes. Although the low-latitude DTR fluctuates greatly, the grid is less before 1950, and therefore it cannot have too much influence on the average large-scale DTR in East Asia. The DTR trend reversal phenomenon occurs only in the mid-latitude of 20-40°N. This trend reversal phenomenon is similar to that reported for the global DTR by Sun et al. (2019).
With regard to DTR spatial distribution in different periods, there are also obvious differences in DTR changes. To minimize the impact of short-time series stations on the trend estimation of DTR, we only select stations with records longer than 25 years in the periods 1901-1950and 1951. After 1950, the DTR over East Asia shows a decreasing trend at widespread locations, especially at broad land stations. Before 1950, the grid boxes constructed by stations with a record length of more than 25 years are mainly distributed in East China, Russia, Japan, and the Philippines. Before 1950, the low latitude shows a significant decreasing trend. In the middle latitudes, 86% of the grids show a significant increase in DTR. The grid boxes with a significantly decreased and increased trend in high latitude account for 50% and 50%, respectively. Furthermore, we also evaluated the effect of T max and T min on DTR changes in these two periods. Figure 4b shows the relationship between T max and T min trends and DTR trends in 1901-1950 and 1951-2018. The warming rate of T min has been more extensive than that of T max in the periods 1901-1950 and 1951-2018, which directly caused the significant decline in the DTR. The DTR's dependence on T max shows differences in different periods (Fig. 4c,d red  lines). Before 1950, the DTR increases significantly with the increase in T max , but after 1951, the DTR's dependence on the increase in T max weakened. These conclusions indicate that nonuniform changes of T max dominate the difference of DTR trend in different periods. Simultaneously, the long-term rapid increase in T min caused the decline in the DTR.
To determine whether the DTR change is seasonal, we also calculate DTR trends over the whole study period (Fig. 5). In general, the long-term decrease of the DTR exists in all seasons, but the magnitude of the changes is different. The decreasing trend in winter (DJF) is often larger than that in the other three seasons. These conclusions are also consistent with the characteristics of DTR changes over East Asia since 1950 reported by Liu et al. (2016Liu et al. ( , 2018.

Relationship between the DTR and precipitation over East Asia
Although precipitation has no direct effect on DTR, it could indirectly affect DTR by affecting cloud cover and solar radiation, etc. To determine the indirect role of precipitation in changes of the DTR over East Asia, we examined the association between DTR and precipitation. In general, there is a significant negative correlation between the DTR and precipitation over East Asia during 1901-2018. Furthermore, we also found that the association between the DTR and precipitation is not significant in arid and semiarid regions, but correlations are significantly Fig. 4 (a, b) Spatial distribution of the East Asian DTR trend in 1901-1950 and 1951-2018. The grid boxes marked with X represent the trend is not statistically significant (p > 0.05). (c, d) The relationship between T max (red), T min (blue) trends, and the DTR trends in 1901-1950 and 1951-2018. The red and blue dot lines represent the linear trends of the relationship between T max , T min , and the DTR, respectively correlated in other regions (Fig. 6b). This is also similar to the conclusion that the correlation between global precipitation and the DTR in arid areas is not significant after 1950 (Zhou et al. 2009). Dai et al. (1999) indicated that the DTR in arid areas was more affected by solar radiation than by evaporative cooling. This can theoretically explain why the long-term changes in the DTR over East Asia had no significant relationship with precipitation since 1901.
To further determine the relationship between T max , T min , and DTR's long-term changes to annual mean precipitation over East Asia (Fig. 6b), it is noted that from arid regions to humid regions, T max and T min trends show an increasing trend, but the warming rate of T max relative to T min is more significant from 1901 to 2018. However, the difference between T max and T min  warming rate is gradually smaller, from arid regions to humid regions. The above causes a slower decline in DTR, from arid regions to humid regions. Although there is no direct connection between precipitation and DTR in physical processes, precipitation and DTR changes are statistically related from 1901 to 2018. The above shows that even if the direct influence factor data of DTR is lacking, the indirect factor of precipitation can also reflect the possible change characteristics of DTR.

Long term decline of DTR in East Asia
Due to data limitations, previous studies mainly started from the 1950s, and few studies focused on the change of the DTR since 1901. Recent studies indicated that the rapid increase in T min in the Northern Hemisphere and the globe caused the DTR decrease since 1901 (Thorne et al. 2016;Sun et al. 2019). Table 2 shows a comparison of the global, Northern Hemisphere, and East Asian DTR changes in different studies. Since 1901, the decline of the DTR in East Asia (-0.6°C) is significantly lower than that in globe (-0.41°C) and the Northern Hemisphere (-0.42°C). After 1951, the DTR also declined rapidly in East Asia.
A previous study indicated that global temperature trends could be estimated using only 172 "independent stations" (Jones 1994). Still, the conclusion was drawn based on a basic understanding of global climate change characters. Although there is significant uncertainty in the surface temperature series before 1950, it does not mean that the temperature in that period cannot reflect the primary climate change because the temperature change can reflect the climate of a broad range (Tang et al. 2009). In the statistical, better spatial coverage can reduce uncertainty in time series (Thorne et al. 2005;Brohan et al. 2006).
The reason for the decreased global and regional DTR is generally considered to be the synergistic effects of solar radiation, cloud cover, aerosols, precipitation, water vapor, and land use (Dai et al. 1999). It is also found that the solar radiation reaching the ground shows a significant negative correlation with cloud cover and aerosol at a global or regional scale (Dai et al. 1997;Liu et al. 2016). Dai et al. (1997Dai et al. ( , 1999 found that the increase of cloud cover after 1950 could explain the global DTR's spatial distribution by over 80%. However, cloud cover in China and East Asia cannot explain the DTR decline after 1950 (Shen et al. 2014;Liu et al. 2016). In East Asia, there were few clouds and aerosol observation stations in the first half of the twentieth century, and therefore it is not easy to evaluate these  1901-2018 1951-2018 1901-2018 −0.53 −0.60 factors' impact on the DTR. This paper found that the DTR in East Asia has a significant correlation with precipitation during 1901-2018. Although there is no direct connection between precipitation and DTR in physical processes, precipitation and DTR changes are statistically related from 1901 to 2018. This implies that that even if the direct influence factor data of DTR is lacking, the indirect factor of precipitation can also reflect the possible change characteristics of DTR.

The impact of urbanization on DTR during 1951-2018
Urbanization is a crucial source of uncertainty in temperature trend estimation. Due to the influence of human activities, it is found that the cloud cover and aerosol level showed a significant increase in the past decades (Croke et al. 1999;Wang et al. 2009), especially in urban areas (Romanov 1999). The impact of urbanization on the DTR can generally be attributed to two interrelated mechanisms (Arnfield 2003;Oke et al. 2017;Varquez and Kanda 2018). In the daytime, the albedo of the underlying urban surface is lower than that of rural areas, and the sensible heat of the city is greater than that of the countryside due to more solar radiation being absorbed in the city. At the same time, the cooling effect of evapotranspiration in urban areas is smaller than in rural areas (Varquez and Kanda 2018). These caused the city's temperature to be much higher than that in the countryside. On the other hand, due to the city's complex three-dimensional structure, urban buildings absorb and reflect infrared radiation at night time. In addition, urban thermal inertia causes the heat absorbed in the city during the daytime and released at nighttime. The synergistic effect of the two causes leads to a more substantial heat island effect at nighttime than during daytime. The intensity of the above two causes is related to the city size (Arnfield 2003;Oke et al. 2017). Under the influence of more substantial nighttime urban heat island at night, T min in urban areas decreases more slowly than in rural areas (Ren and Zhou 2014; Jiang et al. 2020). As a result, the DTR of urban stations is usually smaller than surrounding rural stations (Yang et al. 2013;Ren and Zhou 2014;Wang et al. 2018). East Asia is one of the regions most clearly affected by urbanization globally during the past decades (Jones et al. 2008;Ren and Zhou 2014). For example, the impact of urbanization on the annual mean temperature in China can reach 27% (Zhang et al. 2010), which is higher than the 10% global average assessed by Jones et al. (1990) and the IPCC (2013). However, stations with early observation records are mainly concentrated in large cities; thus, it is difficult to assess the impact of urbanization on the DTR before 1950 due to the lack of data from rural stations. Therefore, we only calculated the contribution rate of urbanization to T max , T min , and the DTR after 1951, applying the method by Zhang et al. (2021). The results show that the impact of urbanization on T max (25.9%) is smaller than that on T min (28.2%). This is also consistent with the conclusion that urbanization has a more noticeable impact on the nighttime temperature or T min . The impact of urbanization on the DTR reaches 43.6%, which indicates that the DTR is more sensitive to urbanization (Fig. 7). These limitations need to be addressed through early data rescue and reasonable correction for urbanization bias in future studies. Therefore, these indicate that the contribution of urbanization leads to underestimating the DTR trend. However, it cannot affect the decreasing trend of the DTR in the second half of the twentieth century. This also suggests that the DTR over East Asia decreased faster than that of the globe and the Northern Hemisphere in the second half of the twentieth century (Table 2), which is mainly due to the larger scale and more rapid urbanization in East Asia than in other regions of the global.
We selected the observation data of 1968 stations over East Asia during 1901-2018 from the CMA-LAST v1.1 dataset to analyze the climatological and long-term change characteristics of the DTR over East Asia. Furthermore, we also analyzed the association between the DTR change and precipitation. The main conclusions are as follows: (1) The observed annual mean DTR over East Asia is approximately 10.0°C, which increases gradually from coastal to continental Asia. The DTR of the marine continent (0-20°N) and coastal grid boxes are significantly lower than those of inland grid boxes. (2) The DTR changes over East Asia during 1901-2018 show two dominant characteristics.
The first is a significant long-term decrease in the DTR, with a 0.60°C decrease from 1901 to 2018. The second is noticeable interdecadal fluctuations in the DTR before the 1940s over East Asia. Before 1950, the high latitudes and mid-latitudes, where the grid boxes coverage rate is relatively high, show nonuniform decreased and increased trend. These two dominant characteristics of the East Asian DTR are consistent with the global changes since 1901 and accompanied by prominent regional characteristics. However, compared to the globe, Northern Hemisphere, and East Asia, the decline of the DTR in East Asia (-0.6°C) is significantly lower than that in the globe (-0.41°C) and the Northern Hemisphere (-0.42°C).
(3) This study found that the long-term spatial patterns of DTR change show a significant negative correlation with mean precipitation patterns except in arid and semi-arid areas during 1901-2018. In addition, the decreasing DTR trend is gradually smaller from arid regions to humid regions during 1901-2018, mainly because the difference between T max and T min warming rate is gradually smaller. (4) Additionally, it should be noted that the DTR is highly sensitive to urbanization, and the impacts of urbanization contribution on the current trend estimation of T max , T min , and DTR in East Asia reached 25.9%, 28.2%, 43.6% from 1951 to 2018, respectively. This indicates that the DTR of East Asia decreased faster than that of the Northern Hemisphere and globe in the second half of the twentieth century, mainly due to the larger scale and more rapid urbanization in East Asia than in other regions of the globe.