Recent changes in temperature extremes in subtropical climate region and the role of large-scale atmospheric oscillation patterns

Understanding the recent variations in temperature extremes is crucial to anticipate the forthcoming incidences of extreme phenomena. However, knowledge of temperature extremes’ spatial and temporal patterns, as well as their links to atmospheric oscillation and topography, is scarce in Bangladesh. To this end, this research intends to analyze the spatial and temporal trends in recent extreme temperatures and their relationships with oscillation indices and the topography of Bangladesh. Daily temperature data obtained from 20 meteorological stations for 1980–2017 were employed for this purpose. Results revealed increasing trends in summer days (SU25), tropical nights (TR20), warm days (TX90p), warmest days (TXx), and warm nights (TN90p), while decline in the coldest days (TNn), cold days (TX10p), and cold nights (TN10p) was observed in Bangladesh. Spatial distribution of trends revealed an increase in SU25 and TN90p by 1.9–2.38, 2.33–2.90 days/decade, and a decrease in TX10p and TN10p by 1.7–3.3 days/decade in most regions. Besides, TR20 showed an increase of 3.22–4.17 days/decade in all sub-regions. The temperature extremes of Bangladesh showed a significant connection with multivariate ENSO index (MEI) and Sea Surface Temperature (SST). Besides, the extremes in most regions of the country showed a significant connection with Southern Oscillation Index (SOI) and Indian Ocean Dipole (IOD). The influence of atmospheric oscillation indices was more evident on cold days/nights than on warm days/nights. TN10p and SU25 also showed a significant correlation with elevation, which suggests an increase in cold night and summer day temperature with the increase in elevation in Bangladesh. Large-scale climate mode reanalysis revealed that a strong (weak) wind speed, enhancing (decreasing) geopotential height, and fast warming (cooling) over the northwestern (southeast) region have attributed to the variations in extreme temperature in Bangladesh to several extents. These findings will assist the policymakers in disaster mitigation and climate change adaptation in Bangladesh.


Introduction
The global mean surface temperature increased by 0.74 °C during the previous century and is predicted to climb by 1.84 °C by the end of this one (IPCC 2014). Only a slight shift in the mean temperature results from significant variations in the frequency of temperature extremes (Hansen et al. 1988;Khan et al. 2020;Mallick et al. 2021). As a result of global warming, temperature extremes are already increasing over the world. Therefore, the scientific community is more concerned about increasing climate extremes because of higher societal vulnerability to climatic extremes than the accumulated mean climate (Katz and Brown 1992). Numerous studies reported variations in intensity, frequency, regional extent, length, and timing of temperature extremes and their implications for agricultural practices and human mortality (Piao et al. 2010;Dey et al. 2021). Widespread increases in temperature extremes are anticipated to remain owing to global warming, making them a key policy concern for governments, the general people, and the climate research scientists (Almazroui et al. 2014;Durre et al. 2000;Sun et al. 2014;Viola et al. 2014;Khan et al. 2019;Das 2021).
Bangladesh is one of the top ten nations most susceptible to climatic change in the world Eckstein et al. 2017;Das and Islam 2021). Increased daily temperatures and temperature-related extreme phenomena are the most substantial effect of climate change in the country, as they are elsewhere in the world (Shahid et al. 2016). Temperature extremes have severe consequences for agriculture (Sikder and Xiaoying 2014), ecosystem (Islam et al. 2021a), and other sectors in the country (Shahid et al. 2016). Temperature variability is most likely to reduce agricultural yields in Bangladesh . Plant development, pollination, and reproductive processes are all affected by higher temperatures (Sacks and Kucharik 2011;Klein Tank et al. 2006). Even short-term extreme high and low temperatures can decline the growth and productivity of crops (Mearns et al. 1984). Due to extreme temperature variability, total rice production in Bangladesh may decrease by 7.4% per year in the future (Sarker et al. 2012). Therefore, it is vital to estimate changes in the mean temperature sequence and the frequency, magnitude, and extent of extreme temperature occurrences (Alexander et al. 2006;Easterling et al. 2000;Moberg and Jones 2005). However, the features of the extreme climate-related phenomenon at the regional level in Bangladesh are little understood. So, it is essential to monitor changes in temperature extremes in Bangladesh frequently.
In recent times, Bangladesh has experienced a colder winter with a significant increase in the low temperature and a hotter summer (Nishat and Mukherjee 2013). The monthly mean temperature increased at more than a few agro-ecological zones both on a seasonal and annual scales (Mia 2003). The monthly maximum temperature showed an increasing trend, while the monthly minimum temperature fluctuated between rising and falling (Islam et al. 2020a). Earlier research solely looked at monthly and annual minimum and maximum trends to assess extreme changes. The Expert Team on Climate Change Detection and Indices (ETCCDI) has been extensively employed in climate change studies all around the globe to identify extreme temperature indices that better represent temperature extremes (Zhang et al. 2011). However, these indices are used by very few scholars for studying on Bangladesh context to date. Previous studies also mainly focused on either the shift in mean value or the changes in pattern (Kamruzzaman et al. 2019a;Nowreen et al. 2012;SMRC 2009;Shahid 2009;Abdullah et al 2020;Islam et al. 2021b). The temporal distribution of six threshold temperature indices of Bangladesh and its associated effects on several fields have been assessed by Shahid et al. (2016). However, the results presented in their study were not sufficient to understand the variations in regional temperature extremes. It is possible to describe temperature-related climate extremes in Bangladesh better using the broad range of extreme temperature indices defined in ETCCDI.
Large-scale atmospheric circulations directly impact local climate fluctuations at both short-term and long-term climate time scales. An insight into their connections with temperature extremes can better inform the drivers of temperature extremes (Ghose et al. 2021a, b). A few studies investigated the relationship of large-scale atmospheric circulations to temperature and precipitation over the country. According to Wahiduzzaman and Luo (2020), the El Nino-Southern Oscillation (ENSO) affects the temperature and precipitation in Bangladesh. The relationship varies depending on the location and season. However, so far, no research has been conducted on the temporal and spatial trends of the connection of extreme temperature phenomena to the large-scale climatic indices in Bangladesh. To fill the voids in the existing studies, the current study aims to analyze the changes in Bangladesh's ETCCDI extreme temperature indices between 1980 and 2017 and the links between the extremes and atmospheric oscillation indices and the country's geography. The novelty of this work is that the trends of extreme temperature indices and their connections to largescale climate indices were analyzed in Bangladesh for the first time on a regional scale to understand their spatial and temporal variability. The key contribution is that this is the primary comprehensive study appraising recent changes in temperature extremes in the country and linkage with largescale atmospheric oscillation patterns, which have considerable implications on agro-economical sector, water resource, and ecology. This research may be used for climate change mitigation and adaptation planning, as well as for establishing an early warning system.

Study area description
Bangladesh is a low-lying subtropical nation in South Asia, with latitudes ranging from 20° 34 N to 26° 38 N and longitudes varied from 88° 01 E to 92° 41 E. Except for few hilly regions in the southeastern and eastern portions, the country's entire area is 147,570 km 2 , with flood plains covering its majority. Seasonal variations in rainfall and temperature are indistinguishable characteristics of its climate. The country's climate is characterized by a hot and humid summer with heavy rain and a dry and mildly cold winter. The four major seasons in the nation are winter (December to February), pre-monsoon (March to May), monsoon (June to September), and postmonsoon (October to November) (Kamruzzaman et al. 2019b). Bangladesh has an average daily mean relative humidity of 80% and evapotranspiration of 3.72 mm per day, respectively (Salam and Islam 2020). The coldest January and hottest months in Bangladesh between April and October. The maximum rainfall occurs in the northeastern Sylhet district. The present study divided the whole of Bangladesh into three regions as northern (region 1), southern (region 2), and central (region 3) according to its geographical location, hydrological settings, climatic variation, and soil type. The northern region consists of the districts of Bogura, Rajshahi, Rangpur, Mymensingh, and Sylhet. The southern part consists of Cox's Bazar, Rangamati, Sandwip, Barishal, Bhola, Feni, Khulna, Patuakhali, Sitakundu, and Teknaf. The central region covers Jashore, Cumilla, Faridpur, Madaripur, and Dhaka ( Fig. 1).

Data sources and quality control check
There are 43 meteorological sites in the Bangladesh Meteorological Department (BMD) distributed homogeneously across the country. However, most of them start operating after 1990 (Jerin et al. 2021). The availability of a longer period dataset, the number of missing documents, and the spatial distribution of the stations were used to secure the records from these 43 stations. Because several stations were built after 1990, long-term data was not accessible. Furthermore, at certain sites, a missing climate record was discovered at a very high level. As a result, only 20 locations with data dating back to 1980 and covering the whole country were used in this study. The daily maximum and minimum temperature datasets obtained at those 20 locations from Fig. 1 The location of Bangladesh, as well as its elevation spatial distributions and meteorological station locations 1980 to 2017 were used to reflect the country's climate uniformly, with missing values less than 4% and geographical distribution.
Missing data at each site was supplemented by data from neighboring sites. In addition, the stations that were eliminated owing to a lack of data for a longer period were used to fill in the gaps. Supplementary Tables S1 and S2 provide further details on the filling-in of missing datasets. The BMD recorded and collected meteorological data following the World Meteorological Organization's (WMO) criteria. However, before examining climate extremes, quality control of the dataset is still necessary since erroneous outliers have a substantial impact on the extremes (Gao et al. 2015).
First, station observations were checked for quality by systematic means, such as looking for positive records of climatic variables like Tmin being lower than Tmax and temperatures being below 45 °C. In addition, the SNHT (Standard Normal Homogeneity Test) at a significance threshold of p < 0.05 was carried out at all sites to show any irregularities in the datasets (Islam et al. 2020b) by R package "snht." The time series data were found to be homogeneous and consistent at all locations (Hans 1986). All data records were authorized by the BMD staff after a quality check was performed on them. One of the most difficult issues in trend analysis is serial autocorrelation (Praveen et al. 2020). Only a few instances of serial autocorrelation in time series were found to have a p value of 0.05 or higher.
Extreme climate indices were calculated from daily data using RClimDex (V1.0) software package, a specially built and publicly available software. ETCCDMI (Expert Team on Climate Change Detection, Monitoring, and Indices) produced this software, which can be obtained from the ETC-CDI website (http:// etccdi. pacifi ccli mate. org/ softw are. shtml). The ETCCDI has developed a set of 27 core climate indices, including 16 temperature indices that are mainly related with extreme events. Eight core temperature indices were examined in this study: warm days (TX90p), warm night (TN90p), cold days (TX10p), cold nights (TN10p), hottest days (TXx), coldest days (TNn), summer days (SU25), and tropical night (TR20). These chosen extreme climate indices include both percentile and absolute value-based. They were picked through a comprehensive literature analysis and found significant for Bangladesh's climatology (Khan et al. 2019;Mallick et al. 2021;Abdullah et al 2020;Islam et al. 2020a;Islam et al. 2020b). Table 1 presents a detailed description of these eight indices.
Five climatic indices of the atmospheric and oceanic circulation patterns and the elevation of the respective stations were used to analyze their possible connections with temperature extremes. These indices include North Atlantic Oscillation (NAO), multivariate ENSO index (MEI), Indian Ocean Dipole (IOD), Southern Oscillation Index (SOI), and Sea Surface Temperature (SST). Monthly values of SST, IOD, NAO, MEI, and SOI were derived from the National Oceanic and Atmospheric Association (NOAA) Climate Prediction Center (CPC) (www. cpc. ncep. noaa. gov) during the period 1980-2017.

Mann-Kendall test
Temperature extremes follow a nonlinear behavior (Shahid et al. 2016;Praveen et al. 2020;Islam et al. 2021a, b;Kamruzzaman et al. 2019b). Thus, a commonly used nonparametric Mann-Kendall test has adopted to explore the trend of the temperature extremes (Kendall 1975;Mann 1945;Ullah et al. 2018;Islam et al. 2020a;Kamruzzaman et al. 2019b). It is widely utilized because of its robust characteristics and minimal sensitivity to rapid changes (Li et al. 2018a, b). The Mann-Kendall statistic (S) is calculated as follows: where The probability related to S for sample size, n, is calculated to measure the significance of the trend statistically. Normalized test statistic Z is computed as follows: If |Z| is higher than or equal to 2.575, 1.96, or 1.645, the null hypothesis of no trend is rejected at the 99%, 95%, and 90% significance levels, respectively. To get the sequential form of the MK test, the number n i of lower elements x j x j< x i preceding it (j < i) for each element The test statistics t is given by In the lack of any trend (null hypothesis), t is asymptotically normal data distribution It has a normally distributed, with an expected value E(t) and variance var 2 (t) stated by, and The null hypothesis can, thus, be rejected for elevated values of |u (t)|.
There is a possibility that serial autocorrelation influences trend patterns, with positive autocorrelation pointing to an increased number of false-positive MMK test results (Ullah et al. 2018). This is why the serial autocorrelation was eliminated prior to implementing the MMK test (Islam et al. 2020b). To reduce autocorrelation, the TFPW (trend free pre-whitening) technique was used in this study (Li et al. 2018a, b). Besides, the sequential form of the MK test comprises of the test, commencing from the first term and finishing in the i-th term, then in those beginning in the i-th term and terminating in the last term (regressive analysis). The graphic depiction of the direct (ui) and the backward (u′i) series derived from this method yields curves that can overlap one another in the absence of any trend. In the event of a significant trend (the 95% confidence level), |ui |≥ 1.96, the intersection of the curves makes the time of change roughly possible to discern (Shahid 2011).

Sen's slope estimators
To evaluate the changes in temperature extreme indices, a common nonparametric test called Sen slope (Sen 1968) was used. Sen's slope approach estimated the magnitude of the trend (Sen 1968). A time series dataset is necessary for this procedure. The technique is calculated by computing the path as a unit time change (Shahid 2011), where, Q΄ = slope between data points x t΄ and x t x t΄ = data measurement at time t΄ x t = data measurement at time t The median slope simply gives Sen's estimator of the slope, N is the number of computed slopes.

Correlation matrix
The Pearson correlation was used to discover the link between temperature and atmospheric oscillation indices. Pearson's correlation measures the direct relationship between two variables, where positive correlation denotes by the value 1; if the value of r is 0, it means no correlation; and negative correlation indicates by the value − 1. X and Y are considering two variables, each having n values X 1 , X 2 ,. …, Xn and Y 1 , Y 2 ,..., Yn, respectively. Let the mean of X be X and the mean of Y beY . Pearson's r is calculated following Tomar et al. (2016): The Pearson correlation coefficients have been utilized in the present study to explore the spatial (northern, southern, and central) and temporal (1980-1998, 1999-2017, and 1980-2017) correlation of eight extreme temperature indices with five large-scale oceanic indices. The relationship of the temperature indices with the elevation was explored by performing the least-squares method (Casella and Berger 2002;Moore and McCabe 2003). Table 2 presents the results of the MK trend test of 8 temperature indices in each region and the whole country for 1980-2017. The results showed an increasing trend in TN10p and TX10p at (p < 0.1) in all three areas and the entire country (except TN10p). The TNn exhibited an insignificant rising trend in the central region and whole

Trends of the extreme temperature indices
Bangladesh. The results also showed an increase in TN90p (p < 0.01 and whole Bangladesh p < 0.05), SU25 (p < 0.05 and the Northern region p < 0.1), and TR20 (p < 0.01) during 1980-2017. The TX90p (p < 0.05) and TXx (p < 0.1) were increasing only in the northern region. The rest of the indices showed an insignificant increasing trend in all regions and the whole country. Figures 2 and 3 demonstrate the rate of the change of extreme temperature indices found by Sen's slope of the estimator. The blue circles indicate increasing trends, and the red circle represents decreasing trends. TN10p showed an increase in Dhaka at the rate of 2.19 days/decade. The changes in TN10p in Rangpur, Rajshahi, Bogura, Mymensingh, Madaripur, and Cumilla stations were in the range of 0-1 day/decade. The rest of the stations showed a decreasing trend in TN10p at the rate of − 1 to − 2 days/ decade (Fig. 2a). TN90p showed an increase at all stations from 0.62 to 9.91 days/decade, except at Patuakhali (− 1.15 days/decade) (Fig. 2b). TX10p showed a decreasing trend in Bangladesh from − 0.04 to − 2.31 days/decade (Fig. 2c), except at Rangpur (3.34 days/decade), Patuakhali (0.56 day/decade), Sandwip (1.57 days/decade), and Rangamati (1.54 days/decade). The TX90p (Fig. 2d) and TR20 (Fig. 3d) revealed a similar trend pattern. The TX90p and TR20 were decreasing at Rangamati and Sandwip from − 0.13 to − 1.25 days/decade. The other stations exhibited an increasing trend in TX90p (0.21 to 4.83 days/ decade) and TR20 (1.25 to 10 days/decade). On average, the decreasing trend in TNn was dominant in Bangladesh at a rate of − 0.09 to − 0.26 °C/decade (Fig. 3a). All stations in the southern region showed an increasing trend in TXx from 0.11 to 0.76 °C/decade (Fig. 3b). The SU25 presented a variation from − 1.5 to 6.67 days/decade at different locations. Overall, the results showed an increasing trend in warm temperature and a decreasing trend in cold temperature indices, indicating hot days and nights are increasing and cold days and nights are decreasing in Bangladesh.

Correlation between extreme temperature indices and ENSO indices
The associations of the extreme temperature indices with the ENSO circulation indices in two different periods, 1980-1998 and 1999-2017, are shown in Table 3. The TR20 showed a negative correlation with SOI (p < 0.05) and NAO (p < 0.05) during 1980-1998. The TXx, TN90p, and TX90p were negatively associated (insignificant) with SOI, IOD, MEI, and NAO as well as TN10 and TX10p displayed significant positive correlation with all the ENSO indices (Table 3). Among the five atmospheric and oceanic circulation patterns, MEI, SST, and IOD showed a comparatively higher positive correlation with temperature indices than SOI and NAO.
The correlations of temperature extremes with circulation indices were inconsistent for the two periods (Table 3). None of the extreme temperature indices, except TX10p and TR20, was significantly correlated with circulation indices during 1980-1998. TR20 was negatively correlated with SOI and NAO (p < 0.01). On the other hand, TX10p displayed significant positive association with MEI and SST (p < 0.01), as well as the SU25 also significant negatively correlated with MEI (p < 0.01) during 1980-1998.  The warm temperature extremes, TR20, SU25 (except SOI and NAO), TN90p, and TXx, were negatively correlated with SOI, IOD, and NAO during 1999-2017 (Table 3). In contrast, the cold temperature extremes (TN10p and TX10p) were positively correlated with SOI, IOD, and NAO. Furthermore, TX90p was positively correlated with all circulation indices except IOD, while TNn was negatively correlated with all the ENSO indices except the MEI and SST. Additionally, TN10p and TX10p were positively associated with IOD (p < 0.01), whereas TX90p was negatively associated with IOD (p < 0.05) during 1999 − 2017. Figure 5 represents the association of temperature indices with IOD circulation indices during 1980-2017. Similarly, the association of temperature indices with IOD circulation indices is also displayed in Supplementary Figs. S1-4. A significant positive relationship of cold extremes with circulation patterns was observed in this study. In contrast, warm extremes and circulation patterns showed a statistically significant negative association in the study area during the study period. The stations showing negative correlation were more than the stations showing positive correlation. TXx, TNn, SU, TR, TN90p, and TX90p were negatively correlated with SOI, MEI, and NAO in the north, south, central, and whole Bangladesh, and a negative correlation in the northern region. Among them, TN10p, TN90p, and TR exhibited a significant negative correlation. In the northern area, all of the warm extremes except TX90p exhibited a negative correlation with SST. However, the correlation was significant only for TR. All the warm and cold extremes except SU showed an insignificant positive association with SST in the southern region. All cold extremes and two warm extremes (TXx and TX90p) positively correlated with SST in the central area. However, the positive correlation was significant only for TN10p. Two temperature extremes, TR and SU, showed a negative association with SST for the whole country.
In the south, central, and whole Bangladesh, all warm extremes exhibited a negative association with IOD, whereas all cold extremes showed a positive correlation with IOD. In the northern region, all cold and warm extremes except TX10p showed a negative correlation. TN90p exhibited a significant correlation with IOD in all regions. Only TR exhibited a significant negative connection with MEI in the northern area, and all warm extremes except TXx showed a negative link with MEI. TX90p, TXx, and SU showed a negative association with MEI. However, the correlation was significant only for SU in the southern region. Four indices, TN10p, TX10p, TX90p, and TXx, showed a positive correlation, and the others showed a negative correlation with MEI in the central region. All of the warm extremes except for TN90p demonostrated a negative association, TX10p exhibited a significant positive, and SU showed a significant negative association with MEI for the whole country. Table 4 shows that negative correlation was observed more in the different sub-regions with ENSO indices than the positive correlation. In this table, extreme temperature indices exhibited more negative association than positive association with the ENSO indices. The significant positive or negative association was mainly limited to the cold (TN10p and TX10p) and hot (TN90p and TR20) extreme temperature indices with the MEI, NAO, and SOI (Table 4). In the northern region, TR20 indices displayed significant  . 5 Mapping of the association of extreme temperature indices with IOD negative correlation with the MEI (r = − 0.411, p < 0.05) and NAO (r = − 0.454, p < 0.01) indices and significant positive for SOI (r = − 0.411, p < 0.05). The cold extreme, TN10p, showed also significant positive association with NAO (r = 0.447, p < 0.01). In the southern region, NAO exhibited significant negative correlation with TR20 (r = − 0.363, p < 0.05) and significant positive with cold extreme like TX10p (r = 0.376, p < 0.05). However, in the central region, cold extreme (TN10p) indices displayed a significant negative and positive correlation with SOI (r = − 0.465, p < 0.01) and MEI (r = 0.466, p < 0.01), respectively. Moreover, warm extreme indices like TN90p showed a significant negative correlation with NAO (r = − 0.399, p < 0.05). At last, over the whole, TR20 exhibited a significant negative correlation with NAO (r = − 0.450, p < 0.01). The results exposed that almost all of the positive correlation with ENSO indices was displayed by cold extreme indices (TX10p and TN10p) and most negative correlation was displayed by hot extreme indices (TN90p and TR20) in almost all over the different sub-regions and the whole country. Figure 6 shows the relationship between the trend magnitude in temperature extremes and elevation. The results indicated an inconsistent association between the trends in temperature extremes and altitude. Except for tropical nights (TR20) and warm nights (TN90p), a significant positive relationship of elevation with warm temperature indices was observed in this study. On the other hand, the cold temperature indices exhibited a significant negative association with elevation except for cold nights (TN10p). The warm days (TX90p), warmest days (TXx), summer days (SU25), and cold nights (TN10p) exhibited a significant positive relationship (Fig. 6). The trends of warm nights (TN90p), tropical nights (TR20), and coldest days (TNn) showed a substantial decrease with the decrease of elevation (p < 0.05).

Large-scale atmospheric circulation dependent temperature extreme
The temperature indices had been significantly and abruptly changed found by the sequential MK test. However, the abrupt change point was varied for various temperature indices. After the abrupt change point, the temperature indices trends were either noteworthy upward or downward, and the climatic indices considerably affected the temperature extremes. In general, an abrupt change point of annual average temperatures in Bangladesh for 1980-2017 was in 1995 (Fig. 7). Therefore, the change in atmospheric circulation was estimated by subtracting the ERA-interim and ECMWF reanalysis dataset before and after the breaking point, 1980 to 1995, from 1996 to 2017. The differences in air temperature, wind speed (vectors), and geopotential height (shaded) at 850 hPa between 1980-1995 and 1996-2017 are illustrated in Fig. 7. Figure 7a revealed an evident rising in air temperature in the northwestern region, whereas the southern and southeastern areas were subjected to a decline in air temperature. Similarly, Fig. 7b represents a decrease in wind speed in the northwest of Bangladesh and an increase in the southeast. Again, the highest geopotential height differences (around 20 gpm) were observed in the northern and northwestern regions. In contrast, the least geopotential height dissimilarities (about 0 gpm) were noticed in the southeastern and southern regions.

Discussion
The current study has evaluated the spatial and temporal trends of the eight temperature indices in Bangladesh from 1980 to 2017. The results revealed that the warm temperature indices (TXx, TX90p, TN90p, TR20, and SU25) were increasing and a decreasing trend (meaning lower temperature was gradually decreasing more than past) in cold temperature indices (TNn, TX10p, and TN10p). The rate of change in different temperature indices was different from each other. Khan et al. (2019) have also shown that the lowtemperature trend has been dropping and that maximum temperature trends have been rising in Bangladesh. Abdullah et al. (2020) observed significant warming in Bangladesh's coastal and inland locations, which indicated the increasing trends in high-temperature indices. Khan et al. (2020) found similar results to those found in this study. Day-warming was shown to be larger than night-warming, with high-temperature extremes rising faster than low-temperature extremes, contradicting the findings of the result of Khan et al. (2019). The datasets and techniques utilized in the current analysis differ significantly from those used by Khan et al. (2019). The major explanation for the larger rise in hot extremes is owing to a smaller magnitude of warming in summer compared to winter (Abdullah et al. 2020), as well as substantial warming due to greenhouse gas (GHG) radiative effects in summer (Aguilar et al. 2009). The observed warming trend is consistent with findings from studies conducted at both global (Easterling et al. 2000;Alexander et al. 2006) and regional (Sheikh et al. 2015;Klein Tank et al. 2006;Shrestha et al. 2017;Shi et al. 2018;Tong et al. 2019) scales. Temperature indices in the northwest and western areas have been significantly higher recently, but changes in the northeast and southwest have been small. This might be linked to the Sundarban forest's (the world's biggest mangrove forest) ability to store carbon and mitigate the effects of climate change in the southern region of the country . Some cited works performed in northeast and North Bengal of India are in line with our findings. A study performed in the northeast part of India which is adjecent to our research area implying that extreme temperature is rising in the northeast area of India as well (Dash et al. 2012). The findings of Jhajharia andSingh (2011), Jhajharia et al. (2014), and Sharma et al. (2018) in extreme temperature studies under northeastern and northern regions of India are in accord with our research. The climate in the northwest and western areas has been heavily impacted by unfavorable land use/land cover changes, urbanization, groundwater depletion, and irrigation modifications (Shahid 2011).
Atmospheric and ocean oscillations are critical drivers of extreme climate and climate change. Several studies have revealed that the ENSO phenomenon is intimately linked to the interannual monsoon shift in East Asia (Wang and He 2012;Wahiduzaman et al. 2020;Ghose et al. 2021a). The large-scale atmospheric oscillation has been a probable reason for the interdecadal fluctuation of cold winter temperatures in China (Han et al. 2014). The variation of the climate mode over the North Atlantic region, regional temperature variations over Northern Eurasia is the results of ENSO, which disturbs the air temperature in winter in Northern China . The meridional Hadley circulation (MHC) in the western Pacific deteriorated dramatically from the late 1970s. Thus, the response of Asia's atmospheric oscillation to ENSO declined, resulting in a decrease in the responsiveness of China's winter cold temperature to ENSO (Han et al. 2014;Shi et al. 2018).
This study showed a positive correlation of SST with all eight extreme temperature indices in Bangladesh, followed by MEI. In contrast, the extreme temperature indices showed an overall negative correlation with SOI, IOD, and NAO. There was a strong correlation between two temperature indices (TX10p and SU25) and MEI during 1980-1998. The correlation between three temperature indices (TN10p, TN90p, and TX10p) and IOD was significant during 1999-2017. This indicates a greater impact of ENSO on the faster decrease in temperature during cool days and cool nights than the increasing temperature of warm days and nights. Han et al. (2018) reported that the association between the NAO and cool days (nights) in Northeast China varies over time. The variance could be ascribed to NAO-related climate anomalies in different interdecadal backdrops, according to the researchers.
Elevation-dependent temperature extreme (EDTE) has become a hot topic of research for insights into the climate change evidence. A significant number of studies reported that temperature extreme at high altitude is more evident than at low elevation Guo et al. 2016;Li et al. 2017;You et al. 2017). By contrast, several studies also revealed no evidence of EDTE (You et al. 2008). The uncertainty in the EDTE may be attributed to diverse meteorological station density, remote areas, and inadequate datasets Pellicciotti et al. 2012). Selected temperature indices revealed significant geographic heterogeneity between the higher southeastern and low-lying regions in the current study. Cold night and summer days were the solely minimum and maximum temperature-based indices that showed a significant positive connection with elevation (p < 0.01). The related studies also showed similar findings (Adams et al. 2020;Olmo et al. 2020;Jhajharia et al. 2014;Jhajharia et al. 2021). However, the findings contradict those of Sun et al. (2017) and You et al. (2017), who discovered a definite negative trend in TN10p and FD. The elevation of the eastern region is comparatively higher than other regions of the country. Although, the effects of extreme phenomena on water balance and river flow at the higher elevation remain still uncertain due to high climatic variability, remoteness, and the composite action between lithology, climatic, and hydrologic processes (Zhou and Jumei Huang 2020). The present study indicates EDTE in the study region and Bangladesh; however, this evidence is still unclear, and the drivers of the EDTE need to be systematically examined at regional and national levels.
The large-scale atmospheric and oceanic oscillation variables, such as geopotential height, air temperature, and wind fields, influence the local climate. Both spatially and temporally, both the negative and positive effects of large-scale atmospheric oscillation patterns on surface variables and viceversa. We examined the connection of surface temperature with the changes mentioned above; geopotential height, air temperature, and wind speed are frequently utilized (Tong et al. 2019;Ullah et al. 2018;Jhajharia and Sing, 2011). This study implies that increases in air temperature might also impact the regional as well as local temperature and temperature extremes (Khan et al. 2008;Grotjahn et al. 2016;Ren et al. 2017). Besides, our study reports that the rising geopotential height with the reduction in wind speed may be linked to a rising trend in annual mean temperature and their influence on the incidence of extreme temperature phenomena in the northwestern region to some extents (Shahid and Khairulmaini 2009;Xu et al. 2018;Ullah et al. 2018;Shi et al. 2018). The strong wind speeds move towards the southern coastal region from the Bay of Bengal (BoB), which might influence the long-range temperature at a regional scale (Pathak et al. 2017;Sinha et al. 2015). The strong southeast winds transport a considerable amount of water vapors from the BoB to the southern and southwestern coastal regions, which may ultimately influence the regional temperature and result in a cold trend and vice versa (Pathak et al. 2017). Long-term forecasting based on ENSO conditions is critical for planning and management in climate-sensitive industries like the environment and hence deserves more investigation. This study also demonstrated an increasing trend of warm extremes in the northwestern regions of Bangladesh, which are in line with the air temperature patterns and atmospheric circulation fields. Interestingly, the large-scale atmospheric circulations exhibited a dipole tendency across the northwestern and southeastern regions of the country. Although, exploring the causes and mechanism of this dipole tendency is out of scope in the present study. It deserves further examinations.
The findings of this study might have potentials for Bangladesh. First, this research showed that Bangladesh may become warmer and drier sooner than previously predicted. The salinity problem in the southwestern area may worsen in the coming days as a result of a considerable increase in severe heat indices. Warming and drying conditions may cause drought in northern and western locations at the same time. To adapt to the drying circumstances, drought-tolerant crop varieties and efficient use of surface water resources may be needed. Second, increased severe temperature events may have an impact on vegetation dynamics, leading to an increase in pests and disease. As a result, this area merits further attention and in-depth research.
There are some shortcomings in this work that must be addressed. First, the transition sites between climatic extremes were not extensively examined, which might be due to a lack of metadata support. More information about the homogenization process may result in improved results. Second, GHG emissions and anthropogenic aerosols were not taken into account, potentially increasing the probability of climate extremes. Despite these constraints, our research examines the spatiotemporal variations in temperature extremes and their links to climate mode. The findings of this research can aid policymakers in formulating strategies for developing an early warning system and reducing the affects of growing temperature extremes in Bangladesh.

Conclusion
The current study looked at the spatial and temporal changes in recent temperature extremes in Bangladesh and their teleconnection to atmospheric oscillation indices during the study period . All temperature indices exhibited a difference in the line of the global warming trend. The study revealed decreases in cool nights, coldest days, cold days, and the increases of tropical nights, warmest days, warm nights, warm days, and summer days in nearly all sub-regions of Bangladesh. The magnitudes of the change in warm days/nights were higher than those on cold nights/days. Cold temperature indices were highly associated with large-scale climate indices, while warm temperature indices were weakly associated with atmospheric oscillation indices during the observation period. The areal mean warm temperature extremes of Bangladesh were linked to SST and MEI, and the cold extremes were related to SOI, IOD, and NAO. Atmospheric oscillation indices influenced the decreasing temperature of cold days/nights more than the warm days/nights. The variations of the cold nights and tropical nights were related to NAO and SOI in the north, south, and central regions of Bangladesh, while the cold days were associated with IOD for the whole country.
The findings may contribute to assessing climate changerelated disaster management and infrastructural development planning of the country both for policy and implementation levels. The findings would contribute to enriching knowledge on temperature fluctuations with large-scale atmospheric oscillation indices and elevations over Bangladesh. This study suggests that the natural and human-induced factors of temperature extremes and their probable connections with climatic mode should deserve further study.