Unprecedented recent late-summer warm extremes recorded in tree-ring density on the Tibetan Plateau

Two up-to-date reconstruction series of August–September mean temperature based on tree-ring maximum latewood density (MXD) on the TP (Duan et al 2018, 2019) are used to identify the late summer temperature extremes over the past four centuries. One of the August–September temperature series covering the period of 1572–2014 was reconstructed based on the actual August–September mean temperature anomalies with respect to 1986–2005 (hereafter T_Aug–Sep-ano), and thus it remains the long-term trend (Duan et al 2019). The other reconstruction covers the interval of 1620–2014 (Duan et al 2018), but this reconstruction was performed based on the instrumental August–September temperature deviations in each year relative to the preceding 5 year mean (hereafter T_Aug–Sep-dev). Thus, the T_Aug–Sep-dev does not remain the long-term trend of August–September temperature changes. Both reconstructions show a good spatial representativeness for August–September temperature variability over the TP (i.e. 79-station average) (Duan et al 2018, 2019).

Variation of tree-ring MXD depends on the thickening of latewood cell walls, and thus it is generally very sensitive to light or surface air temperature (Briffa et al 1988, Schweingruber et al 1991), especially in the high-elevation and high-latitude regions (e.g. the TP and the northern high latitudes) (Stine and Huybers 2014, Duan et al 2017). Both reconstructions used in this study were obtained using tree-ring MXD data from high-elevation sampling sites (the average elevation is about 3800 m a.s.l.) on the TP, and the statistics of calibration and verification of regression models from both reconstructions show robust signal of August–September temperature. The correlation coefficients between tree-ring indices and instrumental data is 0.75 over the common period of 1956–2014 for T_Agu-Sep_dev and 0.80 in the common period of 1951–2014 for T_Agu-Sep_ano, respectively. Both the reduction of error and the coefficient of efficiency are positive in all split-period verifications for the two reconstructions. Moreover, all sign tests reach the significance level of 0.001. All these statistics of calibration and verification of the regression models indicate very robust signals of August–September temperature from both reconstructions. Five years less in the calibration period of T_Agu-Sep_dev reconstruction than in T_Agu-Sep_ano reconstruction is due to T_Agu-Sep_dev reconstructin was calculated as a deviation relative to the preceding 5 year mean. The good response relationship of tree-ring MXD to August–September temperature extremes can also be demonstrated by the good match of the cross-dating results using COFFECHA software (Holmes 1983), i.e. a coherently low MXD among the 17 sites occurs in those extremely cold years and a coherently high MXD among the 17 sites occurs in those extremely warm years.

The gridded land surface air temperature dataset CRU TS4.01 Mean Temperature (hereafter CRUT4.01) with a spatial resolution of 0.5° by 0.5° (Jones and Harris 2008) was used to present the spatial characteristics of late summer temperature extremes occurred in the instrumental period. The instrumental period is defined as the interval of 1951–2014, in which most of the grids and meteorological stations have data on the TP. Gridded geopotential height data obtained from the NOAA/OAR/ESRL PSD (http://esrl.noaa.gov/psd/) with a spatial resolution of 2° by 2° are used to analyze the related atmospheric circulations.

According to the recommendation of World Meteorological Organization (WMO), generalized extreme value (GEV) distributions (Klein Tank et al 2009) was applied to define extreme temperature events during 1620–2014 for the two reconstructions. A 25 year return period was chosen as the threshold of extreme events of the reconstructions, i.e. temperature anomalies respond to a return period more than 25 years were defined as extreme events. To reveal the frequency characteristics of late summer tempeature extremes, the number of both warm and cold extremes were counted in each 50 year interval. Composite analyses of the geopotential height fields at 500 hpa were performed to reveal the large-scale atmospheric circulations resulting in the late summer tempeature extremes. The significance of composite responses was tested using a Monte Carlo resampling procedure based on ensembles of 10 000 random surrogates derived from permutation of the composite matrices.

Based on the GEV distributions, August–September temperature anomalies in a return period of 25 years are 0.29 °C and −1.48 °C for T_Agu-Sep_ano reconstruction and 0.86 °C and −0.79 °C for T_Agu-Sep_dev reconstruction (figure 1). Thus, August–September temperature anomalies greater/smaller than 0.29 °C/−1.48 °C were identified as August–September warm/cold extremes for the T_Agu-Sep_ano reconstruction, while August–September temperature anomalies greater/smaller than 0.86 °C/−0.79 °C were identified as August–September warm/cold extremes for T_Agu-Sep_dev reconstruction. Identified August–September tempeature extremes from both reconstructions show changes in frequency (figure 2). The reconstruction series of T_Agu-Sep_ano shows that most of the late summer warm extremes occurred in the past 40 years (14 out of 16 warm extremes occurred in this period), while only 1 cold extremes occurred in the instrumental period (figure 2(b)). The most frequent warm extremes and the few cold extremes in recent decades respond to the unprecedented recent late summer warming over the past four centuries (figure 2(a)). Compared to the extreme events identified from the T_Agu-Sep_ano reconstruction series, late summer tempeature extremes identified from the T_Agu-Sep_dev reconstruction series show different characteristics of frequency changes (figures 2(c) and (d)). Figure 2(d) shows that it appears to be less T_Agu-Sep-dev extremes occurred in the instrumental period than the preceding period, i.e. no warm extreme and three cold extremes in the instrumental period. Overall, different frequency characteristics of late summer temperature extremes exist between the T_Agu-Sep-ano (figure 2(b)) and T_Agu-Sep-dev (figure 2(d)) reconstruction series over the past four centuries.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To reveal the long-term frequency changes of late summer temperature extremes on the TP, especially for a comparison between the instrumental period and the pre-instrumental period, the number of extremes occurred in the period 1620–2014 were counted in each 50 year interval both for the T_Agu-Sep_ano and T_Agu-Sep_dev reconstruction series (figure 3). A 50 year interval used to count the number of extremes was based on the consideration that this interval is near the length of the observations. The results show that the frequency of late summer warm extremes derived from the T_Agu-Sep_ano resonstruction (red line in figure 3(a)) is unprecedented in the last 50 year interval. However, the frequency of warm extremes identified from the T_Agu-Sep_dev reconstruction appears to decrease since 1865 (magenta line in figure 3(a)). Simutaneously, the difference of the frequency of late summer warm extremes between the two situations (i.e. derived from the T_Agu-Sep_ano reconstruction and the T_Agu-Sep_dev reconstruction) became larger after the 1865, and the largest difference occurred in recent 50 years (figure 3(b)). Both frequencies of late summer cold extremes identified from the T_Agu-Sep_ano and T_Agu-Sep_ano reconstructions show relatively smaller changes (figures 3(c) and (d)) than the warm extremes (figure 3(a) and (b)) over the past four centuries. The frequency of late summer cold extremes identified from the two reconstruction series appears to show a decrease trend in the last century (figure 3(c)).

Zoom In Zoom Out Reset image size

Besides the frequency changes of late summer temperature extremes with time, the spatial scale of the extremes covered is the other important aspect to understand the TP temperature extremes. There is no gridded August–September temperature reconstruction available, we here use the observed grid dataset of CRUT4.01 to discuss the spatial characteristics of late summer temperature extremes on the TP during the instrumental period. In order to highlight the spatial characteristics of the top extremes, we chose the first five of the twelve warm extremes (i.e. 1994, 2005, 2006, 2007, 2009) identified from the T_Aug–Sep-ano reconstruction (remaining the long-term trend) to compared with the one warm extreme (i.e. 1994) identified from T_Aug–Sep-dev reconstruction (not remaining the long-term trend) (figure 4). It can be found that year 1994 is included in both situations, and the spatial characteristics are basically same (figures 4(a) and (b)). The other four extremely warm years identified only from the T_Aug–Sep-ano reconstruction show warming in a larger spatial scale than the year 1994. This indicates that the extremely warm year 1994 was not resulted from the recent warming trend, but the other four years were done. These results imply that those warm extremes induced by the warming trend generally occur in a larger spatial scale than those warm extremes unrelated to the warming trend. The extremely cold years 1962 identified from the T_Aug–Sep-ano reconstruction series was also found in the T_Aug–Sep-dev reconstruction series. Both situations show a similar spatial characteristic (figure 5). All those extremely cold years show a relatively smaller spatial scale (figure 5) than those extremely warm years (figure 4), suggesting that the extremely cold late summers on the TP were unrelated to the large-scale trend.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Compared to other studies in other regions, the unprecedented frequency of the recent warm extremes has also been found in the northern high-latitudes over the past 600 years (Tingley and Huybers 2013). However, extreme temperature events are sometimes regional. For example, Europe experienced the unprecedented heatwave in 2003 during 1500–2003 (Luterbacher et al 2004), but it did not occur in other regions or it was not unprecedented. Similarly, the record-breaking coldwave occurred in Southeastern China in 2008 (Duan et al 2013), but it did not occur in other regions. In this study, our analyses results also indicate the regional characteristic of August–September temperature extremes on the TP (figures 4 and 5).

In view of the observed records of August–September mean temperature near our tree-ring sampling sites (varies from 6.8 °C (minimum) to 16.2 °C (maximum)), August–September cold extremes on the TP can definitely limit the growth of tree rings, but the warm extremes are still favorable to trees growth. For other ecosystems, such as meadows and shrubs, on the TP, some studies reported that warming can induce a degradation of meadows (Wang et al 2011), but may increase the area of shrubs (Gao et al 2016).

The most distinctive feature of the frequency changes of late summer tempature extremes on the TP is the unprecedented recent warm extremes over the past four centuries. Extreme temperature events can be triggered by either an increase/decreased trend (warm/cold extremes) or the increased variability (i.e. the increased variance). Comparisons between the late summer warm extremes identified from the T_Aug–Sep-ano and T_Aug–Sep-dev reconstructions (a large difference) demonstrates that the unprecedented frequency of the recent late summer warm extremes on the TP is a result of the persistent warming in recent decades. Moreover, we can also find that the frequency of late summer warm/cold extremes derived from the T_Aug–Sep-dev reconstructions has decreased since 1865/1900 (figures 2 and 3), which are earlier than the instrumental interval. This indicates a post-industrial decrease of late summer tempature variability on the TP. Therefore, we conclude that the unprecedented frequency of late summer warm extremes on the TP in recent decades is linked to the warming trend rather than an increased variability found in other regional. However, the decreased frequency of late summer cold extremes can be related to both the warming trend and the decreased variability of late summer tempature on the TP since the 19th Century (figures 2(a) and (c)). Based on previous studies (Chen et al 2003, Duan et al 2006), the increased trend of surface temperature on the TP in recent decades can be attributed to increased concentrations of greenhouse gases. However, the occurrence of specific August–September warm extremes on the TP are related to specific atmospheric circulations.

Composite analyses of 500 hpa geopotential filed can provide dynamic mechanisms for the late summer temperature extremes related and unrelated to the warming trend (figure 6). Considering the large difference of the number of extremes in the instrumental period derived from the two reconstruction series (i.e. 14 warm extremes were identified in the T_Agu-Sep_ano reconstructions and no one warm extreme was identified in the T_Agu-Sep_dev reconstructions), we performed the composite analysis of 500 hpa geopotential field using extremes years identified in the instrumental period derived from the two reconstruction series in the instrumental interval (i.e. 1951–2014). Because too few extreme events (sometimes only one) are identified using a 25 year return period based on the GEV distributions for the instrumental August–September data to perform the composite analysis of 500 hpa geopotention height with a significance test, the extreme warm/cold years in the instrumental period are identified as the mean of August–September temperature over the period of 1951–2014 plus/minus the 1.5 SD. As a result, seven extremely warm year (i.e. 1994, 2003, 2005, 2006, 2007, 2009 and 2010) and three extremely cold years (i.e. 1956, 1959, 1965) were identified for the instrumental T_Aug–Sep-ano, while four extremely warm years (i.e. 1981, 1989, 1994, 2005) and seven extremely cold years (i.e. 1956, 1959, 1965, 1976, 1984, 1986 and 2008) were identified for the instrumental T_Aug–Sep-dev. We clarify that although more extreme years obtained using instrumental data than reconstruction series over the instrumental interval (1951–2014), the general spatial characteristics of the occurrence of the late summer temperature extremes do not changes between them (figure not shown). The composite analyses show that the occurrence of late summer warm extremes derived from the T_Aug–Sep-ano links to large-scale positive anomalies of 500 hpa geopotential height in the low latitudes of the North Hemisphere, while there is no a significantly large-scale circulation anomalies was found for those late summer cold extremes (figures 6(a) and (b)). However, the occurrence of late summer warm extremes identified from T_Aug–Sep-dev significantly links to the anomalous high in northern mid-high latitudes, while those late summer cold extremes significantly links to the anomalous low in the northern mid-high latitudes (figures 6(c) and (d)). These results suggest that the late summer warm extremes related and unrelated to the recent warming trend presenting different spatial scales and links to different atmospheric circulations.

Zoom In Zoom Out Reset image size