Evaluation of ERA5-Land reanalysis datasets for extreme temperatures in the Qilian Mountains of China

An increase in extreme temperature events could have a significant impact on terrestrial ecosystems. Reanalysis temperature data are an important data set for extreme temperature estimation in mountainous areas with few meteorological stations. The ability of ERA5-Land reanalysis data to capture the extreme temperature index published by the Expert Team on Climate Change Detection and Indices (ETCCDI) was evaluated by using the observational data from 17 meteorological stations in the Qilian Mountains (QLM) during 1979–2017. The results show that the ERA5-Land reanalysis temperature data can capture well for the daily maximum temperature, two warm extremes (TXx and TX90p) and one cold extreme (FD0) in the QLM. ERA5-Land’s ability to capture temperature extremes is best in summer and worst in spring and winter. In addition, ERA5-Land can capture trends in all extreme temperature indices except the daily temperature range (DTR). The main bias of ERA5-Land is due to the difference in elevation between the ground observation station and the ERA5-Land grid point. The simulation accuracy of ERA5-Land increases with the decrease of elevation difference. The results can provide a reference for the study of local extreme temperature by using reanalysis data.


Introduction
Between 2011 and 2020, the average temperature of the global surface was 1.1°C warmer than it was at the end of the 19th century (before the Industrial Revolution) (IPCC, 2021). Since the 1950s, all inhabited regions have seen more frequent and more intense heat waves, while extreme cold temperatures have become fewer and milder, which affects people's lives and property safety (Shtober-Zisu and Wittenberg, 2021). Global warming has also led to the spread of extreme events to places where they have not been common before. The extent of warming projected by models after the 2050s is associated with taking action against climate change. If carbon dioxide emissions are drastically and rapidly reduced from now on and throughout the 21st century, warming will stop around the middle of the century, and warming will be around 1.5°C or 2°C by the end of the century. With global warming, heat waves, heavy rainfall and droughts will continue to become more severe and frequent (Ruml et al., 2017;Sun et al., 2017;Supari et al., 2017).
Generally, previous research on extreme temperatures utilized observational data (Lin et al., 2017;Ruml et al., 2017;Sun et al., 2017;Supari et al., 2017). However, in mountainous 2. Data and methods 2.1. ERA5-Land temperature data (T e ) ERA5-Land hourly data were downloaded from the website of ECMWF (link: 1 ). We used ERA5-Land hourly 2 m temperature data for years 1979 to 2017 in this study. The location of cell grid points ranged from 35.8 to 40°N and from 93.5 to 104°E, which covered the entire QLM region (Figure 1). According to the latitude and longitude coordinates of the 17 stations, the ERA5-Land cell closest to each station is selected to avoid spatial interpolation (Zhao and He, 2022a). ERA5-Land hourly reanalysis data is given for UTC time, with an 8-h difference from Beijing time in China; we matched the times of observation and ERA5-Land reanalysis with a timedifference conversion method (Zhao P. et al., 2020;Xu et al., 2022). ERA5-Land grid elevation is extracted from the Digital Elevation Model (DEM) of Geospatial data cloud, 2 with the information of ASTER GDEM 30 m. The spatial range of DEM used in this study is 33. 94-45.29°N, 88.81-107.24°E, which can cover the entire the Qilian Mountains (Table 1; Figure 1).

Observations (T o )
In this study, observational daily temperature data including daily mean temperature (T_mean), daily maximum temperature (T_max) and daily minimum temperature (T_min) at 17 meteorological stations in the QLM were obtained from China Meteorological Data Sharing Service of National Meteorological Information Center. 3 These data can be directly used in scientific research after strict quality control by provider from five aspects: climate limit value and allowable value checking, extreme value checking, internal consistency checking between timing value, daily average, and daily extreme value, time consistency checking, and space consistency checking (Zhao C. et al., 2020;Zhao P. et al., 2020;Zhao and He, 2022a). Spatial distribution and specific information on the meteorological stations can be found in Figure 1 and Table 1, respectively.

Evaluation methods
In this study, we used DISO (the distance between indices of simulation and observation) indicator combined with correlation coefficient (r) and mean absolute error (MAE) to assess the reliability of ERA5-Land in capturing daily temperature data and 11 extreme temperature indices in the QLM (Hu et al., 2019;Zhou et al., 2021;Hu et al., 2022;Xu et al., 2022). The equation of DISO is shown in Table 3. DISO is more flexible in expressing the quality of data than other indices (Zhou et al., 2021). The results of DISO value can be divided into two categories in this study: above 0.5 and 0.0-0.5 to analyze the reliability of ERA5-Land in capturing extreme temperature events in the QLM (Hu et al., 2019). It is suggested that ERA5-Land can capture extreme temperature events well when DISO is less than 0.5 (Hu et al., 2019). DISO greater than 0.5 indicates a poor simulation (worst if greater than 1.0), and less than 0.5 indicates a good simulation (best if close to 0.0). In addition, RB was divided into two groups: below 0.0 and above 0.0. The RB below 0.0 or above 0.0 was considered to indicate that ERA5-Land relatively underestimates or overestimates the observational data, respectively. NRMSE shows the mean error of ERA5-Land. NMAE can eliminate the effect of positive and negative offsetting of ERA5-Land errors. The CC is a important indicator of the correlation between ERA5-Land and Observations .

Comparisons of daily temperatures
The correlation coefficient (r) between ERA5-Land and observed T_mean, T_max, and T_min was all larger than 0.9 (Table 4), the determination coefficient (R 2 ) between ERA5-Land and observed T_mean, T_max, and T_min was 0.99, 0.99, and 0.98, respectively ( Figure 2), indicating excellent agreement between ERA5-Land daily temperature and observational daily temperature. The average MAE of T_mean, T_max, and T_min at all stations was 2.24, 3.52 and 2.69, respectively (Table 4). Station No.16 has the smallest r value and the largest MAE, which may be due to the larger altitude difference (1,309 m) between T e and T o at Station No.16 (Table 1). The DISO value for T_mean and T_max was <0.5 for 47 and 53% of the stations, respectively ( Table 5). The DISO value of T_min at all stations is greater than 0.5. The average DISO at all stations of T_mean, T_max, and T_min was 2.44, 0.62, and 2.19, respectively. The spatial distribution of DISO value for T_mean and T_max is shown in Figure 3. In general, the DISO value was higher within the QLM than in the Hexi Corridor, indicating that the ability of ERA5-Land to capture T_mean and T_max in this mountainous area was worse than that in the oasis area. In general, reliability of ERA5-Land to capture T_max is best in the QLM.

Comparisons of extreme temperatures
High correlation coefficient (r) of 11 extreme temperature indices between T e and T o indicates that ERA5-Land can capture the extreme temperature indices well (Table 6). Only the average r at all stations of DTR is less than 0.8, and the average r at all stations of other extreme temperature indexes is above 0.85. After comparing the MAE of 11 extreme temperature indices between T e and T o , we can learn that TX90p have MAE smaller than 1 at all stations ( Table 7). DISO of DTR at individual stations ranged from 0.28 to 0.69 with an average across the stations of 0.46, indicating that ERA5-Land simulates DTR  (Zhao P. et al., 2020;Xu et al., 2022). Figure 3 shows the spatial distribution of DISO values of extreme temperature index. In general, FD0 has the best performance within the Qilian Mountains region.  Distribution of ERA5-Land grid points and 17 meteorological stations in the QLM.

Trends in daily and extreme temperatures
Frontiers in Ecology and Evolution 05 frontiersin.org 1979-2017 in the QLM. Daily temperature data obtained from ERA5-Land reanalysis and observation datasets showed significant positive trends from 1979 to 2017. This indicated that the warming trend in observed data at daily scale was captured accurately by the ERA5-Land reanalysis temperature. TXx (warmest day), TNx (warmest night), TX90p (warm days), and TN90p (warm nights) exhibited a significant positive trend. With global warming, extreme heat events are increasing in intensity and frequency . Also, extreme high-temperature indices were captured by ERA5-Land well. In addition, TX10p (cool days), TN10p (cool nights), ID0 (ice days), and FD0 (frost days) based on observations and ERA5-Land exhibited negative trends. Both TNn (coldest night) and TXn (coldest day) calculated from T e and T o showed a positive trend, showing that the intensity of extreme low temperature events in the QLM increases with global warming. Zhou et al. (2016) analyzed the variation in extreme temperature indices over China using high-resolution gridding data (CN05), and found that the coldest day (TXn) had less variability than the coldest night (TNn), and the warmest day (TXx) had less variability than the warmest night (TNx), similar to the results of this study. Our conclusion is that the trend in extreme    (Table 10). Generally, all tendencies in extreme temperature events except DTR were captured well by ERA5-Land in the QLM.

Possible bias analysis
Observational data in the QLM were underestimated by ERA5-Land for daily temperature, DTR, and absolute indices of extreme temperature (Figures 4, 5). Our results are similar to previous studies. (Zhao P. et al., 2020;Huai et al., 2021;Zou et al., 2022;Zhao and He, 2022a). Many studies indicated that the main reason for bias in reanalysis temperature data is the elevation difference between T e and T o (Gao et al., 2014(Gao et al., , 2018Xu et al., 2022). The linear relationship between T e and T o bias and elevation difference is shown in Figure 8, the R 2 values of T_mean, T_max and T_min were 0.541, 0.454, and 0.469, respectively. The daily biases were related to the elevation difference between T e and T o , especially for T_mean. The simulation reliability of ERA5-Land increased with a decrease in the elevation difference (Zhao and He, 2022a). Therefore, it is possible to correct ERA5-Land reanalysis temperature through a temperature lapse-rate model to further reduce the error and improve the precision of temperature reanalysis (Dee and Uppala, 2009;Gao et al., 2012;Di Giuseppe et al., 2013;Gao et al., 2017). The number of meteorological stations used in this study is small, especially above 3,500 m. However, many studies have shown that the limited number of meteorological stations is sufficient to effectively and accurately evaluate the applicability of reanalysis data (Huai et al., 2021;Jiao et al., 2021). We believe that the bias is mainly caused by the elevation difference between T e and T o . Other factors, such as the interpolation method, terrain complexity, and glaciers, are also potential sources of bias (Huang et al., 2022;Zhao and He, 2022a Scatter plots of ERA5-Land daily temperature and observed values of T_mean (A), T_max (B), and T_min (C). The solid line is 1 :1 line.
Frontiers in Ecology and Evolution 07 frontiersin.org   points, the accuracy of simulation is influenced .
Large DISO values have been found at those stations located within the QLM. Therefore, the bias between the observed values and the reanalysis values within the QLM is higher than that in the surrounding areas such as Hexi Corridor. This may be due to the complicated terrain inside the Qilian Mountains (Zhao and He, 2022a).

Conclusions
In this study, we evaluated the ability of ERA5-Land reanalysis to capture extreme temperatures in the Qilian Mountains. The R 2 between ERA5-Land and observed T_mean, T_max, and T_min was 0.99, 0.99, and 0.98, respectively, which indicates that ERA5-Land can capture observational daily temperature data well. However, an average DISO at all stations of T_mean, T_max, and T_min was 2.44, 0.62, and 2.19, respectively, indicating that the reliability of capturing the T_max of ERA5-Land reanalysis temperature data is better than that of the T_mean and T_min in the QLM. ERA5-Land performed well in capturing extreme temperature indices TXx, TX90p and FD0. DISO for TXx was <0.5 at 12 stations accounting for approximately 71% of all stations. DISO for TX90p was <0.5 at 9 stations accounting for approximately 53% of all stations. ERA5-Land captured FD0 very well, with DISO <0.5 for 15 meteorological stations and an average of 0.27. The mean DISO values of the warmest days (TXx) in spring, summer, autumn and winter were 0.31, 0.16, 0.32 and 0.93, 131 respectively, showing that evaluation results of the extreme temperature index on the seasonal scale show that the capture results of ERA5-Land  Trends in daily temperature and DTR in observation and ERA5-Land datasets from 1979 to 2017 in the QLM.  Trends in threshold indices of extreme temperature in observation and ERA5-Land datasets from 1979 to 2017 in the QLM.  between ERA5-Land reanalysis temperature data and observation data may be the main cause of temperature bias. Generally, ERA5-Land reanalysis temperature data were reliable in simulating temperature data and identifying extreme temperature events; we conclude that ERA5-Land can be applied in the QLM for climate change research.

Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number (s) can be found below: https://cds.Climate.copernicus.eu/ cdsapp#!/dataset/reanalysis-era5-land?tab=form.