Assessment of Temperature Extremes in China Using RegCM4 and WRF

HTML

--> --> -->

1. Introduction

Over the past three decades, temperature extremes associated with global climate change have caused great concern, because living creatures and ecosystems, as well as human society, are sensitive to the severity, frequency and persistence of extreme temperature events (Wigley, 1985; Frich et al., 2002; Wang and Fu, 2013). The Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC, 2013) reported that changes in extreme weather and climatic events have been observed since 1950, and the changes in the frequency and severity of those extreme events have induced a lot of climatic disasters. Both changes in the extreme temperature and their impacts on human society and ecosystems usually vary with the type of extremes, regions, and seasons (IPCC, 2012). Therefore, understanding the response of recent climate and extreme climatic change on regional and local scales is essential and necessary, ultimately helping us to make policy decisions on climate disaster mitigation and climate change adaptation.
In order to overcome the in-situ observation limitations due to inhomogeneities and spatiotemporal discontinuities, general circulation models (GCMs) have been extensively employed to simulate climatic extremes (Zhai et al., 2005). For example, (Toreti et al., 2013) indicated a spatially homogeneous behavior of climate extreme means simulated by state-of-the-art GCMs during 1966-2005 over the middle and high latitudes. However, the relatively coarse horizontal resolution of GCMs makes it difficult to adequately capture regional and local spatial features, as well as the interannual variations of extreme events. Small-scale meteorological features usually interact with the complex terrain or land cover to play a considerable role in determining the climate and weather at regional and local scales (Mass et al., 2002). Therefore, dynamical downscaling based on regional climate models (RCMs) is an essential tool to overcome the deficiency of the coarse horizontal resolution of GCMs (Qian and Leung, 2007; Wang et al., 2015).
With their relatively high horizontal resolution, RCMs can refine the physical processes related to detailed information on local climatic characteristics and extreme events, which are usually not resolved well in GCMs (Girogi and Gutowski, 2015). The Coordinated Regional Climate Downscaling Experiment (CORDEX) project is the first worldwide coordinated framework for regional climate modeling and downscaling (Giorgi et al., 2009), and its aim is to develop a coordinated framework for evaluating and improving RCMs over 14 sub-regions globally (http://cordex.org/domains/). Numerous investigations based on various RCMs have been involved in different CORDEX regional projects (e.g., Jacob et al., 2014; Feng et al., 2015; Gao et al., 2017). However, most of those works have focused on mean climate simulations over different regions. The performance of extreme climatic events in RCMs, in terms of their climate means, variability, and trends, have not been quantitatively assessed to a sufficient extent.
East Asia is one of the CORDEX domains and is also a vulnerable region to the future increase in climatic extremes under global warming (IPCC, 2012). Many previous researchers have investigated climate extremes based on the simulations of RCMs over China or the whole of Asia. Their conclusions indicate that RCMs perform generally well in reproducing the spatial distribution and trends of climatic extremes (Gao et al., 2002; Ji and Kang, 2015; Park et al., 2016). Because successfully simulating extreme climatic events depends greatly on the performance in different regions of the adopted RCM, evaluating different RCMs for a specific region is crucial to model selection for dynamic downscaling. For instance, (Park et al., 2016) applied five RCMs in the simulation of summer climate extremes in East Asia, and found that extreme climatic events based on the multi-model ensemble mean was better captured by the RCMs compared with the driving GCM. However, few studies have focused on the quantitative assessment of different RCMs in terms of temperature extremes in China, where the climate over most regions is affected by the complex East Asian monsoon system (Hui et al., 2018; Niu et al., 2018). To gain a better understanding of RCM performance with respect to extreme climatic events, this study evaluates and intercompares various extreme-temperature indices derived from the simulations of two widely used RCMs for the period 1981-2010.
The rest of the paper is organized as follows. Section 2 describes the RCMs' physical processes and the experimental design. Section 3 introduces the observed and modeled data, as well as the extreme-temperature indices. Section 4 presents the results of comparisons and evaluations in terms of regional means; plus, sub-regional analyses to understand the interannual variations of the temperature extremes over the Tibetan Plateau (TP) are also performed, and the large-scale circulations examined. Conclusions are presented in section 5.

3. Observational data, extreme-temperature indices, and methods

3.1. Observational data

-->

3.1. Observational data

The construction of extreme-temperature indices requires long-term daily maximum temperature (T_max) and minimum temperature ($T_{min}$) datasets. CN05 is a transformed daily gridded (0.5^°× 0.5^°) dataset derived from in-situ observations made at weather stations on mainland China (Wu and Gao, 2013). CN05 is available from 1961, and is updated on a regular basis. The construction of CN05 applies the "anomaly approach" in the interpolation method, which largely maintains the spatial and temporal pattern of station observations (Wu and Gao, 2013). Because the network of observed stations is relatively homogeneous, stable and numerous (more than 2400 stations) after 1980, our analyses are performed for the period 1981-2010.
CN05 has been compared with four other gridded datasets (Chinese daily homogenized datasets, HadEX2, GHCNDEX, and APHRODITE), and it was found that the trends of temperature extremes among the five datasets are coherent (Hong et al., 2015). Moreover, CN05 has been popularly used as a reference dataset for climate model validation in previous studies (Wang et al., 2015; Gao et al., 2017; Liang et al., 2018). Notably, due to the difficulties involved in data collection in vast remote regions, and the related missing data, station records are sparse over Northwest China and the TP, where the gridded CN05 contains relatively large uncertainties.

2

3.2. Extreme-temperature indices and methods

-->

3.2. Extreme-temperature indices and methods

Analyses of climatic extreme events are usually based on aggregated indices, which are statistically calculated from daily extreme climate data. The Expert Team on Climate Change Detection and Indices (ETCCDI) have constructed a set of extreme-climate indices, which have been widely used in climate change studies (e.g., Karl et al., 1999; Easterling et al., 2000; Frich et al., 2002). Here, we select 12 extreme-temperature indices, including six T_max-related indices and six ($T_{min}$)-related indices (Table 2). In general, these indices can be divided into three categories: (1) fixed threshold indices, which count the number or percentage of days during which time the variable (temperature) is under or over a predefined threshold, such as TX90p, TX10p, TN90p, TN10p, FD, and ID; (2) duration indices, which calculate the length of time the variable can be characterized in a specific way, such as WSDI and CSDI; and (3) absolute indices, which quantify the minimum/maximum magnitude of the variable during a certain period, such as TXx, TXn, TNx, and TNn. Thus, our selected indices in this paper potentially cover all aspects of extreme-temperature variations.
In our approach, the percentile indices are computed using a slightly different algorithm from the standard ETCCDI package, which is freely available at http://etccdi.pacificclimate.org/software.shtml. For the daily $T_{max}/T_{min}$ at each grid on each day, the percentage indices are computed based on the reference time period of 1981-2010. For instance: TX90p is the percentage of days when the daily T_max is greater than the 90th percentile of the calendar day centered on a five-day window for the base period 1981-2010; and CSDI is calculated as the number of days with at least six consecutive days when ($T_{min}$) is less than the 10th percentile during 1981-2010. Other procedures for the computation of extreme indices are the same as the standard processes of the ETCCDI package.
The simulated daily T_max and ($T_{min}$) are obtained from the simulations of RegCM and WRF. Then, the simulated extreme indices are calculated following the above algorithms. To allow a convenient comparison between observations and simulations, the T_max and ($T_{min}$) from both models are interpolated into a common grid of 0.5^°× 0.5^°, to be consistent with CN05, before all the indices are computed. The observed topography in this study is from the data in CN05, provided by (Wu and Gao, 2013). A bilinear interpolation algorithm with consideration of the lapse rate is used, due to the different elevations between the observational dataset and the models in each grid cell. (Zhang et al., 2009) showed that the actual temperature differences are rather small when the model-generated lapse rates (i.e., 2-7^°C km^-1) or standard lapse rate (i.e., 6.5^°C km^-1) are used for interpolation. Thus, for our calculation, the standard lapse rate of 6.5^°C km^-1 is used.
We assess three different aspects of the extreme-temperature indices. First, we compare their observed and simulated annual means, interannual variabilities, and linear trends over mainland China for the period 1981-2010. Second, their interannual variabilities and corresponding trends are computed from area-weighted averaged indices in China. And finally, detailed sub-regional analyses are performed to evaluate the simulations of their climate means, variabilities, and trends over the TP. Besides, the large-scale circulations simulated by the two RCMs are also compared to the driving ERA-20C reanalysis products.

4. Results

4.1. Means of extreme-temperature events

-->

4.1. Means of extreme-temperature events

Figure 2 displays the spatial distributions of the annual mean from CN05 and the annual mean biases (model-simulation departures from CN05) of four absolute temperature indices (TXx, TXn, TNx, and TNn) during 1981-2010 over China. The observed TXx decreases from south to north, and from southeast and southwest in China (Fig. 2a). TXx reaches its maximum with a magnitude above 30^°C in South China, and is generally below 15^°C in Northeast China and in the TP. Both RCMs reproduce well the spatial patterns and regional details of TXx (figure not shown). However, there are obvious regional differences of TXx biases in model simulations (Figs. 2b and c). For example, the TXx biases in RegCM over most regions vary between -1^°C and 1^°C, except for the Sichuan Basin and Xinjiang Province, where biases show obviously positive values (Fig. 2b). The TXx bias pattern of WRF is similar to that of RegCM. Nevertheless, an obviously cold bias of WRF prevails over the entire TP (Fig. 2c). The area-weighted average TXx bias over the whole of China is 0.06^°C from RegCM and -1.59^°C from WRF, while that from CN05 is 19.60^°C.
The observed pattern of TXn (Fig. 2d) is similar to that of TXx. The bias of TXn in RegCM is positive in South China but negative in West China (Fig. 2e). An obvious underestimation of TXn, with the magnitude as low as -4^°C, can be seen in North China and the TP for WRF (Fig. 2f). The mean bias of TXn over China is -1.14^°C and -3.81^°C for RegCM and WRF, respectively. From the bias patterns of TXx and TXn, we can see that a warm bias is relatively apparent in the Sichuan Basin in RegCM, while a cold bias over the TP is prominent in WRF.
TNx gradually decreases from south to north in China (Fig. 2g). The value of TNx is above 21^°C in both Guangdong and Guangxi Province. RegCM underestimates TNx in both South China and the TP, but slightly overestimates it along the north edge of the TP, in Northeast China, and over the Tianshan Mountains (Fig. 2h). Besides, in Northwest China, WRF simulates TNx as lower than in CN05 (Fig. 2i). Overall, RegCM has comparatively smaller negative TNx biases than WRF, which implies that RegCM outperforms WRF in simulating TNx in China. The observed TNn (Fig. 2j) decreases from southeast to northwest in China, which is similar to the pattern in CN05. TNn in RegCM exceeds its observed values in some parts of Northeast and Northwest China, and negative biases are located in South China and over the TP (Fig. 2k). The WRF simulation substantially underestimates TNx, with a bias magnitude under -4^°C over most regions (Fig. 2l). In summary, although the two models underestimate both TNx and TNn, the RegCM simulation performs better, with smaller biases than those in WRF in China.

Figure2. Spatial patterns of TXx, TXn, TNx, and TNn based on CN05 (left-hand panels), and the bias (departures from CN05) of TXx, TXn, TNx, and TNn for RegCM (middle panels) and WRF (right-hand panels) during 1981-2010 over China. The value in the top-right corner is the area-weighted average value of each plot. Units: ^°C.

The above results also show that the biases of TXx are smaller than those of TXn, and the discrepancies of TNx are smaller than those of TNn simulated by the two models. Besides, it seems that the models perform worse for nighttime than daytime regarding surface temperature. The T_max usually occurs during daytime and the ($T_{min}$) at nighttime. Because both models use the same boundary conditions, the physical mechanisms that control these two quantities in the models are partly responsible for the discrepancies. According to surface energy balance theory, the surface temperature is mainly impacted by shortwave radiation during daytime and longwave radiation at night. Therefore, the above differing results might be related to the processes of surface albedo, soil moisture, and cloud shading in the two RCMs (Dai et al., 2004; Zwiers et al., 2011).
Figure 3 presents the ID, FD, WSDI, and CSDI results from CN05, along with the annual biases from the two RCMs. For both ID (Fig. 3a) and FD (Fig. 3d) in CN05, the lowest values appear in South, Southwest China, and the Taklimakan Desert, while the maximum values occur in Northeast China and over the TP. The observed area-weighted average ID and FD are 62 days and 159 days, respectively. The Asian low, blocking highs, strong moisture transport, and an inversion layer contribute to the occurrence and persistence of ID in China (Wang et al., 2014). For FD, water vapor plays a significant role (Liu et al., 2008). Clearly, WRF (Fig. 3c) produces greater systematical biases of ID than RegCM (Fig. 3b) over Northeast China and the TP. For FD, the two models show positive biases along the Yangtze River valley (Figs. 3e and f). RegCM exhibits better spatial structure than WRF in Northeast and Northwest China. Notably, overestimation of FD is found over the south region of the TP for WRF, with an overestimated magnitude of up to 60 days. The ID and FD results are consistent with previous simulations based on RegCM4 (Ji and Kang, 2015) and WRF (Yu et al., 2015). The mean biases of the simulated ID and FD also indicate that RegCM outperforms WRF (e.g., the bias value of ID is 17 days for RegCM but 35 days for WRF).

Figure3. As in Fig. 2 but for ID, FD, WSDI and CSDI. Units: d.

The highest WSDI values occur in Southeast and West China, where the index exceeds 21 days (Fig. 3g). RegCM underestimates WSDI over Southeast China and the TP, but overestimates it over North China and Northwest China (Fig. 3h). WRF overestimates the index in most areas of China, with an exception over a small area in Northeast China and the coastal region of Southeast China (Fig. 3i). The area-weighted average bias over the whole of China is 0.09 days for RegCM and 3.81 days for WRF, indicating RegCM performs better than WRF in terms of simulating WSDI. Large values of observed CSDI appear in Yunnan Province, Northeast China, and Northwest China, while small values exist over the TP, with an area-weighted average of 17.62 days (Fig. 3j). Both models simulate similar patterns, with negative bias in South China and Northwest China but positive bias over the southeastern TP and North China. Compared to RegCM (Fig. 3k), WRF (Fig. 3i) performs worse overall, with relatively large CSDI bias (0.73 days) in magnitude over the whole of China.
Figure 4 is a Taylor Diagram (Taylor, 2001), produced to quantitatively display the performances of the simulated extreme-temperature indices. Figure 4 indicates that RegCM is more capable than WRF, expect with respect to FD. Compared to WRF, RegCM shows slightly larger correlation coefficients, but lower normalized standard deviation, for TXx, TXn, TNx, and TNn (Fig. 4a). For instance, the correlation coefficients for TNn are 0.95 and 0.92, and the normalized standard deviations 1.06 and 1.25, for RegCM and WRF, respectively. Similarly, RegCM correlates more strongly than WRF with observed ID, CSDI, and WSDI (Fig. 4b). The correlation coefficient and normalized standard deviation of WSDI are 0.95 and 1.46 for RegCM, respectively, while they are 0.52 and 1.57 for WRF. For FD, both models show almost identical performance (Fig. 4b). As such, overall, RegCM performs better than WRF.

Figure4. Taylor diagram of (a) TXx, TXn, TNx, and TNn, and (b) FD, ID, CSDI, and WSDI, for RegCM (blue) and WRF (red) during 1981-2010 over China. Different numbers represent different indices.

Figure 5 plots the probability distribution function (PDF) of biases in extreme-temperature indices, as well as annual-mean T_max and ($T_{min}$), from both models. For the T_max and derived indices (first row in Fig. 5), RegCM shows relatively smaller negative biases. Meanwhile, obviously positive biases in both ID and WSDI exist, which are also consistent with the above results. WRF reproduces dominant negative biases for both TXx and TXn, while remarkable positive biases exist for ID and WSDI. As shown in the left-hand column of Fig. 5, RegCM has weaker negative biases for TXx and TXn, and weaker positive biases of ID and WSDI, than WRF, which indicates quantitatively the better skill of RegCM. For the PDF pattern of ($T_{min}$) and derived indices, cold biases can be seen in the two models, but with a more severe discrepancy for WRF. Moreover, positive biases of FD are also apparent in the two RCMs, but with worse skill on show in WRF. The CSDI bias in RegCM shows a near normal-distribution pattern, which is also indicative of better skill compared with WRF.

Figure5. PDF distributions (%) of bias (departures from CN05) of T_max and ($T_{min}$), and the 12 selected extreme-temperature indices over China, derived from RegCM (blue) and WRF (red). The value at 10% and 90%, calculated by the cumulative distribution function, of the extreme-temperature indices is also shown in each plot.

2

4.2. Interannual variability and trends of extreme-temperature events

-->

4.2. Interannual variability and trends of extreme-temperature events

Figure 6 shows the time series of annual anomalies of the extreme-temperature indices from the three products during 1981-2010. The anomalies are computed as the departure from mean values for the period 1981-2010, based on the individual products. The two models generally capture well the interannual variability of every extreme-temperature index. In most cases, the temporal correlation coefficient (TCC) of the detrended indices is larger than 0.7 in both RCMs over China. The TCC values for TXx, WSDI, CSDI and TX90p are higher for RegCM than they are for WRF. For instance, the TCC for TX90p averaged over the whole of China is 0.73 and 0.62 for RegCM and WRF, respectively. The other indices show that WRF has a higher correlation coefficient than RegCM, which indicates a better performance than RegCM.

Figure6. Time series of anomalous extreme-temperature indices over China based on CN05 (black line), RegCM (blue line) and WRF (red line). Values are the TCCs between RegCM/WRF and CN05 according to the removed-trend values (before: calculated between RegCM and CN05; after: calculated between WRF and CN05). An asterisk indicates the correlation coefficient is statistically significant at the 95% confidence level.

Next, the mean annual linear trends of the different indices based on CN05 and the two simulations for the period 1981-2010 are compared. Figure 7 shows that the trends for all the indices are consistently the same sign for CN05 and the simulations, although the magnitudes of the trend differences are still visible. The two RCMs succeed in simulating the observed trends, with statistical significance at the 95% confidence level. For TXx, TXn, TNx, and TNn, both CN05 and the two models yield obvious increases of 0.3-0.6^°C (10 yr)^-1, although the simulated trends are smaller than in CN05. ID and FD decrease at a statistically significant rate of 1-5 d (10 yr)^-1 during 1981-2010 in all three datasets. A significant increase in WSDI [3.5 d (10 yr)^-1] in CN05 is successfully reproduced by the RCMs. RegCM reproduces more accurate trends of FD and CSDI than WRF, but for the simulated trends of ID and WSDI, WRF outperforms RegCM. The increasing linear trends of TX90p and TN90p are kept consistent from CN05, RegCM and WRF. Meanwhile, the two models can capture the smaller trend in TX90p than TN90p, which is consistent with the findings of a previous study (Wang and Fu, 2013). According to CN05, the average TX10p and TN10p decreases by 1.2% (10 yr)^-1 and 2.1% (10 yr)^-1, with statistical significance at the 5% level. Both RegCM and WRF reproduce these results, but RegCM better represents the decreasing trend of TX10p and TN10p compared with WRF.

Figure7. Temporal linear trends for CN05 (grey), RegCM (blue), and WRF (red), computed from the time series of area-weighted average extreme-temperature indices in the period 1981-2010 over China. The units in (a-c) are ^°C (10 yr)^-1, d (10 yr)^-1, and % (10 yr)^-1, respectively. All values are statistically significant at the 95% confidence level. The indices are defined in Table 1.

2

4.3. Sub-regional analyses

-->

4.3. Sub-regional analyses

To quantitatively evaluate the extreme-temperature events in detail on the regional scale, we select the TP as a typical sub-region (shown in Fig. 1), where broadly remarkable biases are apparent in the two models, as analyzed above. The TP has high topography, with a mean elevation above 4000 m, and is of great importance in terms of regional and local climate change, with mechanisms involving both dynamic and thermodynamic forcings (Liu and Chen, 2000). Taylor Diagram analyses (not shown) support RegCM as the more skillful in reproducing the spatial distribution of extreme-temperature events, as compared with WRF.
Figure 8 presents the time series of annual anomalies of the extreme-temperature indices over the TP. Apparently, both simulations and observations are in agreement with respect to the interannual variability, albeit the spreads of the different indices are much larger over the TP than the whole of China (Fig. 6). From the corresponding linear trends of those indices (Fig. 9), we find that the temperature extremes over the TP experience more dramatic change than in other regions of China (Fig. 7). Furthermore, RegCM tends to underestimate the trends of the extreme-temperature indices, as compared with CN05, while WRF overestimates them, over the TP. For example, the linear trends of TXx are 0.53^°C (10 yr)^-1, 0.44^°C (10 yr)^-1, and 0.75^°C (10 yr)^-1 for CN05, RegCM, and WRF, respectively. The above result might be related to elevation-related amplification warming over the TP, which has been reported widely in the literature (e.g., Pepin et al., 2015; Guo et al., 2016). Besides, over the TP, the 50-km grid cell is relatively coarse for representation of the topography. (Xu et al., 2018) investigated the role of horizontal resolution over the TP by use of an RCM, and showed that the horizontal resolution affect almost all the dynamic and thermodynamic processes. Therefore, the horizontal resolution partly affects the simulation of climate extremes.

Figure8. Time series of anomalous extreme-temperature indices over the TP based on CN05 (black line), RegCM (blue line) and WRF (red line). The values are the TCCs between the corresponding detrended simulations by the two models (before: calculated between RegCM and CN05, after: calculated between WRF and CN05). Correlation coefficients with an asterisk are statistically significant at the 95% confidence level.

Figure9. Temporal linear trends for CN05 (grey), RegCM (blue) and WRF (red), computed from the time series of area-weighted average extreme-temperature indices in the period 1981-2010 over the TP. The units in (a-c) are ^°C (10 yr)^-1, d (10 yr)^-1, and % (10 yr)^-1, respectively. All values are statistically significant at the 95% confidence level. The indices are defined in Table 1.

2

4.4. Patterns of large-scale circulations

-->

4.4. Patterns of large-scale circulations

Large-scale circulation is important to climate extremes. Figure 10 shows the spatial patterns of 500-hPa height for both winter and summer based on ERA-20C, RegCM, and WRF for the period 1981-2010. In winter, the East Asian trough is strong over East China according to ERA-20C (Fig. 10a). In summer, this intense trough is weakened and displaced by the subtropical Pacific high over East China (Fig. 10d). The large-scale circulations of the mid-troposphere simulated by RegCM (Figs. 10b and e) and WRF (Figs. 10c and f) show similar structures and characteristics. However, the two models' 500-hPa height present discrepancies during winter and summer. For instance, the East Asian trough in WRF is stronger than in RegCM, and RegCM is relatively close to ERA-20C during winter. Furthermore, during summer the simulated subtropical Pacific high in RegCM is more intense than in ERA-20C, while the result in WRF is slightly weaker than in ERA-20C. The spatial patterns of averaged 200-hPa height based on ERA-20C and the two RCMs in both winter and summer also present similar structures and characteristics, albeit with some simulated biases (figure not shown). These discrepancies partly lead to the biases of the lower troposphere. In other words, the discrepancies of the mid and upper troposphere are associated with the biases of surface air temperature.

Figure10. Spatial patterns of averaged 500-hPa height based on (a, d) ERA-20C, (b, e) RegCM, and (c, f) WRF, in (a-c) winter and (d-f) summer, during 1981-2010 over China. The interval in winter is 40 gpm, and in summer it is 20 gpm. Units: gpm.

5. Conclusions

In this study, we run both RegCM and WRF, driven by ERA-20C, with a 50-km horizontal resolution, for the period 1900-2010 over continental China. Based on both RCMs' simulated T_max and ($T_{min}$), we calculate 12 extreme-temperature indices of the ETCCDI for the period 1981-2010. The simulation-based indices are then assessed by those derived from in-situ observational data (from the CN05 dataset) during the 30-year study period. We evaluate the climate means, interannual variabilities and trends of the extreme-temperature indices in depth. Because the forcing data (ERA-20C) lack daily extreme temperature, the dynamic downscaling in this work provides a new, high resolution, with more than 100 years of added-value data for the community to use.
Overall, both RegCM and WRF reproduce well the climate means of the extreme-temperature indices and large-scale circulations in comparison with CN05 and ERA-20C, respectively. When we compare the mean biases and interannual variabilities of the extreme indices, most have smaller biases in RegCM compared with WRF, especially over the TP. Analyses using both the Taylor Diagram and PDF distribution also support this result. However, both RCMs show cold biases in terms of the simulated mean T_max and ($T_{min}$). These biases tend to induce the cold biases of the derived absolute extreme-temperature indices, i.e., TXx, TXn, TNx, and TNn, and lead to the positive biases of ID, FD, and WSDI. Meanwhile, the biases are more substantial in WRF than in RegCM. However, the signs of the biases in both CSDI and ($T_{min}$) are inconsistent with one another, and show complicated behaviors (Fig. 5). Previous studies have also reported such close relationships between the means of temperature and extremes (Liu et al., 2008; Park et al., 2016).
The interannual variabilities of the extreme-temperature indices are captured well by the RCMs in a typical sub-region (the TP) and China as a whole. From the temporal correlation coefficients of the detrended indices from CN05 and the simulations, WRF performs better than RegCM with respect to interannual variability. The observed temperature extremes show substantially and statistically significant trends in China. The two RCMs robustly simulate the correct magnitudes of the linear trends of temperature extremes over the TP and China, such as the increasing trend of TXn and decreasing trend of TN10p, in the period 1981-2010. However, RegCM underestimates the observed trends in the extreme-temperature indices, while WRF overestimates them, over the TP. This also indicates that, even though the climatology of temperature extremes is quite well reproduced by the RCMs, the challenge for simulating the linear trend remains. The question of which RCM outperforms the other in terms of extreme-temperature events remains unanswered, since it strongly depends on the time scale and region.
Since the dynamical core, vertical levels, and physical process schemes are different in different models, the sources of cold biases in RCMs are still unclear. Table 1 shows the differences between the land surface schemes of the two RCMs, and the representation of land surface processes will critically affect the simulation of the near-surface climate (such as the surface albedo) over land. The albedo affects the surface temperature, latent/sensible heat fluxes, and shortwave/longwave radiation, hence having an effect on the stability of the planetary boundary layer and therefore subsequently has a direct relationship with the nocturnal near-surface air temperature (Argüeso et al., 2011). These processes are also relevant to the microphysical and convective schemes of cloud formation, which affects the transportation of radiation and thus the near-surface temperature. The interaction between the physical schemes and dynamical core also affects model performance (Awan et al., 2011). The radiation schemes in our simulations are as follows: CCM3 in RegCM, and the Goddard scheme in WRF. The radiation schemes affect the solar component and the longwave radiation. In the shortwave radiation, the effect of aerosols and trace gases (such as ozone, water vapor, carbon dioxide, and oxygen) will affect the surface air temperature. Also, the parameterization of cloud effects like scattering and absorption effect the climatology of climate extremes too (Kiehl et al., 1996). These complex relationships induce a challenging interpretation of model deficiencies and uncertainties in the simulation of extreme-temperature events. Besides, the observational dataset used in the present paper, CN05, contains potential uncertainties——in particular, over regions with a sparse distribution of weather stations——due to the processing algorithm and the data sources (Yu et al., 2015). For the forcing fields, (Hersbach et al., 2015) found that there are some clear biases of the circulation above 500 hPa with respect to ERA-Interim. In general, this deviation tends to increase from the upper troposphere towards the surface. Therefore, the biases from the driving fields may transmit to the RCMs and lead to some cold biases.
This study suggests that, in the climatology of extreme-temperature events, RegCM outperforms WRF; whereas, in terms of the interannual variability and changes of extreme-temperature events, WRF is the more skillful model. Future work should analyze the whole simulation period (1900-2010) and include the results for precipitation.

本站小编 Free考研考试/2022-01-02

Corresponding author: Aihui WANG,wangaihui@mail.iap.ac.cn;

HTML

3.1. Observational data

3.2. Extreme-temperature indices and methods

4.1. Means of extreme-temperature events

4.2. Interannual variability and trends of extreme-temperature events

4.3. Sub-regional analyses

4.4. Patterns of large-scale circulations

相关话题/Assessment Temperature Extremes

领限时大额优惠券,享本站正版考研考试资料!

Global Monsoon Changes under the Paris Agreement Temperature Goals in CESM1(CAM5)

A 31-year Global Diurnal Sea Surface Temperature Dataset Created by an Ocean Mixed-Layer Model

A Limb Correction Method for the Microwave Temperature Sounder 2 and Its Applications

High-Order Statistics of Temperature Fluctuations in an Unstable Atmospheric Surface Layer over Gras

Assimilation of Sea Surface Temperature in a Global Hybrid Coordinate Ocean Model

Further-Adjusted Long-Term Temperature Series in China Based on MASH

Assessment and Assimilation of FY-3 Humidity Sounders and Imager in the UK Met Office Global Model

Skillful Seasonal Forecasts of Summer Surface Air Temperature in Western China by Global Seasonal Fo

Reduced Sensitivity of Tropical Cyclone Intensity and Size to Sea Surface Temperature in a Radiative

Impacts of Anthropogenic Forcings and El Niño on Chinese Extreme Temperatures