1. Introduction
An increase in the magnitude of both temperature and precipitation extremes is one of the most consistent changes being currently observed (Donat et al. 2013; Fischer and Knutti 2016) and predicted for the future by global climate models (Bao et al. 2017; Easterling et al. 2000; IPCC 2013). Intensification of extreme climates, instead of gradual changes in means of temperature or precipitation, is recognized as an immediate threat to humans and their properties (Sillmann and Roeckner 2008). Extreme temperature and precipitation have drawn a great deal of attention given their large socioeconomic and ecological impacts (Diao et al. 2015; Smith 2011; Ye et al. 2016).
The present study focuses on extreme temperature and precipitation years since their frequency and intensity are predicted to increase in a warmer climate (Fischer and Knutti 2016; IPCC 2013; Seneviratne et al. 2016; Ye et al. 2013). Aside from the annual mean temperature and annual amounts of total precipitation, extreme years differ in many aspects on the daily scale, including maximum day and night temperatures (Brohan et al. 2006) and numbers of heavy or very heavy precipitation days (Easterling et al. 2000; Fischer and Knutti 2016; Min et al. 2011). Using five daily precipitation indices, Knapp et al. (2015) suggested that extremely large events (events above the 99th percentile of all events) are the best index for distinguishing extreme wet from extreme dry years. However, no analysis has been conducted to identify the best temperature index for distinguishing extreme warm from extreme cold years. We aim to extend such analysis by using 28 daily indices of both precipitation and temperature. Such analysis is needed because the most important daily indices that characterize extreme years may be most sensitive to changes in precipitation amount or annual temperature, and these daily indices should be considered together with total precipitation amount or annual warming rate when designing experiments and model simulations that focus on the response of ecosystems to extreme climate.
Our study includes various daily indices of precipitation and temperature that are recommended by the Expert Team on Climate Change Detection and Indices (ETCCDI) and Knapp et al. (2015). The ETCCDI, which is jointly sponsored by the World Meteorological Organization (WMO) commission of Climate and Ocean: Variability, Predictability and Change project, proposed 27 core indices derived from daily temperature and precipitation data (Sillmann and Roeckner 2008; Zhang et al. 2011). These indices include percentile-, threshold-, and duration-based indices (Sillmann and Roeckner 2008; Zhang et al. 2011). Indeed, observed changes in these indices worldwide already support predictions of increased temperature and precipitation extremes (Bao et al. 2017; Donat et al. 2013; Fischer and Knutti 2016; Zhang et al. 2011). In most regions of the globe, the annual occurrence of cold nights has significantly decreased and the occurrence of warm nights has significantly increased over the last five decades (Alexander et al. 2006; Donat et al. 2013); extreme daily maximum and minimum temperatures have also risen in the last century (Brohan et al. 2006). An increasingly warm climate possibly leads to precipitation extremes, including high total precipitation, more heavy and very heavy wet days, fewer rain events, and extended drought spans (Easterling et al. 2000; Fischer and Knutti 2016; Min et al. 2011; Ye 2014).
On the basis of long-term (100 yr; 1911–2010) observations worldwide, we calculate 28 extreme daily indices of temperature and precipitation recommended by ETCCDI and Knapp et al. (2015). Using these daily indices, we aim to characterize how extreme wet and dry years differ from one another and from average years, and how extreme cold and warm years differ from one another and from average years in terms of their daily extremes. We also evaluate whether these differences are consistent across mean annual precipitation (MAP) and mean annual temperature (MAT) gradients and across terrestrial ecoregion domains globally. Events that are currently rare (or extreme) may become commonplace in the future. Therefore, we also discuss the implications of our analysis for future climate change and for the design of field manipulation and model simulation experiments that focus on ecosystem responses to extreme climate.
2. Data and methods
a. Data selection criteria and quality control
Station-based daily precipitation and minimum and maximum air temperature records are downloaded from the Global Historical Climatology Network–Daily (Menne et al. 2012). The network comprises approximately 25 000 temperature stations and over 100 000 precipitation stations globally (https://www.ncdc.noaa.gov/ghcn-daily-description). However, only records from sites that meet the following criteria are used in the study: 1) records span a common 100-yr period (1911–2010); 2) the probability distribution of total annual precipitation or annual mean temperature during the 100-yr period fits normal distributions, based on a Shapiro–Wilks test; and 3) station temperature data are homogenous during at least an 80-yr period, as tested with the RHtestsV4 software package (Wang 2003). The daily temperature and precipitation data are quality controlled. Both daily maximum Tmax and minimum Tmin temperatures are set to missing values if Tmax < Tmin in that day (Alexander et al. 2006). Outliers in daily maximum and minimum temperature are defined as those values outside five standard deviations (std) of the climatological mean of the value for the day (i.e., mean ± 5 std; Brohan et al. 2006; Hansen et al. 1999). Outliers are also set to missing values. Finally, we select stations that have at least 80-yr data with no more than 15 days in missing values in daily records of each year. On the basis of these criteria and quality control, 1154 precipitation stations and 561 temperature stations are used in the study (Fig. 1).
Geographical distributions of 561 temperature stations and 1154 precipitation stations used in the study.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
b. Indices of extreme temperature and precipitation
The ETCCDI and Knapp et al. (2015) proposed 29 different indices to quantify the daily extremes of precipitation and temperature. These indices are selected in the present study because they fit our objective, that is, to identify key indices for characterizing the difference between extreme and average years. Similar to the works of Sillmann and Roeckner (2008) and Donat et al. (2013), this study does not use one precipitation index (Rnn: number of days above nn millimeters, where nn is a user-defined threshold) with a user-defined threshold. Thus, we include a total of 28 indices (Table 1) that quantify many aspects of climate extremes including intensity, frequency, and minimum and maximum of temperature and precipitation (Knapp et al. 2015; Sillmann and Roeckner 2008). The indices from ETCCDI are calculated in the RClimDex1.1 package (available online at http://etccdi.pacificclimate.org/software.shtml). The calculation of the ETCCDI percentile-indices (such as days when maximum/minimum temperature <90th percentile and annual total precipitation from days >95th percentile) needs a base period for determining the percentile. The ETCCDI recommends 1961–90 as base period for easy comparison of indices across stations with records of various lengths, for easy updates once new daily data are available, and because of the greater availability of data during this period (Zhang et al. 2005).
The 28 extreme temperature and precipitation indices included in this study. The first 26 indices are recommended by the ETCCDI (http://etccdi.pacificclimate.org/list_27_indices.shtml). The last two indices, which are each marked with an asterisk, are proposed by Knapp et al. (2015) but are not included in the ETCCDI list. Indices marked with two asterisks (SDII and CDD) are included in both the ETCCDI and the work of Knapp et al. (2015). The threshold for defining a precipitation event is 0.3 mm in Knapp et al. (2015) and 1 mm in the ETCCDI. Our study uses the 1-mm threshold for all precipitation indices.
c. Definitions of extreme and average years
For each of the precipitation and temperature datasets, annual total precipitation and annual mean temperature are calculated, respectively. Extreme dry (cold) years are defined as annual precipitation (mean temperature) less than the 10th percentile of the 100-yr distribution, extreme wet (warm) years are defined as annual precipitation (mean temperature) greater than the 90th percentile, and average years are defined as those between the 45th and 55th percentiles (Knapp et al. 2015).
d. Identifying best temperature and precipitation indices for distinguishing differences between extreme and average years
To assess how temperature and precipitation indices differ among extreme and average years, we use a subset of precipitation and temperature data to calculate and compare all 28 indices. First, these indices are normalized to allow for the comparison of indices with different units (this process is described in appendix A and illustrated in Fig. A1). Principal component analysis (PCA) is then used to visualize the differences among the extreme and average years when all precipitation or temperature indices are considered collectively. Permutational multivariate analysis of variance (perMANOVA) is applied to determine whether extreme and average years differ significantly when all daily precipitation or temperature indices are considered collectively. If perMANOVA suggests that extreme and average years differ significantly, a Euclidean distance-based similarity percentage (SIMPER) analysis is then used to calculate the percentage contribution of each of the daily indices to the divergence among extreme and average years. Finally, the mean values of the percentage contribution for each of the temperature or precipitation indices are compared among the extreme and average years using analysis of variance (ANOVA), with individual temperature or precipitation stations as random variables. Detailed description of these analyses is given below.
We assign each station to one of Bailey’s four global ecoregion domains (Bailey 1983) based on station latitude and longitude, and then randomly sample a subset of 15 stations in each domain. Ecoregion domains are subcontinental areas identified by broad climatic similarity, including dry, humid-temperate, humid-tropical, and polar climates (Bailey 1983; Bailey and Hogg 1986). Climate exerts an overriding effect on the composition and productivity of ecosystems at a subcontinental scale, and thus it is used to delineate ecoregion domains (Bailey 1983). By sampling the same number of stations in each ecoregion, we minimize the bias in the geographical distributions of the available weather stations. The humid-tropical ecoregion domain only has 15 stations for precipitation and 1 station for temperature. We randomly sample 15 stations for the other three domains (i.e., polar, humid-temperate, and dry). A subset of 60 precipitation stations (15 stations for each of the four ecoregions) and 46 temperature stations (1 station for the humid-tropical ecoregion and 15 stations for each of the other three ecoregions) are sampled.
The perMANOVA is a nonparametric multivariate statistical method to test whether the centroids and dispersions of some groups (extreme and average years in this study) are equivalent. The test requires no assumption concerning the number of variables (or their distributions) in the groups (Anderson 2001).
PCA, perMANOVA, SIMPER, and ANOVA are analyzed separately for precipitation and temperature indices, and for all four ecoregions (combined and separately).
e. Variation of precipitation and temperature extremes across the precipitation and temperature gradients
A linear fit and an exponential decay fit are applied to describe how these deviations in extreme temperature and precipitation amounts vary with MAT and MAP, respectively. Using a similar approach, the deviations are also calculated for the two temperature and precipitation indices that contribute to the largest divergences among extreme and average years (identified by the SIMPER analysis).
f. Trends in the occurrence of extreme temperature and precipitation
We calculate the trends in the occurrences of extreme cold and warm years, and extreme dry and wet years during 1911–2010 for all precipitation and temperature stations. Trends in the two temperature and precipitation indices contributing to the largest divergence among extreme and average years (identified by the SIMPER analysis) are also estimated.
3. Results
a. Contributions of indices to the divergence among extreme and average years
With all 46 stations combined, extreme cold, extreme warm, and average years differ significantly when the 16 daily temperature indices are considered collectively (perMANOVA; F = 471, p < 0.001); the first two axes of the PCA explain about 60% of the variance among the three types of years (Fig. 2a). The x axis explains 48% of the variance, and it is negatively related to the number of cool nights and cool days and positively related to the number of warm nights and warm days. The y axis explains 12% of the variance, and it is positively related to diurnal temperature range (Fig. 2a). Extreme wet, extreme dry, and average year types also differ significantly when the 11 daily precipitation indices are considered collectively (perMANOVA; F = 874, p < 0.001), with the first two axes of the PCA explaining about 75% of the variance among the thee year types (Fig. 2b). The x axis explains 62% of the variance, and it is positively related to consecutive dry days and negatively related to the other 10 daily precipitation indices. The y axis explains 13% of the variance, and it is positively related to the numbers of all precipitation events and events ≥10 mm.
PCA of 27 daily indices of temperature and precipitation. (a) Sixteen daily temperature extreme indices in extreme cold years (annual mean temperature below the 10th percentile), extreme warm years (above the 90th percentile), or average years (between 45th and 55th percentiles) for 46 stations. Vectors of each index are calculated by PCA. Indices with longer vector lengths are more important in contributing to the principal components. Circles are drawn to facilitate the comparison among the vector lengths of indices. The perMANOVA suggests that extreme cold, extreme warm, and average years differ significantly when the 16 daily temperature indices are considered collectively (F = 471, p < 0.001). (b) Eleven daily precipitation extreme indices in extreme dry years (annual precipitation below the 10th percentile), extreme wet years (above the 90th percentile), or average years (between 45th and 55th percentiles) for 60 stations. Extreme wet, extreme dry, and average year types also differ significantly when the 11 daily precipitation indices considered are collectively (perMANOVA; F = 874, p < 0.001). Indices include TN10p, TX10p, TN90p, TX90p, DTR, R10mm, R20mm, CDD, and WD (see Table 1). Other extreme indices are identified as less important and thus are not described here.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
As extreme and average years differ significantly when all precipitation or temperature indices are considered collectively, SIMPER analysis is applied to estimate their contributions to the divergence among the three year types. When all ecoregions are combined, the numbers of cool days and warm nights become the two most important indices for distinguishing extreme cold versus extreme warm years. Remarkably, similar results are found for the four individual ecoregion domains (Fig. 3a). The number of cool days is also the most important index for distinguishing extreme cold years from average years when the four ecoregions are combined and analyzed separately (Fig. 3b). The number of warm nights is the most important index for distinguishing extreme warm years from average years when the four ecoregions are combined and analyzed separately (Fig. 3c).
Percentage contributions of 16 daily temperature indices to the divergence among extreme cold, extreme warm, and average years across four global ecoregion domains (ecoregion names are given an the top of each panel), showing results (a) between extreme cold and extreme warm years, (b) between extreme cold and average years, (c) between extreme warm and average years based on the Euclidean distance-based SIMPER analysis. The bars are the average percentage contributions of temperature stations (n = 15 in each of the polar, humid-temperate, and dry ecoregion domains; n = 1 in the humid-tropical domain; n = 46 for all ecoregions combined). The error bars are the standard errors of the 46 (for all ecoregions combined) or 15 (for each ecoregion) stations. The humid-tropical domain only has one station, and thus no error bars are given. Only the five indices with highest contributions are shown. Indices with same gray color capital letters are statistically insignificant from one another according to individual ANOVA (with individual station as random variables).
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
Among the 11 daily precipitation indices, the number of events ≥10mm is the most important index for distinguishing among extreme dry, extreme wet, and average years when the four ecoregions are combined and analyzed separately (Fig. 4). The number of events ≥20 mm is also among the top five most important indices for distinguishing among the three year types when the four ecoregions are combined and analyzed separately (Fig. 4).
Percentage contributions of 11 daily precipitation indices to the divergence among extreme wet, extreme dry, and average years across four global ecoregion domains (ecoregion names are given on the top of each panel), showing results (a) between extreme wet and extreme dry years, (b) between extreme dry and average years, and (c) between extreme wet and average years. The bars are the average percentage contribution of precipitation stations (n = 15 in each of the four ecoregion domains; n = 60 for all ecoregions combined).
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
To evaluate whether using another random subset changes the results of SIMPER analysis, we randomly sample a new subset of 45 precipitation stations (15 stations each for arid, temperate, and cold ecoregions) and run the SIMPER analysis again (described in appendix B and illustrated in Fig. B1). The contributions of the 11 precipitation indices are relatively consistent between the two subsets (Figs. 4 and B1). Therefore, the results of SIMPER analysis are generally reliable. To test whether the result of the SIMPER analysis depends on the number of indices used, we delete the five precipitation indices with the least divergence contributions. The remaining six indices are reanalyzed by SIMPER. The results are remarkably similar to those using all 11 indices (Fig. B1) and thus SIMPER analysis does not seem to be dependent on the number of indices used.
b. Differences of the most important precipitation and temperature indices in extreme and average years
To further investigate the difference of extreme cold versus extreme warm and extreme dry versus extreme wet years across the climate gradients, we calculate the deviations of the most important indices (identified by the SIMPER analysis described above) in the extreme years from average years (Fig. 5). The deviations of mean temperature in both extreme cold and extreme warm years from average years are higher in low temperature regions than in high temperature regions (Fig. 5a). This decreasing pattern is also evident in the number of cool days (Fig. 5b). Furthermore, a clear pattern of asymmetry is found between extreme cold and extreme warm years with regard to their deviation in the number of cool days from average years (Fig. 5b). Extreme cold years have 8.6 ± 0.7 (uncertainties are 95% confidence intervals) number of cool days more than average years at −7°C MAT, whereas extreme warm years have 4.6 ± 0.4 number of cool days fewer than average years. In high temperature regions of 25°C MAT, the deviations of the number of cool days decrease to 4.2 ± 0.6 more for extreme cold years and 3.4 ± 0.3 fewer for extreme warm years (Fig. 5b). A clear asymmetric pattern is also observed in the deviation of the number of warm nights (Fig. 5c). Extreme warm years have 6.6 ± 0.7 number of warm nights more than average years at −7°C MAT, and this value decreases to 5.2 ± 0.7 at 25°C MAT (Fig. 5c). By contrast, extreme cold years have 4.2 ± 0.5 number of warm nights fewer than average years and this deviation does not change across the temperature gradient (Fig. 5c).
Deviations in extreme temperature and extreme precipitation from average climate: (a) mean temperature, number of (b) cool days and (c) warm nights for extreme cold and extreme warm years compared with the mean of average years across 561 stations worldwide, (d) annual precipitation amount, and number of events (e) ≥10 mm and (f) ≥20 mm for extreme wet and extreme dry years compared with the mean of average years across 1164 stations worldwide. As precipitation change is usually expressed as relative value (Knapp et al. 2015), the relative deviation (%) is used. The insets are the relation between mean annual temperature and number of cool days and warm nights, and the relations of mean annual precipitation and number of events ≥10 mm and ≥20 mm. The lines represent the best fits; only equations with significant regressions (p < 0.05) are given in the plot.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
Slightly asymmetric relative deviations in annual precipitation are observed between extreme wet and extreme dry years in comparison with precipitation amount from average years (Fig. 5d). In general, for stations with MAP above 1000 mm, annual precipitation in extreme wet years is 38% ± 2% greater than that in average years, and this value increases as MAP decreases in more arid stations (up to 65% ± 2% at 100 mm MAP). Annual precipitation in extreme dry years is 32% ± 2% below average for stations with MAP > 1000 mm, and this deviation increases to 51% ± 2% as MAP decreases to 100 mm (Fig. 5d). Clear asymmetric patterns are observed in the number of heavy precipitation events (events ≥ 10 mm and ≥20 mm) between extreme wet and extreme dry years relative to average years (Figs. 5e,f). For stations with MAP above 1000 mm, the relative deviations of the number of events ≥ 10 mm are 32% ± 3% greater in extreme wet years than in average years and 31% ± 3% less in extreme dry years than in average years. As MAP decreases to 100 mm the relative deviations increase to 89% ± 3% and 71% ± 3% for extreme wet and extreme dry years, respectively (Fig. 5e). For sites with MAP above 1000 mm, the deviations of the number of events ≥ 20 mm in extreme wet and extreme dry years are 49% ± 6% and 41% ± 3% greater and lower relative to average years, respectively. As MAP decreases to 100 mm the deviations increase to 205% ± 6% and 99% ± 3% for extreme wet and extreme dry years, respectively (Fig. 5f).
c. Spatial and temporal trends in the occurrence of temperature and precipitation extremes
To assess the current trends in the occurrences of the key temperature and precipitation indices identified in the above section, we present the spatial and temporal trends in Fig. 6. During 1911–2010, about 60% of the temperature stations show significant trends in the occurrences of cool days and warm nights (Figs. 6a,b). As shown from these 60% stations, a trend toward warm conditions can be observed: more stations show significant increases in the occurrences of warm nights compared with the number of stations showing significant decreasing trends (Fig. 6b). However, not all areas on the globe show trends in the same direction. The number of cool days increase and the number of warm nights decrease in the southeastern and central United States (Figs. 6a,b); the region is denoted as the “warming hole” (Kunkel et al. 2006; Pan et al. 2004). Trends in the occurrences of extreme warm and cold years also show similar spatial patterns (appendix C, Fig. C1).
Trends in the occurrences of temperature and precipitation extremes over the whole 100-yr period of 1911–2010. (a),(b) Annual counts of cool days and warm nights. (c),(d) Annual counts of precipitation events ≥10 mm and events ≥20 mm. Stations with no significant changes at the 5% level are not shown.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
Among the 25% precipitation stations that show significant changes in events ≥ 10 mm and ≥20 mm, most exhibit a wetting trend during 1911–2010 (Figs. 6c,d). This trend is consistent with the increasing frequency of extreme wet years during the recent decades (Fig. C1). Spatially, stations showing negative trends in the occurrences of event ≥ 10 mm are mostly located in the southeastern United States and southern Australia (Figs. 6c,d).
4. Discussion
a. Indices of extreme climate and implications for future climate change
The importance of climate extremes has been widely recognized due to their broad ecological and socioeconomic impacts (Diao et al. 2015; Smith 2011; Ye et al. 2016). Using 28 extreme indices proposed by the ETCCDI (Zhang et al. 2011) and Knapp et al. (2015), we identify key indices of extreme climate. Our analysis suggests that high occurrence of events ≥10 mm and events ≥20 mm are the key indices for distinguishing extreme wet years from average and extreme dry years (Fig. 4). This result is different from those of Knapp et al. (2015); they suggest that the presence of extremely large events (daily precipitation amount >99th percentile) is the key index for distinguishing extreme wet years from average and extreme dry years. This percentile-based index is also used in our study, but it is less important than the threshold-based index of the number of events ≥10 mm. The different results may be attributed to the following: 1) R10mm and R20mm were not examined in the work of Knapp et al. (2015), and 2) the threshold for defining a precipitation event is 0.3 mm in the work of Knapp et al. (2015), whereas it is 1 mm in our study (1 mm is recommended by the ETCCDI). These differences suggest that the analysis based on few indices may bias the result and a consensus on the threshold is needed to define precipitation events. Fixed threshold–based indices are usually thought to not be applicable everywhere on the globe (Alexander et al. 2006). Surprisingly, the importance of the number of events ≥10 mm and events ≥20 mm is remarkably coherent across four ecoregions worldwide (Fig. 4). Previous studies of temperature indices also showed that fixed threshold–based indices, such as FD0 (annual count when daily minimum temperature <0°C), exhibit consistent trends over much of the Northern Hemisphere (e.g., Kiktev et al. 2003). These fixed threshold–based indices are convenient to use, and their changes can have profound impacts on societies or ecosystems (Alexander et al. 2006). Therefore, we recommend the number of events ≥10 mm and events ≥20 mm as the key indices for distinguishing extreme wet and extreme dry years.
Consistent with the work of Knapp et al. (2015), our analysis also suggests that compared with high precipitation regions, low precipitation regions require greater proportional deviations from average precipitation in order to achieve extreme dry or wet, and the deviations of precipitation amounts are larger in extreme wet years (65% ± 2%) than in extreme dry years (−51% ± 2%) with regard to average years in arid regions (Figs. 5d–f). Such an asymmetric pattern should not be caused by the underlying asymmetric distribution of the 100-yr annual precipitation data because all the precipitation datasets fit normal distributions. The fact that the decrease of precipitation has an upper limit (i.e., −100%) may result in smaller deviation in dry years. This should not influence our result since both deviations in extreme dry and wet years are generally <100%, while attention should be paid to such upper limit. Moreover, our analysis shows that relative deviations in the number of events ≥10 mm (89% ± 3% in extreme wet and −71% ± 3% in extreme dry years) are much higher than those of precipitation amounts. This result suggests that these heavy events are extremely sensitive to changes in extreme climates.
SIMPER analysis is also used to identify key indices of temperature that distinguish extreme cold from extreme warm years. In contrast to precipitation, the percentile-based temperature indices (number of cool days and number of warm nights) best explain the difference between extreme cold and extreme warm years (Fig. 3). An asymmetric pattern is observed in the deviations of the number of cool days and the number of warm nights between extreme warm and extreme cold years from average years, although the deviations of the annual mean temperature itself is symmetric between the two types of years (Figs. 5a–c). The deviation of the number of warm nights in extreme warm years exceeds that in extreme cold years. Such patterns of asymmetry predict that a warmer world in the future will be primarily characterized by many more warm nights. In fact, significant increases in the frequency of warm nights have already been observed globally, and this condition is consistent with climate warming worldwide (Donat et al. 2013; Kiktev et al. 2003).
b. Implications for climate change experiments
The patterns that emerge from our analysis have important implications for the design of both field experiments and model simulations for evaluating the response of various ecosystems to climate change, including those that simulate extreme wet or dry and extreme warm or cold years. Our analysis supports the recommendation by Knapp et al. (2017) that in order to impose comparable extreme precipitation treatments across all ecosystems, sites characterized by higher precipitation variability will require higher alterations in precipitation amounts than those in ecosystems characterized by lower precipitation variability. Usually, precipitation variability (measured as coefficient of variation) increases across sites with decreasing mean annual precipitation (Davidowitz 2002; Ye et al. 2013). Therefore, dry sites require higher proportional changes of precipitation to achieve extremity (as shown in Fig. 5d). Experiments of alterations in annual precipitation totals need to be combined with changes in the number of extreme events (Knapp et al. 2015, 2017); our analysis suggests that these are events ≥10 or ≥20 mm in all ecoregions. Incidentally, Thomey et al. (2011) applied the threshold of 20-mm rainfall events in an arid grassland to simulate a more extreme precipitation regime. They showed that plots receiving a single 20-mm event each month maintain significantly higher productivity and soil respiration than those receiving four 5-mm rain events. Changes in the number of heavy precipitation events, alone or combined with alteration in total amount, are recognized to be of great importance, especially in more arid ecosystems (Knapp et al. 2008; Thomey et al. 2011; Ye et al. 2016). In study areas where long-term precipitation data are not available, the equations shown in Figs. 5d–f provide a convenient way to calculate the precipitation amount and the number of heavy events needing to be manipulated.
The variability of temperature increases with decreasing MAT (Fig. 5a); this relation is expected given that tropical locations by nature experience smaller interannual variability than high latitude locations, and vice versa. To impose comparably extreme warming across ecosystems, cold ecosystems require a higher increase in temperature than warm ecosystems (Seneviratne et al. 2016; Wallace et al. 2012). Furthermore, both field manipulations and model simulation warming experiments need to be combined with an increase in the number of warm nights (Fig. 5c), which is often ignored in most studies of terrestrial ecosystem responses to warming (Peng et al. 2013). Passive open-top chambers have been widely used to experimentally increase temperature in ecosystems, which mostly increases daytime temperature above ambient (Marion et al. 1997). However, such devices are inefficient in increasing the number of warm nights. Infrared heaters for warming ecosystem field plots (Kimball et al. 2008) are able to incorporate both increasing warm nights and decreasing cool days. The equations in Figs. 5a–c calculate the warming rate and the number of warm nights need to be manipulated.
In summary, using a statistical approach (SIMPER analysis) borrowed from ecology, we identify key indices of extreme precipitation and temperature. The numbers of events ≥10 mm and events ≥20 mm are the two most important precipitation indices for distinguishing extreme wet and extreme dry years; the numbers of cool days and warm nights are key indices for distinguishing extreme cold and extreme warm years. These patterns are consistent across all ecoregion domains worldwide. Some of the patterns shown in our analysis, such as the importance of warm nights in extreme warm years, have been widely noticed. However, no statistical comparison has been made to identify the importance of different daily temperature indices in distinguishing extreme years from average years, particularly on a global scale. Close attention should be paid to these key indices when analyzing climate model projections and designing regional- and global-scale climate change experiments.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grants 31570467 and 31770480) and the project of State Key Laboratory of Grassland Agro-ecosystems of Lanzhou University (Grant SKLGAE201705). We are grateful to the two anonymous reviewers.
APPENDIX A
The Flowchart of Different Steps of Analyses Used in the Study
The stepwise process of analysis utilized in this study is indicated in Fig. A1.
Flowchart of different steps of analyses used in the study.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
APPENDIX B
Robustness of the Euclidean Distance-Based Similarity Percentage Analysis
We include two robustness tests:
The first and second columns of Fig. B1 (Figs. B1a–f) show the results of similarity percentage (SIMPER) analysis based on two different subsets of stations that randomly sampled from 1154 precipitation stations globally. All 11 precipitation indices are included in the SIMPER analysis. The first subset (Figs. B1a–c) includes 60 stations (15 stations each in the polar, humid-temperate, dry, and humid-tropical ecoregion domains). The second subset (Figs. B1d–f) includes 45 stations (15 stations each in the polar, humid-temperate, and dry ecoregion domains). There are only 15 stations in the humid-tropical ecoregion domain, so the second subset does not include station in this domain. The percentage contributions of 11 precipitation indices to the divergence between extreme wet and extreme dry years, extreme dry and average years, and extreme wet and average years. The results of the SIMPER analysis between the two subsets are consistent.
The right column of Fig. B1 (Figs. B1g–i) tests the dependence of SIMPER analysis on the numbers of precipitation indices used. Only the six indices with highest contributions (identified in the first robustness test) are included in the SIMPER analysis. SIMPER analysis is used to calculate the percentage contributions of the six precipitation indices to the divergence among extreme wet, extreme dry, and average years. The results of the SIMPER analysis using 6 indices are remarkably similar to those using all 11 indices (Figs. B1a–c and B1g–i). Thus the SIMPER analysis does not seem to depend on the number of indices used.
Robustness of the SIMPER analysis for calculating the percentage contribution of individual daily index to the divergence between extreme and average years. Results of SIMPER analysis based on (a)–(c) 11 precipitation indices calculated for a subset of 60 stations (15 stations each in the polar, humid-temperate, dry, and humid-tropical ecoregion domains); (d)–(f) 11 precipitation indices calculated for a different subset of 45 stations (15 stations each in the polar, humid-temperate, and dry ecoregion domains); and (g)–(i) as in (a)–(c), but based only on the six indices with the highest contributions identified in (a)–(c). The percentage contributions of precipitation indices to the divergence between (top) extreme wet and extreme dry years, (middle) extreme dry and average years, and (bottom) extreme wet and average years. The bars are the average percentage contributions of the 60 or 45 precipitation stations. The error bars are the standard errors of the 60 or 45 precipitation stations.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
APPENDIX C
Spatial and Temporal Trends in the Occurrence of Extreme Years
During 1911–2010, about 25% of the stations show significant trends in the occurrences of extreme cold and extreme warm years (Figs. C1a,b). A trend toward warm conditions can be observed: more stations show significant increases in the occurrences of extreme warm years compared with the number of stations showing significant decreasing trends. Extreme cold years increase in the southeastern and central United States. Among the 10% of precipitation stations that show significant changes, most exhibit a wetting trend during 1911–2010 (Figs. B1c,d). Spatially, stations showing negative trends in the occurrences of extreme dry years are mostly located in southern Australia.
Trends in the occurrences of extreme temperature and precipitation years over the whole 100-yr period of 1911–2010. (a),(b) Frequencies of extreme cold and extreme warm years. (c),(d) Frequencies of extreme dry and extreme cold years. Stations with no significant changes at the 5% level are not shown.
Citation: Journal of Climate 31, 1; 10.1175/JCLI-D-17-0377.1
REFERENCES
Alexander, L. V., and Coauthors, 2006: Global observed changes in daily climate extremes of temperature and precipitation. J. Geophys. Res., 111, D05109, doi:10.1029/2005JD006290.
Anderson, M. J., 2001: A new method for non-parametric multivariate analysis of variance. Aust. J. Ecol., 26, 32–46, doi:10.1111/j.1442-9993.2001.01070.pp.x.
Bailey, R. G., 1983: Delineation of ecosystem regions. Environ. Manage., 7, 365–373, doi:10.1007/BF01866919.
Bailey, R. G., and H. C. Hogg, 1986: A world ecoregions map for resource reporting. Environ. Conserv., 13, 195–202, doi:10.1017/S0376892900036237.
Bao, J., S. C. Sherwood, L. V. Alexander, and J. P. Evans, 2017: Future increases in extreme precipitation exceed observed scaling rates. Nat. Climate Change, 7, 128–132, doi:10.1038/nclimate3201.
Brohan, P., J. J. Kennedy, I. Harris, S. F. B. Tett, and P. D. Jones, 2006: Uncertainty estimates in regional and global observed temperature changes: A new data set from 1850. J. Geophys. Res., 111, D12106, doi:10.1029/2005JD006548.
Clarke, K. R., 1993: Non-parametric multivariate analyses of changes in community structure. Aust. J. Ecol., 18, 117–143, doi:10.1111/j.1442-9993.1993.tb00438.x.
Davidowitz, G., 2002: Does precipitation variability increase from mesic to xeric biomes? Global Ecol. Biogeogr., 11, 143–154, doi:10.1046/j.1466-822X.2002.00271.x.
Diao, Y., S.-P. Xie, and D. Luo, 2015: Asymmetry of winter European surface air temperature extremes and the North Atlantic Oscillation. J. Climate, 28, 517–530, doi:10.1175/JCLI-D-13-00642.1.
Donat, M. G., and Coauthors, 2013: Updated analyses of temperature and precipitation extreme indices since the beginning of the twentieth century: The HadEX2 dataset. J. Geophys. Res. Atmos., 118, 2098–2118, doi:10.1002/jgrd.50150.
Easterling, D. R., G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and L. O. Mearns, 2000: Climate extremes: Observations, modeling, and impacts. Science, 289, 2068–2074, doi:10.1126/science.289.5487.2068.
Fischer, E. M., and R. Knutti, 2016: Observed heavy precipitation increase confirms theory and early models. Nat. Climate Change, 6, 986–991, doi:10.1038/nclimate3110.
Hansen, J., R. Ruedy, J. Glascoe, and M. Sato, 1999: GISS analysis of surface temperature change. J. Geophys. Res., 104, 30 997–31 022, doi:10.1029/1999JD900835.
IPCC, 2013: Summary for policymakers. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1–30.
Kiktev, D., D. M. H. Sexton, L. Alexander, and C. K. Folland, 2003: Comparison of modeled and observed trends in indices of daily climate extremes. J. Climate, 16, 3560–3571, doi:10.1175/1520-0442(2003)016<3560:COMAOT>2.0.CO;2.
Kimball, B. A., M. M. Conley, S. Wang, X. Lin, C. Luo, J. Morgan, and D. Smith, 2008: Infrared heater arrays for warming ecosystem field plots. Global Change Biol., 14, 309–320, doi:10.1111/j.1365-2486.2007.01486.x.
Knapp, A. K., and Coauthors, 2008: Consequences of more extreme precipitation regimes for terrestrial ecosystems. BioScience, 58, 811–821, doi:10.1641/B580908.
Knapp, A. K., and Coauthors, 2015: Characterizing differences in precipitation regimes of extreme wet and dry years: Implications for climate change experiments. Global Change Biol., 21, 2624–2633, doi:10.1111/gcb.12888.
Knapp, A. K., and Coauthors, 2017: Pushing precipitation to the extremes in distributed experiments: Recommendations for simulating wet and dry years. Global Change Biol., 23, 1774–1782, doi:10.1111/gcb.13504.
Kunkel, K. E., X.-Z. Liang, J. Zhu, and Y. Lin, 2006: Can CGCMs simulate the twentieth-century “warming hole” in the central United States? J. Climate, 19, 4137–4153, doi:10.1175/JCLI3848.1.
Marion, G. M., and Coauthors, 1997: Open-top designs for manipulating field temperature in high-latitude ecosystems. Global Change Biol., 3, 20–32, doi:10.1111/j.1365-2486.1997.gcb136.x.
Menne, M. J., I. Durre, R. S. Vose, B. E. Gleason, and T. G. Houston, 2012: An overview of the Global Historical Climatology Network–Daily database. J. Atmos. Oceanic Technol., 29, 897–910, doi:10.1175/JTECH-D-11-00103.1.
Min, S.-K., X. Zhang, F. W. Zwiers, and G. C. Hegerl, 2011: Human contribution to more-intense precipitation extremes. Nature, 470, 378–381, doi:10.1038/nature09763.
Pan, Z., R. W. Arritt, E. S. Takle, W. J. Gutowski Jr., C. J. Anderson, and M. Segal, 2004: Altered hydrologic feedback in a warming climate introduces a “warming hole.” Geophys. Res. Lett., 31, L17109, doi:10.1029/2004gl020528.
Peng, S., and Coauthors, 2013: Asymmetric effects of daytime and night-time warming on Northern Hemisphere vegetation. Nature, 501, 88–92, doi:10.1038/nature12434.
Seneviratne, S. I., M. G. Donat, A. J. Pitman, R. Knutti, and R. L. Wilby, 2016: Allowable CO2 emissions based on regional and impact-related climate targets. Nature, 529, 477–483, doi:10.1038/nature16542.
Sillmann, J., and E. Roeckner, 2008: Indices for extreme events in projections of anthropogenic climate change. Climatic Change, 86, 83–104, doi:10.1007/s10584-007-9308-6.
Smith, M. D., 2011: The ecological role of climate extremes: Current understanding and future prospects. J. Ecol., 99, 651–655, doi:10.1111/j.1365-2745.2011.01833.x.
Thomey, M. L., S. L. Collins, R. Vargas, J. E. Johnson, R. F. Brown, D. O. Natvig, and M. T. Friggens, 2011: Effect of precipitation variability on net primary production and soil respiration in a Chihuahuan desert grassland. Global Change Biol., 17, 1505–1515, doi:10.1111/j.1365-2486.2010.02363.x.
Wallace, J. M., Q. Fu, B. V. Smoliak, P. Lin, and C. M. Johanson, 2012: Simulated versus observed patterns of warming over the extratropical Northern Hemisphere continents during the cold season. Proc. Natl. Acad. Sci. USA, 109, 14 337–14 342, doi:10.1073/pnas.1204875109.
Wang, X. L., 2003: Comments on “Detection of undocumented changepoints: A revision of the two-phase regression model.” J. Climate, 16, 3383–3385, doi:10.1175/1520-0442(2003)016<3383:CODOUC>2.0.CO;2.
Ye, J.-S., 2014: Trend and variability of China’s summer precipitation during 1955–2008. Int. J. Climatol., 34, 559–566, doi:10.1002/joc.3705.
Ye, J.-S., J. F. Reynolds, G. J. Sun, and F. M. Li, 2013: Impacts of increased variability in precipitation and air temperature on net primary productivity of the Tibetan Plateau: A modeling analysis. Climatic Change, 119, 321–332, doi:10.1007/s10584-013-0719-2.
Ye, J.-S., J. F. Reynolds, F. T. Maestre, and F.-M. Li, 2016: Hydrological and ecological responses of ecosystems to extreme precipitation regimes: A test of empirical-based hypotheses with an ecosystem model. Perspect. Plant Ecol., 22, 36–46, doi:10.1016/j.ppees.2016.08.001.
Zhang, X., G. Hegerl, F. W. Zwiers, and J. Kenyon, 2005: Avoiding inhomogeneity in percentile-based indices of temperature extremes. J. Climate, 18, 1641–1651, doi:10.1175/JCLI3366.1.
Zhang, X., L. Alexander, G. Hegerl, P. Jones, A. Klein Tank, T. Peterson, B. Trewin, and F. W. Zwiers, 2011: Indices for monitoring changes in extremes based on daily temperature and precipitation data. Wiley Interdiscip. Rev.: Climate Change, 2, 851–870, doi:10.1002/wcc.147.
