1. Introduction
It has been shown that with an increase or decrease of total precipitation, disproportionate changes occur in the upper end of the precipitation frequency distribution (Karl and Knight 1998; Groisman et al. 1999, 2001, 2004; Kunkel et al. 1999; Easterling et al. 2000; Folland and Karl 2001; Semenov and Bengtsson 2002; Kunkel 2003). Over the conterminous United States, this feature has become prominent since circa 1970 (Soil and Water Conservation Society 2003). In particular, upward trends in the amount of precipitation occurring in the upper 0.3% of daily precipitation events are statistically significant for the past hundred years within the central regions of the United States (Groisman et al. 2004). A time series of the frequency of events in the upper 0.3% averaged for these regions shows a 22% increase over the period since 1893 with all of this increase occurring over the last third of the twentieth century (Fig. 1). These upward trends are primarily a warm season phenomenon when the most intense rainfall events typically occur.
Karl and Knight (1998) show that most (87% of the variance nationwide for the conterminous United States) of the increase in seasonal/annual precipitation can be ascribed to changes in the number of days with precipitation.1 The tendencies, which emerged during the past 35–40 yr with a disproportional increase in precipitation coming from intense rain events (Groisman et al. 2004, 2005), should lead to discontinuities in the parallel increase/decrease of both total precipitation and precipitation frequency. For the United States this discontinuity was first reported by Sun and Groisman (2004) and for the northeastern quadrant of the conterminous United States confirmed by Groisman et al. (2005). Specifically, for the northeastern quadrant of the United States, they reported an increase (or no change) in precipitation totals but a decrease in the number of days with precipitation. If continued, this decrease in precipitation frequency may lead to an increase in the frequency of another potentially hazardous type of extreme event: prolonged periods without precipitation (even when the mean seasonal rainfall totals increase). Below we investigate whether this development is already occurring during the past several decades over the conterminous United States, for the same period when we begin observing changes in frequency of intense precipitation events (i.e., since circa 1970).
2. Methodology
a. Data
For our analyses we use the same daily precipitation dataset of the U.S. Cooperative Observer Program (COOP) stations described in Groisman et al. (2004) but updated to 2006. The National Weather Service (NWS) COOP is truly the United States’s weather and climate observing network of, by, and for the people. More than 11 000 volunteers take observations on farms, in urban and suburban areas, national parks, seashores, and mountaintops. The data are truly representative of where people live, work, and play.
b. Approach
Paleoclimatic reconstructions (e.g., Herweijer et al. 2007) provide a large-scale picture of drought frequencies during the past millennium, and in situ observations (e.g., Dai et al. 2004; Andreadis et al. 2005) can now deliver quite detailed information about the dry conditions during the past 100 years. If not driven by a large-scale storm system, a precipitation mosaic from summer storms leaves numerous dry spots across the country (cf. http://www.drought.unl.edu/dm/monitor.html). It is not our intention to study this mosaic or individual spectacularly dry years. Instead we focus on systematic changes of dry conditions on the nationwide scale during the past 40 yr using only precipitation information from our station network. We are looking for large-scale changes in the annual areal summation of the duration of nonrain episodes over the nation and are not very much interested in a particular pattern of these dry episodes. The rationale for this selective interest is a potential of linking these changes with changes in global-scale processes (e.g., global warming), which might affect continental dryness (Manabe et al. 1981, 2004). The GCM simulations forced by different scenarios of changes in atmospheric composition (e.g., Manabe et al. 2004; Kharin et al. 2007; McAvaney et al. 2001) hint at the potential for significant changes in “summer” dry conditions but are not yet able to produce a detailed picture of these changes.
c. Focus on the warm season and prolonged no-rain episodes
We know that the number of rainy days over the country has increased during the past 100 years and that a tendency toward opposite trends shows up in some regions during the last 35–40 yr (Karl and Knight 1998; Groisman et al. 2001, 2005). A decrease in the number of rainy days itself does not represent an obvious hazard for society and ecosystems and cannot be considered as an extreme event per se. However, an increase in the frequency (and length) of prolonged periods without sizable rainfall during the warm season when water is intensively used for transpiration may represent a hazard for both terrestrial ecosystem health and agriculture. Therefore, we select the warm season only (here we define it as a period when mean daily temperature remains persistently above the 5°C threshold) and count strings of dry days defined as the days when daily precipitation is absent or is below 1 mm. We consider the string as “broken” when its growth is stopped on the day when the station daily rainfall is 1 mm or above. The string growth is suspended (but the tally saved) at a given day when the daily temperature occasionally falls below 5°C. This dry day is not counted, but the string can continue growing when temperature rises. In other words, a brief late spring or early fall cold spell will not break the “season.” As a result of this procedure, for each year at each station we calculate (i) the number of days in the “warm” season, (ii) the number of strings with dry days, and (iii) the string lengths (in days). Our focus is not on the short strings but only those that are longer than given thresholds, X. The number of days in these long strings is counted and a percentage of the warm season with thus defined prolonged dry periods with duration of X days or longer is calculated. The number of days above a chosen threshold (X) of dry day strings is tallied. The count is then divided by the season day length, which becomes a percent of the warm season length with prolonged dry day strings. This approach gives us a chance to express in one number the severity of dry periods above the X day threshold (later described by the notation DryX+, e.g., Dry20+, Dry30+, Dry60+, etc.) during a specific warm season at a given station and allows for area averaging of these quantities, first over 1° grid cells and second over the large areas of the conterminous United States and, finally, to consider their trends for the past several decades.
d. Regional averaging routine used
We are focusing on the last few decades and are hoping to reveal late changes in the dry conditions on the background of large weather variability and geographic “noise.” Therefore, we are using area averaging so as to suppress this noise (Groisman et al. 2005). Meteorological stations are not uniformly distributed and missing years are present in most of the records. Both factors had to be addressed to properly represent regional averages of (i) the frequency and/or duration of prolonged dry episodes, (ii) duration of the warm season (DWS), and (iii) frequency of very heavy (very light) precipitation derived from in situ observations. Area-averaged calculations presented in this paper all use the same method. First, we selected a reference period with the greatest availability of data to estimate the long-term mean values for each quantity studied. The period selected was 1961–90, but the use of other reference periods (1951–2005 and 1967–2006) was also tested to ensure that this selection does not affect our results. For each station, for each quantity (very heavy precipitation, warm season duration, duration of the dry episodes above X thresholds, and for the dry episode percentage within the warm season), we determined their climatological mean values during the reference period. For each quantity, region, and year we calculated the anomalies from the long-term mean value of these quantities for each station and then arithmetically averaged these anomalies within 1° × 1° grid cells. These anomalies were regionally averaged with the weights proportional to their area. Data from the large regions use the regional area weights to form a national average when those analyses are presented (e.g., for the warm season duration and days with very heavy and very light precipitation). The long-term mean values (normals from reference period years 1961–90) were area averaged in a similar fashion and used to restore actual regional quantity values from regional anomalies. This approach emphasizes underrepresented parts of the region/country because a region, even with a relatively low percentage of grid cells with data, will receive the full weight comparable to the region’s area relative to other regions. It also allows preservation of the regional time series unaffected by the changing availability of data with time.2
e. Interpretation of linear trend estimates
We used (in addition to the linear trend assessment) a nonparametric test to check for a monotonic change of the time series. Once a statistically significant trend has been discovered, we characterize it by the mean rate of change. A linear trend estimate is an essential characteristic in this case. We tested the presence of systematic change in the time series using two standard methods: least squares regression (Draper and Smith 1966; Polyak 1996) and a nonparametric method based on the Spearman rank order correlation (Kendall and Stuart 1967). We used two-tailed tests at the 0.05 or higher significance level (except in Table 2 where a one-tailed test was also used). We tested for autocorrelation of the detrended time series, but the residuals of the regional warm season duration, frequencies of prolonged dry episodes, and very heavy precipitation were never found to be significantly autocorrelated. All area averaging and trend estimation procedures are linear and allow transposition: We can calculate linear trends at each station and then area average them or we can construct area-averaged time series of the regional percentage of the days with prolonged dry day strings and then calculate the trends. We selected the latter way to present our results in this paper. Peculiarities of processing the missing observations, the effects of selection of the X threshold, the 1-mm threshold for inclusion of a day with “sizeable” precipitation, and the beginning year for trend analyses are discussed in the appendix.
3. Climatology and regional partition
a. Long-term mean values of the duration of the warm season and percentage of prolonged no-rain episodes
Figure 2 shows the long-term mean percentage of days with strings of different length as calculated for the 1951–2005 period. The station mean values were arithmetically area averaged within 1° × 1° grid cells. If no one station within the grid cell has a single string with dry days during the entire 1951–2005 period, the grid cell was left blank. For example, the blue dots in Fig. 2a (<0.8%) indicate that during the past 55 yr in these 1° × 1° grid cells there were from one to four 30-day-long or longer, no rain strings observed. Figure 2d shows the climatology of the warm season duration (with mean daily temperatures above 5°C) for the same period. These periods vary from 103 days (in the North and in the mountainous West) to 365 days in southern Florida and California. The duration of this period has systematically changed during the past 100 yr with nationwide and global warming (cf. Easterling 2002; Shein 2006). While not a subject of this study, we, nevertheless, estimated trends in this quantity for the past century (no significant trends were found) and during the past 40 yr.
The Upper Great Lakes region (Minnesota, Wisconsin, and Michigan) is a very humid region with frequent rainfall in the warm season. We observe there a very small percentage of month-long dry strings (only 0.5%). Therefore, we separated this region as a special entity and generally excluded it from further analyses of dry episodes. Thereafter, we partitioned the rest of the contiguous United States into several regions as shown in Fig. 3. In the eastern United States (the states east of 95°W without Minnesota, Wisconsin, and Michigan), we looked for month-long dry strings. West of 95°W we looked for 60-day-long dry strings (in the southwestern United States that includes Texas, Oklahoma, New Mexico, Arizona, California, and Nevada) and for both 30- and 60-day dry strings in the northwestern quadrant of the country that includes the Great Plains, Central Rockies, and northwestern United States. We divided the western United States into southern and northern parts using a well-known anticorrelation between rainfall in the northwest and southwest of the country (e.g., Groisman and Easterling 1994) due to variations in annual rainfall anomalies associated with ENSO-imposed latitudinal variations of the winter storm tracks (Ropelewski and Halpert 1996).
Figures 2c and 2d show that the California climate stands apart from the national climate: strings of dry days 60 days and longer constitute more than 15% (actually ∼40% on average) of days during the warm season. However, consideration should be made that in the valleys where most of the COOP stations are located the warm season can last the entire year. The mean statewide percentage of Dry90+ strings in this state is 28% and in some years nearly half of the warm season lies within these 3-month-long strings. Considering percentages of these frequently occurring events goes against the idea of looking at them for extremes. Therefore, in addition to assessing the frequency of Dry60+ strings over the entire southwestern United States (with the mean occurrence of dry strings of 15% during the past 55 yr), we analyzed the abridged southwestern region (without California and Nevada). Here, the mean occurrence of such strings is more rare (7%) and its changes, therefore, qualify for consideration as changes in extreme events.
b. The driest warm seasons during the past 99 years
The seven driest warm seasons during the past 99 years over the eastern United States (since 1908) were 1963, 1924, 1953, 2000, 1908, 1939, and 2001 with more than twofold exceedance of the average regional percentage of Dry30+ episodes (Table 1). While the no-rain episode of 1963 was the most extended during the past 99 yr, only once during this period (in 2000 and 2001) did we observe a sequence of two extraordinary dry warm seasons over the eastern United States. In the southwestern United States, the driest in the century was the warm season of 2002, which also was observed to a slightly lesser degree in the region extending from California through Texas. Year 1969 was characterized as a year with the most expansive no-rain episode in the northwestern part of the United States, including the Great Plains. It is interesting that the instrumental summer Palmer drought severity index (PDSI) (available online at http://www.ncdc.noaa.gov/paleo/pdsiyear.html) does not report any unusual dry conditions in this region. Generally, the durations of prolonged dry episodes (or their percentage in the warm season) represent new indices of summer dryness and are related to agricultural droughts, and, therefore, it was instructive to compare these episodes with the most popular indices that characterize droughts. PDSIs (Willeke et al. 1994) accumulated in archive (NCDC 2007) were compared (to the extent possible) to the extreme characteristics of Dry30+ and Dry60+ indices (including those shown in Table 1). We found that, while in most cases there is a reasonable correspondence, however, these indices are not congruent and may at times be quite different (e.g., in 1969 in the northwestern United States). For example, this could happen when unusually heavy precipitation is followed by a prolonged no-rain period and thereafter by another heavy rain event. This kind of situation would not trigger a drought condition in some of the PDSI calculations. These situations may be of interest in a follow-up study.
4. Results
a. The eastern United States
Changes in the percentage of the warm season occupied by prolonged (–one month or longer) dry periods over the eastern United States for the past 40 yr are shown in Fig. 4. One-month-long dry intervals are infrequent in the region and on average constitute only 1.5% of the warm season duration (over Florida and Louisiana about 3%). During the past 100 years this duration has not been substantially changed and in the first half of the twentieth century it was even somewhat higher than in the second half (although the differences were insignificant). However, the situation changed during the past 40 yr. Figure 4 shows a systematic near-twofold (by 1.0% per 40 yr) increase in the duration of prolonged (one month and more) no-rain episodes. To more clearly describe this change, let us assume that these no-rain episodes occur in the area with the 200-day-long warm period (e.g., Washington, D.C., area) and that the duration of dry episodes is exactly 30 days. Then a change from 1% to 2% of the duration of the no-rain 30-day-long or longer episodes means that instead of a return period of 15 to 20 yr of such episodes (2–3 of them per 40 yr) we are now facing a return period of 6 to 8 yr for such episodes (5–6 of them per 40 yr). Our analyses show that during the past 99 yr, we have not encountered systematic changes in the dry days frequency over the entire eastern United States, but the strings of dry days (in particular month-long strings of dry days) became more frequent during the past 40 yr. Furthermore, for this 40-yr period, the selections of the beginning of the trend as well as the string length itself are not of crucial importance for this conclusion (Table 2).
b. The southwestern United States
Changes in the duration (in percent of the warm season) of prolonged dry periods 60 or more days without rain during the past 40 yr over the southwestern United States are shown in Fig. 5. The entire region was additionally partitioned into two parts (i) California and Nevada and (ii) Arizona, New Mexico, Oklahoma, and Texas because of very different frequencies of prolonged dry periods (Fig. 2). Taking into account the scale of frequency of occurrence of dry episodes of this kind, a regionwide increasing trend of 3.6% per 40 yr in duration of 2-month or longer periods without rain is observed in both parts of the southwestern United States. This change constitutes different relative changes in California (where on average 40% of the warm season, or approximately 100 days statewide, belong to dry episodes 60 days or more) and Oklahoma (∼1% of the warm season belonging to such episodes, which occurred on average once in 25 yr during the twentieth century). Significance of the trends for the past 40 yr for the southwestern United States is quite robust regarding the change of the beginning year for the trend evaluation as well as regarding the X threshold selection (Table 2). However, if we were looking for the 3-month-long dry episodes in the southwestern United States, the formal results would be similar, but not representative for Oklahoma and most of Texas and New Mexico, where 3-month-long dry periods have been extremely rare or nonexistent during the past century.
c. The northwestern United States
For the northwestern United States, for any duration of the dry episodes considered, we did not find statistically significant trends at the 0.05 level (using two- or one-tailed t tests) in the percentage of the warm season consumed by these episodes.
d. Duration of the warm season and dry episodes
The century-long warming over the contiguous United States that became most pronounced during the post–World War II period (Groisman et al. 2005) caused a general increase in the duration of the warm season (Fig. 6, Tables 3 and 4). The increase was especially strong over the southwestern part of the country and is supported here by earlier snowmelt (cf. Groisman et al. 2001; McCabe and Clark 2005), increase in the frost-free period length (Easterling 2002; Kunkel et al. 2004), and earlier dates of flower blooming (Cayan et al. 2001). However, only over the southwestern United States (and nationwide) were these increases statistically significant.3 If, instead of considering the percentage of the warm season consumed by dry episodes, we take into account their actual duration, the result will not be different. In Table 4, for 1- and 2-month durations of dry episodes during the warm season, we present statistics for time series of the product of mean regional percentage of the dry season episodes, mean regional duration of the warm season, and (in parentheses) direct estimates of regionally averaged absolute durations of such episodes.4
5. Discussion
In the above presentation, we focus on the statistical significance of the results. But, how practically significant are they? Climatology (Table 4) indicates that Dry30+ episodes have been quite infrequent in the eastern United States (with a return period of approximately 15 yr forty years ago). But analysis of trend results in Table 4 shows that “now” the return period of such an event is reduced to 6–7 yr. At the same time, Dry30+ episodes in the southwestern United States and especially in its westernmost parts, such as California, occur each year several times. Therefore, a 20-day increase in duration of these events in California and Nevada means a 15% increase of the duration of the events that occupy more than half (55%) of the warm period in this part of the nation. For the southwestern United States, Dry60+ episodes are also not rare. These episodes materialize every second year regionwide and each year in California and Nevada. The observed 11- to 12-day increase in duration of these episodes for the Southwest consumes the entire increase in the warm season duration and causes their return period to change from 2 yr (40 yr ago) to 1.25 yr “now” (which means every 4 of 5 years).
Contemporary GCM projections of the climate change in various scenarios of atmospheric composition changes over the conterminous United States (e.g. McAvaney et al. 2001; Hegerl et al. 2004; Kharin et al. 2007) show (i) a gradual increase in surface air temperature with the rate close to that of the Northern Hemisphere, (ii) no significant total precipitation changes during the next several decades, and (iii) a significant increase in intense (extreme) precipitation. All of this together could lead to higher stress on water supply/demand during the warm season between the extreme rain events. For the northeastern quadrant of the United States, there are model indications (e.g., Semenov and Bengtsson 2002; Groisman et al. 2005) that in the last decades of the twentieth century this scenario has already materialized (both in the model run and in observations). Our analysis shows that during the past several decades a similar scenario was observed over a significant part of the country. It would be of interest (although beyond the scope of this paper) to assess the existing composite of prognostic GCM runs for their ability to reproduce this feature of the warm season rainfall distribution dynamics over the nation.
6. Summary
Analyzing the duration of prolonged periods without sizeable rainfall during the vegetation period (roughly approximated by the threshold +5°C) over the conterminous United States, we found the following:
During the past four decades the warm season duration has significantly increased nationwide and over the southwestern United States (by 3%–4%, respectively). The largest increase in the warm season duration was observed over California and Nevada where it comprised 15 days (or a 6.4% increase during the past 40 yr).
During the past four decades the duration of prolonged dry episodes has significantly increased over the eastern and southwestern United States in absolute numbers (counting the number of days within such dry periods) and as a percentage of the warm “vegetation” period. The changes are interesting because they are observed on the background of the relatively “wet” period around the nation but do not cover the entire country. They are consistent with the notable change in rainfall rate distribution over the country (increase in intense rainfall frequencies while mean precipitation grows slower) and with modern GCM projections (and nowcast, cf. Groisman et al. 2005), of a warmer climate caused by the increase in greenhouse gases concentration in the atmosphere.
We presented results based on observations during the past several decades only. When we looked for century-long tendencies in the quantities described in this study, we did not see statistically significant systematic changes toward increase/decrease of the prolonged dry episodes over both the eastern and southwestern United States, and the first 60 yr of the twentieth century were even somewhat drier than the last 40 yr (Andreadis et al. 2005). The situation is somewhat similar to discovered trends in intense rainfall frequency (Soil and Water Conservation Society 2003; Groisman et al. 2005). We first constructed the century-long time series that led us to discover that the main (and the only) signal containing systematic changes in the frequency of prolonged dry episodes is provided by the last decades of observations (cf. section f of the appendix).
This study provides observational evidence only: it shows what has happened during the past decades. Extrapolation of trends based on empirical estimates is not a legitimate procedure unless theoretical considerations support it. In the discussion, we cite some examples of expected developments actually observed during the past few decades. Speaking about the future, we can state only that the tendencies reported in this study do not contradict and (to some extent) support the contemporary GCM projections of ongoing climatic changes.
Acknowledgments
NOAA/Climate and Global Change Program (Climate Change and Detection Element) provided support for this study. The thoughtful recommendations of three anonymous reviewers helped us to significantly improve the manuscript.
REFERENCES
Andreadis, K. M., E. A. Clark, A. W. Wood, A. F. Hamlet, and D. P. Lettenmaier, 2005: Twentieth-century drought in the conterminous United States. J. Hydrometeor., 6 , 985–1001.
Cayan, D. R., S. A. Kammerdiener, M. D. Dettinger, J. M. Caprio, and D. H. Peterson, 2001: Changes in the onset of spring in the western United States. Bull. Amer. Meteor. Soc., 82 , 399–415.
Coles, S. G., 2001: An Introduction to Statistical Modeling of Extreme Values. Series in Statistics, Vol. XIV, Springer, 208 pp.
Dai, A., K. E. Trenberth, and T. Qian, 2004: A global data set of Palmer Drought Severity Index for 1870–2002: Relationship with soil moisture and effects of surface warming. J. Hydrometeor., 5 , 1117–1130.
Draper, N. R., and H. Smith, 1966: Applied Regression Analysis. Wiley, 407 pp.
Easterling, D. R., 2002: Recent changes in frost days and the frost-free season in the United States. Bull. Amer. Meteor. Soc., 83 , 1327–1332.
Easterling, D. R., J. L. Evans, P. Ya Groisman, T. R. Karl, K. E. Kunkel, and P. Ambenje, 2000: Observed variability and trends in extreme climate events: A brief review. Bull. Amer. Meteor. Soc., 81 , 417–425.
Folland, C. K., and T. R. Karl, 2001: Observed climate variability and change. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 99–181.
Frich, P., L. V. Alexander, P. Della-Marta, B. Gleason, M. Haylock, A. M. G. Klein Tank, and T. Peterson, 2002: Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Res., 19 , 193–212.
Groisman, P. Ya, and D. R. Easterling, 1994: Variability and trends of total precipitation and snowfall over the United States and Canada. J. Climate, 7 , 184–205.
Groisman, P. Ya, and D. R. Legates, 1995: Documenting and detecting long-term precipitation trends: Where we are and what should be done. Climatic Change, 31 , 601–622.
Groisman, P. Ya, and Coauthors, 1999: Changes in the probability of heavy precipitation: Important indicators of climatic change. Climatic Change, 42 , 243–283.
Groisman, P. Ya, R. W. Knight, and T. R. Karl, 2001: Heavy precipitation and high streamflow in the contiguous United States: Trends in the twentieth century. Bull. Amer. Meteor. Soc., 82 , 219–246.
Groisman, P. Ya, R. W. Knight, T. R. Karl, D. R. Easterling, B. Sun, and J. M. Lawrimore, 2004: Contemporary changes of the hydrological cycle over the contiguous United States: Trends derived from in situ observations. J. Hydrometeor., 5 , 64–85.
Groisman, P. Ya, R. W. Knight, D. R. Easterling, T. R. Karl, G. C. Hegerl, and V. N. Razuvaev, 2005: Trends in intense precipitation in the climate record. J. Climate, 18 , 1326–1350.
Hegerl, G. C., F. W. Zwiers, P. A. Stott, and V. V. Kharin, 2004: Detectability of anthropogenic changes in annual temperature and precipitation extremes. J. Climate, 17 , 3683–3700.
Herweijer, C., R. Seager, E. R. Cook, and J. Emile-Geay, 2007: North American droughts of the last millennium from a gridded network of tree-ring data. J. Climate, 20 , 1353–1376.
Kagan, R. L., 1997: Averaging of Meteorological Fields. Kluwer Academic, 279 pp.
Karl, T. R., and R. W. Knight, 1998: Secular trends of precipitation amount, frequency, and intensity in the United States. Bull. Amer. Meteor. Soc., 79 , 231–241.
Keetch, J. J., and G. M. Byram, 1968: A drought index for forest fire control. USDA Forest Service Research Paper SE-38, 35 pp. [Available online at http://www.srs.fs.fed.us/pubs/.].
Kendall, M. G., and A. Stuart, 1967: Inference and Relationship. Vol. 2, The Advance Theory of Statistics, Ch. Griffin and Co., 690 pp.
Kharin, V. V., F. W. Zwiers, X. Zhang, and G. C. Hegerl, 2007: Changes in temperature and precipitation extremes in the IPCC ensemble of global coupled model simulations. J. Climate, 20 , 1419–1444.
Kunkel, K. E., 2003: North American trends in extreme precipitation. Nat. Hazards, 29 , 291–305.
Kunkel, K. E., K. Andsager, and D. R. Easterling, 1999: Long-term trends in extreme precipitation events over the conterminous United States and Canada. J. Climate, 12 , 2515–2527.
Kunkel, K. E., D. R. Easterling, K. Hubbard, and K. Redmond, 2004: Temporal variations in frost-free season in the United States: 1895–2000. Geophys. Res. Lett., 31 .L03201, doi:10.1029/2003GL018624.
Manabe, S., R. T. Wetherald, and R. J. Stouffer, 1981: Summer dryness due to an increase of atmospheric CO2 concentration. Climatic Change, 3 , 347–386.
Manabe, S., R. T. Wetherald, P. C. D. Milly, T. L. Delworth, and R. J. Stouffer, 2004: Century-scale change in water availability: CO2-quadrupling experiment. Climatic Change, 64 , 59–76.
McAvaney, B. J., and Coauthors, 2001: Model evaluation. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 471–523.
McCabe, G. J., and M. P. Clark, 2005: Trends and variability in snowmelt runoff in the western United States. J. Hydrometeor., 6 , 476–482.
NCDC, cited. 2007: Time bias corrected divisional temperature–precipitation–drought index. Documentation for dataset TD-9640. [Available online at http://www1.ncdc.noaa.gov/pub/data/cirs/drought.README.].
Polyak, I. I., 1996: Computational Statistics in Climatology. Oxford University Press, 358 pp.
Ropelewski, C. F., and M. S. Halpert, 1996: Quantifying Southern Oscillation–precipitation relationships. J. Climate, 9 , 1043–1059.
Semenov, V. A., and L. Bengtsson, 2002: Secular trends in daily precipitation characteristics: Greenhouse gas simulation with a coupled AOGCM. Climate Dyn., 19 , 123–140.
Shein, K. A., 2006: State of the Climate in 2005. Bull. Amer. Meteor. Soc., 87 , 6. S1–S102.
Soil and Water Conservation Society, 2003: Conservation implications of climate change: Soil erosion and runoff from cropland. SWCS Rep., 24 pp.
Stone, D. A., A. J. Weaver, and F. W. Zwiers, 2000: Trends in Canadian precipitation intensity. Atmos.–Ocean, 38 , 321–347.
Sun, B., and P. Ya Groisman, 2004: Variations in low cloud cover over the United States during the second half of the twentieth century. J. Climate, 17 , 1883–1888.
Vincent, L. A., and E. Mekis, 2006: Changes in daily and extreme temperature and precipitation indices for Canada over the twentieth century. Atmos.–Ocean, 44 , 177–193.
Willeke, G., J. R. M. Hosking, J. R. Wallis, and N. B. Guttman, 1994: The National Drought Atlas. Institute for Water Resources Rep. 94-NDS-4, 587 pp.
APPENDIX
Description of Technical Details and Supplementary Experiments
This appendix contains descriptions of technical details of the study as well as some justifications for choices that we made in the design of the processing system. Testing different ways to handle missing data and our decision to use a minimum precipitation threshold are examples of the choices tested. We solidified the design when it was shown that precautions and additional testing and experiments were generally in vain and did not change the results presented in the paper.
Handling of missing days
A missing day (days) may create havoc in statistics of uninterrupted strings of dry days. To mitigate this problem, we selected only 4165 stations from the entire set of more than 8000 currently operational U.S. cooperative stations network according to the following criteria: both station temperature and precipitation must have a minimum of 83% (25 yr) of nonmissing data during the reference period 1961–90. For precipitation we used a very strict missing day tolerance: if the total annual missing day count was greater than five, the year was not used. Thereafter, we eased this requirement and repeated our calculations keeping the station data for each year if the total annual missing day count was less than 20. However, in these relaxed conditions of a notable amount of missing data we allow the count processing of no-rain day duration to be suspended temporarily if one missing day is encountered and thereafter continue the tally until the next rain day. This modification allowed us to (i) substantially increase the number of years with valid data at the stations used in our analyses, thus reducing the noise level of the area-averaged time series, and (ii) receive practically identical results when we used strict and relaxed missing day tolerance criteria. By definition, accumulated data will always be preceded by one or more missing days in the data file. This could cause a detrimental effect by eliminating many years. However, in a separate study using a similar dataset we found that out of the 153 million observations checked, only 0.29% were accumulations.
Summary of mean duration of dry day episodes versus a percentage of dry day episodes in the warm season
To characterize regional dry day episodes extent, we area averaged the percentages of these episode durations within the individual warm seasons (Dry30+ and Dry60+) and thereafter multiplied them by the regionally averaged duration of the warm season. But, it was possible to area-average absolute values of day counts in these episodes directly. We used both approaches and compared them (Fig. A1). The resulting time series correlated extremely well with R2 = 0.98 for the eastern United States and 0.95 for the southwestern United States. In both regions, we found only some systematic biases that arise due to different weights of the area averaging (see footnote 4). Except for this difference, the results after area averaging brought us to the same conclusions.
Selection of the “sizeable rain” threshold
One of the physical reasons to ignore small daily rainfall totals in assessment of the dry episode length is that a small amount of rain either does not reach the soil (stays on and evaporates from the vegetation) or does not replenish soil moisture. In the calculation of the most popular index of forest fire danger, the Keetch–Byram drought index (KBDI), the first 5 mm (0.2 in.) of each rain event are ignored and do not affect the KBDI values (Keetch and Byram 1968). However, there also exists an “observational” reason to skip the daily “drizzle totals”: throughout the history of many national networks these totals are not observed consistently. This was observed first at the Norwegian network, when letters of appreciation sent by the National Meteorological Service to observers caused a doubling of the number of 0.1-mm precipitation reports nationwide (Groisman et al. 1999). In the countries where accuracy of observations was changed throughout time (e.g., Russia and Canada), the lowest “nonzero” precipitation amounts might be assigned to “traces” or to measurable amounts depending upon changes of the gauge resolution and/or observational practice (Stone et al. 2000; Groisman et al. 1999). Ignoring this matter may lead to “discoveries” of nationwide “statistically significant” trends in each part of the country in frequency of precipitation that are not real but are a result of switching—let’s say from British to metric rainfall reporting (e.g., Frich et al. 2002; Vincent and Mekis 2006). During the past 100 years, the U.S. cooperative network has measured precipitation with the same gauges, 8-in. nonrecording rain gauges with an accuracy of (and increments in) observations equal to 0.01 inch (0.254 mm). However, even this network was impacted throughout time because the percentage of ignored small precipitation events changed with time (Fig. A2). This figure presents the nationally averaged counts of precipitation events between 0 and 0.5 mm and between 0.5 and 1 mm during the past 99 yr (since the beginning year when we are comfortable presenting such averages, 1908). It shows that the reporting of the lowest nonzero precipitation (0.01 in. = 0.254 mm) has never been stable during the entire twentieth century (their number steadily grows from 6 to 13 yr−1). Furthermore, the reporting of the next nonzero increments (0.02 and 0.03 in.) stabilized only after 1948, when the first digital archives were introduced, but prior to these years they were also reported less frequently. During the 1961–90 period on average, there were 88 days with nonzero precipitation nationwide and ∼19 of them (22%) reported from 0.01 to 0.03 in. of precipitation. In the beginning of the twentieth century, these numbers were close to 11 days and 100 years later exceed the 20-days threshold, that is, nearly doubled. Therefore, when focusing on the changes in precipitation frequency, we (as in Groisman et al. 1999) selected 1 mm as a safe threshold to start our data analyses.
Experiments with different X thresholds and beginning year of the trend estimates
There were no magic numbers, for example, 30 and 60 days, when we selected X thresholds to present the results for strings of days without sizeable rain. Therefore, we varied X values in broad boundaries. Also, we selected 40 yr (instead of 37, which is the start year 1970) just for convenience of rounding. But, we also tested the robustness of our results against these arbitrary choices. Table 2 presents the summary of these tests and shows that an exact selection of X and/or the beginning year for the linear trend calculations is not so important for the statements about increases in duration of prolonged dry episodes during the past several decades over the eastern and southwestern United States to hold. There are quite broad windows of X thresholds and “beginning” years when the trends in the duration are positive and statistically significantly different from zero at the 0.05 level in two-tailed and/or one-tailed statistical testing. For the past 50 yr, statistically significant negative trends were never encountered in prolonged dry episodes over the conterminous United States. Table 2 also shows that the trends in counts of dry days have never been statistically significant at the 0.05 level during the past 50 yr for both regions, the eastern and southwestern United States, no matter which beginning year was selected. This remains the case for the northwestern United States too.
Distribution of mean duration of dry day episodes above the X threshold
If the X threshold is sufficiently high and the region where we are analyzing dry episodes is spatially homogeneous, the duration of dry day episodes above the X threshold should be distributed according to “generalized Pareto” distribution (Coles 2001). However, the regions that we are assessing cannot be considered as “homogeneous” (cf. Oklahoma and California or Maine and Florida) and it is difficult to select a “sufficiently high” X threshold in the southwestern United States. Therefore, the distribution of duration of extended dry day episodes above the X threshold can (and do) behave differently. Figure A3 shows these distributions for the eastern United States (30-day or longer dry episodes) and for the southwestern United States (60-day or longer episodes). This figure shows that, while it is extremely rare to encounter a 2-month-long no-rain episode at the stations of the eastern United States (only 5 cases during the past 40 yr in the sample of ∼1400 stations), at the 750 stations of the southwestern United States, 4-month-long, no-rain episodes number in the hundreds.
Assessment of century-long time series of prolonged dry episodes
The dense network of meteorological stations across the conterminous United States allowed us to construct a near-century-long time series of Dry30+ and Dry60+ episodes across the country for the 1908–2006 period (not shown). Analyses of these time series clearly show that
on a century time scale there were no systematic changes in the duration of these episodes,
there were years (cf. Table 1) when the prolonged no-rain intervals were more widespread across the eastern United States prior to 1967 than in the 2000s, and
in the southwestern and northwestern United States on average in the pre-1967 period Dry60+ episodes were slightly (i.e., statistically insignificant at the 0.05 level) more widespread than in the past 40 yr.
But, the trend assessment of these century-long time series shows that (i) whichever 40-yr-long period during the period of instrumental observations since 1908 and (ii) whichever interval YB-2006 (where YB is a beginning year) is selected, only in the last periods of 40+ years are there statistically significant systematic changes within two of these regions (cf. Table 2). Will these changes continue and, if so, what is their attribution? These are legitimate questions but they should be addressed to climate modelers. We simply show that they have happened and would strongly argue that the quantity that we have assessed is of real practical importance, its increase may result in hazardous hydrometeorological conditions and, therefore, tendencies revealed in the past several decades are worthy of interest. We see a systematic change that coincides in time with a change in frequency of intense precipitation and with an increase in mean precipitation. It is not contradictory with GCM projections of changes in rainfall frequencies [e.g., Semenov and Bengtsson (2002) for the northeastern quadrant of the United States] but, because of a simultaneous increase in precipitation [7% per 100 yr across the conterminous United States (Groisman et al. 2004)], the observed tendencies are not as prominent as with the discovered earlier increases in intense precipitation, which were enhanced by the growing mean precipitation (Groisman et al. 2005).
Regions of the contiguous United States (hatched) where statistically significant annual increases in very heavy precipitation for the 1908–2002 period were reported by Groisman et al. (2004) and very heavy precipitation (upper 0.3% of daily rain events with a return period of 4 yr) over these regions of the central United States and their linear trends. Linear trends for the 1893–2006 and 1967–2006 periods (solid lines) are equal to 22% per 114 yr and 27% per 40 yr, respectively, and are statistically significant at the 0.05 level or higher (updated from Groisman et al. 2005).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Regions of the contiguous United States (hatched) where statistically significant annual increases in very heavy precipitation for the 1908–2002 period were reported by Groisman et al. (2004) and very heavy precipitation (upper 0.3% of daily rain events with a return period of 4 yr) over these regions of the central United States and their linear trends. Linear trends for the 1893–2006 and 1967–2006 periods (solid lines) are equal to 22% per 114 yr and 27% per 40 yr, respectively, and are statistically significant at the 0.05 level or higher (updated from Groisman et al. 2005).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Regions of the contiguous United States (hatched) where statistically significant annual increases in very heavy precipitation for the 1908–2002 period were reported by Groisman et al. (2004) and very heavy precipitation (upper 0.3% of daily rain events with a return period of 4 yr) over these regions of the central United States and their linear trends. Linear trends for the 1893–2006 and 1967–2006 periods (solid lines) are equal to 22% per 114 yr and 27% per 40 yr, respectively, and are statistically significant at the 0.05 level or higher (updated from Groisman et al. 2005).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
(a)–(c) Climatology of the mean percentage of the warm season included in the string lengths of dry days of (a) ≥30 days, (b) ≥60 days, and (c) ≥90 days in the strings. When no such strings have been observed during the past 55 yr (1951–2005), the area is left blank. A few 1° × 1° grid cells in the western United States (e.g., in Nevada) are also left blank due to insufficient long-term station data coverage. (d) Climatology of the warm season duration (days with mean daily temperature above 5°C) over the conterminous United States.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
(a)–(c) Climatology of the mean percentage of the warm season included in the string lengths of dry days of (a) ≥30 days, (b) ≥60 days, and (c) ≥90 days in the strings. When no such strings have been observed during the past 55 yr (1951–2005), the area is left blank. A few 1° × 1° grid cells in the western United States (e.g., in Nevada) are also left blank due to insufficient long-term station data coverage. (d) Climatology of the warm season duration (days with mean daily temperature above 5°C) over the conterminous United States.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
(a)–(c) Climatology of the mean percentage of the warm season included in the string lengths of dry days of (a) ≥30 days, (b) ≥60 days, and (c) ≥90 days in the strings. When no such strings have been observed during the past 55 yr (1951–2005), the area is left blank. A few 1° × 1° grid cells in the western United States (e.g., in Nevada) are also left blank due to insufficient long-term station data coverage. (d) Climatology of the warm season duration (days with mean daily temperature above 5°C) over the conterminous United States.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Station map and the regional partition selected for this study.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Station map and the regional partition selected for this study.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Station map and the regional partition selected for this study.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 1-month or longer duration during the warm season during the past 40 yr area averaged over the eastern United States. Linear trend (dashed line; 1.1% per 40 yr) is statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 1-month or longer duration during the warm season during the past 40 yr area averaged over the eastern United States. Linear trend (dashed line; 1.1% per 40 yr) is statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 1-month or longer duration during the warm season during the past 40 yr area averaged over the eastern United States. Linear trend (dashed line; 1.1% per 40 yr) is statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 2-month or longer duration during the warm season during the past 40 yr area averaged over the southwestern quadrant of the United States (dots) and the two groups of states that constitute it, California and Nevada (triangles) and Texas, Oklahoma, New Mexico, and Arizona (squares). For the entire region, linear trend (dashed line; 4.3% per 40 yr) is statistically significant at the 0.05 level. For the region encompassing four states (Arizona, New Mexico, Oklahoma, and Texas) an increase (by 3.1% per 40 yr) is not statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 2-month or longer duration during the warm season during the past 40 yr area averaged over the southwestern quadrant of the United States (dots) and the two groups of states that constitute it, California and Nevada (triangles) and Texas, Oklahoma, New Mexico, and Arizona (squares). For the entire region, linear trend (dashed line; 4.3% per 40 yr) is statistically significant at the 0.05 level. For the region encompassing four states (Arizona, New Mexico, Oklahoma, and Texas) an increase (by 3.1% per 40 yr) is not statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Percentage of the dry day episodes with 2-month or longer duration during the warm season during the past 40 yr area averaged over the southwestern quadrant of the United States (dots) and the two groups of states that constitute it, California and Nevada (triangles) and Texas, Oklahoma, New Mexico, and Arizona (squares). For the entire region, linear trend (dashed line; 4.3% per 40 yr) is statistically significant at the 0.05 level. For the region encompassing four states (Arizona, New Mexico, Oklahoma, and Texas) an increase (by 3.1% per 40 yr) is not statistically significant at the 0.05 level.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Warm season duration over the conterminous United States (dots), southwestern United States (filled diamonds), and northwestern United States (squares). Linear trends (dashed lines) for the conterminous United States and its southwestern part are statistically significant at the 0.05 level or above.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Warm season duration over the conterminous United States (dots), southwestern United States (filled diamonds), and northwestern United States (squares). Linear trends (dashed lines) for the conterminous United States and its southwestern part are statistically significant at the 0.05 level or above.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Warm season duration over the conterminous United States (dots), southwestern United States (filled diamonds), and northwestern United States (squares). Linear trends (dashed lines) for the conterminous United States and its southwestern part are statistically significant at the 0.05 level or above.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A1. Southwestern United States. Dry60+ episode durations estimated using area-averaged absolute values of day counts in these episodes directly and using the product of area-averaged percentage of the dry days within these episodes and the warm season duration: R2 = 0.95. Systematic difference (bias) = 3.4 days.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A1. Southwestern United States. Dry60+ episode durations estimated using area-averaged absolute values of day counts in these episodes directly and using the product of area-averaged percentage of the dry days within these episodes and the warm season duration: R2 = 0.95. Systematic difference (bias) = 3.4 days.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A1. Southwestern United States. Dry60+ episode durations estimated using area-averaged absolute values of day counts in these episodes directly and using the product of area-averaged percentage of the dry days within these episodes and the warm season duration: R2 = 0.95. Systematic difference (bias) = 3.4 days.
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A2. Mean number of days with nonzero very light daily precipitation over the conterminous United States. Days with nonzero daily precipitation below 0.5 mm (dots) and days with daily precipitation in the range of 0.5 and 1 mm (filled diamonds).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A2. Mean number of days with nonzero very light daily precipitation over the conterminous United States. Days with nonzero daily precipitation below 0.5 mm (dots) and days with daily precipitation in the range of 0.5 and 1 mm (filled diamonds).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A2. Mean number of days with nonzero very light daily precipitation over the conterminous United States. Days with nonzero daily precipitation below 0.5 mm (dots) and days with daily precipitation in the range of 0.5 and 1 mm (filled diamonds).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A3. Distribution function of prolonged dry episode durations >30 (eastern United States) and >60 days (southwestern United States).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A3. Distribution function of prolonged dry episode durations >30 (eastern United States) and >60 days (southwestern United States).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Fig. A3. Distribution function of prolonged dry episode durations >30 (eastern United States) and >60 days (southwestern United States).
Citation: Journal of Climate 21, 9; 10.1175/2007JCLI2013.1
Seven driest warm seasons during the past 99 yr over the conterminous United States. Partition was defined using percentages of 1-month or longer dry episodes (Dry30+) for the eastern United States and 2-month or longer dry episodes (Dry60+) within the warm season for western regions of the country.
Indicators of statistically significant positive (1) and insignificant (0) linear trend estimates for the period beginning year–2006 at the 0.05 level using a one-tailed t test (in italic letters) and a two-tailed t test (in bold letters) for the frequency of dry days and dry-day strings with length above the X threshold for the eastern and southwestern United States. This was the only place throughout the paper when a one-tailed t test was also used.
Duration of the warm season defined as the period with mean daily temperatures above 5°C over the conterminous United States and its southwestern part. Mean values and trend estimates are presented for the 1967–2006 period. All linear trend estimates are statistically significant at the 0.05 or better levels; R2 is the percent of time series variance described by linear trend and d is days.
Summary of the linear trend assessment for the conterminous United States from 1967 to 2006: duration of the warm season (DWS) percentages of 1-month or longer dry episodes (Dry30+) and 2-month or longer dry episodes (Dry60+) within the warm season, and duration of the dry episodes estimated as product of DWS × Dry30+ (or Dry60+) and directly by area averaging of these durations at individual stations (estimates are given in parentheses). Trend estimates that are statistically significant at the 0.05 level or better are shown in bold; d is days.
Karl and Knight (1998) and subsequent studies (Easterling et al. 2000; Stone et al. 2000; Groisman et al. 2004, 2005) all show that most of the precipitation increase over the United States and Canada occurs due to an increase in the frequency of intense precipitation, while the frequency of days with average and light precipitation does not change or decreases.
We used this area-averaging routine during the past decade for various climate variables (e.g., Groisman and Legates 1995; Karl and Knight 1998; Groisman et al. 2001, 2004, 2005) after extensive testing regarding the robustness of the algorithm. The results of its implementation are close to those based on area-averaging procedures built on optimal interpolation and optimal averaging with normalizing weights (Kagan 1997). “Optimal” procedures (i.e., those that deliver the minimal standard error of area averaging) are much more computationally expensive and preserve their optimal properties only when specifics of the statistical structure of the meteorological field to be averaged are well known. This is not the case for many of the quantities we analyzed and, thus, we opted not to use them.
The Web site “Global Climate at a Glance” (http://www.ncdc.noaa.gov/gcag/gcag.html) indicates that over the entire conterminous United States we observe a systematic increase (e.g., April through October) in temperatures during the past 40 yr. However, if one is considering only statistically significant results (at 0.1 or 0.05 levels), these results will be restricted only to the southwestern corner of the United States.
Mean regional durations of prolonged dry periods estimated directly and as products of DWS × Dry30+ (or Dry60+) were not expected to be exactly the same because (i) durations of the extended dry periods and of the warm period can be correlated and (ii) the day counts having different weights in these two area-averaging procedures. By area averaging the percentages of the vegetation period with prolonged dry episodes, we assigned smaller weights to lowland and southernmost sites compared to the mountains and northernmost sites. When area averaging of actual duration of prolonged dry episodes is made, the sites with a longer vegetation period receive “an advantage” of both a longer vegetation period and a decline in orographic-induced rainfall. Both approaches are legitimate. In this paper, we have chosen the first approach trying to estimate more carefully the regional percentage of time when evapotranspiration might be suppressed by prolonged dry episodes. The regional duration of extended dry episodes estimated using the second approach is also presented in Table 4 in parentheses (see also section b of the appendix).
