1. Introduction
Water vapor plays the central role in the atmospheric branch of the global hydrologic cycle and is the most abundant greenhouse gas. Climate models used for estimating effects of increases in greenhouse gases show substantial increases in water vapor as the globe warms and this increased moisture would further increase the warming (e.g., Houghton et al. 1996). However, the distribution and variability of water vapor in the troposphere is not completely understood. Radiosonde observations of temperature and humidity are the primary source for investigating temporal variations in the tropospheric moisture record because of the long record length. The purpose of this work is to provide estimates of trends in tropospheric water vapor for the Northern Hemisphere; first, for the 1973–95 period and second, where possible, for the 1958–95 period.
Previously, Ross and Elliott (1996a, hereinafter REa) computed trends in several water vapor variables over North America from 1973 to 1993. Increases in surface to 500-mb precipitable water (W) were found over most of the domain except northern and eastern Canada where weak decreases were apparent. Some other radiosonde-based water vapor trend analyses include Gaffen et al. (1992) who found W increases for most of a 50-station global network for the period of 1973–90, Gutzler (1996) who reported increases in tropospheric specific humidity at four stations in the western tropical Pacific over 1973–93, and Zhai and Eskridge (1997) who found predominately increases in surface–200-mb W over China for 1970–90.
All the above trend studies considered the homogeneity of the time series in the selection of stations and the choice of data period. Homogeneity of a record can be affected by changes in instrumentation or observing practice. For example, since relative humidity typically decreases with height through the atmosphere, a fast-responding humidity sensor would report a lower relative humidity than one with a greater lag in response. Thus, the change to faster-response humidity sensors at many stations over the last 20 years could produce an apparent, though artificial, drying over time (e.g., Elliott 1995).
Before 1973, more stations were affected by observing system changes. Furthermore, some humidity sensors respond poorly in colder temperatures making data at higher levels more erratic (e.g., Elliott 1995). In fact, in the pre-1973 period some soundings did not have humidity data all the way to 500 mb, reducing the W record. To compensate partially for these shortcomings over the longer period, we used the 850-mb specific humidity as a surrogate for total column water vapor changes. The choice of surrogate was based on correlations between W and other humidity variables that are summarized in section 3. However, the 850-mb level is not immune to inhomogeneities. Our procedures for determining the longest homogeneous segment in the time series are described in section 4. In section 5 we present the estimates of Northern Hemisphere changes in tropospheric moisture for two periods within the last four decades. We estimate the impact of possible hidden inhomogeneities on the trends in section 6 and finish with a summary.
2. Data processing
We have used data from two sources: the Air Resources Laboratory (ARL) humidity data based on the National Center for Atmospheric Research (NCAR) upper-air archive and described in REa, and a 437-station subset of the Comprehensive Aerological Research Data Set (CARDS; Eskridge et al. 1995). The CARDS subset consists of the 238-station “core” subset (Wallis 1998) and 199 stations selected for completeness and homogeneity in the 1973–95 period. The NCAR data include soundings from all radiosonde data since 1973, and the CARDS subset has a limited number of stations but extends back as far as 1948. The soundings were treated to the same quality control procedures as used by REa and Ross and Elliott (1996b). These included gross error checking for surface pressure and dewpoint depression values as well as a temperature (T) check against climatological thresholds developed for each station at the surface and mandatory levels from 850 to 300 mb. Dewpoint temperature Td, specific humidity q, and precipitable water in layers were computed as described in REa. For each variable, data at the two main observing times (0000 and 1200 UTC) were averaged separately to form monthly means. At least 10 observations had to be available, otherwise that month’s mean for that observation time was considered missing. The stations whose trends are reported in section 5 had many more observations than this requirement; the average number of observations per month over the 1973–95 period was greater than 20 days month−1 at every station and usually exceeded 25 days month−1; over the 1958–73 period the average was greater than 18 days month−1 at each station and usually exceeded 25 days month−1.
We have considered two time periods: 1973–95 (23 yr) using the ARL humidity data and 1958–95 (38 yr) using the CARDS subset. We omitted pre-1958 data from consideration because of the change in time of synoptic observations (e.g., from 0300 to 0000 UTC). For each period, we required that a station’s time series contain no more than 36 missing nonconsecutive months. More than half of the stations in section 5 had fewer than 12 missing months. Furthermore, we eliminated stations with gaps greater than 24 months in the middle of the record. Although we processed the global set of stations, there were so few stations in the Southern Hemisphere of sufficient length or data quality that we limit this discussion to Northern Hemisphere stations.
3. Data homogeneity
In REa we relied on visual inspection along with station history information to assess data record homogeneity. Here, we added an objective statistical test to station history information and visual inspection that identifies abrupt changes in the time series median (hereinafter, change points) and is described by Lanzante (1996). It involves iterative applications of the nonparametric Wilcoxon–Mann–Whitney test for equality of medians. The statistical test was applied to monthly anomaly time series of W and 850-mb Td, and lists of change points that exceeded the 99% confidence level were generated at each station. These change points can disclose either artificial changes or true climate changes so additional information is required to distinguish between the two. For example, as discussed in REa, the climate shift near 1977 (Trenberth 1990) appears as change points in some North American humidity time series, showing that the atmosphere can change rather abruptly.
Station history information (Gaffen 1996) was used to compare change points with observing system changes to indicate where statistically detected change points might represent natural variability instead of artificial change. Because the accuracy of the calculated date of the statistical change point depends on the variability in the time series, we assumed the change was a nonclimate artifact if a historical change occurred less than three years either side of a statistical change. When a particular station’s history information was incomplete or unavailable, we assumed, conservatively, that all statistical change points represented artifacts. For those stations that we felt fairly confident of the completeness of histories, such as U.S. stations, U.S.-run Pacific island stations, or stations in China and the former Soviet Union, we ignored statistical change points that had no historical match (i.e., we assumed the change point resulted from natural variability).
The ignored change points at U.S.-run, Chinese, and Russian stations were largely clustered around 1977, 1981, 1984, and 1988. Change points near 1984 are neglected because there was no evidence of historical changes at that time and this statistical test will sometimes identify the midpoint of a time series containing a trend as a change point (J. R. Lanzante 1997, personal communication). Change points near 1977 and 1988 are ascribed to climate regime shifts (Trenberth and Hurrell 1994). Changes near 1981 are less easily explained. As discussed in REa, a modification to the U.S. sensor occurred in 1980, though its impact was thought to be a small increase in water vapor at most. It is possible that a portion of the increases at U.S. stations was due to this change, but because the change point date is somewhat uncertain, these change points could possibly be related to the large El Niño event in 1982–83 and the El Chichón eruption in 1982.
Figure 1 shows four examples of our assessment of homogeneity. Each example shows the time series of monthly anomalies of 850-mb Td, the statistically identified change points, and any known historical changes. At Sapporo, Japan, the change point in 1981 corresponds with a change in humidity sensor from hair sensor to carbon hygristor, thus we eliminated this station as being temporally inhomogeneous. At Yeniseysk, Russia, no statistical change points were identified near the time of instrumentation changes in 1986 and 1990. The only change point identified is in 1980 and because there was no corresponding historical change we assumed this change point represented natural variation and retained the time series for trend analysis. Similarly, at Hailer, China, there was no historical change near the statistical change point in 1988 and at St. Paul Island, Alaska, no change points are identified near the historical changes. The statistical change points in 1988 (China) and in 1977 (Alaska) are probably due to the climate regime shifts in the North Pacific at these times (Trenberth and Hurrell 1994), and both these time series were retained for trend analysis. Interestingly, no change point is visually apparent in either station time series at the other climate regime shift date (1977 in China or 1988 in Alaska).
These examples show that the combination of historical and statistical information can identify some known instrument changes. However, we caution that the separation of artificial (e.g., instrument changes) and natural variability is inevitably somewhat subjective. For instance, the same instrument change at one station may not show as large an effect at another location or time of day. Elliott and Gaffen (1991) showed an example of the effect of the change in 1965 on the humidity record at Hilo, Hawaii, where the redesign of the duct on the radiosonde allowed daylight to warm the humidity element and so led to an apparent drying in the daytime observations. The same design affected all U.S. radiosondes, but the impact was less apparent at stations where the observation times occurred when the sun was not so high (e.g., 0600 and 1800 LST at 90°W longitude). Variations in the effect of the same instrument change that are related to longitude, latitude, season, and local time of observation hinder simple attempts to estimate a change’s impact by ensemble averaging of the affected station time series.
Differences in ambient conditions can affect radiosondes differently. Schmidlin and Ivanov (1998) showed that RH reported from VIZ sondes was drier than that from Vaisala, Inc., RS80 sondes when RH was less than 50% but was more moist for RH above 50%. Also, VIZ sondes reported drier RH when T was less than 5°C but higher RH for warmer temperatures.
Furthermore, the ability of the statistical method to detect abrupt changes depends on the variability of the record, so that the same effect of an instrument change could be obscured in a very noisy record. In this case, the same change detected at one station may not be detected at another station containing more variability. However, it is possible to calculate an upper bound on the magnitude of an undetected change, including its contribution to the trend (see the appendix). For this reason, when no statistical change points were found at a station, it was included as homogeneous even if a known instrument change occurred at the station. The potential impact of unidentified change points to the trend results are presented in section 6.
4. Surrogates for total precipitable water
The introduction of improved instruments over the data period often affects the temporal homogeneity of the midtropospheric measurements (700–500 mb) more than those of the lower troposphere because of slower sensor response in cold temperatures (e.g., Elliott 1995). Also, early data at some stations did not reach the 500-mb level as consistently as more recent observations. Thus it is especially difficult to find stations with long high-quality records of W. To minimize these quality and homogeneity issues we surmised that temporal changes in lower-level moisture variables might be adequate surrogates for temporal changes in the total precipitable water.
To determine the correspondence between W and the single-level humidity variables, correlation coefficients were computed using the nonparametric Spearman rank correlation method (Wilks 1995) on seasonally averaged anomaly time series of W and T, Td, q, and RH at the surface, 850-, and 700-mb levels. The seasonal averages were based on monthly anomalies. Correlations were computed over 1973–95 using the ARL dataset for 192 Northern Hemisphere stations with homogeneous W records. Table 1 shows the median and the 0.10 quantile correlations by variable, level, and season. The 0.10 quantile indicates that at 90% of the stations the correlation equaled or exceeded that value. Median correlations with W among the stations were highest (≥0.92) for q at 850 mb, although the correlations for 850-mb Td were nearly as high, ranging from 0.85 to 0.92. A similar conclusion was reached by Sinha and Sinha (1981) for India and by Gaffen (1992) for a 63-station global network.
Also important for a good W surrogate is a relatively small spread of the correlations about the median. Seasonal 0.1 quantile values (i.e., 90% of the correlations are greater than this value) for 850-mb q exceed 0.8; those for 850-mb Td exceed 0.7. Surface variables do not correlate particularly well with W; note the lower medians, larger seasonal variability, and larger spread of the correlations. For example, the 0.1 quantile value for surface temperature Tafc ranges from 0.05 to 0.52 depending on season. The larger spread of the correlations at the surface is due to geographic variations in the relationship of W with surface variables as was found by Gaffen et al. (1992). Conversely, the consistency of the 850-mb q and Td correlations with W indicates there is little geographic variation in these relationships and, combined with the high correlations, suggests that these two variables are useful surrogates for W.
5. Trends
For the trend analysis, monthly anomalies were generated as deviations from the mean monthly values over the data period. The anomalies for seasonal trends were calculated as the average of the three-monthly anomalies for that season [December–January–February (DJF), MAM, JJA, and SON] with at least two-months data required for each seasonal anomaly. Similarly, anomalies for annual trends were calculated as the average of the 12-monthly anomalies (January–December) with at least 10 months required for each annual anomaly.
a. 1973–95
After homogeneity testing, we identified a set of 214 stations at 0000 UTC and 215 at 1200 UTC for 850-mb Td and q as well as 192- and 202-station subsets for W. The fewer stations for W is due to change points found in W but not in Td, probably because of inhomogeneities at upper levels but also possibly at the surface. Figure 2 depicts the trends in yearly averaged 0000 and 1200 UTC W over the period of 1973–95 (two more years than REa). Above North America, the trends are almost always positive, and most of them are significantly so. Some negative trends are found over northeast and northern Canada. This picture is nearly identical to that of REa.
Trends in annually averaged W over Eurasia and the western tropical Pacific Ocean, on the other hand, show statistically significant increases only in portions of China and the Pacific islands. The increases over China are consistent with those of Zhai and Eskridge (1997) for surface–200-mb W over the 1970–90 period. The northeastern and northwestern Chinese stations where the increases are generally statistically significant are areas that showed precipitation increases in Zhai and Eskridge (1997, their Fig. 13). A fairly coherent region of W decreases (though not significant) is found over Europe. The rest of Eurasia shows mainly positive but with a few negative trends.
The trends in yearly averaged 0000 UTC 850-mb q (Fig. 3) show a similar pattern to the W trends except there are somewhat fewer increases that are significant over the northeast United States and Alaska. The spatial correlation coefficient between the W and q trends is 0.93. The trend magnitudes (both are expressed as % of annual mean decade−1) are fairly similar, with about 60% of the stations showing less than 0.5% per decade difference between the W and q trends. Thus, for the 1973–95 period, the 850-mb q trends capture the general picture of the W trends.
Trends in 850-mb Td (Fig. 4, top) are similar to those of 850-mb q. A few stations in eastern Asia show trends of q and Td that differ in sign, but this occurs only where the trends are near zero. Over most of Eurasia and North America there is usually agreement in the sign of the Td and T trends and most stations show increases. The correspondence of changes in T and Td suggests a moisture response to warming. However, the weak Td decreases over Europe are in contrast to the general increases seen in these trends of annual 850-mb T (Fig. 4, bottom). The recent trend toward positive North Atlantic oscillation (NAO) index has been associated with warmer conditions over Europe along with less precipitation over southern Europe and more precipitation over northern Europe (e.g., Hurrell and van Loon 1997). Although the weak Td decreases over northern Europe are not consistent with this NAO precipitation response, the southern Europe pattern of warmer temperatures with lower dewpoints is suggestive of an NAO influence.
Table 2 categorizes the number of stations over Eurasia and the western Pacific and separately over North America at each pressure level according to the sign of the T and Td trends; both positive, both negative, or of different sign. Most of the trends are positive in both quantities, as was found by REa for North America. When compared with North American stations, there are more Eurasian stations with negative Td and positive T trends, and the number increases for pressure levels higher in the atmosphere. However, most of these trends are not statistically significant. The last two lines of the table further subdivide the stations with positive trends of both T and Td into those for which temperature trends are larger than dewpoint trends and the reverse. Over North America, the T trends are usually smaller in magnitude, as found by REa. Over Eurasia, that is not the case at the surface (more T trends exceed Td trends) and is less frequent above the surface than over North America.
Seasonal trends (not shown) in W over North America are consistent with REa (see their Fig. 11) with summer showing widespread increases similar to Fig. 2. Seasonal increases over China are largest in winter; the European area of decreases in Fig. 2 is apparent also in the summer and autumn trends. As with the annual trends, seasonal trends over Eurasia show few statistically significant trends.
To indicate the time series variability and to summarize the trends, we have averaged the W anomaly time series of the stations in nine regions listed in Table 3. The choice of regional boundaries shown in Fig. 2 was based on the areas where the annual trends were mostly of the same sign. Figure 5 shows the regionally averaged time series of annual anomalies of W. Regional trends are small or near zero except over the United States and tropical Pacific. Note the nonlinearity of the change over this period. In the western United States and tropical Pacific time series, the steplike climate increase near 1977 is evident. A smaller increase near 1987 is discernable for China. Also, note the strong drying in the Pacific time series in 1983 and 1992 consistent with warm phase ENSO events, perhaps enhanced by volcanic eruptions in 1982 and 1991.
Annual and seasonal trend values for each region are listed in Table 3. These regionally averaged time series and trend values reiterate the message of Fig. 2, namely, that strongest humidity increases occur over the United States, the tropical Pacific, and China. For seasons, the significant increase over China in winter agrees with Zhai and Eskridge (1997), who found winter to show the largest change. As in REa, the largest increases over North American regions are in summer.
b. 1958–95
From the 437 stations in the CARDS subset, much smaller sets of stations (68 at 0000 UTC and 63 at 1200 UTC) were judged to be sufficiently complete and homogeneous over the period of 1958–95. For these stations, annual trends in 850-mb q (as a percentage of the annual mean for the 1973–95 period) are shown in Fig. 6. The general pattern of increases and decreases is similar to that of Fig. 3, although the magnitudes are usually reduced. Note that trends over the United States at 0000 UTC are often larger than at 1200 UTC, especially over the western United States and Alaska, which may indicate an apparent moistening influence from the 1973 change in sensor housing.
The regional-average time series of annual anomalies of 850-mb q are shown in Fig. 7 for eight of the nine regions (the Pacific was omitted because there were only two stations). We show the observation time closest to local nighttime (0000 UTC for Eurasia and 1200 UTC for North America), which should be less affected by any hidden impact from instrument change. Although the records show considerable nonlinear variability, 850-mb q is generally increasing in China, Alaska, and the eastern United States over this period.
Table 4 is a comparison of the sign of T and Td trends for the 1958–95 period. As for the shorter period (Table 2), the trends agree in sign at most stations. However, where trends are positive, the Td trends over North America are always larger than those of T and, for Eurasian stations with positive trends, the Td trends are usually smaller.
Seasonal trends are similar to those of the 1973–95 period except that the decreases over Europe are strongest in winter and autumn rather than summer and autumn. Another difference is that increases over Alaska are larger and more significant than in REa. A further indication that the 0000 UTC trends were somewhat influenced by the 1973 apparent moistening is that the winter 0000 and 1200 UTC trends over Alaska agree fairly well while in summer the 0000 UTC trends are noticeably larger. Over Eurasia there is little difference between the 0000 and 1200 UTC trends. The effect of undetected changes in instruments on the trends is discussed in section 6.
6. Discussion
Given the uncertainties in determining the homogeneity of the record at any station, we have tried to quantify how the trend results of section 5 might be affected by undetected instrument changes. Because the statistical test for identifying a change point depends on the variability of the time series itself, stations with noisy records, either from natural variability or less consistent data quality, may contain a larger undetected instrument change than station records with less noise.
However, we can estimate the maximum value of an undetected instrument change in each time series. Change points are defined as shifts in the median that exceed a critical value, so the magnitude of an undetected change must be less than or equal to this critical value. This bound on the size of an undetected change allows us to estimate how much smaller or larger the trend could be (depending on whether an undetected change is in the same direction as the trend or in the opposite sense) and it depends on the magnitude of the change, its temporal position within the record, and the variability of the time series (see the appendix for details).
Bounds on the 0000 UTC 1973–95 W trends of Fig. 2 range from 0.5% to 1.5%, with a median value of 0.9%. Because the bound depends on the variability of the time series, smallest values are found in the Tropics and values generally increase with latitude. Because the significant trends in Fig. 2 generally have values greater than 3%, even assuming the hidden change is as large as the upper bound, the trends still show essentially the same pattern, and this result adds confidence that our results are not unduly affected by unknown changes. For the 1958–95 period, bounds on the 0000 UTC q trends increase with latitude from 0.25% to 0.9%, with a median of 0.5%. Again, reducing the trends by these bounds would still show a pattern consistent with the trends in Fig. 6.
Although we ignore statistical changes when there is reason to think the changes may have been natural, we have not attempted to estimate the contribution of these climate shifts to the trends presented in section 5. The determination of the magnitude of the change is a problem, as discussed by Gaffen et al. (2000) who demonstrated that estimates of the change magnitude for upper-air temperature records can be very sensitive to the estimation procedures used. This sensitivity to procedure means that trends computed from adjusted records carry an additional measure of uncertainty related to the adjustment. Thus, attempts to adjust records for known change points to get better trend estimates are somewhat problematic and are beyond the scope of this study.
7. Summary
We have examined the radiosonde records for the best-quality data to give some idea of how, if at all, the moisture content of the troposphere has changed globally. We found that a global estimate could not be made because reliable records from the Southern Hemisphere were too sparse; thus we confined our analysis to the Northern Hemisphere. Even there, the analysis was limited by continual changes in instrumentation, albeit improvements, so we were left with relatively few records of total precipitable water over the era of radiosonde observations that were usable. To compensate partially, we used the 850-mb observations, arguing that these should be less affected by the quality of the earlier instruments, and showed that 850-mb moisture was very highly correlated with total precipitable water when the observations were of higher quality. This approach excluded data from stations whose elevations are near or above the 850-mb level, but it is unlikely that such stations would have substantially different moisture trends.
We tried to be conservative in evaluating stations for homogeneity, but even so some judgments were necessarily subjective. The overall conclusion, that there has been an increase in Northern Hemisphere tropospheric moisture at most of these stations in the past three or four decades and especially since 1973, seems to be fairly sound, but quantifying the increase remains difficult. One reason for this confidence is that almost all changes in radiosondes lead to an apparent drying. Because most of the trends are increases (albeit often small over Eurasia) and because artifacts often result in water vapor decreases (a notable exception is the 1973 duct change at U.S. stations, which appears as a moistening with time), it is conceivable that a stronger signal of water vapor increases has been partially masked. With these caveats, the conclusions of this study are summarized as follows.
Northern Hemisphere water vapor has increased at most stations in areas where we have reasonable-quality radiosonde observations over the period of 1973–95. Over the period of 1958–95, for a smaller station set, there have been generally smaller increases, and most of the increase occurred since 1973.
Water vapor increases are larger, more uniform, and more significant over North America than over Eurasia. In contrast to the fairly uniform increases in North America, Eurasia shows more variation, with increases over China as well as small decreases over Europe. Much of the differences in trend magnitude and sign over the two regions may be attributable to abrupt changes in the late 1970s that affected North America more than Eurasia.
If these results are accepted for the regions that have been sampled, what can be said for the entire Northern Hemisphere? The unsampled regions, mainly the Atlantic, Northern Pacific, and parts of Africa, would have to have decreased in moisture for the Northern Hemisphere to show little change. However, further progress toward understanding tropospheric moisture trends prior to about 1980 (the start of the satellite era) will probably hinge on the development of more reliable methods of adjusting the old observations.
Acknowledgments
This research was partially supported by the NOAA Office of Global Programs. We thank John Augustine, Brian Eder, and the anonymous reviewers for helpful comments and suggestions.
REFERENCES
Cryer, J. D., 1986: Time Series Analysis. Duxbury Press, 286 pp.
Elliott, W. P., 1995: On detecting long-term changes in atmospheric moisture. Climatic Change,31, 349–367.
——, and D. J. Gaffen, 1991: On the utility of radiosonde humidity archives for climate studies. Bull. Amer. Meteor. Soc.,72, 1507–1520.
Eskridge, R. E., O. A. Alduchov, I. V. Chernykh, P. Zhai, A. C. Polansky, and S. R. Doty, 1995: A Comprehensive Aerological Reference Dataset (CARDS): Rough and systematic errors. Bull. Amer. Meteor. Soc.,76, 1759–1775.
Gaffen, D. J., 1992: Observed annual and interannual variations in tropospheric water vapor. NOAA Tech. Memo. ERL ARL-198, 162 pp.
——, 1996: A digitized metadata set of global upper-air station histories. NOAA Tech. Memo. ERL ARL-211, 38 pp.
——, W. P. Elliott, and A. Robock, 1992: Relationships between tropospheric water vapor and surface temperature as observed by radiosondes. Geophy. Res. Lett.,19, 1839–1842.
——, M. Sargent, R. Habermann, and J. Lanzante, 2000: Sensitivity of tropospheric and stratospheric temperature trends to radiosonde data quality. J. Climate,13, 1776–1796.
Gutzler, D., 1996: Low-frequency ocean–atmosphere variability across the tropical western Pacific. J. Atmos. Sci.,53, 2773–2785.
Hurrell, J. W., and H. van Loon, 1997: Decadal variations in climate associated with the North Atlantic oscillation. Climatic Change,36, 301–326.
Houghton, J. T., L. G. Filho, B. A. Callander, N. Harris, A. Kattenberg and K. Maskell, Eds., 1996: Climate Change 1995: The Science of Climate Change. Cambridge University Press, 572 pp.
Lanzante, J. R. 1996: Resistant, robust and non-parametric techniques for the analysis of climate data: Theory and examples, including applications to historical radiosonde station data. Int. J. Climatol.,16, 1197–1226.
Ross, R. J., and W. P. Elliott, 1996a: Tropospheric water vapor climatology and trends over North America: 1973–93. J. Climate,9, 3561–3574.
——, and ——, 1996b: Tropospheric precipitable water: A radiosonde-based climatology. NOAA Tech. Memo. ERL ARL-219, 132 pp.
Schmidlin, F. J., and A. Ivanov, 1998: Radiosonde relative humidity sensor performance: The WMO intercomparison—September, 1995. Preprints, 10th Symp. on Meteorological Observations and Instrumentation, Phoenix, AZ, Amer. Meteor. Soc., 68–71.
Sinha, S., and S. K. Sinha, 1981: Precipitable water and dew-point temperature relationship during the Indian summer monsoon. Pure Appl. Geophys.,119, 913–921.
Trenberth, K. E., 1990: Recent observed interdecadal climate changes in the Northern Hemisphere. Bull. Amer. Meteor. Soc.,71, 988–993.
——, and J. W. Hurrell, 1994: Decadal atmosphere–ocean variations in the Pacific. Climate Dyn.,9, 303–319.
Wallis, T. W., 1998: A subset of core stations from the Comprehensive Aerological Reference Dataset (CARDS). J. Climate,11, 272–282.
Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.
Zhai, P., and R. E. Eskridge, 1997: Atmospheric water vapor over China. J. Climate,10, 2643–2652.
APPENDIX
Estimating Change Point Effects on Trends
The largest effect on the trend occurs for a change point at the center of the record, so we use k = n/2. The statistical change point test (section 3) does not provide a critical value that is easily related to D, so it was estimated for each station using the t-test critical value, tcrit = D/(σ2/n)1/2, where a value of tcrit corresponding to the 5% level of a two-tailed test was used, and σ2 is the time series variance. The difference between the calculated trend (using all monthly anomalies;nominally n = 12 × 23 = 276 months) and bjump defines the trend bounds. [Although the trends in section 5 are based on the annually averaged time series (nominally n = 23 yr), there is generally little difference between those trends and the trends using the monthly anomalies.]
Four examples demonstrating the determination of homogeneous data records. Each panel shows monthly anomalies of 850-mb Td (°C) at Sapporo, Japan (top panel), Yeniseysk, Russia (second panel), Hailer, China (third panel), and St. Paul Island, Alaska (bottom panel). Arrows pointing upward show the location of statistical change points and arrows pointing downward indicate instrumentation changes based on the station history information
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Annually averaged trends in W (% decade−1) at (top) 0000 and (bottom) 1200 UTC for the period of 1973–95. Positive trends are indicated by triangles and negative trends by circles. Filled symbols indicate the trends were statistically significant at the 5% level according to the Spearman test. The small symbols indicate trend magnitudes of 0%–3% and the large symbols denote trend magnitudes greater than 3%. The numbered regions delineated in the top panel are used in Table 2 and Figs. 5 and 7
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Same as Fig. 2 but for annually averaged trends in 0000 UTC 850-mb q (% decade−1)
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Same as Fig. 2 but for annually averaged trends in (top) 0000 UTC 850-mb Td (°C decade−1) and (bottom) 850-mb T (°C decade−1). The small symbols indicate trend magnitudes of 0°–0.25°C decade−1 and the large symbols denote trend magnitudes greater than 0.25°C decade−1
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Time series of regionally and annually averaged anomalies of 0000 UTC W (mm) in nine regions. Region boundaries are shown in Fig. 2. Vertical lines show two standard errors of each region’s station anomalies about the regional mean. Linear regression lines are also shown for each region
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Same as Fig. 2 but for annually averaged trends in 850-mb q (% of 1973–95 annual mean decade−1) for the period of 1958–95: (top) 0000 and (bottom) 1200 UTC
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Same as Fig. 5 but for the 1958–95 period. The number of stations in each region is in the upper-right corner. The Pacific region is omitted because only two stations were included in the 1958–95 analysis. The observation time shown is that closest to local night: 0000 UTC for Eurasia and 1200 UTC for North America
Citation: Journal of Climate 14, 7; 10.1175/1520-0442(2001)014<1602:RBNHTW>2.0.CO;2
Median (left value) and 0.10 quantile (right value) correlation coefficients (×100) between seasonal values of 0000 UTC W and T, Td, q, and RH for 192 Northern Hemisphere stations over 1973–95 at the surface and 850 and 700 mb
Number of stations for which annual trends of T and Td at 0000 UTC are both positive, both negative, or of different sign for 1973–95 period. Upper section is for North American stations and lower section is for Eurasian and western Pacific stations. The last row in each section is the number of stations for which dT/dt > dTd /dt (dTd /dt > dT/dt) when both trends are positive
Regionally averaged annual and seasonal trends in 0000 UTC surface–500-mb precipitable water (mm decade−1) for 1973–95. Trends in bold indicate statistical significance at the 5% level
Number of stations for which annual trends of T and Td at 0000 UTC are both positive, both negative, or of different sign for 1958–96 period. The last row is the number of stations for which dT/dt > dTd /dt (dTd /dt > dT/dt) when both trends are positive
