Abstract

Global and large regional rainfall variations and possible long-term changes are examined using the 27-yr (1979–2005) Global Precipitation Climatology Project (GPCP) monthly dataset. Emphasis is placed on discriminating among variations due to ENSO, volcanic events, and possible long-term climate changes in the Tropics. Although the global linear change of precipitation in the dataset is near zero during the time period, an increase in tropical rainfall is noted in the dataset, with a weaker decrease over Northern Hemisphere middle latitudes. Focusing on the Tropics (25°S–25°N), the dataset indicates an upward linear change (0.06 mm day−1 decade−1) and a downward linear change (−0.01 mm day−1 decade−1) over tropical ocean and land, respectively. This corresponds to an about 5.5% increase (ocean) and 1% decrease (land) during the entire 27-yr time period. The year 2005 has the largest annual tropical total precipitation (land plus ocean) for the GPCP record. The five highest years are (in descending order) 2005, 2004, 1998, 2003, and 2002. For tropical ocean the five highest years are 1998, 2004, 2005, 2002, and 2003.

Techniques are applied to isolate and quantify variations due to ENSO and two major volcanic eruptions during the time period (El Chichón, March 1982; Mount Pinatubo, June 1991) in order to examine longer-time-scale changes. The ENSO events generally do not impact the tropical total rainfall, but rather induce significant anomalies with opposite signs over tropical land and ocean. The impact of the two volcanic eruptions is estimated to be about a 5% reduction in tropical rainfall over both land and ocean. A modified dataset (with ENSO and volcano effects removed) retains the same approximate linear change slopes, but with reduced variances, thereby increasing the statistical significance levels associated with the long-term rainfall changes in the Tropics. However, although care has been taken to ensure that this dataset is as homogeneous as possible, firm establishment of the existence of the discussed changes as long-term trends may require continued analysis of the input datasets and a lengthening of the observation period.

1. Introduction

Exploring global climate variability and change has an immense environmental and societal significance (e.g., Kumar et al. 2004). Previous studies showed interannual variability and interdecadal/longer-term changes in various climate components, for example, surface air temperature, sea surface temperature (SST), land rainfall, etc., specifically during recent decades (e.g., Cane et al. 1997; Chen et al. 2002; Simmons et al. 2004). The El Niño–Southern Oscillation (ENSO) generally dominates the global variability on interannual time scale (e.g., Trenberth et al. 1998; Wigley 2000; Trenberth et al. 2002; Smith et al. 2006). Other mechanisms such as the Pacific decadal oscillation (PDO) and the Arctic Oscillation (AO) are found to account for most of the interdecadal variability (e.g., Zhang et al. 1997; Hurrell et al. 2001).

Longer-term change and/or trend are also widely recognized, and are considered to be a consequence of probably both natural and human-induced effects (e.g., Chu and Wang 1997; Bates and Jackson 2001). The gradually warming climate is a good example (e.g., Yang et al. 2003; Neelin et al. 2003; Kumar et al. 2004). However, whether there is a coherent long-term variation within the entire hydrological cycle accompanying this warming is still an outstanding question (e.g., Chen et al. 2002; Simmons et al. 2004; McCarthy and Toumi 2004; Yin et al. 2004). This is mostly due to the lack of long-record, consistent observations of fundamental climate components over the globe, particularly precipitation. With the availability of the 27-yr (1979–2005) monthly rainfall product from the Global Precipitation Climatology Project (GPCP; Adler et al. 2003), it is thus worthwhile to examine global rainfall variability on both interannual and interdecadal/longer-term time scales. The 27-yr record is still relatively short for establishing firm conclusions about long-term changes and for separating these changes from interdecadal variability, but an improved understanding of the general mechanisms controlling the responses of the global hydrological cycle may be possible. The purpose of this study is therefore to make efforts along this direction, specifically to discriminate among rainfall variations due to ENSO, volcanic events, and possible long-term changes, particularly in the Tropics. A recent study by Smith et al. (2006) showed, by means of an empirical orthogonal functions (EOF) analysis, that there are probably tropical trendlike rainfall changes with spatial variations. They further suggested that these changes might be correlated with the interdecadal variability of the tropical SST. The tropical interdecadal variability may significantly contribute to the trendlike changes during the current GPCP time period because the two effects may not be properly separated due to the length of the data record. However, it is arguable that the tropical SST interdecadal variability may function, at least partly, in a similar way as the ENSO events, that is, shifting the major rainfall zones across the Tropics with little effect on the tropical total rainfall change (Su and Neelin 2003). Trenberth and Shea (2005) showed a contrasting temperature–precipitation relationship over tropical land and ocean. Therefore, it is worthwhile to further examine the net rainfall changes in the entire tropical region, and within large regions such as tropical ocean and land. Here, we hence primarily focus on large-domain (specifically oceanic, land, and total) rainfall anomalies within different latitude bands to emphasize and evaluate these large-regional rainfall changes, and to possibly contrast precipitation changes between land and ocean, particularly in the Tropics.

2. Brief description of the data

The monthly precipitation dataset from the Global Precipitation Climatology Project (GPCP) is a community-based analysis of global precipitation under the auspices of the World Climate Research Program (WCRP) from 1979 to the present. On a global 2.5° × 2.5° grid, the data are combined from various information sources: microwave-based estimates from the Special Sensor Microwave Imager (SSM/I), infrared (IR) rainfall estimates from geostationary and polar-orbiting satellites, and surface rain gauges. The analysis procedure is designed to take advantage of the particular strengths of the individual input datasets, especially in terms of bias reduction. For example, the microwave information is used to adjust the bias of the IR-based estimates over the ocean and land, and the gauge information is used to adjust the bias of the merged satellite estimates over land. The dataset therefore should have the low bias of the input information deemed to be best, with the superior sampling on a monthly scale produced by multiple satellites. The period from 1988 to the present is homogeneous in terms of input datasets; however, to extend the dataset back to 1979, the satellite data are limited to infrared-based estimates. Although these pre-1988 estimates are trained on the later period to reduce possible differences, there is still a time inhomogeneity in the analysis, in terms of satellite input datasets. These features of the dataset are considered in the following analysis, though Smith et al. (2006) indicate that the impact of this time inhomogeneity is not a major concern. Further information about the GPCP monthly dataset is detailed in Adler et al. (2003) and Huffman et al. (1997).

3. Results

Tables 1 –3 provide a summary of the mean rain rates around the globe, and their standard deviations from corresponding mean seasonal cycles. Various latitudinal belts over both land and ocean are applied to give a global and large-regional description of the mean precipitation and variability. Large variability (>10%) is generally seen in the mid- to high latitudes except over land along the latitude band 50°–90°S. Interestingly, the deviations in the global total are much weaker than these within various latitude bands, suggesting a tendency toward precipitation conservation, with opposite rainfall anomalies in different regions across the globe at longer-than-seasonal time scales, particularly on the interannual time scale.

Table 1.

Mean rain rate and interannual standard deviation (mm day−1) over land and ocean during Jan 1979–Dec 2005.

Mean rain rate and interannual standard deviation (mm day−1) over land and ocean during Jan 1979–Dec 2005.
Mean rain rate and interannual standard deviation (mm day−1) over land and ocean during Jan 1979–Dec 2005.
Table 3.

Mean rain rate and interannual standard deviation (mm day−1) over land during Jan 1979–Dec 2005.

Mean rain rate and interannual standard deviation (mm day−1) over land during Jan 1979–Dec 2005.
Mean rain rate and interannual standard deviation (mm day−1) over land during Jan 1979–Dec 2005.

The 27-yr time history of the data indicates no significant long-term change in the annual global mean precipitation (solid line in Fig. 1), resulting from possible weak, opposite changes over land and ocean (Fig. 1). An upward (downward) linear change is found in the oceanic (land) mean precipitation, but is not statistically significant. Evidence of regional changes seems stronger. Regional contributions to the global change in precipitation volume are computed by multiplying the estimated linear fits for each region and latitude band by the fraction of the globe that covers the region, and further dividing them by the global mean rain rate (2.62 mm day−1) (Fig. 2). A simple accumulation of volume changes across all subregions is equal to the volume change for the global total rainfall. It is evident that the large volume changes are in the Tropics and Northern Hemisphere midlatitudes. Zonally averaged GPCP data combining land and ocean indicate generally increasing precipitation (about 3.7% over 27 yr) in the deep Tropics (25°S–25°N), with the increase focused on the ocean (5.5% increase), and a smaller (1%) decrease over land. In the northern subtropics (25°–50°N), a decrease over both land and ocean is noted in the dataset.

Fig. 1.

Annual mean global rainfall anomalies over ocean, land, and both ocean and land. Also shown are the estimated slopes of the linear fits.

Fig. 1.

Annual mean global rainfall anomalies over ocean, land, and both ocean and land. Also shown are the estimated slopes of the linear fits.

Fig. 2.

Volume contributions to long-time change/linear fit during 1979–2005.

Fig. 2.

Volume contributions to long-time change/linear fit during 1979–2005.

Since space-borne microwave observations of precipitation became available only since 1987, linear changes are also estimated during the 1988–2005 period for comparison (Fig. 3). As during 1979–2005, tropical rainfall variability, particularly over the tropical oceans, dominates. The changes for this shorter period are similar to the longer period in the Tropics, although the small change over tropical land changes sign. We also note that there is a much more intense change in the global total precipitation, primarily due to a much weaker rainfall change along 25°–50°N during this time period. Certainly, these results warrant at least a careful analysis of the confidence limits we can determine for these statistics. Also, possible mechanisms have yet to be further explored, though the changes of the Hadley circulation strength might be an interpretation (e.g., Chen et al. 2002).

Fig. 3.

Volume contributions to long-time change/linear fit during 1988–2005.

Fig. 3.

Volume contributions to long-time change/linear fit during 1988–2005.

In the next three subsections, we will focus on the Tropics (25°S–25°N) because the most intense long-term rainfall changes occur in this region (Figs. 2 and 3) and also because of the greater confidence of the authors in the quality of the precipitation analysis in the Tropics (Adler et al. 2003). The impact of ENSO and volcanoes will be investigated first to isolate, as much as possible, their effects on surface rainfall. We then reexamine the long-term rainfall variability with and without the ENSO and volcanic impacts. Also, to limit high-frequency noises, annual mean rainfall anomalies will be the primary variable examined for long-term changes.

a. ENSO and volcanic impact

The time series of tropical monthly mean rainfall anomalies are depicted in Fig. 4a, which is similar to Fig. 8 in Adler et al. (2003). The ENSO impact is evident by comparing the rainfall anomalies to the standard Niño-3.4 SST index (Fig. 4b). Generally, the warm events correspond to more (less) rainfall over ocean (land) with the opposite variation for cold events. Strong variabilities can be seen in tropical total rainfall, but are weaker than rainfall variations over land and ocean separately. This is due to the opposite effects of ENSO on rainfall over land and ocean (e.g., Neelin et al. 2003).

Fig. 4.

(a) Tropical rainfall anomalies (mm day−1; 3-month running mean), (b) Niño-3.4 (°C), and (c) global mean stratospheric aerosol optical thickness (τst) during 1979–2005.

Fig. 4.

(a) Tropical rainfall anomalies (mm day−1; 3-month running mean), (b) Niño-3.4 (°C), and (c) global mean stratospheric aerosol optical thickness (τst) during 1979–2005.

Two major volcanic events occurred during the data record. The El Chichón volcano erupted in March 1982, and Mount Pinatubo in June 1991. Following these two large eruptions, abrupt changes showed up in the global mean stratospheric aerosol optical thickness (τst) due to the injection of aerosol particles (Fig. 4c; Sato et al. 1993). These changes effectively influence the radiation budget of Earth (e.g., Robock 2000), causing significant global cooling and drying, and also impacting precipitation efficiency through radiative forcing and aerosol particles injected into the atmosphere (e.g., Spencer et al. 1998; Robock 2000; Soden et al. 2002; Yang and Schlesinger 2002). These impacts may last at least 1–2 yr (e.g., Robock 2000; Yang and Schlesinger 2002).

Volcanic impact is discernible in the tropical rainfall anomalies (Fig. 4a), though it is compounded with the influence of ENSO. To emphasize the volcanic impact, annual mean rainfall anomalies1 over ocean (PO) versus land (PL) are plotted in Fig. 5. Volcano years are identified by choosing a threshold of τst, that is, at least 6 months with τst ≥ 0.02 for each year (horizontal dotted line in Fig. 4c), yielding six volcano years, that is, 1982, 1983, 1984, 1991, 1992, and 1993. A linear regression relation is further estimated between PO and PL during the normal (nonvolcano) years (dashed line in Fig. 5). As expected, PO always tends to be opposite of PL, and the scatter about the regression line is modest except in 1983 and 1991 in which the volcanic impact is strong. The usual compensation between land and ocean is the major reason for a much weaker year-to-year fluctuation in the tropical total precipitation (solid lines in Figs. 4a and 10a), compared to ocean or land values alone. Volcanic signals can also be seen in 1992, though much more weakly. Quantitatively, these two eruptions may have induced up to a 5% rainfall decrease in the Tropics during 1983 and 1991. The effect is evident over both land and ocean. In the other three volcano years defined by τst—that is, 1982, 1984, 1993—PO and PL tend to approach the regression line as in most normal years, likely suggesting that the rainfall anomalies were more sensitive to the lower boundary forcing for some reasons, such as a comparatively strong SST forcing at the earlier stage of volcanic eruption (1982), and/or the completion of an atmospheric vertical adjustment of the thermodynamic state to aerosol forcing at the later stage of volcanic eruption (1984, 1993). However, detailed mechanisms have yet to be investigated.

Fig. 5.

Annual mean rainfall anomalies over tropical ocean vs land. The dashed line represents the regression between these two during all years except 1982, 1983, 1984, 1991, 1992, and 1993 (dot–crosses) in which an intense volcanic impact is expected (at least 6 months with τst ≥ 0.02).

Fig. 5.

Annual mean rainfall anomalies over tropical ocean vs land. The dashed line represents the regression between these two during all years except 1982, 1983, 1984, 1991, 1992, and 1993 (dot–crosses) in which an intense volcanic impact is expected (at least 6 months with τst ≥ 0.02).

Fig. 10.

Tropical annual mean rainfall anomalies (a) with and (b) without the ENSO and volcano impacts. Also shown are the slopes of the linear fits (significance levels are given in Table 4).

Fig. 10.

Tropical annual mean rainfall anomalies (a) with and (b) without the ENSO and volcano impacts. Also shown are the slopes of the linear fits (significance levels are given in Table 4).

We now focus on assessing the impact of ENSO and volcanoes using the time series of monthly rainfall anomalies. Again, these time series are first detrended for this particular analysis as we assume that these impacts will occur on the interannual-to-interdecadal time scale. Interestingly, the two volcanic eruptions coincidently occurred with El Niño events (Figs. 4b and 4c). We thus have to discriminate their impact during this period. As mentioned above, the whole months during 1979–2005 are first categorized into two time periods: the normal (τst < 0.02) and volcanic impact (τst ≥ 0.02) months (Fig. 4c). Our strategy is to first estimate the possible relations between rainfall anomalies and ENSO during the normal period, then apply them to the volcano period to isolate the impact of ENSO, and finally estimate the relations between rainfall anomalies and τst after the ENSO impact is removed from the rainfall anomalies during the volcano period. Because ENSO has opposite effects on rainfall anomalies over land and ocean (Fig. 5), we assume that the ENSO signals are reasonably weak in the tropical total rainfall anomalies (Fig. 4a). This is consistent with Su and Neelin (2003), who showed that the tropical total precipitation is not a simple, linear response to tropical SST (or ENSO), but is controlled by other factors. Obviously, the success of our strategy heavily depends on whether and to what extent the impacts of ENSO and volcano are separable given their likely nonlinear relations (e.g., Wigley 2000), such as the possible modulations of ENSO by volcanic eruptions through radiation-forced global temperature change (e.g., Mann et al. 2005).

The rainfall anomalies over ocean, land, and the entire Tropics are plotted as function of the ENSO index (Niño-3.4) during these two (normal and volcano) periods in Fig. 6. Over ocean (Fig. 6a), a linear relation is seen between rainfall and Niño-3.4 with a slope of 0.0535 mm day−1 °C−1 during the normal (nonvolcano) period, though there is considerable scatter (correlation = 0.46). This relation confirms that tropical oceanic rainfall responds to the SST anomalies in the central Pacific Ocean associated with ENSO. In contrast, during the volcano period no significant response in rainfall shows up, which is probably a consequence of the opposite impacts of the El Niño and volcano as they occurred almost simultaneously (Figs. 4b and 4c). Over land (Fig. 6b), rainfall anomalies decrease with the increase of Niño-3.4 during both the normal and volcano periods, with linear slopes of the same order: −0.1372 and −0.1441 mm day−1 °C−1, respectively. The similar rainfall changes during these two periods indicate that both ENSO and the volcano tend to suppress surface rainfall over land. For tropical total precipitation (Fig. 6c), there is a near-zero rainfall change with Niño-3.4 (0.0017 mm day−1 °C−1) during the normal period; however, a significant downward slope (−0.0384 mm day−1°C−1) can be seen during the volcano period. The results confirm that the tropical total precipitation does not systematically respond to the ENSO forcing as mentioned above, but they do indicate that volcanic eruptions significantly reduce surface rainfall in the entire Tropics, and are possibly felt across the world (a seemingly simultaneous decrease appearing in the global precipitation total, as well as in the oceanic and land precipitation totals in 1983 and 1991 as shown in Fig. 1).

Fig. 6.

Tropical rainfall anomalies (mm day−1; 3-month running mean) as a function of Niño-3.4 (°C) over (a) ocean, (b) land, and (c) both ocean and land. Dots (crosses) denote the months in which τst< ( ≥ ) 0.02, with solid (dashed) lines representing their corresponding linear fit lines.

Fig. 6.

Tropical rainfall anomalies (mm day−1; 3-month running mean) as a function of Niño-3.4 (°C) over (a) ocean, (b) land, and (c) both ocean and land. Dots (crosses) denote the months in which τst< ( ≥ ) 0.02, with solid (dashed) lines representing their corresponding linear fit lines.

To further focus on the volcanic impact on rainfall changes, the tropical rainfall anomalies during the volcano period are plotted as a function of τst (Fig. 7a). The most intense response is over land due to the same-sign impact of both the warm ENSO and the volcano (dashed line). For the tropical total precipitation, a negative slope can be seen (solid line). As shown in Fig. 6a, a very weak rainfall response occurs over ocean caused by the opposite effects of the El Niño and the volcano. A further assumption is made that the ENSO events consistently impact the rainfall changes during both the normal and volcano periods.

Fig. 7.

Tropical rainfall anomalies (mm day−1; 3-month running mean) as a function of τst during the intense volcanic impact period (a) with and (b) without the ENSO impact. Dots denote the tropical total rainfall anomalies, crosses the anomalies over land, and diamonds the anomalies over ocean. Their corresponding linear fits are represented by solid (dots), dashed–dotted (diamonds), and dashed (crosses) lines.

Fig. 7.

Tropical rainfall anomalies (mm day−1; 3-month running mean) as a function of τst during the intense volcanic impact period (a) with and (b) without the ENSO impact. Dots denote the tropical total rainfall anomalies, crosses the anomalies over land, and diamonds the anomalies over ocean. Their corresponding linear fits are represented by solid (dots), dashed–dotted (diamonds), and dashed (crosses) lines.

The linear regression relations of rainfall anomalies versus Niño-3.4 over ocean (0.0535 mm day−1 °C−1) and land (−0.1372 mm day−1 °C−1) (Figs. 6a and b), respectively, estimated earlier for the normal (nonvolcano) period, are then applied to estimate the rainfall response to the warm ENSO during the volcano occurrence. After the ENSO impact is removed in this way, the relationships between rainfall anomalies and τst are depicted in Fig. 7b. Though still very scattered, the linear fits for rainfall changes over ocean (−1.0752), land (−1.687), and the entire Tropics (−1.2184) are similar. This supports the idea that volcanic eruptions simultaneously reduce rainfall over both land and ocean in the Tropics.

b. Interannual-to-interdecadal rainfall variability in the Tropics

The linear regressions of rainfall anomalies with Niño-3.4 and τst obtained in the preceding section (Figs. 6 and 7) are applied to estimate the rainfall responses due to these factors. The ENSO events generally force opposite anomalies over land and ocean as discussed above (Fig. 8b). The two volcanic eruptions forced a similar negative tendency of rainfall change over land and ocean, although the reduction of rainfall during 1991 seems to begin prior to the eruption, indicating other possible factors (Figs. 8a and 8c). These responses are further subtracted from the rainfall monthly time series (Fig. 8d). The variances of these three residual time series become much smaller, although it seems that partial ENSO signals may still be included in the anomalies over land and ocean. This is likely because the simple linear regression relations used here are not completely consistent during the entire period, especially for the strongest ENSO event of the period during 1997–2001. The ENSO events may themselves undergo long-period variations (e.g., Vimont et al. 2003; Sun et al. 2004). Interestingly, even after the impacts of ENSO and volcano are removed, interannual oscillations can still be seen, particularly in the tropical total rainfall. This tends to agree with Su and Neelin (2003).

Fig. 8.

(a) Tropical rainfall anomalies (mm day−1; 3-month running mean), (b) linear response of tropical rainfall to Niño-3.4 (°C), (c) linear response of tropical rainfall to volcanic eruptions, and (d) tropical rainfall anomalies with ENSO and the volcanic impact removed.

Fig. 8.

(a) Tropical rainfall anomalies (mm day−1; 3-month running mean), (b) linear response of tropical rainfall to Niño-3.4 (°C), (c) linear response of tropical rainfall to volcanic eruptions, and (d) tropical rainfall anomalies with ENSO and the volcanic impact removed.

We further clarify the ENSO and volcano signals in tropical rainfall anomalies, and demonstrate the success/effectiveness of the method used here by applying spectrum analyses to various time series with and without the impact of ENSO and the volcano (Fig. 9). Three power peaks appear in the tropical oceanic rainfall anomalies at periods of 2–3, 5, and 7–11 yr. Similar spectrum features can also be seen in rainfall anomalies over land with a much stronger peak at the 5-yr period (not shown). For tropical total rainfall anomalies, the power peak at the period of 5 yr disappears, further confirming the weak impact of ENSO. The existence of oscillations in tropical oceanic rainfall within the period of 2–7 yr is consistent with previous studies focusing on SST and surface winds in the equatorial Pacific (e.g., Rasmusson and Carpenter 1982). Within the period of 7–11 yr, the spectral signals might not be only ascribed to the ENSO-type interdecadal changes (e.g., Zhang et al. 1997; Vimont et al. 2003), but are probably contributed to by the volcanic eruptions as they appear in both the oceanic and total rainfall (solid and dashed lines in Fig. 9). It should be noted that the spectral properties in the higher-period (7–11 yr) end might not be properly resolvable due to the length of the data record.

Fig. 9.

Spectral power of the tropical total and oceanic rainfall anomalies (mm day−1) with and without the impact of ENSO and the volcano.

Fig. 9.

Spectral power of the tropical total and oceanic rainfall anomalies (mm day−1) with and without the impact of ENSO and the volcano.

Spectral analyses are also applied to the time series of the tropical oceanic and total rainfall anomalies without the ENSO or volcano signals. The signals around the 5-yr period are greatly suppressed for the oceanic rainfall, partly suggesting the success of the regression method we used. The signals around two other peaks that also appear in the tropical total rainfall are suppressed too, but are still very evident. Thus, it seems that other than ENSO, the signals at the 2–3-yr period may be related to the tropospheric biennial oscillation (TBO) (e.g., Meehl 1997). On the other hand, within the 7–11-yr period band, the appearance of the peak, though becoming much weaker, tends to suggest that the volcanic signals still exist to a certain degree in the time series, implying the difficulties of eliminating their impact.

c. Long-term rainfall change

After the impacts of ENSO and volcanoes are isolated, we now reexamine the long-term rainfall change in the Tropics. Figure 10a depicts the original time series of the annual mean rainfall anomalies over ocean (PO), land (PL), and the entire Tropics (PLO). As noted before, large variations occur in PL and PO, and they are mostly opposite. Compared to them, PLO shows much weaker interannual variability as shown in the previous subsection. Long-term rainfall changes represented by the slopes of the regression lines are computed for PLO, PO, and PL, respectively. These linear changes are the same as in Fig. 2 but are denoted by different units (Fig. 10a). It is interesting to note that 2005 has the largest annual tropical total precipitation (land plus ocean) for the GPCP record. The five highest years are (in descending order) 2005, 2004, 1998, 2003, and 2002. For tropical ocean the five years are 1998, 2004, 2005, 2002, and 2003.

The significance levels of these changes are estimated based on the Student’s t test (e.g., Woodward and Gray 1993). The degrees of freedom (dof’s) are estimated by means of dof = n(1 − γ)/(1 + γ) − 2 (e.g., Livezey and Chen 1983; Chu and Wang 1997; Bretherton et al. 1999), where n is the length of time series and γ is the lag-one correlation coefficient of the residual time series (Santer et al. 2000; Feidas et al. 2004). For PO, the slope (0.0624 mm day−1 decade−1) is at the 99% level (Table 4), showing a relatively strong upward rainfall increase; a negative slope is seen over land (−0.0119 mm day−1 decade−1), but is not statistically significant; tropical total rainfall shows an upward increase (0.0422 mm day−1 decade−1), significant at the 98% level.

Table 4.

Linear trends in annual mean rainfall anomalies (mm day−1 decade−1) in the Tropics (25°S–25°N) during 1979–2005. The significance levels are estimated based on the Student’s t test; the dof’s are estimated by means of dof = n(1 − γ)/(1 + γ) − 2, γ the lag-one correlation coefficient of the regression residuals, and n(= 27) the length of the time series.

Linear trends in annual mean rainfall anomalies (mm day−1 decade−1) in the Tropics (25°S–25°N) during 1979–2005. The significance levels are estimated based on the Student’s t test; the dof’s are estimated by means of dof = n(1 − γ)/(1 + γ) − 2, γ the lag-one correlation coefficient of the regression residuals, and n(= 27) the length of the time series.
Linear trends in annual mean rainfall anomalies (mm day−1 decade−1) in the Tropics (25°S–25°N) during 1979–2005. The significance levels are estimated based on the Student’s t test; the dof’s are estimated by means of dof = n(1 − γ)/(1 + γ) − 2, γ the lag-one correlation coefficient of the regression residuals, and n(= 27) the length of the time series.

We again apply the linear relations of monthly rainfall anomalies with the ENSO and volcano effects to remove/limit their signals in the annual mean rainfall anomalies. Our purpose is to examine how the ENSO and volcano signals may influence the statistical significance level given their strong existence in annual mean rainfall anomalies (Fig. 10a).

After the impact of ENSO and volcano is removed/limited in the monthly time series, the annual mean time series are reconstructed (Fig. 10b). It is clear that the interannual variability is reduced. As expected, the linear change becomes smaller for PLO because of the removal of negative anomalies associated with volcanoes in the earlier time period (Fig. 5). However, the significance level for the combined ocean–land signal increases from 98% to 99.5% (Table 4) resulting from an increase in the dof’s and reduced variance. Over land, we remain unable to find any significant long-term linear change, though the magnitude of the slope increases. The significance level for tropical oceanic rainfall becomes 99.95%, in spite of a slightly smaller slope, again due to a larger number of dof’s and reduced variance. Thus, an upward trend in the tropical oceanic rainfall appears in the GPCP dataset, which dominates the tropical total rainfall change. The years with the largest rainfall also shift somewhat with the modified dataset. For the tropical total rain, the highest five years (in descending order) are still 2005, 2004, 1998, 2003, and 2002, while for the tropical ocean the years are 1998, 2005, 2004, 2003, and 2002. In either the original or the modified datasets the recent years seem to dominate the high end of the distribution of the annual tropical rainfall, except for 1998, in which a very strong warm ENSO occurred.

The sensitivity of the linear changes of the tropical oceanic rainfall with the ENSO and volcanic impacts removed, and the corresponding dof and significance level, is further tested in terms of the length of the data record (Fig. 11). We construct a set of subset time series by varying the lengths of the time series in two ways, simply changing either the starting year or the ending year of the time series. For each subset time series, the linear change and the corresponding dof’s and their significance level are estimated using the method detailed above. The results from these two sets of calculations are well separated for shorter time series, but approach each other once their length is larger than 23 yr. The slope always remains positive, varying approximately from 0.04 to 0.075 mm day−1 decade−1. Particularly, the linear changes become statistically significant (≥99%) if the time series is larger than 25 yr (Fig. 11c). However, the significance for the more homogeneous data during the 1988–2005 period is only 87%, that is, not significant due to the short length of the record. Consistent with Smith et al. (2006), these results seem to support the idea that there may be an upward mean rainfall increase during the last 27 yr over the tropical oceans, and this increase is statistically significant, particularly after the noise at interannual-to-interdecadal time scales is reasonably suppressed, even though the data record is relatively short.

Fig. 11.

Sensitivity of (a) long-term linear changes, (b) degrees of freedom, and (c) corresponding significance levels for tropical oceanic precipitation with the ENSO and volcanic impacts removed, to varying lengths of time series. Dots denote the starting year of the time series changing from 1979 (27) to 1990 (16), squares represent the ending year changing from 2005 (27) to 1994 (16), and triangle represents the time period of 1988–2005.

Fig. 11.

Sensitivity of (a) long-term linear changes, (b) degrees of freedom, and (c) corresponding significance levels for tropical oceanic precipitation with the ENSO and volcanic impacts removed, to varying lengths of time series. Dots denote the starting year of the time series changing from 1979 (27) to 1990 (16), squares represent the ending year changing from 2005 (27) to 1994 (16), and triangle represents the time period of 1988–2005.

In terms of the existence of the tropical increase in precipitation, the statistical results assume that the data are correct, at least to the point of not affecting the resulting significance tests. What is important in this regard in the dataset is that any bias in the analysis (e.g., over tropical ocean) be unchanged as a function of time. A constant high or low bias will not affect the statistics or our interpretation of them. However, a bias changing through the decades could produce a false trend, or eliminate a real trend.

As stated before, care has been taken in the construction of the GPCP monthly dataset to minimize the time-varying bias (Adler et al. 2003). For example, the analysis technique for the pre-1988 period is cross calibrated with the later period via an overlap period. In addition, during the 1988-to-the-present period only data from one SSM/I instrument at a time, using a brightness temperature input dataset cross calibrated from satellite to satellite during the period and flying at about the same equator crossing time, is used throughout the analysis to minimize any impact from varying the number of satellites and their equator-crossing times. The one microwave sensing satellite is used to calibrate or adjust the geosynchronous IR data covering the diurnal cycle to obtain a consistent estimate during the SSM/I period.

However, there may still be issues that could affect the dataset in ways that might affect the estimated changes during the period. One of these may be a subtle change to the equator crossing time in the “early morning” series of SSM/I carrying satellites used in this study. Because of the diurnal cycle of rainfall over the ocean (and the nonsinusoidal shape of the diurnal curve), a slight shift in observation time of the satellite due to orbit drift could cause a small change in the bias and an error in the trend calculation. Although we believe that this should not affect the main results related to the calculated changes, it emphasizes the need for additional analysis of the input datasets and the merging techniques in order to increase our confidence in the results.

4. Summary and concluding remarks

Global and large-regional rainfall variabilities are examined using the 27-yr (1979–2005) GPCP monthly dataset. Attempts have been made to discriminate among the variations resulting from ENSO, volcanoes, and possible long-term climate changes.

ENSO and two major volcanic eruptions during the data record heavily modulated the tropical rainfall variabilities. ENSO has opposite effects over land and ocean, with only a negligible influence on the tropical total rainfall. The warm events tend to increase (decrease) oceanic rainfall, but decrease (increase) rainfall over land with the opposite effects for the cold events. Thus, ENSO is not a major contributor to the interannual variability in tropical total rainfall (Su and Neelin 2003), though it can cause intense rainfall anomaly features by shifting the major rainy zones in the Tropics (e.g., Wigley 2000; Curtis and Adler 2003; Neelin et al. 2003). The volcanic eruptions generally suppress tropical rainfall in the entire Tropics resulting from their impact on the global radiation budget and precipitation efficiency (e.g., Spencer et al. 1998; Soden et al. 2002). The two eruptions hence project a tropical rainfall oscillation within a periodicity band peaking at about 9 yr. Therefore, on the interannual-to-interdecadal time scales the tropical total rainfall variability is mostly controlled by the volcanic eruptions and factors discussed in Su and Neelin (2003). Also, the tropospheric quasi-biennial oscillation seems to be another important component at the high-frequency end of the interannual time scale (Fig. 8; e.g., Rasmusson and Carpenter 1982; Jiang et al. 1995; Meehl 1997), and has to be further quantified in the future.

Regarding the long-term rainfall change, globally it tends to approach zero. However, an upward linear change (0.0624 mm day−1 decade−1) is found in the dataset over the tropical ocean with a 99% significance level. A downward change (−0.0119 mm day−1 decade−1) is found in the rainfall anomalies over tropical land, though it is not statistically significant (lower than the 90% significance level). For the tropical total rainfall, the upward increase (0.0422 mm day−1 decade−1) can reach the 98% significance level. These changes correspond to about a 5.5% rainfall increase over the tropical ocean and a 1% decrease over land during the entire data record. The years in the GPCP record with the highest total annual tropical (and tropical ocean) rainfall have mainly occurred after the year 2001. The year 2005 (the last year in the dataset) had the highest total annual tropical rainfall during the period. A linear regression method is applied to the dataset to extract the relations of rainfall anomalies with ENSO and volcanoes, respectively, by assuming that the impacts of ENSO and volcanoes are separable to a certain extent. After their impacts are assessed, isolated, and removed from the rainfall anomalies, the changes for the annual mean rainfall over the tropical ocean, land, and the entire Tropics are reexamined. The linear trend for tropical land rainfall is still not statistically significant. However, the linear changes for the tropical total rainfall and oceanic rainfall both reach higher significance levels, 99.5% and 99.95%, respectively, though the magnitudes of the linear change become slightly smaller. This is primarily because of larger dof’s and less variances in the modified dataset. Although strenuous efforts have been made to make this dataset as homogeneous as possible for the longest possible period, firm establishment of the existence of the discussed trends requires still further evaluation of the input datasets, analysis of the merger techniques, and a lengthening of the data record into the future.

However, the results here suggest that an increase in tropical oceanic rainfall is occurring. It tends to support ideas that there is an intensifying hydrological cycle in the Tropics given a warming environment (e.g., Yang et al. 2003; Neelin et al. 2003; Kumar et al. 2004).

Table 2.

Mean rain rate and interannual standard deviation (mm day−1) over ocean during Jan 1979–Dec 2005.

Mean rain rate and interannual standard deviation (mm day−1) over ocean during Jan 1979–Dec 2005.
Mean rain rate and interannual standard deviation (mm day−1) over ocean during Jan 1979–Dec 2005.

Acknowledgments

Mr. David Bolvin prepared the GPCP rainfall data. The global mean stratospheric aerosol optical thickness data were provided by NASA GISS from their Web site (http://data.giss.nasa.gov/). Comments and suggestions from two anonymous reviewers greatly improved the manuscript. This research is supported under the NASA Energy and Water-cycle Study (NEWS) program.

REFERENCES

REFERENCES
Adler
,
R. F.
, and
Coauthors
,
2003
:
The version 2 Global Precipitation Climatology Project (GPCP) Monthly Precipitation Analysis (1979–present).
J. Hydrometeor.
,
4
,
1147
1167
.
Bates
,
J. J.
, and
D. L.
Jackson
,
2001
:
Trends in upper-tropospheric humidity.
Geophys. Res. Lett.
,
28
,
1695
1698
.
Bretherton
,
C. S.
,
M.
Widmann
,
V. P.
Dymnikov
,
J. M.
Wallace
, and
I.
Bladé
,
1999
:
The effective number of spatial degree of freedom of a time-varying field.
J. Climate
,
12
,
1990
2009
.
Cane
,
M. A.
,
A. C.
Clement
,
A.
Kaplan
,
Y.
Kushnir
,
D.
Pozdnyakov
,
R.
Seager
,
S. E.
Zebiak
, and
R.
Murtugudde
,
1997
:
Twentieth-century sea surface temperature trends.
Science
,
275
,
957
960
.
Chen
,
J.
,
B. C.
Carlson
, and
A. D.
Del Genio
,
2002
:
Evidence for strengthening of the tropical general circulation in 1990s.
Science
,
295
,
838
841
.
Chu
,
P-S.
, and
J-B.
Wang
,
1997
:
Recent climate change in the tropical western Pacific and Indian Ocean regions as detected by outgoing longwave radiation records.
J. Climate
,
10
,
636
646
.
Curtis
,
S.
, and
R. F.
Adler
,
2003
:
The evolution of El Niño–precipitation relationships from satellites and gauges.
J. Geophys. Res.
,
108
.
4153, doi:10.1029/2002JD002690
.
Feidas
,
H.
,
T.
Makrogiannis
, and
E.
Bora-Senta
,
2004
:
Trend analysis of air temperature time series in Greece and their relationship with circulation using surface and satellite data: 1955–2001.
Theor. Appl. Climatol.
,
79
,
185
208
.
Huffman
,
G. J.
, and
Coauthors
,
1997
:
The Global Precipitation Climatology Project (GPCP) version 1 dataset.
Bull. Amer. Meteor. Soc.
,
78
,
5
20
.
Hurrell
,
J. W.
,
Y.
Kushnir
, and
M.
Visbeck
,
2001
:
The North Atlantic Oscillation.
Science
,
291
,
603
605
.
Jiang
,
N.
,
J. D.
Neelin
, and
M.
Ghil
,
1995
:
Quasi-quadrennial and quasi-biennial variability in COADS equatorial Pacific sea surface temperature and winds.
Climate Dyn.
,
12
,
101
112
.
Kumar
,
A.
,
F.
Yang
,
L.
Goddard
, and
S.
Schubert
,
2004
:
Differing trends in the tropical surface temperatures and precipitation over land and oceans.
J. Climate
,
17
,
653
664
.
Livezey
,
R. E.
, and
W. Y.
Chen
,
1983
:
Statistical field significance and its determination by Monte Carlo techniques.
Mon. Wea. Rev.
,
111
,
46
59
.
Mann
,
M. E.
,
M. A.
Cane
,
S. E.
Zebiak
, and
A.
Clement
,
2005
:
Volcanic and solar forcing of the tropical Pacific over the past 1000 years.
J. Climate
,
18
,
447
456
.
McCarthy
,
M. P.
, and
R.
Toumi
,
2004
:
Observed interannual variability of tropical troposphere relative humidity.
J. Climate
,
17
,
3181
3191
.
Meehl
,
G. A.
,
1997
:
The South Asian monsoon and the tropospheric biennial oscillation.
J. Climate
,
10
,
1921
1943
.
Neelin
,
J. D.
,
C.
Chou
, and
H.
Su
,
2003
:
Tropical drought regions in global warming and El Niño teleconnections.
Geophys. Res. Lett.
,
30
.
2275, doi:10.1029/2003GL018625
.
Rasmusson
,
E. M.
, and
T. H.
Carpenter
,
1982
:
Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño.
Mon. Wea. Rev.
,
110
,
354
384
.
Robock
,
A.
,
2000
:
Volcanic eruptions and climate.
Rev. Geophys.
,
38
,
191
219
.
Santer
,
B. D.
,
T. M. L.
Wigley
,
J. S.
Boyle
,
D. J.
Gaffen
,
J. J.
Hnilo
,
D.
Nychka
,
D. E.
Parker
, and
K. E.
Taylor
,
2000
:
Statistical significance of trends and trend differences in layer-average atmospheric temperature time series.
J. Geophys. Res.
,
105
,
D6
.
7337
7356
.
Sato
,
M.
,
J. E.
Hansen
,
M. P.
McCormick
, and
J. B.
Pollack
,
1993
:
Stratospheric aerosol optical depths, 1850–1990.
J. Geophys. Res.
,
98
,
22987
22994
.
Simmons
,
A. J.
, and
Coauthors
,
2004
:
Comparison of trends and low-frequency variability in CRU, ERA-40, and NCEP/NCAR analyses of surface air temperature.
J. Geophys. Res.
,
109
.
D24115, doi:10.1029/2004JD005306
.
Smith
,
T. M.
,
X.
Yin
, and
A.
Gruber
,
2006
:
Variations in annual global precipitation (1979–2004), based on the Global Precipitation Climatology Project 2.5° analysis.
Geophys. Res. Lett.
,
33
.
L06705, doi:10.1029/2005GL025393
.
Soden
,
B. J.
,
R. T.
Wetherald
,
G. L.
Stenchikov
, and
A.
Robock
,
2002
:
Global cooling after the eruption of Mount Pinatubo: A test of climate feedback by water vapor.
Science
,
296
,
727
730
.
Spencer
,
R. W.
,
F. J.
LaFontaine
,
T.
DeFelicee
, and
F. J.
Wentz
,
1998
:
Tropical oceanic precipitation changes after the 1991 Pinatubo eruption.
J. Atmos. Sci.
,
55
,
1707
1713
.
Su
,
H.
, and
J. D.
Neelin
,
2003
:
The scatter in tropical average precipitation anomalies.
J. Climate
,
16
,
3966
3977
.
Sun
,
D-Z.
,
T.
Zhang
, and
S-I.
Shin
,
2004
:
The effect of subtropical cooling on the amplitude of ENSO: A numerical study.
J. Climate
,
17
,
3786
3798
.
Trenberth
,
K. E.
, and
D. J.
Shea
,
2005
:
Relationship between precipitation and surface temperature.
Geophys. Res. Lett.
,
32
.
L14703, doi:10.1029/2005GL022760
.
Trenberth
,
K. E.
,
G. W.
Branstator
,
D.
Karoly
,
A.
Kumar
,
N-C.
Lau
, and
C.
Ropelewski
,
1998
:
Progress during TOGA in understanding and modeling global teleconnections associated with tropical sea surface temperatures.
J. Geophys. Res.
,
103
,
C7
.
14291
14324
.
Trenberth
,
K. E.
,
J. M.
Caron
,
D. P.
Stepaniak
, and
S.
Worley
,
2002
:
Evolution of El Niño–Southern Oscillation and global atmospheric surface temperatures.
J. Geophys. Res.
,
107
.
4065, doi:10.1029/2000JD000298
.
Vimont
,
D. J.
,
J. M.
Wallace
, and
D. S.
Battisti
,
2003
:
The seasonal footprinting mechanism in the Pacific: Implication for ENSO.
J. Climate
,
16
,
2668
2675
.
Wigley
,
T. M. L.
,
2000
:
ENSO, volcanoes and record-breaking temperatures.
Geophys. Res. Lett.
,
27
,
4101
4104
.
Woodward
,
W. A.
, and
H. L.
Gray
,
1993
:
Global warming and problem of testing for trend in time series data.
J. Climate
,
6
,
953
962
.
Yang
,
F.
, and
M. E.
Schlesinger
,
2002
:
On the surface and atmospheric temperature changes following the 1991 Pinatubo volcanic eruption: A GCM study.
J. Geophys. Res.
,
107
.
4073, doi:10.1029/2001JD000373
.
Yang
,
F.
,
A.
Kumar
,
M. E.
Schlesinger
, and
W.
Wang
,
2003
:
Intensity of hydrological cycles in warmer climates.
J. Climate
,
16
,
2419
2423
.
Yin
,
X.
,
A.
Gruber
, and
P.
Arkin
,
2004
:
Comparison of the GPCP and CMAP merged gauge–satellite monthly precipitation products for the period 1979–2001.
J. Hydrometeor.
,
5
,
1207
1222
.
Zhang
,
Y.
,
J. M.
Wallace
, and
D. S.
Battisti
,
1997
:
ENSO-like interdecadal variability: 1900–93.
J. Climate
,
10
,
1004
1020
.

Footnotes

Corresponding author address: Guojun Gu, NASA GSFC, Code 613.1, Greenbelt, MD 20771. Email: ggu@agnes.gsfc.nasa.gov

1

The time series of the annual mean rainfall anomalies are detrended as we first focus on interannual-to-interdecadal variability.