TRMM and the global interannual variability of rain over the past five decades

Until 1979, the evidence linking El Niño with changes in rainfall around the world came from rain gauges measuring precipitation over land and a handful of islands. Before the launch of the Tropical Rainfall Measuring Mission (TRMM) in November 1997, the remote sensing evidence gathered since 1979 was confined to ocean rainfall because of the very poor sensitivity of the instruments over land. In this paper we analyze the first five years of the first global land and ocean remote-sensing record of rainfall. We distill the information into a few objective indices, the first principal components of the rain anomaly, and extend them back in time to show how the global remote-sensing record implies that El Niño is indeed the major driver of the global interannual variability of rainfall. The El Niño / Southern Oscillation (ENSO) phenomenon has been characterized using a few indices calculated from those observables which are most directly related to the physical mechanisms that govern it, namely • the “El Niño” indices representing an average of the sea surface temperature anomaly within a specified region of the equatorial Pacific Ocean, such as the rather popular “Nino-3” index (see e.g. Trenberth and Stepaniak, 2001), • the “Southern Oscillation” indices representing the normalized anomaly in the difference between the atmospheric pressure over the Eastern Pacific and that over the Western Pacific / Indian Ocean, and almost always calculated using the measured sea-level pressures (SLP’s) at Tahiti and at Darwin, as in the case of the “Troup SOI” (Troup, 1965), • the indices directly representing the anomaly in the Walker Circulation (Bjerknes, 1969), such as the 850-mb trade wind index (representing the behavior of the near-surface wind, and calculated from the “reanalysis” of the atmospheric fields estimated from large-scale data-assimilating models – see Latif et al., 1998) or the 200-mb zonal wind at the equator (representing the uppertropospheric wind, and whose anomalies in the tropics have a most direct effect on the global circulation – see Arkin, 1982). Thus, the most pertinent observables are the sea surface temperature, the sea-level pressure, and the boundary-layer or upper-tropospheric winds, all of them observed in the tropical Pacific. Also useful are the depth of the thermocline in the Eastern Pacific, and the outgoing longwave radiation 2 over the tropical Pacific (as a gauge of the frequency and depth of the convective systems). Yet the observable which has the most immediate impact on people around the globe is neither the strength of the trade winds nor the sea surface temperature nor the atmospheric pressure. Rather, it is the effect of ENSO on the regional change in local rainfall patterns. Numerous studies have documented the link between ENSO and rainfall in many regions of the globe, associating the warm phase with drought conditions in some cases, unusually abundant precipitation in others. The most extensive and detailed study of this kind is undoubtedly Ropelewski and Halpert’s (1987, 1988), in which the change in the rainfall sampled over land and island stations within several regions around the globe is carefully analyzed depending on the prevailing ENSO conditions. Indeed, consistent correlations are found between the rain anomaly and the ENSO phase in most of the regions considered (Ropelewski and Halpert, 1987). One could contemplate synthesizing these observations into a global ENSO precipitation index, which would be calculated by adding the rainfall anomalies in all areas which experience excess rain during warm ENSO phases and subtracting the anomaly in those areas which experience a deficit. The problem with such a proposition is that regions which experience excess rain during warm phases do not always experience rain deficits during cold phases and vice versa, as Ropelewski and Halpert (1988) observed. In other words, the maps of the rainfall anomalies during warm or cold ENSO phases do not appear to be mirror images of one another. An equally serious problem with the proposition of subtracting deficit areas from excess areas is that, by subjectively selecting only those areas which have a consistently sustained correlation with ENSO, one would be ignoring those regions which are less significantly affected by the phenomenon, and which could be responsible for a large proportion of the global rainfall variability. This problem was addressed by the objective study of Dai et al (1997), in which a global set of yearly rainfall compiled from land/island station data from 1900 to 1988, was analyzed. After subtracting from the values for each station their mean from 1900 to 1988, and normalizing by the corresponding standard deviation to prevent regions with a large overall variation from overwhelming the subtle change in regions with low rainfall, a principal component analysis of the resulting normalized anomaly was performed. Dai et al found that the first principal component of the normalized annual rain anomaly over the period 1900-1988 was very well-correlated indeed with the bi-monthly average sea-surface temperature anomaly over the equatorial Pacific. While these results are based exclusively on land/island station data which leave vast expanses of ocean unrepresented, they are compelling indicators that ENSO is a very important factor in the variability of rainfall. Thus the accumulated evidence begs the question: how can one objectively quantify the importance of ENSO in the global (land and ocean) variability of surface rainfall? 3 Until the work of Arkin (1979), Xie and Arkin (1997) and that of Adler et al. (1993) and Huffman et al. (1997), this question had remained unaddressed largely because the systems required to monitor precipitation over the oceans simply did not exist. This dire situation changed dramatically in the 1980’s with the availability of data from low-earth-orbiting multiple-frequency microwave radiometers such as the Special Sensor Microwave Imagers (SSMI), and from geostationary visible/infrared (Vis/IR) imagers. The latter are useful in the sense that they can gauge the height of the cloud tops (and hence, at least in convective systems, the depth of the clouds, and hence, allowing for a quite large uncertainty in one’s estimates, the amount of rain which these clouds are producing – see Arkin, 1979), with frequent updates. With less frequent updates, the low-earthorbit microwave radiometers provide a handful of radiances in which the surface emissivity effects and the competition between the absorption/emission and the scattering from rain and ice can be approximately sorted out to produce an estimate of the rainfall amount at rather poor resolution. Acknowledging the limitations of SSM/I and geostationary IR imagers, Adler et al. (1993) sought to combine them in order to take advantage of the strengths of each and build a “merged” IRSSMI/surface-gauge dataset of truly global rainfall, the Global Precipitation Climatology Project (GPCP; see Huffman et al., 1997). An “ENSO precipitation index” (ESPI) is currently calculated from GPCP, essentially by subtracting the precpitation anomaly over the region around the Maritime Continent (10◦S – 10◦N × 90◦E – 150◦W) from that over the eastern Pacific (10◦S – 10◦N × 160◦E – 100◦W) – the exact boundaries of the boxes are “dynamically” calculated in real-time to maximize the contrast. By design, ESPI correlates very well with the “El Niño” and “Southern Oscillation” indices (Curtis and Adler, 2000). Going one step further, Xie and Arkin (1997) folded in numerical model predictions as well, and produced the “CMAP” global dataset of monthly surface rainfall estimates from 1979 to 1995 on a 2.5◦ grid. Their maps of the seasonal difference (warm phase cold phase) of the rainfall anomaly averaged over the 17 years of data incorporated in CMAP confirmed many of the results of Ropelewski and Halpert, and Dai and Wigley’s principal component analysis of the normalized annual rain anomaly (Dai and Wigley, 2000) yielded a 20-point time series (CMAP had by then been updated to 1998) which matches the SOI over that period remarkably well. These first truly global results depend ultimately on the reliability of the sources of the data, namely the IR and SSMI estimates. As we have already noted, the former relies on the statistically-derived correlation between cloud top heights and surface rain, which has a large intrinsic uncertainty and whose applicability depends on precipitation type. While SSMI is more directly sensitive to the rain itself, the poor resolution of the instrument forces one to make homogeneity assumptions about the precipitation which are likely to introduce large biases in the estimates (because the average rain quantities one would like to estimate are related in a very nonlinear way to the average radiances one measures). Most important, over land, the relation between 4 either the IR or the microwave radiances and the surface precipitation is tenuous at best. It is precisely to remedy the shortcomings of these systems (their poor resolution and their lack of much direct sensitivity to the vertical structure of precipitation) that the Tropical Rainfall Measuring Mission (TRMM) was conceived and the TRMM satellite launched in November, 1997 (Simpson et al., 1988). In addition to having a very low resolution-enhancing orbit (originally 350 km), TRMM’s advantage is that it carries the first spaceborne precipitation-profiling radar (PR), in addition to a nine-channel microwave radiometer (TMI) and a visible/infrared imager. Although the clutter from the overwhelming surface echo severely limits the swath of the PR and, therefore, limits its ability to sample the precipitation as frequently as a radiometer, the vertical detail with which it can probe the atmosphere, its insensitivity to the characteristics of the surface, and its high horizontal resolution (' 4 km) make it an ideal instrument with which to “calibrate” the rain retrievals of the radiometer within the narrow common swath of the radar (Haddad et al., 1997), and subsequently carry this calibration over to the TMI-only retrievals over the wide swath of the radiometer (Adler et al., 2000). Of particular interest are the surface rainfall estimates produced by this “TRMM-combined” radar/radiometer algorithm from December 1997 until the present. These estimates are available in the form of monthly rain maps over the region between 40◦S and 40◦N at a resolution of 5◦×5◦. To synthesize the information in these maps objectively, we performed principal component analyses of the rainfall estimates and of their anomalies. Figure 1 illustrates the results. It displays the coefficients −150 −100 −50 0 50 100 150 40