Anthony S. Kiem, Stewart W. Franks, and George Kuczera
Recent research has highlighted the persistence of multidecadal epochs of enhanced/reduced flood risk across New South Wales (NSW), Australia. Recent climatological studies have also revealed multi-decadal variability in the modulation of the magnitude of El Niño/Southern Oscillation (ENSO) impacts. In this paper, the variability of flood risk across NSW is analysed with respect to the observed modulation of ENSO event magnitude. This is achieved through the use of a simple index of regional flood risk. The results indicate that cold ENSO events (La Niña) are the dominant drivers of elevated flood risk. An analysis of multidecadal modulation of flood risk is achieved using the interdecadal Pacific Oscillation (IPO) index. The analysis reveals that IPO modulation of ENSO events leads to multidecadal epochs of elevated flood risk, however this modulation appears to affect not only the magnitude of individual ENSO events, but also the frequency of their occurrence. This dual modulation of ENSO processes has the effect of reducing and elevating flood risk on multi-decadal timescales. These results have marked implications for achieving robust flood frequency analysis as well as providing a strong example of the role of natural climate variability. INDEX TERMS: 1821 Hydrology: Floods; 4215 Oceanography: General: Climate and interannual variability (3309); 4522 Oceanography: Physical: El Niño; KEYWORDS: Climate variability, El Niño/Southern Oscillation (ENSO), flood frequency, Inter-decadal Pacific Oscillation (IPO), Pacific Decadal Oscillation (PDO), multidecadal. Citation: Kiem, A. S., S. W. Franks, and G. Kuczera, Multi-decadal variability of flood risk, Geophys. Res. Lett., 30(2), 1035, doi:10.1029/2002GL015992, 2003.
The quantification and understanding of hydrological variability is of considerable importance for the estimation of flood risk. At present, traditional methods are largely empirical in that annual maximum floods are assumed to be independently and identically distributed [Franks and Kuczera, 2002]. Despite the development of rigorous Bayesian frameworks to assess the uncertainty of flood risk estimates, these techniques have not acknowledged the possibility of serial correlation within periods of elevated or reduced flood risk [cf. Kuczera, 1999]. However, recent research has highlighted the persistence of multi-decadal epochs of enhanced/reduced flood risk across New South Wales [Erskine and Warner, 1988; Franks, 2002a, 2002b; Franks and Kuczera, 2002]. In particular, Franks and Kuczera  demonstrated that a major shift in flood frequency occurred around 1945. Previous authors have noted that the mid-1940’s corresponded to a change in both sea surface temperature anomalies as well as circulation patterns [Allan et al., 1995]. Franks [2002b] showed that the observed change in flood frequency could be objectively identified as corresponding to this shift in climate parameters. Furthermore, it was shown through the use of a simple index of regional flood risk that the observed shift in flood frequency was statistically significant at the <1% level.
In addition to hydrological observations of changing flood risk, recent climatological studies have also revealed multi-decadal variability in the modulation of the magnitude of El Niño/Southern Oscillation (ENSO) impacts. Power et al.  have investigated marked temporal changes in ENSO correlations to Australian rainfall records. The temporal stratification of the rainfall sequences was achieved according to what has been termed the Inter-decadal Pacific Oscillation (IPO). The IPO was defined by anomalous warming and cooling in the Pacific Ocean and is similar to the Pacific Decadal Oscillation or PDO [Mantua et al., 1997; Franks, 2002a]. Importantly, Power et al.  demonstrated that individual ENSO events (ie. El Niño, La Niña) had stronger impact across Australia during the negative phase of the IPO, implying that there exists a multidecadal modulation of the magnitude of ENSO events.
The study aims to extend the analysis of Franks [2002b] to assess the role of ENSO processes and their multi-decadal modulation, in dictating flood risk across New South Wales (NSW), Australia. In this paper, a derived regional index of flood risk [Franks, 2002b] stratified according to ENSO classifications based on the NINO3 index. The index is then further stratified according to the multi-decadal IPO classifications. The stratified flood frequency data are analysed using Bayesian flood frequency analysis to quantify uncertainty on quantiles and thus elucidate the key controls on NSW flood risk.
The streamflow data used in this study were obtained from the PINEENA database, developed and managed by the NSW Department of Land and Water Conservation. 40 records spanning 1924 to 1999 were deemed suitable in terms of the length and continuity of record. If the flood gauges were perfectly correlated, treating them as entirely uncorrelated would imply 40 independent records with any inferred change having unwarranted statistical support. To avoid the issue of spatial correlation, the 40 flood records were collapsed into a regional flood index following Franks [2002b].
Stratification of the regional flood index record according to ENSO classifications was made using the monthly NINO3 index. Every year from 1924 to 1999 was given an ENSO classification based on the six-month October to March average NINO3 value. This method and index combination has previously been demonstrated to be the most robust for the time period being investigated [Kiem and Franks, 2001].
The Inter-decadal Pacific Oscillation (IPO) is the coherent pattern of sea surface temperature (SST) variability occurring on inter-decadal time scales over the Pacific Ocean [Folland et al., 1999; Power et al., 1998, 1999; Allan, 2000]. In classifying the different IPO phases, Power et al.  used the thresholds of ±0.5 to distinguish positive, neutral and negative phases. Figure 1 shows the time series of the IPO over the period of flood data employed in this study. As can be seen, during this period there have been three major phases of the IPO: Two positive phases (IPO > 0.5) and a negative phase (IPO < -0.5).
The regional flood index was stratified according to El Niño and La Niña extremes, as defined by the NINO3 index. The stratified data were then subjected to a Bayesian flood frequency analysis in order to properly account for parameter uncertainty [Kuczera, 1999].
To assess the role of ENSO extremes Figure 2 presents the flood frequency under El Niño and La Niña conditions along with the associated 90% confidence limits. From this plot it can be readily seen that much higher flood risk must be associated with La Niña events as opposed to El Niño. Also immediately apparent is the degree of separation of the confidence limits indicating a highly statistically significant difference between the two ENSO extremes. Although not shown in Figure 2 for the sake of clarity, the flood frequency distribution associated with neutral ENSO events lies between the two extremes.
Given the clear role of La Niña events in flood risk identified in Figure 2, to test the hypothesis that the IPO modulates the magnitude of La Niña events, as suggested by Power et al. , a stratification on La Niña under different IPO phases is required. To achieve this test, the regional index is stratified according to La Niña events occurring under negative IPO phase (<-0.5) and then according to La Niña events occurring under neutral and positive IPO phases (>-0.5). Figure 3 shows the resultant flood frequency curves. As can be seen, the frequency curve associated with La Niña events under IPO negative (<-0.5) is markedly higher than the flood frequency associated with all other La Niña events. The 90% quantiles marginally overlap suggesting significant difference between the conditioned distributions.
Finally, given the observed persistence of IPO phases, it is desirable to assess the variability of flood risk under the different IPO phases irrespective of interannual ENSO events. Figure 4 shows the flood frequency curves for IPO negative (<-0.5) against non-negative IPO phases. Again, it can be seen that IPO negative phase corresponds to a much increased flood risk when compared to the non-negative phases of IPO. It is therefore clear that monitoring of the multi-decadal IPO phase may provide valuable insight into flood risk on multi-decadal scales, whilst the joint occurrence of inter-annual La Niña events within the IPO negative phase represents further elevated flood risk.
Given the strong control on flood risk exerted by La Niña events and modulated in their magnitude by multidecadal IPO processes, it is intuitive to examine the frequency of occurrence of such high magnitude events. Table 1 shows the IPO phases that have occurred between 1924 to 1999, and the frequency of ENSO events under each of these phases. Note that the IPO phases are as defined earlier in this paper (with 1924–43 denoted IPO > 0.5(1) and 1979–97 denoted IPO > 0.5(2)).
Immediately apparent from Table 1, it can be seen that IPO negative phases tend to be biased towards an increased frequency of La Niña events. It therefore appears that the multi-decadal processes as represented by the IPO may modulate the frequency of ENSO events as well as the magnitude of their impact.
To test the statistical significance of the dependence of ENSO event frequency on IPO phase a simple test of proportions was applied [Hogg and Tanis, 1988]. It is assumed that the sampling distribution of the proportion of El Niño events occurring within any IPO phase can be approximated by a normal distribution with a mean of p and a variance of p(1-p)/n, where p is the proportion of El Niño events that have occurred within each IPO phase, calculated using p = y/n, where y is the number of El Niño events that have occurred and n is the total number of years in the IPO phase being investigated. This was repeated for La Niña and Neutral events.
In order to determine whether the probability (P1) of a given ENSO event occurring during one IPO phase was significantly different to the probability (P2) of the same ENSO event occurring during a different IPO phase the following statistical test was used. Let y1 represent the number of El Niño events, for example, that occurred in the n1 years when IPO was positive and y2 the number of El Niños that occurred in the n2 years when IPO was negative. A test statistic used to test the hypothesis that P1 equals P2 is:
Table 2 shows the results obtained when the probability of El Niño, La Niña and Neutral events occurring in different IPO phases were compared. The p-value in Table 2 indicates the probability that the frequency at which a given ENSO event occurs in one IPO phase is equal to the frequency at which the same ENSO event occurs at in a different IPO phase.
Table 2 shows that when the negative IPO phase is compared with the positive IPO phases, the frequency at which La Niña events occur is significantly higher when the IPO is negative. Table 2 also demonstrates that the number of Neutral events that occur when the IPO is positive is significantly higher than when the IPO is negative, indicating a higher rate of occurrence of the ENSO extremes (El Niño or La Niña) when the IPO is negative. Table 2 also shows that no significant difference is observed between the two positive IPO phases in either El Niño, La Niña or Neutral events.
It is therefore apparent from these results that the IPO negative phase, representing cool anomalies in the midlatitude Pacific Ocean SST, contain a statistically significant proportion of La Niña events. This indicates a predisposition of the negative (cool) IPO phase towards increased frequency of cool La Niña events. Thus the IPO modulation of flood risk across NSW appears due to its modulation of the magnitude and frequency of strong La Niña events.
This dual modulation has the effect of reducing and elevating flood risk on multi-decadal timescales. Indeed, the 100 year average return interval derived by traditional empirical analysis returns a value of 3.17 for the regional index. However, within the IPO negative phase the regional flood of this magnitude occurs with a return period of 15 years. Given the observed persistence of IPO phases beyond this period, it seems that the ‘100 year flood’ is most likely to occur during this period. Instrumental evidence bears testament to this in the occurrence of clusters of high magnitude floods. This apparent clustering, statistically anomalous under the traditional paradigm, is entirely intuitive within the concept of multi-decadal modulation of ENSOinduced flood extremes.
This paper has sought to explain the temporal changes previously observed in NSW flood risk over the period 1924– 99. This has been attempted through an analysis of ENSO processes and their modulation via multi-decadal SST as represented through the IPO index. The results have shown that La Niña events predominate the long-term flood risk. Moreover, multi-decadal modulation of ENSO processes result in extended periods of elevated flood risk. This paper has demonstrated that these multidecadal processes may modulate the frequency of ENSO extremes as well as the magnitude of their impact.
There are a number of important implications associated with these insights;
Finally, it is worthwhile to note that the results shown here represent one manifestation of natural climate variability. The quantification of hydrological variability represents an integrated measure of natural climate variability. Flood risk is a key hydrological variable in terms of social and economic importance. At present it is unclear whether the multi-decadal modes of sea surface temperature variability are an internal artifact of the ocean-atmosphere system, or forced by external variations in ultraviolet irradiance [Latif and Barnett, 1994; White et al., 1997; Reid, 2000; Franks, 2002a]. In either case, the data presented here might be used as a performance indicator for General Circulation Models that attempt to project the influence of anthropogenic factors on climate. If these models can successfully represent such historic variability in a key hydrological variable, then increased confidence might be placed in the simulation of future, anthropogenically forced climate.
Acknowledgments. This research was funded under the Australian Research Council SPIRT grant, ‘Development of a rainfall model for water resources management’ (SWF) with collaborative funding from Hunter Water Corporation.
Allan, R. J., ENSO and climatic variability in the last 150 years, in El Niño and the Southern Oscillation: Multi-scale Variability, Global and Regional Impacts, edited by H. F. Diaz and V. Markgraf, Cambridge University Press, Cambridge, UK, 3-56, 2000.
Allan, R. J., J. A. Lindesay, and C. J. C. Reason, Multi-decadal variability in the climate system over the Indian Ocean region during the austral summer, J. Climate 8(7), 1853-1873, 1995.
Erskine, W. D. and R. F. Warner, Geomorphic effects of alternating flood and drought dominated regimes on a NSW coastal river, in Fluvial Geomorphology of Australia, edited by R. F. Warner, Academic Press, Sydney, Australia; 223-244, 1988.
Folland, C. K., D. E. Parker, A. W. Colman and R. Washington, Large scale modes of ocean surface temperature since the late nineteenth century, in Beyond El Niño: Decadal and Interdecadal Climate Variability, edited by A. Navarra, Springer, Berlin, 73-102, 1999.
Franks, S. W., Assessing hydrological change: deterministic general circulation models or spurious solar correlation?, Hydrol. Proc. 16(2), 559-564, 2002a.
Franks, S. W., Identification of a change in climate state using regional flood data, Hydrol. and Earth System Sci. 6(1), 11-16, 2002b.
Franks, S. W. and G. Kuczera, Flood frequency analysis: Evidence and implications of secular climate variability, New South Wales, Water Resour. Res. 38(5), doi:10.1029/2001WR000232, 2002.
Hogg, R. V. and E. A. Tanis, Probability and Statistical Inference, 3rd Edition, Macmillan Publ., New York, 1988.
Kaplan, A., M. A. Cane, Y. Kushnir, A. C. Clement, M. B. Blumenthal and B. Rajagopalan, Analyses of global sea surface temperature 1856-1991, J. Geophys. Res.-Oceans 103(C9), 18567-18589, 1998.
Kiem, A. S. and S. W. Franks, On the identification of ENSO-induced rainfall and runoff variability: a comparison of methods and indices, Hydrol. Sci. J. 46(5), 715-728, 2001.
Kiem, A. S. and S. W. Franks, Multi-decadal variability of drought risk, Hydrol. Proc. (in review), 2003.
Kuczera, G., Comprehensive at-site flood frequency analysis using Monte Carlo Bayesian inference, Water Resour. Res. 35(5), 1551-1558, 1999.
Latif, M. and T. P. Barnett, Causes of decadal climate variability over the North Pacific and North America, Science 266, 634-637, 1994.
Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace and R. C. Francis, A Pacific Interdecadal Climate Oscillation With Impacts On Salmon Production, Bull. Amer. Meteorol. Soc. 78(6), 1069-1079, 1997.
Power, S., T. Casey, C. Folland, A. Colman and V. Mehta, Inter-decadal modulation of the impact of ENSO on Australia, Climate Dynamics 15, 319-324, 1999.
Power, S., F. Tseitkin, S. Torok, B. Lavery, R. Dahni, and B. McAvaney, Australian Temperature, Australian Rainfall and the Southern Oscillation, 1910-1992 - Coherent Variability and Recent Changes, Aust. Met. Mag. 47, 85-101, 1998.
Reid, G. C., Solar variability and the Earth’s climate: Introduction and overview, Space Sci. Rev., 94, 1-11, 2000.
White, W. B., J. Lean, D. R. Cayan, and M. D. Dettinger, Response of global upper ocean temperature to changing solar irradiance, J. Geophys. Res. 102, 3255-3266, 1997.