1. Authors: Pui, A., Lall, A., Sharma, A.
Acknowledgments: Australian Research Council
2. Positive Phase 1
Negative Phase
Positive Phase 2
The IPO is a coherent pattern of sea surface temperature (SST)
variability over the Pacific Ocean occurring on inter-decadal
timescales
Power, S.B., T. Casey, C Folland, A Colman, and V Mehta, 1999: Inter-decadal modulation of the impact of ENSO
on Australia. Climate Dynamics, 15, 319-324
Recent studies have shown that Flood Risk is not stationary
and is conditioned to the IPO phase
3. Kiem, Anthony S. et al., "Multi-decadal variability of flood risk." Geophysical Research Letters, 2003: 1-4.
Negative Phase :
Increased Flood Risk
Positive Phase :
Decreased Flood Risk
1 in 6 year flood (IPO –ve)
1 in 100 year
flood (IPO+ve)
Flood risk is influenced by the IPO.
Is this caused by changes in design rainfall
or antecedent conditions?
4. 1. Do antecedent wetness conditions influence
the design flood estimate?
2. Does design rainfall vary between opposing
IPO phases?
3. Do antecedent conditions vary between
opposing IPO phases?
5. Pi
Catchment Antecedent Conditions are approximated by
ANTECEDENT PRECIPITATION INDEX (API):
Where :
• P = annual maximum 24 hour rainfall amount
• i = day on which the annual maximum event occurs
• K = API exponential decay factor ( 0.92)
• n is the specified time lag (10)
K Pi-1
K2
Pi-2
K3
Pi-3
Kn
Pi-n
Cordery I . Antecedent wetness for design flood estimation. Civil Eng Trans I E Aust 1970; 12:181–5
6. High P is defined as above 50th
percentile annual rainfall maxima
9. Proportion:
Australia wide
Iratio > 1 = 0.64
East Australia
Iratio > 1 = 0.61
Not significant as per
field significance test
(0.95%)
Iratio > 1
Iratio < 1
11. APIratio > 1
APIratio < 1
Proportion:
Australia wide
APIratio > 1 = 0.78
East Australia
APIratio > 1 = 0.86
Significant as per field
significance test
(0.95%)
12. We have shown:
1. antecedent wetness conditions influence the
design flood estimate
2. Variation in design rainfall between opposing
IPO phases is not statistically significant
3. However, antecedent conditions vary
significantly between opposing IPO phases?
What does this mean for current
approaches to Design Flood Estimation?
15. • Future approaches for flood estimation need
to account for the non-stationary character of
antecedent moisture.
• This seriously compromises the assumption
that design rainfall leads to design floods (‘AEP
neutrality’).
• Also beware of rainfall-runoff models
calibrated to data from a single IPO state.
16. Alexander Pui
School of Civil & Environmental Engineering, UNSW
Email: a.pui@student.unsw.edu.au
Notes de l'éditeur
A very good afternoon, I’m Alex Pui from the University of New South Wales in Sydney, Australia. Today, I will be presenting about how the IPO affects Design Floods in Eastern Australia. Before I begin, I would like to acknowledge fellow contributors to this study – Assoc Prof Ashish Sharma and Allen Lal
I would like to briefly go through some background to put the objectives of this study into perspective. Recent studies have shown that Flood Risk is NOT stationary and is conditioned to the IPO phase. This raises a number of issues because current conventional flood design assumes climate stationarity – which essentially means that while weather from day to day varies randomly, underlying climate stats such as long term mean, variance and extremes are assumed to be constant. However, the discovery of the IPO, which is an ENSO-like coherent pattern of SST variability over the Pacific Ocean occuring on inter-decadal timescales has challenged this assumption. If you look at the time series of IPO shown here – from the 1920s – you will see 2 positive phases straddling a negative phase in between. What this means for East Australian climate – is wetter conditions during the IPO negative phase and drier conditions during the IPO positive phase.
To further illustrate my point about IPO modulation of flood risk – here is a diagram adapted from a study by Kiem et al. based on streamflow data from catchments in NSW state, South East Aust. The Y-axis represents magnitude of Flood, and the X axis shows average return interval (or inverse of AEP) –It is clear that during IPO Negative Phase, you have increased FLOOD RISK. During IPO +ve phase, you have decreased flood risks – note that the confidence intervals do not intersect. On closer inspection – you can also see that the 1 in 100 year flood estimated based on IPO +ve phase coincides with the 1 in 6 year flood for IPO-ve phase – thus amounting to a significant underestimate of flood risk. Now that flood risk has been shown to be influenced by the IPO, are these changes caused by CHANGES in design rainfall or antecedent condtitions?
The main objectives of this study can be set out as 3 hypotheses – they are: Do Antecedent Wetness conditions (antecedent conditions) influence the design flood estimate? Does Design Rainfall vary between opposing IPO phases? Do Antecedent Conditions vary between opposing IPO phases?
In order to investigate whether antecedent conditions influence the design flood estimate – we need to find some way of approximating catchment antecedent conditions. Exact measurements of soil moisture are fraught with difficulty due to difficulty in quantitatively estimating infiltration losses, catchment topography and evapo-transpiration rates. However, we do know that the entire circulation of water within a catchment is largely governed by the spatial and temporal distribution of rainfall. To this end, we use the API to approximate catchment antecedent conditions. The API equation gives rainfall occuring closer to our annual maximum rainfall event (P(i)) more ‘weight’ over catchment conditions compared to RF occuring further back in time. K is exponential decay factor related to evapotranspiration while n is the specified time lag (we used 10 days in this study).
For 128 catchments with daily catchment averaged rainfall and streamflow values, we obtain for each catchment, the annual max P and corresponding Q. We then grouped years where High P (note that we define High P as > 50 th percentile annual maxima) and High Q into one sample set and High P and Low Q into another set. We then estimated the API associated with High P, High Q case and conversely, the API for High P ,but Low Q case.
What we found is that almost all the catchments had API for the High P, High Q case greater than the API for High P , Low Q case – thus leading us to infer that High RF corresponding to High Flows is most likely when the catchment is in a wetter state.
Having established that API is higher for the transformation of higher RF to higher flows – we then move on to the second objective of the study, which was to see if design rainfall varied according to IPO phase, thus serving as the main driver of differences in flood risks from IPO + to – ve phase. For this section of the study, we utilized HQ daily rainfall data from 166 locations across Australia from 1920 - 2001. In particular, at each station we used a ‘ratios test’ to see if the 50 year 24 hour Duration Design Rainfall was larger during IPO – phase compared to IPO +ve phase. So, in other words, our null hypothesis is that no. of stations with I ratio > 1 would be roughly equal to or smaller than No. stations with I ratio < 1 if the IPO was identically distributed. Alternatively, no. of stations with I ratio > 1 should be significantly greater than no. stations with I ratio <1 if the IPO negative phase yielded higher design rainfalls.
Australia wide, the proportion of stations with I ratio > 1 is 64 percent. If we take a look at the Western Part of Australia first, the number of stations with I ratio < 1 and > 1 are roughly equal. However, it should be pointed out that there is limited spatial coverage and that the IPO is not known to affect West Aust RF. Let’s focus on the East – 61 % of stations returned I ratio > 1. While this no. is greater than 50% , it is not significant as per a field significance test based on bootstrap resampling – in other words – this results is not significantly greater than would have occurred by chance.
Now lets move on to the 3 rd objective of this study which is to investigate if the API varies according to the IPO Phase. Again, we repeat the analysis conducted on Design Rainfalls at each station we using ‘ratios test’ to see if the mean API was larger during IPO – phase compared to IPO +ve phase. So, in other words, our null hypothesis is that no. of stations with API ratio > 1 would be roughly equal to or smaller than No. stations with API ratio < 1 if the IPO was identically distributed. Alternatively, no. of stations with API ratio > 1 should be significantly greater than no. stations with API ratio <1 if the IPO negative phase is indeed associated with wetter antecedent conditions.
Australia wide, the proportion of stations returning API ratio > 1 is 78%, which is larger than the 64% obtained for Design Rainfall. However, if we just focus on the East, we can see that 86% of stations returned API ratio > 1. This number is statistically significant as per the field significance test and would not have resulted based on chance alone. It should be noted that this study assumed that the stations were not spatially correlated which is not true, BUT this result still provides strong evidence that API distributions are not identically distributed between IPO phases.
In conclusion, we have found that : Antecedent wetness conditions do influence the design flood estimate. To what degree, we are uncertain due to use of approximation methods on coarse (daily) scale. That any variations in Design Rainfall between opposing IPO phases is not statistically significant. However, antecedent conditions do vary significantly between opposing IPO phases – and may thus contribute to the observed differences in flood risk.
What do these conclusions mean for Design Flood Estimation Practice? In Australia, the conventional way to flood design from rainfall is to use the Design Storm Approach –which involves developing IFD relationships for a given region using Annual Maxima. A flood frequency curve is then derived from these IFD relationships. Alternatively, one could use continuous simulation which involves using a rainfall –runoff model to convert continuous rainfall to continuous flows. The annual max flows is then obtained from the synthetic flow sequence and lastly, a FFC is derived on these annual max Qs.