Creswick Draft Flood Report - a critical review

by Roly Roper

Introduction

It has been a year since the floods, and the Flood Report has finally been published, or to give it its full title the Creswick Flood Mitigation and Urban Drainage Plan Draft Study Report.

All year I have held to a hope that at this point I could send a short e-mail to all concerned thanking them for a good job well done, and for devising workable solutions to the thorny problems I had identified early in the year, but to which I could see no easy solutions; after all, these guys are the experts.

That I could return to my technical pottering and tinkering secure in the knowledge that the required flood mitigation works for Creswick were in good hands and well in train, and that the good folk of Creswick would be spared the repeated flooding that rocked the town over the summer of 2010/11.

Sadly, it was not to be.

On the 11th of August I attended a town meeting where an abstract of the coming report was presented. About mid-evening a map was put up on the screen which perported to show where the flood flows had gone, and loud alarm bells started to sound in my head when I noticed a flow shown where I knew for certain that no such flow had occured; and worse, that I had discussed exactly this anomaly with the NCCMA and WaterTech on the ground early in the year.

This did not bode well. There is a saying that “you never find just one roach in a kitchen”, and experience shows that such a howler was unlikely to be alone.

Now the report is finally released my worst fears have been confirmed; it is colourful, well-presented garbage; by parts good, bad, and ugly.

Some very dubious assumptions have been given a good computer processing through sundry computer models until they look the authoratitive goods - but at the bottom line they are victim of the old computer adage GIGO - Garbage In, Garbage Out.

This critique is not intended to be an exhaustive examination of the report, there are far too many faults and questionable assumptions to fully catalogue, but to highlight sufficient material that casts serious doubt on the conclusions of the report, and sufficient to justify its rejection.

The Draft Report

If the basic input data that informs your computer model is not highly reliable there is a very good chance that computer processing will only reduce it to good looking, but meaningless, crud.

For example, the executive summary states “ The model was calibrated to within 0.1m³/s of the gauged peak flow at Clunes for both events.” (page v). Given the flow at Creswick was 100m³/s this implies an accuracy of 0.1% or 1 part in 1000, until we realise that there is a very loose relationship between the peak flows at Clunes and Creswick, some 23 creek kilometers apart.

This process of making a high accuracy match to a very rubbery number is called “spurious accuracy”. The overall accuracy of the figures produced for Creswick is not given nor estimated, but this review will demonstrate that the error band may be as high as 3-to-1, i.e. anywhere between one-third and three times.

Your results cannot be any better than your base data, and in this case the base data is minimal and consists of:

• the gauge at Cosgrave Reservoir which comprises only about a quarter of the upstream catchment;
• (a raingauge upstream of Cosgrove, 407237, which the report claims wasn't operating in January (2.3) but in fact was - data ignored);
• extrapolated rainfall data from Ballarat Aerodrome some 20-odd km distant ;
• and the Clunes gauging station “approximately 23 km downstream of Creswick” (Sect. 2.3) and operating well outside its calibrated range (sect 4.3.2).

Conditions in Ballarat are well known to often be quite different to conditions at Creswick, and during this year, for example, Ballarat experience a very heavy fall of hail that wasn't seen at Creswick. During the drought prior to the flood events lake Wendouree in Ballarat dried right out and even had a grassfire, while St Georges Lake only fell a couple of metres and the Creswick area was quite obviously more moist than areas only a few kilometres away, such as Ballarat and Clunes. Or as the report puts it “As the Ballarat gauge is on the other side of the divide, there is some uncertainty in the appropriateness of this temporal pattern for the Creswick Creek catchment” (p17, my emph.)

The data given for the flood peaks at the Clunes gauging station for September and January (Figs 4-6 and 4-7) are basically guesswork.

A half-metre dip is shown in the September record “however no anecdotal evidence was found to confirm this.” (p 15, my emph.); i.e. nobody at Clunes can confirm a half-metre drop in level recorded at Clunes, yet this is the data that is used to establish the flow 23km upstream, for a headwater catchment of 85km² of a total upstream catchment of 331.3km² (p 11), or 25% of the whole.

It is therefore quite remarkable that the flow estimate thus produced for the January flood is identical to my own measurement, viz 100 cubic metres per second at Water St bridge. Given that, it is more than passing curious that many other results produced post hoc by the report computer modelling process are at considerable variance to my own measurements and photographic evidence on the ground at the time (and supplied to NCCMA and WaterTech).

The midpoint of this whole process is a series of flood prediction maps (Appendix C), Average Recurrence Interval or ARI maps for floods of various frequency, starting with frequent 1-in-5 year events, and ranging up to rare 1-in-200 year events. These are generated from the data and assumptions at the input end, and are then used as the basis for testing the various mitigation proposals and their cost-benefit ratios.

We can reality test the model that produced these predictions by comparing them to what actually happened in 2010 and 2011, and see how well they match up.

Do they correctly post-dict what actually happened?

 

This is important because some very important reasoning and conclusions about damage and mitigation cost-benefit flow directly from these model outputs, and if they are significantly in error the recommendations that rest on them will also be significantly in error, ranging from having no value at all to being grossly misleading.

 

It is important to note that the report defines the September and January flood events as having an Average Recurrence Interval (ARI) of 25 and 35 years respectively (what is popularly known as a 1-in-25 and 1-in-35 year flood).

Apart from mentioning Creswick suffered several such rare events in only a few months, it is reasonable to ask if these flood frequencies are internally consistant with other parts of the report produced by the same modelling process?

There are actually quite a few places were this report is questionable, but only the more glaring internal inconsistancies need to be explored and highlighted to show the conclusions are worthless and that the report overall is a waste of both time and money.

Firstly we will look at several key points down along the creekline through the town, comparing the predictions output by the computer model with photographic data and witness reports of what actually happened.

The key points are;

• Northcott Park,
• Hammon Park,
• Victoria St where it meets Albert St,
• Park Lake and Castlemaine Rd,
• where Nuggety Gully creek/sludge channel meets Creswick creek.

Secondly we will look at some of the recommended mitigation measures in the light of the actual flood behaviour.

Northcott Park

The report states;

2.6 Other Background Data
High resolution (1 m) aerial images of Creswick were sourced from NearMap. For flood mapping, the most recent aerial imagery (20th January 2011) was used in this study.

As it happens this 20th January picture clearly shows one result of the January flood, three quartz scree deposits across Northcott Park dropped by the flood flows, shown here at X;

(*NCCMA/HSC)

Now given that the September and January floods were defined to be 25ARI and 35ARI respectively it is surprising to see that even the 50ARI prediction from the model still doesn't actually cover all the scree deposits.

(*NCCMA/HSC)

Naturally to drop these deposits the flood waters had to cover these areas, but when we look at the model ARI outputs we have to go to the 100ARI to just cover them, and when we compare ground level photographs it is obvious the actual level was more like that predicted by the 200ARI model output.

4/9/2010 12:24 Note car on Old Melbourne Rd, far right

14/1/2011 08:49 The long grass gives a clue to the depth

In both the September and January events the actual extent and depth of flooding was a great deal more than the model post-dicts.

Hammon Park

A similar observation on the September flood can be made about the model-predicted levels and the reality shown in photographs, but it is when we look at the predictions and the actuality of the January flood that a serious disparity appears.

Even the 50ARI does not predict Water St overtopping (we have to go to the 100ARI for the model to predict that).

50 year ARI (*NCCMA/HSC)

But what actually happened in the January event that is rated as only 35ARI?

Water St overtopping outside the Uni entrance.

Yet again the model fails the reality test. We can liken this to a car speedo that reads only 35KPH when we are actually doing 100KPH.

Victoria St

At the meeting on 11th August a diagram was presented which showed a surface flow down Victoria St into Albert St.

The report states:

5.4
...flows from Nuggetty Gully overtopped the bluestone wall along the primary school and pooled in the low lying north-east corner of the school field before running down Victoria Street. Some of this breakout flowed across Albert Street and flowed east towards Creswick Creek while the remaining flow travelled north-west towards the low lying area near the motel (through the Farmers Arms). (my emph.)

At the time I objected that this was a fantasy, that no such flow occured out of Victoria St into Albert St during either the September or January events.

20ARI (i.e. less than for September or January) (*NCCMA/HSC)

The yellow area shown between “421” and “418” in the middle above, and the flow shown on the left side of Albert St simply didn't happen.

Apart from being at Infolink (where the “418” is) myself during both the September and January floods, I have asked most of the Infolink volunteers and others who were in the area on those days, including an employee of the Farmers Arms, if they witnessed any flow down Victoria St, and so far all agree that it didn't happen. I have also reviewed a number of photographs and videos and many show emergency vehicles parked in that stretch of Victoria St and no surface flow.

4/9/2010 13:42 Victoria and Albert, no flows or mud trail on the western side. (Photo: Andrew Young)

Again, the model predicts a surface flow down Victoria St at 20ARI, yet the September flood which was supposed to be 25ARI in reality shows no trace at all of a flow down Victoria St that was modeled occuring some 10 hours before the above photograph (Table 4-4).

While there was flow down Victoria St in January it did not proceed on the surface past the intersection with Cambridge St.

This is a classic case of “cyber-crud”; placing trust in a computer output when there is clear evidence that it is wrong.

There seems to be some unwillingness to come to grips with the situation around the intersection of Victoria and Cambridge streets.

It is implausable that this entire flow down Victoria St then went via the old (confined, high resistance, and flooded) brick-lined drain to Creswick Creek (Fig A-1, p85), and there is evidence that it proceeded (by currently unknown pathways) to the top end of Cambridge St where it was reported “emerging from drains”, and directly into the cellar of the Farmers Arms hotel where water was reported as “jetting out of the walls”.

The fact that this curious sub-surface flow was first noticed as excessive water arriving at the intersection of Cambridge St and Cushing Ave has serious negative implications for one of the flood mitigation proposals - a levee around the Creswick motel.

Park Lake

On 4th of February 2011 Creswick suffered yet another flood following the landfall of Cyclone Yasi on the Queensland coast the previous day. This became known as the “night flood” and is not covered at all by the flood report, however that doesn't mean it was unknown to the report authors, being the topic of discussion during an on-the-ground inspection and in e-mails between myself and Rowan Hogan of the NCCMA as far back as March.

During this night flood Park Lake overtopped and sent a sheet of water, roughly two to three inches deep and the width of the road, swiftly down and across Castlemaine Rd to the southern side, upstream of the bridge.

Somehow the importance of this observation has been overlooked in the report. It should hardly come as a surprise since Hepburn Council employed two large pumps over the following days to pull the level of Park Lake down.

Still pumping on the 10/2/2011

While this night flood event was similar to the September event it is clear that the possibility of Park Lake overtopping was not considered in the modelling. This is a considerable oversight since Park Lake is part of the historic line of Sawpit Gully Creek, and this overtopping discharge has a very direct and negative impact on one of the mitigation works proposed later in the report.

Nuggetty Gully creek/sludge channel

Again the report states;

5.4
Nuggetty Gully overtopped its left and right banks downstream of Cushing Avenue, inundating Calembeen Park, the Cushing Avenue-Cambridge Street intersection and the low lying area near the motel. The low lying area around the motel and the Cushing Avenue-Cambridge Street intersection was inundated primarily from Creswick Creek backing up Nuggetty Gully and spilling out, the Nuggetty Gully peak flow had passed prior to this area becoming inundated.

This is at best a misrepresentation of what happened.

13/1/2011 15:18

At this time there is no flow incoming across Albert St or from the sludge channel above Cushing Av, the flow is away from the camera towards Creswick creek, and the flooding in the foreground is entirely due to water arriving down Cambridge St, behind.

The floodwater on the far side of Cushing Ave is basically overflow from Creswick creek, into which the sludge channel is flowing.

14/1/2011 09:29 Nuggetty Gully Ck/Sludge Channel downstream of Cushing Ave

The following morning the grass tells the story; this channel had not overtopped its banks directly into Calembeen Park and was still flowing into the “Lap Pool” billabong between the caravan park and the motel before joining the waters escaping into Calembeen Park.

This should not come as any sort of surprise because the flow is severly restricted at the Victoria St crossing by the “Hoover Dam”, and there were no signs that this channel exceeded its banks at any time during any of the events until it arrived at the pool created by Creswick creek backing up into the “Lap Pool” billabong area between the caravan park and motel.

Similarly, the timing of the Nuggetty flood peak is moot since the restriction at Victoria St effectively filtered the unrestricted flood peak to a constant flow.

4/9/2010 13:43 In September there is wide flooding but Nuggetty Gully Creek/Sludge channel can clearly be seen entering at right, still contained within its banks. Calembeen is already well flooded (note car lower right).

While the flow of the sludge channel is right-to-left, the flow escaping the pool above into Calembeen Park is a counter flow from top-left-to-lower-right.

The failure to properly understand the reality of the dynamics of what happened in this area, as opposed to relying on a computer model, has resulted in the construction of the ineffectual “Lap Pool” drain, the triplication of the culvert under the caravan park access road, and the construction of a levee where the floodwater was escaping from the “Lap Pool” billabong, above. The most likely outcome of these various works will be to facilitate flooding and make it worse (deeper).

Put simply, floodwater accumulated in this area because there was nowhere for it to readily escape. In January at the intersection of Nuggetty Gully Creek/sludge channel and Creswick creek the water depth was some two metres above the creek bed, or 1.3 metres above Cushing Ave.

From this observation it would seem the only effective solution is to significantly lower the bed of Creswick creek.

Albert St and overview

The report states;

5.4
As Creswick Creek continued to rise floodwaters broke out of bank in Hammon Park and the section between Water Street and Castlemaine Road Bridge. Creswick Creek overtopped its banks, flooding properties on the east side of Albert Street, North Parade, units in Semmens Village and properties along Castlemaine road upstream of the Castlemaine Road Bridge.

Again, this is at best misleading.

In fact Albert St properties were flooded because Creswick creek broke its banks starting at the obstruction presented by the Castlelmaine Rd bridge, and water pooled back from this obstruction.

On the afternoon of the 13th of January the Castlemaine Rd bridge had clearly reached its flow capacity;

13/1/2011 15:44 Water entering, up to the top of the culverts

...while the creek was still within its banks and crews were frantically sandbagging the western levee.

13/1/2011 15:42 Creek through trees, behind. In each flood this is where the creek first overtopped, just upstream of the Castlemaine Rd bridge.

The report states;

5.4
As Creswick Creek continued to rise further, floodwaters overtopped Albert Street and flowed towards the north-west inundating more properties along Albert Street, Cambridge Street and Cushing Avenue.

This statement is simply wrong in fact. Properties in Cambridge Street and Cushing Avenue were already inundated before the creek overtopped its banks.

The timed photographs above clearly show that the intersection of Cushing Ave and Cambridge St was already flooded at 15:18hrs, quite some time before Creswick creek broke its banks at Castlemaine Rd bridge (the levee was still being sandbagged at 15:44hrs, above).

This again highlights the failure to properly account for where the water at the foot of Cambridge St was coming from, because it certainly wasn't coming across Albert St, it wasn't coming down Cushing Ave from the sludge channel, and it certainly wasn't backup from Creswick creek because it was flowing towards the creek at that time.

Mitigation - Maginot Lines

The Maginot Line has become a catchphrase for defensive works that were facing the wrong way when actually needed.

As part of the mitigation proposals flowing from the modelling work are three levee banks or bunds;

• around Hammon Retirement Village,
• around properties facing Castlemaine Rd,
• and around the Creswick motel.

All of these proposals suffer from the same defect; measures that are supposed to defend against flood waters are in fact more likely to trap and contain flood waters and make the end effect worse for those they are supposed to protect.

Semmens Court

What is intended as a barrier to Sawpit Gully Creek is also a pocket with an opening facing up Moore St and the University grounds.

The proposed Semmens Court levee (fig. D-2, p105)

Looking down Moore St towards the opening in the proposed levee in the distance;

September - note depth of fence at far end

January

This is the fence of the University grounds directly opposite the opening in the proposed levee, demolished again by floodwater;

6/2/2011 Uni fence demolished by floodwaters

The photographs above should hardly require explanation, but what would obviously have happened during the September, January and February flood events is that the pocket formed by the proposed levee would have been rapidly filled by the Creswick creek overflow coming down Moore St.

Again, the proposed mitigation would have resulted in a worse outcome, deeper flooding, than actually occured.

Castlemaine Rd

Once gain, what is intended as a barrier to protect properties along Castlemaine Rd from Sawpit Gully Creek turns out to be another pocket that will trap floodwater, this time overflow from Park Lake (discussed above).

Proposed Sawpit Gully Creek (Castlemaine Rd) levee (fig D-4, p106, *NCCMA/HSC)

While I was unable to obtain photographs of the overflow from Park Lake during the Night Flood in February, I did walk through it, and it left its mark on the bank of the lake where it flowed down and on to Castlemaine Rd (quite overwhelming the local street drainage).

Overflow damage at Park Lake (from Castlemaine Rd, 11/2/2011)

This flow went directly and entirely into the open top of the proposed protective levee and would have quickly filled it to overflowing. Again, for those inside this proposed levee the outcome would have been a greater depth of floodwater than they actually suffered.

Cambridge St, Cushing Ave, Creswick Motel

There are several different proposals for a levee to cut off the Cushing Ave and motel area from Creswick creek, but all suffer from a similar defect of not accounting for flow coming from behind.

This is particlarly significant given the rapid build up of floodwater at the foot of Cambridge St during each of the flood events, and the failure so far to account for where this water was coming from, and therefore how to effectively block or divert it. (This is discussed above concerning the imaginary Victoria St flow.)

Conclusion

The modelling used to generate the theoretical extent of flooding under various conditions is seriously flawed and does not pass the test of post-diction, matching what various records show actually happened.

For example, the report itself admits that even with considerable forewarning they couldn't do a reliable field survey;

5.5 Design Flood Modelling
Field survey cross sections taken in May 2011 were used to model the modified channel sections in Creswick Creek and Nuggetty Gully. There was a significant gap between the surveyed cross sections from the Bowling Club footbridge down to the Ring Road. As the survey through this section was insufficient, the closest surveyed cross section at the footbridge, along with photographs and observations of the post flood channel profile were used to estimate the profile of the widened channel section.

That is one hell of an admission. This section is of critical interest, not least just upstream on the Castlemaine Rd bridge (see Albert St and overview, above), yet following the field survey the model still had to depend on interpolated estimates.

It is therefore unreliable and cannot be depended on to give a true pre-dictive picture of what may happen in future flood events.

As a result the conclusions drawn about the effectiveness of mitigation measures and their cost-benefit ratios derived from these faulty model outputs are themselves faulty and therefore moot and worthless.

It should be clear from the above review that compares the report findings and proposals with photographs of the reality that the report as a whole is very seriously flawed. The first test of any model is that it can correctly reproduce past events, and this modelling fails that test spectacularly.

The predicted ARI flood levels are grossly inaccurate, and therefore forward modelling of flood mitigation works built on those predicted flood levels is worthless.

There are a number of witness accounts that the historic bed of the creek below Water St (from the construction of the channel to the construction of the current bridges) was about a metre lower than today, and that the previous single timber bridge presented much less flow obstruction than the two culvert bridges now in place, yet this report gives a return to such a situation a very low cost-benefit score.

To apply a cost-benefit ratio to this report, it has been an expensive waste of time, producing conclusions that are worthless, and therefore has a cost-benefit ratio of zero.

Copyright

*NCCMA/HSC - material attributed thus is Copyright to North Central CMA and Hepburn Shire Council, and is used here under the Fair Dealing provisions of the Australian Copyright Act 1968 - Section 41 Fair dealing for purpose of criticism or review.

All other material unless otherwise attributed is Copyright to the author.