Transcript for Maryborough water resource management project

[Speaker: Jacqui Lloyd (DEECA)]

Doctor David Robertson is a principal research scientist at CSIRO's Environment Research Unit. David leads the water forecasting team. He has extensive research experience in water forecasting, hydrology and water resources. Thank you for presenting today, David.

[Speaker: David Robertson (CSIRO)]

Thanks for the opportunity. I feel like a bit of an imposter here because I shouldn't be presenting this. It should be Zitian who's done most of the work, unfortunately (inaudible), but I do want to acknowledge the rest of the team that were involved. So there’s Zitian, Francis Celine from my team Pat Russell and John Fredelja associated with Central Highlands Water. And Sandra and Geoff provided useful insights along the way as well.

So interesting. I'm just grab one for one forward scroll 2 forwards. Sorry, yeah, I don't know what happened there.

So I guess the purpose of this piece of work was really to try and take a lot of the pieces of the puzzle that we've done through the research and look at could be applied practically to a real, real-world problem.

So, you know, we know that that climate change will impact water security and that, you know, when you're assessing water resources systems and particularly assessing things like system augmentations, you need to under understand potentially climate change, climate variability, and you need to have an understanding of the baseline that you're working from.

So, you know, in the SEACI, sorry, the VicWaCI reports proves how old I am. We, we see that this, you know, if we look at different periods of record, we, we could derive different baselines from that. And that the guidelines talk about for record baseline and a post 97,1975 baseline and, and a more recent baseline as well. So this study was really about trying to unpick what are the consequences of, of making choices about different baselines for a water resources assessment. And then how does climate change lay overlay on top of that.

So if we move on the we'll to the next slide. Yep. So the system we were looking at is the Maryborough water supply system. Maryborough, we're all Victorians, mainly Victorians. We know where Maryborough is. There's about 11,000 people that this system is supplying. It surprised me. It's quite a complicated system of, of water supply. There's sort of three main water resources. There's a,there's a local supply system that's got a couple of reservoirs with different, different inflows. There's, I'm going to call it a regional surface water system. They can extract water from Tullaroop reservoir. And then when there's shortage in both of those, there's also a groundwater, A groundwater system that can be, can be tapped.

So quite a complicated system, but some background analysis that Central Highlands at water had done and said, you know, maybe this isn't enough if, if and we might be be exceeding our service level specifications. So there's when, when we started this piece of work, we had a, as a, as a model of the system that was coded up in the source modelling platform. So if you look at the system, there's a, there's a map of it on the right. We have a number of key inflows, the Stony Creek and Mccallum's Creek, there's Tullaroop Creek, there's, there's reservoirs and then there's demands that are in Maryborough that, that are driven in this model by, by climate time series.

So what do we do? Well, what we tried to do in, in, we did this work in kind of two stages.The first work was to look at evaluating the performance of the system, assuming different baselines, baseline climates. So to do this, rather than just going and using the historical record, we used a stochastic data generation method to generate data that represented those different baseline.

So we had this stochastic data model. I'll talk about that briefly. We parameterized it over a full the full entire historical climate record, a post 1975 climate record, and a post 1993 record. We used 1993 to get roughly 30 years of which is, you know, a climatology, an approximate climatology. So we used that stochastic model parameterized to generate different stochastic data. We then ran that through should be GR4J and inverted commas It was a modified version of GR4J so that we the rainfall runoff model so that we could simulate long sequences of cease to flow conditions which had been observed, which off the shelf GR4J couldn't do. And then we ran those inflow series and climate data through the water supply system to understand the system performance.

The second part, next part was to look at under future climate change, what the system performance would be and then look at augmenting the system and the consequences of augmenting now versus delaying and augmentation into the future. So to do that, we used we took one of those climate data sets as a baseline. And then rather than doing the traditional let's move, have a stationary future climate, we looked at transitioning the climate from now to some future time.

And we embedded that future trend in that fire in in the stochastic side of generation model. Rather than choosing a single future, we looked at different sensitivities. So we looked at different changes of rainfall and temperatures and so forth. And then we generated our run off and run it through the system. We then having all of those looked at providing a little bit of an augmentation to the system at different times into the future and the consequences on some key metrics.

We move to the next slide. We'll just talk through the first bit and then I'll come to the climate change part forward. Yeah. So don't just not going to talk any detail here.

We had a stochastic data generation model. We ran it at monthly time steps. We're generating 12345678 climate series, both rainfall potential, evapotranspiration and temperature. We were maintaining inter variable correlations between all of those time series. You can see there's some examples of what was generated in the grey lines and some corresponding the observations for a particular. That are just overlaid on top of that.

So we took that stochastic data generation model and we said, is it any good? It looks OK. It's not perfect, but you know, there was a number of different checks we did on it. So that's great. We had our, had our, had our models. We then ran it and we fitted it to those three different periods. We then went and generated some data and just had a look at the data that was being generated.

So here we've got three different time series. We've got a range full time series, we've got an evapotranspiration time series and we've got a temperature time series. And the verbal panel we're showing here the monthly main for each of those. So we've got January to December on the horizontal axis and the quantities on the vertical axis. And we've got three sets of box plots, the green ones which were fitted the data to the whole, fitted our model to the whole record. The orange ones are the post 1975 record and the blue ones to the post 1993 record. And immediately, you know, there's, there's some interesting things in, in evapotranspiration and temperature. But you know, basically there's an increase across those 3 periods.

But if we look at the, the rainfall, we can see some interesting dynamics in what it's, it's picked out. So what we see is that using the post 1993 data to fit our model, it's not surprising given the conversations to here today, we can see that the rainfall for April to November is actually lower in a more recent period and the stochastic data is represented. The other interesting part of it is that we're actually seeing increases in December and January rainfall using that that more recent period.

So that's some, we've got some stochastic climate data. OK, what is that? How does, how do we then translate that to runoff?

Yeah, so we've got the three different runoff time series here. And this is where Pat and John started to get really interested is that using those three different sets of climate forcing, we get very different runoff characteristics. And in fact, if we just use a benchmark period to of, of the most recent 30 years to generate that the runoff, we get a lot, a lot less runoff than if we use the full record. And that's particularly noticeable in the really wet seasons. The, the, the general pattern of the hydro of the runoff though is not, is not changing over those, those different parts. That's great, Phil, just tinkering around with the forcing of the system.

What does the system performance look like? So you know, there's lots of things that we can pull out of that that that model out of the source model. The first thing we started looking at was, you know, the total was water resources and how that broke it down under the different systems.

So here we have some exceedance plots. So the probability depends on the horizontal axis and a reservoir amount level basically on the vertical axis. So what you can see here is that when you just use the more recent period as a baseline, you end up with storages being a lot lower over the over the period of record.

And what we did, we did some comparison with some observations in Tullaroop because we were actually moderate modelling demands of from the irrigation system as well. And we were doing OK with simulating those. But essentially this is telling us, OK, we're going to meet we'll move just using different baselines. The storage responses look quite different. This thing. What's more interesting I think is sources of water. So they have these three sources of water available and we looked at the average extractions, average annual extractions and the maximum annual extractions to supply what Maryborough over a long records. And what we see is that as you and using the different baselines and if you use the more recent baseline to simulate the system, you'll tend to draw a lot more water from Tullaroop Reservoir and from the Moreland groundwater system.

Now, from a system operation perspective, that's not necessarily very desirable because the water may not always be there. The groundwater system has not been proven to be able to deliver the maximum licence volume and there was some water quality have to be dealt with as well. But the, the supply from the local system, the evidence didn't tell what reservoirs is, is to on average declines.

You can see, you know, in, in the best years on the with on the right, right side in the best years when we're extracting the maximum volumes from those, they remain about the same. But as you look at the Tullaroop and Moolort, we actually sort of max out the, the extractions that we can take from those to be able to deliver water supply to the system. So there's really a big difference in the baseline in, in, in choosing a climate baseline on how that system performs. And one of the measures that Pat and John kept on asking for is, is the system yield. We have a system yield. Yeah.

Next slide is very is the amount of water that can be delivered so that the organisation meets service level specifications, which is related to the frequency of water restrictions in in Maryborough. And what we can see is we use the full period of record. We can deliver quite a quite, quite a large volume of water if we only use the recent baseline, recent period as the baseline. That's a lot less that we can deliver without restrictions or, or meeting this the expected levels of restrictions. So that really, I guess sparked some questions about augmentation and I've since spoken to somebody today about, you know, the serious consideration about how to augment the system. But this is just looking at the historical climate.

We also wanted to look at the future climate. So I'll talk through very recently quickly just how we look through at the future climate part. So the first thing we did and I described briefly, it's basically we took our stochastic data generation method and for rainfall, we looked at declines in annual rainfall. So we're just changing the annual mean and we changed the annual mean by at different sensitivity levels. So no change, 5% reduction, 10% production and a 15% reduction by the end of the century. We're generating these out for a long period of time.

We then looked at, we also looked at varying temperature and evapotranspiration and they were going in concert equivalent to a 1° increase in temperature and a two degree increase in temperature by the end of the century. And we've tripped got these being transient from now into the future. So if you think about that, that's quite a number of combinations. I'll show you some graphs that summarise those combinations. But then we took those, those forcing and converted them into rainfall. And it's one of these, these graphs that show all the combinations.

So the top lists in these graphs. So horizontal axis for each panel is, is the year into the future. Well, from 2020 going forward, the vertical axis is the annual runoff volume, our inflow volume effectively into the system. The top of these 12 panels, the top left is the no change scenario and the bottom right is the maximum, the maximum change. And so you can see that from that baseline out to 2100.

But if you look at the maximum change, it's more than a 50% reduction in the main rainfall represented by the black line. But there's a lot of uncertainty about that in the in the annual rainfall now. So we're now dealing with something that's transient. It's not a stationary climate. And all the analysis that we've done, I showed you about the historical stuff has been talking about stationary climates. We can do a yield analysis on a stationary with a stationary climate. It's very hard to do that when the climate's shifting.

So what we did, rather than do try and unpick and, and, and have those same sorts of measures, what we looked at was just the chance of water restrictions occurring and at different levels across each year into the future. And we, because with our stochastic data we had 1000, realise different realisations we could estimate, you know, the chance of water restrictions at each level occurring for each year. So when we do that, starting with the baseline, so the current the, the, the level of service specifications is that they don't want to have any stage 3 or 4 water restrictions at all. And a percentage of years with stage 1 or two is less than 5%. So we can't quite use those, those numbers to, to interpret these. But I think it does help, you know, if, if the 5% chance of one stage 1 or two restrictions, then that's probably that in any given year into the future that probably saying that, that we're breaching those service level specifications.

So we've got the different levels, stages of water restrictions in different colours. Black is stage 4 least, least likely to occur. White is stage 1 softest restriction and the most likely to occur. So that's the upper bound of those panels. Under no change, you can see that there's no change, roughly no change in the, in the characteristics of the water restrictions under the most severe change we might end up with, with climate change that we've considered, we might not with end up the stage for water restrictions 10% of the time on year 10. So you know that, that suggests that perhaps there may be some, some augmentation required.

So we looked at a really simple, right, very, very simple way of augmenting this. We just said, OK, what if we can supply 20% of demand with some alternative water source that is perfectly reliable. So it's a great green situation, but you know, it's, it's just for the sake of testing this and what we did.

So we've got a no augmentation cases. We've just taken a subset of the no climate change and mid range one and A and a severe climate change on the bottom rows on the far right column is the no change. And then the others are if we augment the system in 10 years time, in 30 years time or in 50 years time. So you can see that, you know, depending on what's acceptable in terms of the chance of water restrictions of different levels, it might be possible to and, and, and you know, a midpoint climate change prediction, it might be possible to defer an augmentation by 10 or possibly in 30 years if we're, if we're happy to accept a slightly increased chance of, of severe water restriction. But if we're not, then we probably need to do it now.

But under the worst, worst climate change prediction scenario that we looked at, probably augmenting with 20 with the system with, you know, a perfect water source of its could supply 20% of the demand probably isn't enough and it isn't enough probably in a few years time.

So this was really, I guess, a bit of a, a scoping study to say, well, how can we investigate and, and take some of the science that we've been doing and, and understand the consequences for a water management system of different choices of different baselines. There's a choice that needs to be still be made about what is the baseline to adopt, but we're providing insight into those choices.

It also lets us think about climate change as a process that's unfolding, that a real water management manager is thinking about, rather than thinking 50 years into the future. This is our stationary climate. We're thinking about the transition from now to that 50 year time scale by building in some sort of nice neat linear trend. That's really only been possible by using stochastic data methods which allow us the flexibility to fit the model to a period and then generate a really, really long time, a long time series, 100 years from series that the analysis is potentially done.

The other thing that we found was that we really needed to have our range full runoff model in, in that system. And one that really does reflect the hydrology of the catchments that we're dealing with.

And, you know, to explore the, the, the, the future climates. I guess what, what we've implemented let's you look at the, we could look at the stationary future climate, but it also lets you look at that, that transition and inform the, the timing of, of augmentation decisions. I’ll leave it at that.