They're going to talk a little bit about FastStart, one of the projects we've got going on in relation to that, and answering some cracking questions that we literally had 10 minutes ago. Gentlemen over to you.

Thanks Oliver, I don't know about a double act but because I'm not going to take too long I'm going to leave it for Patto. So all I want to do I guess was introduce the topic and I know and I can't see where Mike Bange is gone to now but, as we toured around with our previous meetings he would ask this question: where does CSD in invest? And I guess he was a little taken aback by um the lack of response, I said don't take don't take it personally it's just that people were more interested in other speakers but that's all right but we do many collaborations and I guess Sarah just talked about obviously the CBA collaboration we have which you all know, Mike talks about and works for the commercial stuff that we do with the Richard Williams Initiative and you'll see a lot more of those things certainly during this week uh with the Disease Initiative and those other things that are happening in the industry, so please come along to Emma's talk and those types of things there.

But the one I wanted to cover briefly obviously was the FastStart collaboration that we have, and FastStart for those who don't know is a collaboration between CSD and Syngenta and you'll see the Syngenta guys here. It's been going now for 15 years. A lot of people don't realize how much has gone into that and you'll see a lot of promotion as we do the morning teas and that, just what's come out of that and certainly uh many of you be used to The Weather Network it's certainly a FastStart collaboration, Vibrance seed treatments but what we want to talk more now is some of the other new pre-emptive tools that we've been working on as well.

And one of those, and Sarah and the theme has been about well how can we predict you know where our soil temperature is going, we can predict air temperature well more or less thanks BOM for 3 days but why not - our seeds in the ground as you saw for the first seven days... can we come up with a tool that helps us predict where our individual soil temperatures are, when we all have weather stations and those types of things. So that's why uh FastStart got together with GoannaAg and did some research and it's a project that um John's going to talk to you about, so I'll hand over to him because he knows all the ins and outs.

I'm not sure that's entirely true Peter and as the least masterful cotton grower in the room it's a rare privilege to stand in front of you and share the project and where we're up to with it. I'd like to begin by thanking CSD and Syngenta for entrusting GoannaAg to think about a different way to solve a well understood issue, we've talked about it already. That's namely building a fit for purpose tool to help make a more confident planning decision.

So what you seeing now as an extract from the FastStart guidelines, I'm sure you'll be pretty familiar with them, which at first glance appear pretty pretty straightforward. If your 8a.m. soil temperature is above 14 degrees, and your assessment of the coming week is that the rising air temperature trend is up, then you've met the minimum guidelines: two green lights and you can get on with a job. But how simple is that in practice?

I'll assume that collectively you have an understanding the important role that soil temperature plays, we've had it spelled out to us very clearly in the last couple of presentations in optimising plant establishment and seedling vigour. And that it's not an overly difficult or onerous chore to measure your 8 am soil temperature. But what about assessing the 7 day or the near-term outlook (the 7 to 10 day outlook)?

Until now that's meant dialing into the weather channel, looking at the temperature forecast and using that essentially as a surrogate for soil temperature. Noting that it is air temperature forecasted, it's a generic one at that, and certainly not site specific.

So we set about creating a site specific soil temperature prediction that reliably and accurately could answer that question. So before I get into the detail, uh quick machine learning 101. So machine learning or ML involves training computers and please I've learned this and machines taught me this it's not something it's not a novel thought, but instead of programming the machine with explicit set of instructions for a particular task, we provide them with lots of relevant data and let them figure out the patterns and trends and rules on their own.

So for instance if we want a machine to recognize an apple we show it thousands of pictures of apples and not apples, and over time the machine learns to work out what an apple looks like. The beauty of machine learning lies in its ability to improve over time, so as the machine gets more and more data and continual feedback it refines its understanding and becomes more accurate. And this ability to learn and adapt is ideally suited for things just like predicting future soil temperature.

Equally importantly because it recognizes the patterns and nuances of a particular site or set of data, it means we no longer need to rely on a one-size-fits-all approach. So know that each site essentially has its own unrelated independent site specific prediction model attached to it.

So that's what we did. With over 200 million weather data points across time and space we had an enormous set of training data - essentially lots of pictures of apples. This training data set allowed us to make an initial prediction, check that against the actual soil temperature, review the outcome and continuously iterate until we reached a worthwhile level of accuracy.

So we then move from a training set of data to act to ingesting actual real time data and for now we're using using data from GoannaAg weather stations that make up the 90 site FastStart weather network that Whitey just alluded to, along with the Bureau's gridded weather forecast. We created a feedback loop that allows us to continuously predict, reflect and iterate and as part of this process we ,measure the performance, or the accuracy if you like, on a site by site basis. And as Illustrated in the graph the green columns do show some modest variation between individual sites with prediction accuracy within half a degree of the measured soil temperature.

So we're really pleased with what we've come up with so far. Our model predicts future soil temperature at 90% accuracy, to within half a degree over a rolling seven days. The graph you're looking at there shows an example of a particular site where the blue line, Oliver the average temperature with that work we just did since the last question we've been on it like white on rice mate... So the graph that you're looking at shows the blue line which is the measured daily average soil temperature and the orange line is our prediction of that temperature. The two parallel lines that you can see provide the context of the 14 degree FastStart 14 degree minimum and 16 degree average.

The repeatability and scalability of this solution has given the CSD and Syngenta FastStart team the confidence to now move to launching the FastStart Field Forecast Tool (which recently won an award for illiteration) to the Australian industry this coming season. We'll also to continue to explore the interaction with other data. So for example what if we looked at early root development, which we can understand through soil moisture probes, data from soil moisture probes, and how much influence does soil temperature have on that root establishment and because we've established a correlation between early root vigour and the ability to extend irrigation interval intervals post first flowering, so what other sets of data can we bring to this over time?

We're currently finalizing the integration between the GoannaAg and FastStart worlds which we expect to be finalized so this tool will be live and available this season via the FastStart portal. And talk to Whitey's team to understand more about that.

Given the model is adaptable we also have the ability to adjust parameters and functionality based on user feedback. In other words this is dynamic and adaptable and it will inevitably evolve over time, and those tweaks can be simply and accurately tested and deployed as required.

I quickly wanted to share some other analysis we did as a part of this project as a bit of a side hustle where we looked at around 300 go field sites it's um using probes and canopy temperature sensor data, where growers had entered their planning date into our platform. We then compared those site specific soil temperature conditions at the at the nominated planting date against the FastStart guidelines to get some sense of the status quo. And based on those numbers 85% of fields were indeed planted subsequent or after those minimum soil temperature thresholds were met. That meant that up 15% or about 60,000 hectares if it was extrapolated across the industry were planted early in fact on average eight days early, so there's a multitude of reasons as to why that happens, but I just wanted to share this with you to provide some scale and context of the issue.

So let's put the ML whiz-bangery to one side and focus on what this actually means to your agronomic decision making, and the performance of your crop. So our view is that by making a more informed planting decision that relates specifically to your conditions, you will set up indeed a healthier and more resilient crop. This is one of a number of similar projects where we're looking to unlock the power of machine learning including several with CSD. And each of those are focused on providing you with tools that make some of your key agronomic decisions a little bit more informed, a little bit simpler and ultimately a little more profitable.

So thanks again for the opportunity, thanks to CSD and to Syngenta for your continued support and I look forward to welcoming you to the GoannaAg site over the next few days. Thanks very much.