← Cynoptec home
CYN-MU-20260819

Cynoptec Climate Intelligence Model: Version 1 Walkthrough (August 2026)

Valid: 19 Aug 2026 – 30 Sept 2026Issued: 19 Aug 2026Author: Stuart Williams
Low riskModerate confidence

Stu walks through the new Cynoptec Climate Intelligence Model: what it is, what it is not, and why Version 1 looks the way it does. The model replaces probability maps with a dated timeline of weather systems and a reasonable rainfall range for the time of year. The walkthrough also looks at the lagging ENSO risk heading into 2027.

Full forecast

Update, October 2026: The Cynoptec Climate Intelligence model is now on version 5.6. This walkthrough shows the original August 2026 release and why it was built the way it was.

Cynoptec Climate Intelligence Model: Version 1 Explained [Watch video for details] 


Since the new model went live the questions have not stopped, and that is exactly how it should be. It is a new system, a new perspective, and in a lot of ways a new way of looking at the weather altogether, so the questions coming in are doing real work in shaping what gets built next. What follows is the full picture: what the model is, what it deliberately is not, why Version 1 looks the way it does, and where it goes from here.


The place to start is with what everyone already knows. The Bureau of Meteorology long-range product runs out roughly four months and tells you the chance of exceeding median rainfall over a set period. For some purposes that is useful. For a great many operators it is not, because even if the outlook verifies exactly as issued, it never tells you when the rain arrives or how it is delivered, and those two things are usually the whole question.


Take direction first. When a model puts a 10 to 30 mm outlook over your district and that rain is arriving from a specific direction, and you know from experience that your place simply does not receive rain from that direction, you can discount the information entirely. Knowing how the rain is delivered matters far more than knowing there is a chance of something falling between the 1st and the 30th. Understand your location and you understand your weather, and no probability map can do that part for you.


Timing is the other half of it, and it is where monthly statistics quietly mislead. If rain falls on 1 September and the month records average to above average, that reads well. If the next fall does not come until 30 October and that month also records average to above average, the record now shows two consecutive months of good rainfall while you have actually gone something like eight weeks without a drop. The growers who ring in say the same thing over and over: you cannot go that long between falls at key points in the year, and mostly you cannot go that long at any point in the year. It is highly detrimental to crops and to whatever process is running at the time. The data says one thing and the paddock says another.


There is a third problem, which is that these products are revised continually. The map issued on 18 August is a good map for right now. If what you need is confirmation on the next two weeks, then continually updated data is precisely what you want, and that is why nowcasting exists and why every weather page and social group posts the latest model run each day. Inside a few days they rarely get it wrong. What they cannot do is see past that window. And it is very hard to commit to a decision when the carrot keeps moving position. That moving carrot is the gap the Cynoptec long-range service exists to close.


The other major models sit in much the same territory. The CFS drops down to a weekly timeframe, which sounds promising, but it is only ever as good as the data feeding it, so in practice it holds up for the next couple of weeks and no further. It changes run after run, with cloud band placement shifting each time. One run puts a strong band across the country, the next backs it off as the date closes in. That behaviour is perfectly fine for nowcasting and close to useless for a long-range decision. The ECMWF monthly is granular and quite black and white, and it refreshes in the first few days of each month, which means it has already absorbed the events about to occur. The current month therefore looks convincing while everything beyond it weakens considerably. Its current run shows dry anomalies through the north and the Arafura moving into summer, in the middle of the wet season, when tropical lows should be in play. That is highly unlikely to verify. And the deeper issue sits above all of them: every one of these major models is producing a completely different scenario, all probability based, all anomaly based, all built on different background states and inputs. How anyone is meant to make a decision on that spread is a fair question.


Part of the answer is to stop chasing the wrong number. Predicting a specific rainfall total is close to impossible no matter how good the forecaster or the computer is. A 50 mm forecast can verify at 5 mm just 500 metres either side of the corridor, and no amount of resolution fixes that. So the Cynoptec approach uses climatology to set a reasonable range for the system and the season instead. In summer, when a tropical low tracks over northeast Queensland, 100 to 300 mm is a safe assumption for anyone in its path. In the dry season, rainfall from a passing system is usually only 5 to 10 mm. A specific weather system, at a specific time of year, in a specific location, delivers a general range. Anything above that range is a bonus. Anything below it is an error.


That principle is what the model is built on. The Cynoptec long-range service has always given a date, a description of the weather, a timeline of how it develops, and to some extent a corridor the weather should transit through. The Climate Intelligence Model takes that same high energy framework and builds calculated rainfall ranges and interactive display on top of it. The high energy periods establish which systems are due and when. The season and the system type establish which corridors those systems generally pass through, which is why particular districts can be named for rainfall in a given setup rather than shaded in at random. Historical data and climatology, the bulk of it already calculated by the Bureau and other scientific bodies, then sets the general range for that system, in that place, at that time of year. The intent is deterministic rather than probabilistic. This is the timeline, this is where, this is when, and this is the reasonable likely outcome.


The mid August event shows how that reads in practice. The model period ran from the 5th to the 13th, and the written timeline had the mid month rainfall window opening from the 5th as a system organised in the west, a front crossing southwest WA between the 5th and the 7th, moisture building around the Gascoyne, and that corridor carrying right through the southern agricultural zones into the southeast. Actual rainfall for the 6th to the 12th came in through the Gascoyne, through southern parts of South Australia and across to the southeast. Distribution differed slightly, the general total ranges were about right, and the colour scaling sat quite close to what verified. The same period was flagged for damaging winds through southern South Australia, western Victoria and the Alps, along with blizzard conditions, which is exactly what that system type delivers at this time of year. The specific totals were never the point. The point is that the system, the window and the corridor were on the board in advance, and most growers already know what a northwest cloud band delivers in their own paddock: 10 to 30 mm on the flat, sometimes only 1 to 5 mm on certain sides of the ranges.


Reading the output is deliberately straightforward, and there is no hidden information buried in it. Start with the timeline, which tells you when a system should form, where it should cross, and any key features such as damaging winds, frost risk or alpine conditions. Then read the corridor, which is the path the system should transit, described at district level. Then read the range rather than the number. The display gives a low to high range and a central value, and that central value is effectively the average of the range. From the recent period, the West Coast district showed 1 to 10 mm with heaviest areas around 9 mm and a central value of 3.5 mm, and that heaviest figure sits where the southern part of the district was expected to out perform the western part. The Murraylands showed 4 to 24 mm, and in reality the heavier end of that would have banked up against the Adelaide Hills while the lighter end sat further east toward the border. What the model will not do is account for one property picking up 100 mm when everyone else gets 30. It is built for the reasonable likely outcome, and the reasonable likely outcome is the 30.


As for why Version 1 looks the way it does, the honest answer is that it had to start somewhere. The last several months have gone into working out how to do this at all, and there is a point where refining in private stops being useful and something has to go out the door. Version 1 is that line in the sand. It sets the limit, which gives Version 2 something concrete to expand from. It is also worth saying plainly that this is a one person build. The forecasting methodology is being translated into code as it goes, and turning forecasting ideas into code language turns out to be difficult work that would ordinarily belong to a software engineer. Learning to code while building the thing is part of the current pace.


The limitations should be stated openly rather than buried. Rainfall indications in Version 1 are a fair way off in places, which is expected, because the model still needs training against actual outcomes before those ranges tighten. The bigger issue is district blanketing. The model currently works at district level and it works broadly, so a small amount of rain in the southern part of a district effectively lights up the whole district. The Goldfields is the clearest example from the recent event, where most of the rain fell across the southern Goldfields and the entire district lit up regardless. The same happened with rain pushing into the pastoral districts, which had not been accounted for at all. The coastal problem is related. The 14 to 17 August period showed up to 15 to 20 mm along the East Coast, and the rain that fell was heavily coastal, with New South Wales pushing past 25 mm right on the coast and far less inland. Blanket washing a district like that puts potential misinformation in front of a lot of users, and fixing it is the single highest priority. There is a good deal of maths still to go into weighting rainfall within a district rather than smearing it across one.


Version 2 is close. Data is finishing training in the background now and the release is due at the end of this week, refining the rainfall indications and the corridors. It still will not give rainfall to the millimetre. It will give a better general range and better placement inside the districts, which is the part that matters.


In the meantime the best way to use the model is alongside the original long-range forecast rather than instead of it. The written long-range forecast continues as normal. Over time the two products converge: at the next monthly release, the data trained into the model gets fed back into the long-range forecast format, so the written breakdown stays exactly as it is now in dot points while the hand drawn maps are replaced by still shots from the model. Same format, better maps. Anyone wanting the animations clicks through to the model section itself.


One last thing worth raising, since much of the months ahead material is already published. El Nino is occurring, it is expected to strengthen, and it is all anyone is talking about. What history shows is that the stronger these events are, the longer the atmosphere takes to couple, and by the time coupling gets underway the wet season is already running. The monsoon then negates a good deal of the ENSO influence, in much the same way it negates the IOD. Once the wet season retracts, the traditional drivers return and rainfall becomes far more dependent on moisture from local oceans than from the equator. That is where the concern sits. Moisture across the north disappears through the end of March, rainfall begins to rebuild in the south but intermittently and not in great volume, and the IOD structure through the first months of 2027 points toward a considerably drier cycle than the last six months have delivered.


This spring looks reasonable enough. Next year looks quite ordinary. Working through 2027 there are no strong broad rainfall signals to speak of: the typical mid June cloud band that turns up every year, broad dry conditions either side of it, a few troughs and fronts through early to mid July, then rainfall largely disappearing again through August. There is plenty of noise online about big floods coming. There are big floods every year somewhere, usually driven by tropical cyclones. Long duration wet cycles of the kind seen in 2011 or 2022 are not showing up anywhere in this framework for next year. The CFS is currently showing a very wet, at times enormously wet April and a similar May, and while it has backed off a little in recent days, it is worth being careful about what sits behind those maps. Against the framework this entire service is built on, once March passes into April there is really nothing there. Those are traditional frontal systems, not national rainfall events, and a national rainfall event in April or May would be unusual in any year. A late coupling ENSO phase has the potential to catch a lot of people out.


Feedback is welcome and ideas are worth more than anything right now. If there is something that would make this more usable at the paddock or the business end, send it through. The more usable this is for the end user, the better the outcome for everyone.


Cynoptec - Know the Future™

Video update

Want forecasts like this every month?

Cynoptec members receive long-range Australian weather forecasts, High-Energy Period schedules and district tables for every state.

View memberships