We’ve already written about (various aspects of) the changing ‘dependability’ of coal on various occasions – including:

On 31st May 2019 we released the GRC2018:
- From that report (on 3rd June 2019) we wrote about ‘Five different aspects of assessing the “dependability” of coal-fired power (in the Generator Report Card)’;
- And then on 14th June 2019 we wrote about ‘17-year trend of aggregated coal unit unavailability as from the Generator Report Card’.
Then on 15th December 2021 we released GenInsights21:
- inside of which we continued to explore the themes of coal unit dependability.
- including via Appendix 18 (Coal Unit Availability)
On 6th March 2023 we wrote that ‘Unavailability of coal units hits 24% across calendar 2022 (but they are *not* collectively tripping more often)’.
On 26th August 2024 we wrote that ‘Coal-fired unit performance has improved markedly from the ‘dog year’ that was 2022’.

There are other articles tagged ‘coal unit outage’ collated here … many of these speak to particular outages.

Prior analysis

Across these successive pieces of analysis:

1) We have consistently seen (and noted) that the following two things are both true … even if they initially might seem contradictory:

Finding #1) Coal units are not, in fact, coming offline more frequently than they have in the past

… in this article we will again demonstrate how this is correct.

Finding #2) When coal units come offline, they are typically offline for longer

… this article indirectly also (indirectly) demonstrates this.

2) Which might be quite different from the claims you might have heard (or learnings you might have misinterpreted) from elsewhere

For other reasons to do with Energy Security Concerns with the current War in the Middle East, we cranked the handle again recently … but in this case slightly differently.

1) With respect to Energy Security, see the article ‘To come soon’; but in addition

2) We thought it was worth publishing more details of this particular analysis here.

Classifying coal unit starts … a.k.a. ‘Starts ain’t starts’

With reference to recent conversations with mechanical engineers who are far smarter than I (and also still much more actively engaged onsite in thermal power generation plant) I’ve constructed this table to help explain (at a high level) how the industry commonly classifies ‘a start’ according to:

How long the unit’s been offline
Hence how much the heavy steel components have cooled down; and
Hence what needs to happen in order for the units to be brought back online.

But readers should be aware that, though smarter people have helped me draw up this table, it is still a simplification – and any errors that remain are well and truly mine:

Classifying Name of Start	Approx Time offline	Brief Notes
Hot restart	Only 1-2 hours offline	This classifier reflects what happens when a unit: Was not planned to be offline (i.e. no maintenance planned); but For whatever reason, *trips offline** unexpectedly. * A reminder for some readers that Hugh Saddler’s earlier article ‘Power station ‘trips’ are normal, but blackouts are not’ will be worth reading. Generally speaking, the generator will seek to bring the unit back online in a short space of time (say under 2 hours). Because all components are already ‘at temperature’ the ‘hot restart’ process is typically the simplest process.
Hot start	Under 12 hours (say)	In this case the unit might have tripped offline unexpectedly, but the generator might need to take some quick remedial action before being able to return the unit to service: that remedial action can’t be in the high-temperature components, by definition it could be something in ‘balance of plant’ Because the high temperature components are large masses of steel, and they are insulated, steps can be taken to ensure they remain at temperature. Which means the ‘hot start’ process (whilst not as simple as a ‘hot restart’) is still relatively simple.
Two-shifting operations	e.g. Down in the morning, back online in the evening	We’ve already written some articles about the increased use of (and broader interest in) ‘two shifting’ operations, in which deliberate steps are taken: to reduce a unit’s output to 0MW (which may, or may not, entail coming fully offline) for a period of a few hours; such that it can be more easily brought back online later in the day when market conditions change. Without going into detail here, readers should note that; This *operation differs from* a standard ‘Hot Start’** But, in our analysis here, we’ve not specifically looked at separating them out in the analysis.
Warm start	Typically 48 hours, but can extend for 5 days	Typically, the longer the unit is offline, the more the high temperature components are cooled. But there are complexities either way: If the generator needs to get inside of high-temperature components (e.g. the boiler) then it might use a ‘force cool’ approach to cool it down more quickly than would otherwise happen … in this case they might need to do a ‘Cold Start’; Alternatively, if it does not need to do this, but wants the return to service to be as quick and smooth as possible, then procedures like ‘hot banking’ might be used … hence the chance of a ‘Warm Start’ up to 5 days after coming offline.
Cold start	Over 5 days … and up to months offline (or years, in extreme cases – though see below).	This will be the case after a long period offline, when the high temperature components are at ambient temperature. In this case, significant effort (and a lengthy elapsed time) is required to gradually warm these components to avoid creep and fatigue (or other more catastrophic) damage. This consumes more fuel in the process.
Special case of ‘Cold Start’	After major rebuild	On rare occasions there are special cases that require even more complex restart processes. For instance … after the Callide C4 Catastrophe: a whole new turbo-alternator was required to be manufactured and supplied and fitted and commissioned. in parallel with this, the extremely long time offline meant that personnel onsite lost familiarity with operations of the C4 and C3 units (given C3 had also been offline … not least because of natural staff churn over that period. In this case the startup procedure was much more complex than a ‘normal’ cold start. I’m not familiar with the specific details of the long outage on YWPS2, but wonder about the extent to which bringing this unit back was not a ‘standard’ process.

Trending a count of coal unit starts

At the time of writing (March 2026) there are currently 44 still-operational coal units across the NEM.

To ensure the following longitudinal trends are comparable over time, we have only looked at the performance of these same 44 units backwards in time to January 2010 (in other words, we omitted Liddell and Hazelwood and the other coal units that have closed in between 2010 and 2026 from these statistics).

All starts

In this trend, we’ve categorised each of the starts* for each of the 44 units unit starts according to the four main categories above – then stacked them and trended them over time:

* i.e. with a ‘start’ being a 5-minute sequence from 0MW to some number>0MW in their output.

There’s quite some complexity in these calculations and so some simplifying assumptions are made:

1) Readers with a particular interest can contact me directly to explain more (time permitting)

2) But useful to note here for all readers that we chose to use Target and not FinalMW to run these calculations, because of the greater simplicity in the numbers (we don’t believe it will have made any tangible difference to the high-level conclusions).

It’s clear to see that (contrary to what some readers might have been believing from listening to, or misinterpreting, other sources) that there’s not been some massive up-tick in number of unit starts across these 44 units over 16+ years.

… so what gives?

Starts by Type

Where the picture gets more interesting (and thankfully aligned with our prior hunches) is when we break each of the colours out, and trend them separately as follows…

Hot Re-Start

Starting at the fastest return to service route (the hot restart) we see the following:

For ease of reference, I’ve added in a 12-month trailing rolling average – which makes it clear that:

1) To January 2017 there was an increase in number of hot restarts

2) But from mid 2020 there was a steep drop …

(a) now averaging 2 per month across all 44 units,

(b) rather than the 6 per month from 2015 to mid 2020

… which correlates well with the increasing influence of ‘the duck curve’ and coal unit challenges.

Hot start

Slowing things down slightly, we see the ‘Hot Start’ category as follows:

Over most years the results here have been quite consistent:

1) Apart from calendar 2011,

2) And a period of 2022 and 2023.

Warm start

In the warm start category (and remember that this will also include the use of two-shifting), we see the following:

In this case we see that the median range of ~4 warm starts per month:

1) With a blip higher in 2016 and 2017;

2) And a drop lower (to ~2.5 warm starts per month) from 2020 onwards.

Cold start

With the slowest starts category, we see the following:

We can see in this picture that:

1) From ~2015 onwards there was a step up

2) And from ~2019 there’s also been a small shift up.

Conclusions

Given the number of times we’ve already looked at this, I did not really expect to see anything different – but it was still pleasing to see that our prior conclusions remain valid:

Conclusion #1 = Coal Units are not tripping or restarting more often …

Conclusion #2 = by virtue of the move to more ‘Cold Starts’ and fewer ‘Warmer Starts and Hot Restarts’ we also see that these same coal units are offline for longer