SpaceX progress towards a 100 launch year

[Robotbeat]

Yeah, remember they have the second constellation partially approved which means being on the hook for nearly twice as many ADDITIONAL satellites than gen1, plus they can only fit like half to a third as many in each F9, so their required launch rate is… about 5 times as much as what they needed for gen1, which they still haven’t finished. That’s if they stick to F9.

So Starlink as a demand source just keeps getting bigger. They won’t be hurting for payloads.

[AnalogMan]

Since several of you seemed interested in how the long term trend in Falcon 9 launches looks when compared to an exponential curve, I decided to produce the following chart.

It shows the cumulative total of F9 launches (all versions of the vehicle) from the first launch up until March 9, 2023 as a function of time.  The horizontal axis is days since the first launch.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

The exponential trend line is the one generated by Excel.  This is not an attempt to make any predictions about future launches, merely to show that on the whole the cumulative launch totals do roughly follow an exponential pattern.

[AmigaClone]

Yeah, remember they have the second constellation partially approved which means being on the hook for nearly twice as many ADDITIONAL satellites than gen1, plus they can only fit like half to a third as many in each F9, so their required launch rate is… about 5 times as much as what they needed for gen1, which they still haven’t finished. That’s if they stick to F9.

So Starlink as a demand source just keeps getting bigger. They won’t be hurting for payloads.

To finish the first constellation using 1.5 Starlinks SpaceX would need 21-24 F9 launches depending on rideshares.

Each of the currently approved gen 2 shells of 2500 Starlinks each would require up to 50 Starlink launches using  F9/V1.5 combination (like the gen2 shell 5 currently in orbit), 125 launches using the F9/V2.0-mini combination, or speculating 45-50 Starship/v2.0 (full sized).

SpaceX has requested approval to switch the current gen 1 constellation Starlinks (1.0 to 1.5) with v2,0. That would imply even more launches.

[meekGee]

Since several of you seemed interested in how the long term trend in Falcon 9 launches looks when compared to an exponential curve, I decided to produce the following chart.

It shows the cumulative total of F9 launches (all versions of the vehicle) from the first launch up until March 9, 2023 as a function of time.  The horizontal axis is days since the first launch.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

The exponential trend line is the one generated by Excel.  This is not an attempt to make any predictions about future launches, merely to show that on the whole the cumulative launch totals do roughly follow an exponential pattern.

The one conclusion is that there's no annual behavior like was proposed upthread.  There's nothing in the data to indicate where one year ends and another begins.

This means that we should expect the rate in each quarter to be lower than the rate in the following quarter, and so 7/month during Jan-Feb is not a sign that 100/yr is unlikely.

[eeergo]

Since several of you seemed interested in how the long term trend in Falcon 9 launches looks when compared to an exponential curve, I decided to produce the following chart.

It shows the cumulative total of F9 launches (all versions of the vehicle) from the first launch up until March 9, 2023 as a function of time.  The horizontal axis is days since the first launch.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

The exponential trend line is the one generated by Excel.  This is not an attempt to make any predictions about future launches, merely to show that on the whole the cumulative launch totals do roughly follow an exponential pattern.

The one conclusion is that there's no annual behavior like was proposed upthread.  There's nothing in the data to indicate where one year ends and another begins.

This means that we should expect the rate in each quarter to be lower than the rate in the following quarter, and so 7/month during Jan-Feb is not a sign that 100/yr is unlikely.

From my second post:

You can have a trend that follows an exponential with coarse/long enough binning [...] a linearized approximation over short time bins might be much preferrable because the short-term increase isn't well-fitted by an exponential, or the exponent is too small to show up over the noise

Note the exponent above is <0.001.

For the nth time: this thread was discussing *annual* behavior. Just as it is silly to try to measure the curvature of the Earth over a few kilometers of ground, yet round it is - so can the cumulative trend be approximated by an exponential over 13 years, yet it will be best described by a linear trend in *annual* periods. We were discussing what "short enough" meant here in the sense of linearizing the "smooth" function we can theorize cumulative launches to follow over time. Turns out a 10-year period is not "short enough", yet a 1-year period is.  It is evident Falcon launches have become more frequent since the first launch, but this acceleration isn't visible over a timescale that's just one year long.

As for discontinuities: there are discontinuities in the data AnalogMan posted, but nobody was claiming "annual behavior" for every year of SpaceX's launch history. We were noting a definite change in slope *from 2022 to 2023*'s provisional launch figures, which roughly coincided with the start of the year.

Way to plush up a strawman while moving the goalposts again. This thread is about *scorekeeping*, as already repeated a dozen times, and having a reasonable (linear) reference against which to measure.

But see, there's even more: attached is a zoom of the function AnalogMan kindly posted the equation for, scaled to both 2022's and this year's 365 days, in blue. I've added a linear trendline for both years in black. It ends on the predicted number of launches for last year (60), and in 100 launches for this one. While both curves are practically coincident last year within the noise, as expectable from what I explained above - there is a significant deficit (about 15) with respect to this year's aim. That implies that either this year's aim won't be fulfilled, if the multiyear exponential behavior keeps on going, or that there will be an *annual* discontinuity precisely this year to slope it up and reach it. Pick your poison.

[Robotbeat]

Nope, the current launch rate is compatible with an exponential curve the has a year on year growth of (1-(100/61))=64%.

Analogman’s curve shows very clearly the growth rate behaves a lot like an exponential curve even over a decade. No linear rate shows anything close to as much agreement with the long term data.

[eeergo]

Nope, the current launch rate is compatible with an exponential curve the has a year on year growth of (1-(100/61))=64%.

Analogman’s curve shows very clearly the growth rate behaves a lot like an exponential curve even over a decade. No linear rate shows anything close to as much agreement with the long term data.

I literally just showed the curves, side by side. You pull out absolute conclusions out of thin air, after failing to show what you claimed a couple of weeks ago, in spite of extensive explanations. You're either purposefully refusing to understand to muddle the waters, or you really can't be bothered to follow through a rigorous answer, preferring to double down until everyonr's patience runs out. And it's not the first time - so again, enjoy lol'ing/having the last word.

[mandrewa]

I think eeergo makes a good point.  We can see it by eye from AnalogMan's exponential plot of the launch record.  Over the span of a year that exponential curve looks a lot like a line.

Another way to look at it would be to ask what is different this year from last year?  The main thing is Vandenberg.  SpaceX is hoping for a much more rapid launch rate from Vandenberg.  I think SpaceX has already shown us what is probably close to the maximum launch rate they can achieve from Pad 40 and Pad 39A currently, and they did it last year.

So let's be optimistic and project that SpaceX does 24 more launches this year from Vandenberg than they did last year.  And let's be optimistic about Starship and add on five launches of the Starship.  That gives 90 launches for 2023.

That would actually be wonderful!

And if we want to be pessimistic we could try to estimate the odds of a launch failing somewhere in there.  And that of course would have a big impact.

[Robotbeat]

Nope, the current launch rate is compatible with an exponential curve the has a year on year growth of (1-(100/61))=64%.

Analogman’s curve shows very clearly the growth rate behaves a lot like an exponential curve even over a decade. No linear rate shows anything close to as much agreement with the long term data.

I literally just showed the curves, side by side. You pull out absolute conclusions out of thin air, after failing to show what you claimed a couple of weeks ago, in spite of extensive explanations. You're either purposefully refusing to understand to muddle the waters, or you really can't be bothered to follow through a rigorous answer, preferring to double down until everyonr's patience runs out. And it's not the first time - so again, enjoy lol'ing/having the last word.

Nah, I Derived it using equations, I just haven’t shared.

I think the curve shown by Analogman is pretty clear, however. Didn’t know about the Gunter’s data source, that’s much nicer. I’ll use that.

[meekGee]

...
I literally just showed the curves, side by side. You pull out absolute conclusions out of thin air, after failing to show what you claimed a couple of weeks ago, in spite of extensive explanations. You're either purposefully refusing to understand to muddle the waters, or you really can't be bothered to follow through a rigorous answer, preferring to double down until everyonr's patience runs out. And it's not the first time - so again, enjoy lol'ing/having the last word.

Eergo, this was your starting point:

Why? Rate is approximately constant (see last year's), why should it accelerate in such a pronounced manner as the year goes by? Moving average pace increase doesn't show such an exponential behavior either: https://forum.nasaspaceflight.com/index.php?topic=58240.msg2455547#msg2455547

You're shifting between claiming that the rate is constant over the year, to linear.

It's easy to show that within a single year, the differences between linear and exponential fits are small.  Guess what - within a month they're even smaller.

But it's equally easy to show that the long-term trend is exponential, and that there are no discontinuities in the slope when the years turn over - which argues strongly that the behavior within the year is exponential, even if you can approximate it linearly.

I'm with RB on this one.

[alugobi]

I'm with RB on this one.

Making a claim and not being bothered to show evidence is not good faith posting.   And borders on trolling.

[Robotbeat]

I'm with RB on this one.

Making a claim and not being bothered to show evidence is not good faith posting.   And borders on trolling.

Did you see what Analogman posted?

Anyway, to preview:

Suppose the instantaneous launch rate is an exponential function of the form:

A*e^(b*t), where b has units of 1/year and A has units of launches/year.

Cumulative launches then are the integral of that function, so integral(A*e^(b*t),t) = (A/b)*e^(b*t) (plus a constant… but that cancels out when you do a definite integral).

To find A and b, we just pick a t=0 to be, say, 2022 and t=1 to be 2022.

We know that integral(A*e^(b*t),t,0,1) = 61, so (A/b)*(e^(b) -1) =61. Solving for A:
A =b*61/(e^(b) -1)

We have two unknowns, so we need another equation.

Now it has been 73 days since the start of the year, so t=1+73/365.
integral(A*e^(b*t),t,1,1+73/365) = 16, so (A/b)*(e^(b*(1+73/365)) - e^b) = 16.
Plugging in A from above yields:
((b*61/(e^(b) -1))/b)*(e^(b*(1+73/365)) - e^b) = 16.

And solving for b: b=~0.466396 1/year

And therefore A= ~47.876687 launches/year (which is the estimated instantaneous launch rate at Jan1, 2022, ie t=0).

Integrating that from t=1 to t=2 gives us a total estimate of 97.2 launches in 2023. Jumps to above 105.6 if SpaceX launches on time tomorrow. So again, the current launch rate so far this year looks completely in line with ~100 launches in 2023.

[steveleach]

Everyone seems to be arguing finer and finer distinctions here. Maybe time to move on?

[meekGee]

I'm with RB on this one.

Making a claim and not being bothered to show evidence is not good faith posting.   And borders on trolling.

The entire post is the justification.  The graphs are upthread, show me any of them that don't agree with what I wrote.

Also not good faith, wtf?  That term implies I'm what, cheating somehow?

[AnalogMan]

Not sure if the following will help or hinder discussion!

I've plotted the number of Falcon 9 launches within 90-day moving windows.  First chart is from the first launch to present - the second shows the last 3 years so you can see more recent detail.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

Values are calculated for each day, and they are a count of the number of launches that took place in the previous 90-day period (the plot day is the last day of that 90-day period).  The steps in the curves are caused by launches dropping in or out of the 90 day windows.

This gives a idea of how the the launch rate has varied over time.  The most recent peak is 22 launches per 90 days.  I could have chosen different time periods, but 90 days smooths the data while still giving reasonable time resolution.

Incidentally I think the original purpose of this thread to simply monitor progress towards the 100 launches target for this year without attempting to make forward predictions was a great idea.

[meekGee]

Not sure if the following will help or hinder discussion!

I've plotted the number of Falcon 9 launches within 90-day moving windows.  First chart is from the first launch to present - the second shows the last 3 years so you can see more recent detail.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

Values are calculated for each day, and they are a count of the number of launches that took place in the previous 90-day period (the plot day is the last day of that 90-day period).  The steps in the curves are caused by launches dropping in or out of the 90 day windows.

This gives a idea of how the the launch rate has varied over time.  The most recent peak is 22 launches per 90 days.  I could have chosen different time periods, but 90 days smooths the data while still giving reasonable time resolution.

Incidentally I think the original purpose of this thread to simply monitor progress towards the 100 launches target for this year without attempting to make forward predictions was a great idea.

Thx!  Quick question: Why are the ticks on the time scale not uniform?

[eeergo]

You're shifting between claiming that the rate is constant over the year, to linear. [...]it's equally easy to show that the long-term trend is exponential, and that there are no discontinuities in the slope when the years turn over - which argues strongly that the behavior within the year is exponential, even if you can approximate it linearly. I'm with RB on this one.

Launch rate has units of [launches/unit time], so it needs a temporal binning. Examples have been shown by AnalogMan and Comga of such binning (variable-width bins such as Comga's "last 10 launches" period, or definite-width bins of a given time period, be it 90 days or 12 months). It is approximately constantly increasing over the recent past (2020 onwards), although it did see what can be described as a discontinuity (marked slope increase) around 2020 if you consider the whole of SpaceX's launch history, as AnalogMan showed above. 

In my first post you quoted, I sloppily referred to "constant rate" while meaning "constant rate of increase" (i.e. linear), not "launch rate". It should be evident from the second half of the quoted message "why should it [rate of increase] be accelerating?", since it's evident the launch rate (what I called "pace" there) increases over time.

Cumulative launches are another metric, which has units of [launches]. It has a clearer exponential trend, but this character only becomes apparent in multi-year (>3) periods. Cumulative launches and launch rates are two different quantities.

The exponential trend in the cumulative launch curve is too feeble in single-year periods, and the difference with a linear trend is not appreciable in annual plots. The analytical exponential function that best fits the whole historical data, calculated by AnalogMan, shows a 15-launch deficit (15%) with respect to this year's aim when extrapolated to Dec 31st, 2023. A mere continuation of the multi-year exponential fit will do a bad predictive job with respect to a linear extrapolation *in a yearlong period*.

Any function can be linearized, as we've incessantly reiterated. But I've also incessantly reiterated it's not useful to insist on measuring the curvature in a (hilly) few-mile-long stretch of land despite the Earth being spherical. Similarly, it's not useful to track year-long launch cadence with exponential fits, even if the 15-year trend approaches one. This just shows progress is too slow to capture in a year-long period, not that it doesn't exist at all!

I've repeated this in practically every post here, so I will not keep on beating this dead horse. I hope you reciprocate by not beating it either when xyv provides an updated version of his plot.

Nah, I Derived it using equations, I just haven’t shared. [...] Suppose the instantaneous launch rate is an exponential function of the form:
[whole lot of calculator-punching]

You are literally analytically imposing two boundary conditions (initial and final values of an imposed function) to solve for two free parameters without looking at how well the postulated function actually fits the intermediate data. You complained earlier we were taking a random cutoff point (New Year's) for the linear considerations, yet now you take it as the only anchor point in your plugging-in of numbers. That's actually what any fitter routine does, but for every datapoint and not just the ones you prefer: it finds the equation for which the error to ALL available points is minimal - not just for two points.

The beauty of wildly extrapolating an analytical function is that you can check how much it errs elsewhere - like in 2021, where it overestimates that total by 20% (31 vs 38), or in 2018, where it estimates 9 launches yet 21 were performed (-55%), or in 2017 where 6 are predicted and 18 happened (-66%), or 2016 (-55%)... plus all the midway points: for example, evaluating it for just the second half of 2021 (July 1st-Dec 31st), it yields a whooping 100% overestimate (21 predicted vs 11 performed). Of course, the reason for that is immediately visible in AnalogMan's plots.

Your 6-decimal point function just evaluates certain periods alright because you've fine-tuned it to fit well there in the first place! Statistics doesn't work that way, sorry - much less if used to criticize a simpler model, or trying to foresee a distant value with a few percent accuracy when you fail by a factor of 2 in another recent period.

Again, a year is too short a period to tea-leaf-read any exponential behavior.

Everyone seems to be arguing finer and finer distinctions here. Maybe time to move on?

Indeed!

[AnalogMan]

Not sure if the following will help or hinder discussion!

I've plotted the number of Falcon 9 launches within 90-day moving windows.  First chart is from the first launch to present - the second shows the last 3 years so you can see more recent detail.

Data courtesy of Gunter's Space Page:  https://space.skyrocket.de/doc_lau/falcon-9.htm

Values are calculated for each day, and they are a count of the number of launches that took place in the previous 90-day period (the plot day is the last day of that 90-day period).  The steps in the curves are caused by launches dropping in or out of the 90 day windows.

This gives a idea of how the the launch rate has varied over time.  The most recent peak is 22 launches per 90 days.  I could have chosen different time periods, but 90 days smooths the data while still giving reasonable time resolution.

Incidentally I think the original purpose of this thread to simply monitor progress towards the 100 launches target for this year without attempting to make forward predictions was a great idea.

Thx!  Quick question: Why are the ticks on the time scale not uniform?

The plotted points are daily, so to to give roughly 3 month or 6 month grid intervals I just set major tick to either 365/4 or 365/2 days.  There may be a better way to force true integer monthly grid intervals - I was just being lazy!

Edit to add:  The ticks are actually uniform, they just don't always coincide with the same day each month.

[kevin-rf]

Okay, I got a little impatient waiting for the next April 1st update.

(Yes, I also peek at my presents under the Christmas tree)

[FutureSpaceTourist]

https://twitter.com/jeff_foust/status/1636094535775207426

Ochinero: no reason we could not increase above 100 launches a year; depends on market and how fast Starship enters service.