Check out the power curve at the bottom of the post, it's composed of two screen caps from the same ride, stacked on the same image:

Apparently on the ride in question the cyclist rode at:

  • 291W for 12min 55s
  • 270W for 9min 10s

But I would have though that if you rode for 291W for 12min 55s, you also rode at least 291W for 9min 10s ... right? ... 🤷‍♂️

What are these dips? How are we meant to interpret this data?

I would expect power curves to either be flat (usually a result of ERG mode on a trainer) or constantly falling as the time period increases.

I reached out to Strava and their facebook rep said it was 'complex' and to send in a query to the support email - they haven't replied.

I had a look here: What method or algorithm is used for computing Power Curve on Strava?

It's good info, but the example given there doesn't have a dip, and doesn't discuss it - but it does make make me think that the answer might just be what I'm trying to avoid suggesting...

enter image description here

Addressing some of the comments:

At a casual glance I'll agree that power curves do show power plotted against time, but don't be mistaken in to thinking that they show instantaneous power over time.

Power curves show the maximum power a rider rode for a duration (averaged via various means, some algorithms more clunky than others), and as has been confirmed to my satisfaction they should be a 'monotonically decreasing function of duration' - but they're not on Strava, (which prompted my original query...)

Here is a visual example of how power curves are different from power with respect to time:

Instantaneous power over time:

enter image description here (This was a Zwift ramp test in ERG mode until failure when it slipped in to free ride)

Power Curve plot from the same data:

enter image description here

Note that:

  • They are under different sections in Strava
  • They don't look alike
  • The 418W at 1 min on the power curve correlates to the 1 minute duration that 418W was held (around the 21min 40s period seen in the instantaneous plot and not 1 minute in to the ride)
  • The power curve has a logarithmic time scale
  • 1
    The measurements shown are instantaneous. At about 9 minutes you were starting to tire, but when you saw that cute girl you had to show off, so you were doing 30 watts better for a few minutes. May 20, 2020 at 1:59
  • 1
    @DanielRHicks, 🤦‍♂️ Not sure that you understand what power curves are. Instantaneous power is presented in 'Analysis' and look like this: i.stack.imgur.com/QjUp4.png. I'm talking about power curves. If you're a subscriber have a read at Strava's glossary.. May 20, 2020 at 2:26
  • 1
    In my experience, the strava estimated power numbers are directly calculated off your speed and your bodyweight (under settings) and modified by the grade at that point. In other words, completely useless.
    – Criggie
    May 20, 2020 at 3:21
  • 2
    @Criggie, I agree somewhat, as there is still use in relative measurements with your own data over time - but in this case the data wasn't estimated, it was from actual power meter data. May 20, 2020 at 3:22
  • 2
    @DanielRHicks No, the greyed-in area is the highest values on record for each duration over many rides, and the dark-colored line is the highest value for each duration during this particular ride. These are called the MMP or "maximal mean power" for all rides and for just this ride. The name tells you exactly what the algorithm is.
    – R. Chung
    May 21, 2020 at 1:06

1 Answer 1


You were probably doing intervals, or at least your effort was at a high effort for several minutes, followed by a recovery for several minutes, and another high effort rather than an extended "steady" output to exhaustion.

In general, you're right, the maximum mean power for a duration should be a monotonically decreasing function of duration. However, for any single ride, especially when you're doing intervals, the MMP for that ride needn't be monotonic.

Here's a simple (too simple) numerical example. Suppose you were doing repeated 5 minute intervals, with 5 minutes at 325 watts followed by a 5 minute recovery at half that, or 162 watts. If you were to calculate the MMP for a duration of 9:00, it would be 252 watts; the MMP for 13 minutes would be higher than that, at 262 watts. You didn't do an extended effort for either 9 or 13 minutes but that doesn't prevent the arithmetic of calculating an average power over 9 or 13 minutes. In your case, it appears that your workout was on an ergometer and involved at least one 1 minute interval at 450 watts. Depending on how long your work and recovery intervals were, and at what levels, you can see non-monotonic MMP curves.

  • Ok, using your 5min 325W / 5min 162.5W example (i.e. starts with the higher Wattage interval) calculated for 9mins: ((325*5) + (162.5*4)) / 9 = 252.7W. And for the 13min: ((325*5) + (162.5*5) + (325*3)) / 13 = 262.5W. (I calculate these to show I'm on the same page as you). So if I 'change the phase' by 180° (i.e. the intervals start with the lower wattage) then we get for 9min: ((162.5*5) + (325*4)) / 9 = 234.7W and for 13min: ((162.5*5) + (325*5) + (162.5*3)) / 13 = 225W ... May 20, 2020 at 1:39
  • I understand the computational complexity of the search is large, so what I'm understanding here is that Strava are only doing a really quick/low res algo that is affected by the 'phase', basically finding local maxima according to the random nature of ... your ride. I wonder how deep Veloviewer goes in its 'splits' functionality? (same algorithmic problem, just a different data set) May 20, 2020 at 1:43
  • 1
    TLDR: Strava have a crappy algo that results in nonsensical power curve data. That they charge for. May 20, 2020 at 1:49
  • 1
    @LamarLatrell It's not quite that bad. The example is artificial to illustrate it's possible to get a non-monotonic MMP for a given ride. If you have the original data file, you can do a running mean over the entire file of 9 minutes and another running mean of 13 minutes. Then look at the maximums for each series. I've seen non-monotonic MMPs from many individual rides -- in fact, I think it's probably more common than seeing a monotonic MMP -- and I don't use Strava. There are lots of nutty things about Strava, but this isn't one of them.
    – R. Chung
    May 20, 2020 at 15:41
  • 2
    @LamarLatrell And, actually, the computational complexity isn't large.
    – R. Chung
    May 20, 2020 at 16:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.