Can Base estimate my FTP without a power meter?

Yes, for outdoor rides with a GPS route. It models power per segment from speed and elevation change, then takes your best sustained twenty minutes — which is what a real field test measures. It is explicitly an estimate, its confidence is capped accordingly, and every number derived from it is labelled as estimated.

Why does Base model power per segment instead of using a ride average?

Because a ride average cannot distinguish twenty minutes of climbing from twenty minutes of freewheeling down the other side, and those are roughly 250 watts and zero. Averaging them produces a number that describes neither. Segment-level modelling is the only way the estimate means anything on terrain that is not flat.

How accurate is estimated cycling power?

Good enough to prescribe training from, not good enough to compare against someone else's meter. The model assumes values for aerodynamic drag and rolling resistance rather than measuring them, and wind is entirely unknowable from a watch — a headwind reads as power you never produced. Treat it as your own consistent scale rather than an absolute.

Why can my indoor trainer rides not be used?

An indoor ride has no GPS route, so there is no elevation data, and without a power meter there is nothing to read either. Base counts those rides as unusable for threshold estimation rather than averaging them into a number that would be meaningless. They still count toward your weekly minutes and ramp rate.

Does a power meter override the estimate?

Yes. When a ride carries real power data, Base reads it straight from the workout's statistics and does not model anything. Metered rides also carry a higher confidence rating than modelled ones, which is reflected in how the threshold estimate is presented.

Estimate Cycling Power Without a Power Meter

Why per segment

A ride average is useless on any terrain that is not flat.

Twenty minutes climbing at 250 watts and twenty minutes descending at zero average to 125 watts — a number describing neither part of the ride and corresponding to nothing you did. The information is entirely in the variation, and averaging destroys exactly the information you need.

So the model runs over segments. Each stretch of the ride gets its own power figure from its own speed and gradient, and the twenty-minute best is found by rolling through those.

climbing segment    18 km/h,  +6% gradient   →  ~280 W
flat segment        32 km/h,   0% gradient   →  ~200 W
descent             48 km/h,  −7% gradient   →  ~0 W

ride average of the three:  ~160 W   (describes none of them)
best sustained 20 min:      ~265 W   (is the number you want)

Twenty minutes is not arbitrary. It is the classic field test protocol, which exists precisely because an hour-long maximal effort is unpleasant enough that almost nobody repeats it.

What the model actually accounts for

The physics of moving a bicycle is well understood and dominated by a small number of terms:

Term	Depends on	Known from the watch?
Aerodynamic drag	Speed cubed, frontal area, air density	Speed yes, area assumed
Rolling resistance	Weight, surface	Weight yes, surface assumed
Climbing	Weight, gradient	Both yes
Acceleration	Speed changes, weight	Yes

Speed and gradient — the two that dominate — are measured. Frontal area and rolling resistance are assumptions rather than measurements. That is where the error comes from, along with the term that is not in the table at all.

Where it stops being trustworthy

Wind is unknowable from a watch.

This is the honest limit and it is worth stating clearly. A headwind makes you slower at the same effort, and the model reads slower-at-this-gradient as less power. A 25 km/h headwind can easily account for a hundred watts of apparent difference that has nothing to do with your legs.

Nothing in the data distinguishes a windy day from a bad day. Which is why:

Confidence on a modelled threshold is capped below a metered one
Every consumer of the number labels it as estimated
The threshold screen shows you which single effort won, so you can recognise a ride that was suspiciously fast with a tailwind

Other real limitations: drafting reads as free power, a heavy bike or bags shift the weight term, and an unusually aerodynamic position beats the assumed frontal area.

What it is good for

Your own consistent scale.

The assumptions are wrong in a consistent direction for you, because your bike, your position and your usual routes do not change much week to week. So the estimate tracks your own change well even where its absolute value is off — which is exactly what prescribing training from it requires.

What it is not good for is comparison. Your modelled 265 watts and a training partner's metered 265 watts are not the same claim.

When a power meter is present

Base reads it directly from the workout's statistics and models nothing. Metered rides carry a higher confidence, reflected in how the threshold is presented on the thresholds screen.

Indoor rides

An indoor ride with no power meter has no GPS route, therefore no elevation, therefore nothing to model from. Those rides are counted as unusable for threshold estimation and reported as such rather than being quietly averaged into a number that would mean nothing.

They still count toward your weekly minutes and your ramp rate — the load was real even where the power is unknowable.

If you ride indoors mostly, set your threshold by hand from a trainer's own power figures. That is more accurate than anything this model could produce.

Common mistakes

Comparing a modelled figure to a friend's meter. Different instruments, different absolute scales.

Setting FTP from a tailwind-assisted ride. Check which effort won before accepting the estimate.

Expecting indoor rides to contribute. No route, no power, nothing to model.

Treating the estimate as static. Re-scan after a block; the model is only as current as the rides behind it.

Assuming a heavy commuter bike gives the same number as a road bike. The weight and drag terms both change.

Ignoring the confidence label. It is capped for modelled power deliberately, and the cap is the honest part.

What to do next

Run the threshold scan and read which ride produced the winning twenty minutes. If it was a fast solo effort on a familiar route, the estimate is about as good as this method gets — if it was a group ride, expect it to be optimistic and set the threshold slightly lower by hand.