Wind and Elevation: How Much of the AI Caddie’s Adjustment Is Real?

You’re on the 7th tee. Laser says 168 to the flag, the ground rises to the green, and Arccos Caddie has decided this plays like 194. You are standing in a gap between two beech trees and the flag looks limp. Somewhere between those two readings is the right club, and the gap between them is two clubs wide.

Here is the thing nobody in the app store listing tells you: your phone has no idea what the wind is doing on that tee. It never has. Every wind figure in every golf app on your phone is a lookup against a regional forecast, and that forecast was computed for a square of ground that might be nine kilometres across.

No app has a sensor where you’re standing

When people search ai golf caddie wind accuracy, they usually want to know whether the model is smart. Wrong question. The model is probably fine. The input is the problem.

Numerical weather models output wind on a grid. The Met Office UKV runs at 1.5 km resolution, ECMWF’s high-res global model sits around 9 km, and NOAA’s GFS is roughly 13 km. Whichever feed an app licenses, the value it hands you is a grid-cell figure for 10 metres above ground over open, short, unobstructed grass. That’s the meteorological standard. It is not a description of a tee box in a corridor of mature trees, and it is definitely not a description of the 4th at a coastal links where the grid cell is 60% open sea and reads high because water has almost no surface roughness.

Apps then take that single number, resolve it against the bearing from your position to the green centre, and convert the head or tail component into yards. Golfshot, 18Birdies, Hole19 and SwingU all draw wind arrows this way. Arccos goes further and folds it into an actual club recommendation. Garmin’s PlaysLike on the Approach S70 and Z82 pulls wind from the paired phone and adds temperature on top. Bushnell, to its credit, stops short: the Pro X3+ “Elements” feature adjusts for temperature and barometric pressure and explicitly does not pretend to know the wind.

Four places the number quietly inflates

The cell isn’t your hole. Tree cover, a ridge, a clubhouse, a stand of poplars along the boundary. All of it changes surface roughness, and roughness is what sets the wind profile near the ground. Run the standard log profile and a 20 mph forecast over open grass gives about 14.5 mph at head height. Drop the same air mass into a tree-lined parkland corridor and head height falls to roughly 6 mph.

Direction gets deflected. Trees and terrain channel air along the line of least resistance. A forecast crosswind from 250° becomes a corridor wind running down the hole, or nothing at all. The app resolves the free-stream vector against the hole bearing and has no mechanism for local deflection.

The adjustments stack additively. Wind, elevation and temperature are each computed independently and then summed. None of them cancels the others, and each is derived from a model carrying its own error in the same direction: longer.

It doesn’t know your ball flight. Wind exposure is a function of apex height and time in the air. A 15-handicap who hits a 7-iron 148 total with a 62-foot apex takes less wind than a model calibrated on a 95-foot tour flight. Arccos learns your distances from thousands of shots. It does not learn your apex.

Rough working figures for a mid-flight club golfer:

Note the asymmetry. Into the wind costs you roughly 0.8% of carry per mph; downwind hands back about 0.5%. Anyone telling you to “take one less, it’s downwind” is giving away half the yardage they think they are.

The 7th, worked properly

Back to 168. Say the forecast is 14 mph from 30° off the flag, the ground rises 8 m to the green, and it’s 9°C on a late September morning.

The app’s arithmetic, reconstructed:

• Headwind component: 14 × cos30° = 12 mph → 12 × 0.8% × 168 = +16 yards
• Elevation: 8 m = 26 ft, at the one-yard-per-three-feet rule = +9 yards
• Temperature, 9°C against a ~21°C baseline = +3 yards
• Plays like 196

Now the version you’d get with instruments. A £110 Kestrel 1000 held up on the tee reads 7 mph of headwind component, not 12, because you’re sheltered and the corridor is deflecting the flow. Your laser’s inclinometer puts the actual rise at 6 m, not 8, because the app’s terrain model sampled a 30-metre grid square and missed that the green sits in a shallow saddle. Redo it: 168 + 9 + 7 + 3 = 187.

So the real answer lives somewhere in 187 to 196. That 196 is the ceiling: the most this hole can play if every assumption breaks against you. It is not an instruction.

What you do with a nine-yard band depends entirely on what’s behind the green, which is the part the app can’t weigh. Grassy bank you can chip off? Take the 5-iron, swing at 85%, accept that you might be pin-high left. Front bunker and a downslope through the back? Six-iron, hit the front third, take your 25-footer and move on. The broader argument about when to hand the decision to software at all is worth reading separately in our pillar on AI caddies and on-course decisions.

Elevation is the term you can actually fix

Most golf apps derive elevation from a free global digital elevation model. SRTM at one arc-second gives roughly 30-metre postings with vertical error commonly in the 6 to 10 metre range over UK terrain. OS Terrain 50 is a 50-metre grid. Neither resolves a green sitting on a 3-metre plateau 20 metres wide, which is most raised greens on most parkland courses in Britain. Layer on phone GPS vertical accuracy, typically two to three times worse than horizontal, and the app’s confident “+9 yards” is really a range from +7 to +20.

A laser with an inclinometer is a genuinely different instrument. It measures the angle to the thing you pointed it at, and the trig is not a model:

line of sight  = 171 yd   (what the laser ranged)
inclination    = +4.2°
horizontal     = 171 × cos(4.2°)  = 170.5 yd
rise           = 171 × sin(4.2°)  =  12.5 yd = 37.6 ft
plays-like add ≈ 37.6 / 3         =  12.5 yd

Bushnell Pro X3+, Garmin Approach Z82 and the Nikon Coolshot ProII all do this properly. Remember to switch slope off in competition; obtaining elevation change is not permitted unless a local rule says otherwise.

Calibrate your own course, then calibrate yourself

Buy a handheld anemometer. A WeatherFlow Weather Meter is £35, a Kestrel 1000 around £110, and either one will teach you more about your home course in five rounds than a season of app arrows. Pick six spots: two exposed tees, two chutes, the high point, the low point. Each round, write down what the app said and what you measured. You’ll finish with a personal shelter multiplier per hole, usually somewhere between 0.4 and 0.8 on the tree-lined holes and near 1.0 on the open ones. That single column of numbers converts the app from a voice in your ear into a sanity check.

Then use the launch monitor you already own for the thing it’s uniquely good at, which is apex height. Wind sensitivity is mostly a function of how high the ball goes, and almost nobody knows their own numbers. A Garmin R10 (£500) or Rapsodo MLM2PRO (£420) will give you apex; both estimate rather than directly measure spin, which barely matters here. A typical session sheet for a 14-handicap looks like this:

Club          Carry  Apex   Land ang   Note
PW  stock      125    82 ft   47°       full swing
7i  stock      155    89 ft   45°       full swing
7i  3/4 knock  146    64 ft   39°       ball back 1 ball, shorter finish
7i  chest-high 138    51 ft   34°       hands stop at sternum
5i  stock      185    91 ft   44°

That 7-iron knockdown is the whole point of this exercise. Twenty-five feet off the apex for nine yards of carry means a 12 mph headwind costs you roughly 8 yards instead of 15, and the ball lands at 39° rather than 45° so it holds a line into a crosswind too. Practise it as a distance, not as a trick: hit fifteen chest-high 7-irons, learn the number, write it on your glove.

Your phone footage does have a role, but a narrower one than the app marketing suggests. Film the knockdown from down-the-line and check three things against your stock swing: ball position moved back no more than one ball, hands finishing at chest height, and the same tempo rather than a decelerating stab. Sportsbox AI and V1 Golf will both measure this reliably enough. What they won’t tell you is whether the apex actually dropped, which is why the launch monitor is doing the real work.

Next time you play your own course, put two numbers beside each par 3 in the yardage book: the ceiling, and where the free miss is. The first takes ten minutes with the app and a spreadsheet. The second takes one walk round the back of the green with your eyes open, and it’s the one that will save you strokes in November.