How accurate is the snow-depth model?

Hand-labeled snow stake images vs. what the fine-tuned Qwen2.5-VL model reads — sampled per resort at roughly 0", 4", 8" and 12".

77
labeled sample images
1.5″
mean absolute error
64%
within 1″ of truth
75%
within 2″ of truth

On the fully held-out test set — 50 labeled images the model never saw during training — the model scores MAE 1.3″, 80% within 1″. Samples below marked training image were part of fine-tuning; unmarked ones are held out.

Model reading vs. measured truth

Each dot is one labeled image. The dashed line is a perfect reading; distance from it is the error.

Per-resort summary

Same data as the cards below, in table form.
ResortimagesMAEbiaswithin 2″
A-Basin 4 2.4″ 1.6″ 75%
Alta 4 2.0″ -2.0″ 75%
Aspen Highlands 2 4.0″ 0.0″ 0%
Aspen Mountain 2 2.0″ 2.0″ 50%
Beaver Creek 4 1.6″ -1.6″ 75%
Breckenridge 4 1.0″ -1.0″ 75%
Brian Head 2 0.5″ -0.5″ 100%
Buttermilk 1 4.0″ -4.0″ 0%
Copper 4 0.5″ -0.5″ 100%
Deer Valley 2 1.5″ -1.5″ 50%
Discovery 2 0.5″ -0.5″ 100%
Eldora 4 3.2″ -3.2″ 25%
Grand Targhee 3 2.7″ -2.7″ 67%
Keystone 4 0.6″ -0.1″ 100%
Loveland 4 0.9″ 0.6″ 100%
Monarch 2 0.0″ 0.0″ 100%
Northstar 2 2.0″ -2.0″ 50%
Park City 3 0.3″ 0.3″ 100%
Powder Mountain 2 2.8″ -1.2″ 50%
Powderhorn 2 2.0″ -2.0″ 50%
Snowbird 4 0.9″ 0.1″ 100%
Snowmass 2 3.0″ -3.0″ 50%
Steamboat 3 1.5″ 0.8″ 100%
Sundance 2 0.5″ -0.5″ 100%
Sunlight 2 2.0″ -2.0″ 50%
Whitefish 3 1.7″ 1.0″ 67%
Winter Park 4 0.4″ -0.1″ 100%

A-Basin

4 labeled samples · MAE 2.4″ · bias 1.6″
A-Basin snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-28
0.0″
measured
0.0″
model
exact
model confidence: high
A-Basin snow stake, measured 5.0 inches
≈4″ conditions · 2026-03-11
5.0″
measured
7.0″
model
+2.0″
model confidence: medium
A-Basin snow stake, measured 6.0 inches
≈8″ conditions · 2026-05-06 training image
6.0″
measured
12.0″
model
+6.0″
model confidence: medium
A-Basin snow stake, measured 12.0 inches
≈12″ conditions · 2026-03-06 training image
12.0″
measured
10.5″
model
-1.5″
model confidence: medium

Alta

4 labeled samples · MAE 2.0″ · bias -2.0″
Alta snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-18
0.0″
measured
0.0″
model
exact
model confidence: high
Alta snow stake, measured 2.0 inches
≈4″ conditions · 2026-03-04
2.0″
measured
0.0″
model
-2.0″
model confidence: high
Alta snow stake, measured 7.0 inches
≈8″ conditions · 2025-12-27
7.0″
measured
6.0″
model
-1.0″
model confidence: medium
Alta snow stake, measured 11.0 inches
≈12″ conditions · 2025-12-27
11.0″
measured
6.0″
model
-5.0″
model confidence: medium

Aspen Highlands

2 labeled samples · MAE 4.0″ · bias 0.0″
Aspen Highlands snow stake, measured 2.0 inches
≈0″ conditions · 2026-04-27 training image
2.0″
measured
6.0″
model
+4.0″
model confidence: medium · reading is a 24-h total stake
Aspen Highlands snow stake, measured 6.0 inches
≈8″ conditions · 2026-03-07 training image
6.0″
measured
2.0″
model
-4.0″
model confidence: medium · reading is a 24-h total stake

Aspen Mountain

2 labeled samples · MAE 2.0″ · bias 2.0″
Aspen Mountain snow stake, measured 2.0 inches
≈0″ conditions · 2026-04-27 training image
2.0″
measured
2.0″
model
exact
model confidence: medium · reading is a 24-h total stake
Aspen Mountain snow stake, measured 6.0 inches
≈8″ conditions · 2025-12-04 training image
6.0″
measured
10.0″
model
+4.0″
model confidence: medium · reading is a 24-h total stake

Beaver Creek

4 labeled samples · MAE 1.6″ · bias -1.6″
Beaver Creek snow stake, measured 0.0 inches
≈0″ conditions · 2026-04-04
0.0″
measured
0.0″
model
exact
model confidence: high
Beaver Creek snow stake, measured 3.5 inches
≈4″ conditions · 2026-02-12
3.5″
measured
3.0″
model
-0.5″
model confidence: medium
Beaver Creek snow stake, measured 7.0 inches
≈8″ conditions · 2025-12-04
7.0″
measured
7.0″
model
exact
model confidence: medium
Beaver Creek snow stake, measured 12.0 inches
≈12″ conditions · 2026-02-21
12.0″
measured
6.0″
model
-6.0″
model confidence: medium

Breckenridge

4 labeled samples · MAE 1.0″ · bias -1.0″
Breckenridge snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-07 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Breckenridge snow stake, measured 4.0 inches
≈4″ conditions · 2025-12-05 training image
4.0″
measured
0.0″
model
-4.0″
model confidence: high
Breckenridge snow stake, measured 8.0 inches
≈8″ conditions · 2026-01-25 training image
8.0″
measured
8.0″
model
exact
model confidence: medium
Breckenridge snow stake, measured 12.0 inches
≈12″ conditions · 2026-01-24 training image
12.0″
measured
12.0″
model
exact
model confidence: medium

Brian Head

2 labeled samples · MAE 0.5″ · bias -0.5″
Brian Head snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Brian Head snow stake, measured 4.0 inches
≈4″ conditions · 2026-02-18 training image
4.0″
measured
3.0″
model
-1.0″
model confidence: medium

Buttermilk

1 labeled sample · MAE 4.0″ · bias -4.0″
Buttermilk snow stake, measured 6.0 inches
≈4″ conditions · 2026-03-07 training image
6.0″
measured
2.0″
model
-4.0″
model confidence: medium · reading is a 24-h total stake

Copper

4 labeled samples · MAE 0.5″ · bias -0.5″
Copper snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Copper snow stake, measured 4.0 inches
≈4″ conditions · 2026-04-27 training image
4.0″
measured
4.0″
model
exact
model confidence: medium
Copper snow stake, measured 8.0 inches
≈8″ conditions · 2026-05-05 training image
8.0″
measured
6.0″
model
-2.0″
model confidence: medium
Copper snow stake, measured 12.0 inches
≈12″ conditions · 2025-12-06 training image
12.0″
measured
12.0″
model
exact
model confidence: medium

Deer Valley

2 labeled samples · MAE 1.5″ · bias -1.5″
Deer Valley snow stake, measured 0.0 inches
≈0″ conditions · 2025-12-01 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Deer Valley snow stake, measured 3.0 inches
≈4″ conditions · 2025-12-06 training image
3.0″
measured
0.0″
model
-3.0″
model confidence: high

Discovery

2 labeled samples · MAE 0.5″ · bias -0.5″
Discovery snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Discovery snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
3.0″
model
-1.0″
model confidence: medium

Eldora

4 labeled samples · MAE 3.2″ · bias -3.2″
Eldora snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-01 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Eldora snow stake, measured 4.0 inches
≈4″ conditions · 2026-03-06 training image
4.0″
measured
0.0″
model
-4.0″
model confidence: high
Eldora snow stake, measured 8.0 inches
≈8″ conditions · 2026-03-07 training image
8.0″
measured
5.0″
model
-3.0″
model confidence: medium
Eldora snow stake, measured 12.0 inches
≈12″ conditions · 2026-03-07 training image
12.0″
measured
6.0″
model
-6.0″
model confidence: medium

Grand Targhee

3 labeled samples · MAE 2.7″ · bias -2.7″
Grand Targhee snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Grand Targhee snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
3.0″
model
-1.0″
model confidence: medium
Grand Targhee snow stake, measured 11.0 inches
≈12″ conditions · 2026-04-03 training image
11.0″
measured
4.0″
model
-7.0″
model confidence: medium

Keystone

4 labeled samples · MAE 0.6″ · bias -0.1″
Keystone snow stake, measured 0.0 inches
≈0″ conditions · 2026-04-14 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Keystone snow stake, measured 4.0 inches
≈4″ conditions · 2025-12-06 training image
4.0″
measured
4.0″
model
exact
model confidence: medium
Keystone snow stake, measured 8.0 inches
≈8″ conditions · 2025-11-25 training image
8.0″
measured
9.0″
model
+1.0″
model confidence: medium
Keystone snow stake, measured 12.0 inches
≈12″ conditions · 2025-12-06 training image
12.0″
measured
10.5″
model
-1.5″
model confidence: medium

Loveland

4 labeled samples · MAE 0.9″ · bias 0.6″
Loveland snow stake, measured 2.0 inches
≈0″ conditions · 2026-04-16 training image
2.0″
measured
1.5″
model
-0.5″
model confidence: medium
Loveland snow stake, measured 5.0 inches
≈4″ conditions · 2026-05-20 training image
5.0″
measured
6.0″
model
+1.0″
model confidence: medium
Loveland snow stake, measured 9.0 inches
≈8″ conditions · 2026-05-07 training image
9.0″
measured
10.0″
model
+1.0″
model confidence: medium
Loveland snow stake, measured 11.0 inches
≈12″ conditions · 2025-10-29 training image
11.0″
measured
12.0″
model
+1.0″
model confidence: medium

Monarch

2 labeled samples · MAE 0.0″ · bias 0.0″
Monarch snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Monarch snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
4.0″
model
exact
model confidence: medium

Northstar

2 labeled samples · MAE 2.0″ · bias -2.0″
Northstar snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Northstar snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
0.0″
model
-4.0″
model confidence: high

Park City

3 labeled samples · MAE 0.3″ · bias 0.3″
Park City snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-03 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Park City snow stake, measured 4.0 inches
≈4″ conditions · 2026-04-03 training image
4.0″
measured
5.0″
model
+1.0″
model confidence: medium
Park City snow stake, measured 8.0 inches
≈8″ conditions · 2026-04-26 training image
8.0″
measured
8.0″
model
exact
model confidence: medium

Powder Mountain

2 labeled samples · MAE 2.8″ · bias -1.2″
Powder Mountain snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
1.5″
model
+1.5″
model confidence: medium
Powder Mountain snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
0.0″
model
-4.0″
model confidence: high

Powderhorn

2 labeled samples · MAE 2.0″ · bias -2.0″
Powderhorn snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20
0.0″
measured
0.0″
model
exact
model confidence: high · reading is a 24-h total stake
Powderhorn snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02
4.0″
measured
0.0″
model
-4.0″
model confidence: medium · reading is a 24-h total stake

Snowbird

4 labeled samples · MAE 0.9″ · bias 0.1″
Snowbird snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-16 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Snowbird snow stake, measured 4.0 inches
≈4″ conditions · 2026-02-20 training image
4.0″
measured
3.0″
model
-1.0″
model confidence: medium
Snowbird snow stake, measured 8.0 inches
≈8″ conditions · 2026-04-01 training image
8.0″
measured
10.0″
model
+2.0″
model confidence: medium
Snowbird snow stake, measured 14.0 inches
≈12″ conditions · 2026-03-06 training image
14.0″
measured
13.5″
model
-0.5″
model confidence: medium

Snowmass

2 labeled samples · MAE 3.0″ · bias -3.0″
Snowmass snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-23
0.0″
measured
0.0″
model
exact
model confidence: medium · reading is a 24-h total stake
Snowmass snow stake, measured 6.0 inches
≈4″ conditions · 2026-04-27
6.0″
measured
0.0″
model
-6.0″
model confidence: high · reading is a 24-h total stake

Steamboat

3 labeled samples · MAE 1.5″ · bias 0.8″
Steamboat snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
1.5″
model
+1.5″
model confidence: medium
Steamboat snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
6.0″
model
+2.0″
model confidence: medium
Steamboat snow stake, measured 7.0 inches
≈8″ conditions · 2026-03-06 training image
7.0″
measured
6.0″
model
-1.0″
model confidence: medium

Sundance

2 labeled samples · MAE 0.5″ · bias -0.5″
Sundance snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-20 training image
0.0″
measured
0.0″
model
exact
model confidence: high
Sundance snow stake, measured 4.0 inches
≈4″ conditions · 2026-01-02 training image
4.0″
measured
3.0″
model
-1.0″
model confidence: medium

Sunlight

2 labeled samples · MAE 2.0″ · bias -2.0″
Sunlight snow stake, measured 0.0 inches
≈0″ conditions · 2026-05-01 training image
0.0″
measured
0.0″
model
exact
model confidence: high · reading is a 24-h total stake
Sunlight snow stake, measured 4.0 inches
≈4″ conditions · 2025-12-15 training image
4.0″
measured
0.0″
model
-4.0″
model confidence: high · reading is a 24-h total stake

Whitefish

3 labeled samples · MAE 1.7″ · bias 1.0″
Whitefish snow stake, measured 0.0 inches
≈0″ conditions · 2026-02-01 training image
0.0″
measured
3.0″
model
+3.0″
model confidence: medium
Whitefish snow stake, measured 2.0 inches
≈4″ conditions · 2026-01-02 training image
2.0″
measured
3.0″
model
+1.0″
model confidence: medium
Whitefish snow stake, measured 6.0 inches
≈8″ conditions · 2025-12-01 training image
6.0″
measured
5.0″
model
-1.0″
model confidence: medium

Winter Park

4 labeled samples · MAE 0.4″ · bias -0.1″
Winter Park snow stake, measured 0.0 inches
≈0″ conditions · 2026-04-08
0.0″
measured
0.0″
model
exact
model confidence: high
Winter Park snow stake, measured 3.5 inches
≈4″ conditions · 2026-03-07
3.5″
measured
4.0″
model
+0.5″
model confidence: medium
Winter Park snow stake, measured 8.0 inches
≈8″ conditions · 2026-04-28 training image
8.0″
measured
7.0″
model
-1.0″
model confidence: medium
Winter Park snow stake, measured 10.0 inches
≈12″ conditions · 2026-05-02 training image
10.0″
measured
10.0″
model
exact
model confidence: medium
Ground truth is a human reading of the stake in each image. Samples badged “training image” were part of the model’s fine-tuning data, so treat those as illustrations rather than proof — the honest benchmark is the held-out test set quoted above, whose error is essentially identical to this demo’s. Model: Qwen2.5-VL-7B with a LoRA fine-tune, run run_20260518_145633. Generated 2026-08-04 15:39 UTC.