Shield correction · reference check
Against Plzeň-Mikulka
The ČHMÚ reference minus the calculated shade temperature on one local calendar day. Choose a date up to 30 days back.
Ten-minute slots, Europe/Prague. Positive means Mikulka is warmer. Gaps mean missing data or an unavailable correction. ČHMÚ observations can arrive later. Days when the clocks change contain 23 or 25 hours.
Last 30 days · independent of selected date
Deviation from Mikulka
1 001 matched observations. Both columns use the same observations, within 90 seconds of a usable Mikulka reading, after calibration validation.
| Difference, °C | Shield sensor | Shade estimate |
|---|---|---|
| Mean absolute deviation | 0,59 | 0,45 |
| Mean signed difference | +0,46 | +0,25 |
| Minimum signed difference | −1,94 | −1,98 |
| Maximum signed difference | +5,71 | +2,33 |
3 148 reference matches excluded because the correction was unavailable.
Signed differences are station minus Mikulka; positive means warmer. Mikulka measures at a different location, so local weather differences are included in these deviations. Historical estimates use only calibrations already validated at the time of measurement.
All calibrations · independent of selected date
Model history
| Model | Status | Trained | Validated | Observations | Previous model MAE, °C | This model MAE, °C | Improvement, °C | Sun-facing improvement, °C | In use |
|---|---|---|---|---|---|---|---|---|---|
| #2 | in use | 2.10.2026 - 15.10.2026 | 16.10.2026 - 18.10.2026 | 431 | 0,43 | 0,42 | +0,01 | +0,05 | 19.10.2026 00:00 - now |
| #1 | accepted | 30.9.2026 - 9.10.2026 | 10.10.2026 - 14.10.2026 | 719 | - | 0,59 | - | - | 15.10.2026 00:00 - 19.10.2026 00:00 |
Mean absolute deviation from Mikulka on each model's validation days, against the model that was in use before it, on the observations where both have an estimate. Improvement is the previous model's MAE minus this model's; positive means this model is closer to Mikulka. A model under validation shows its days so far. A dash under the previous model means none was in use; this model's MAE then covers all its validation observations. The highlighted row corrects the page now.