# Solar Eclipse Event Replay — 12 August 2026

## Conclusion

EDS captured a strong, publication-worthy eclipse record.

The ambient-light result is decisive: the station's minimum reading was **11 lux at 19:32:22 WEST**, inside the same one-minute sample as the predicted **19:32:14** local maximum. The five-minute mean fell **99.0%**, from 1,818.4 lux at first contact to 17.8 lux at maximum, before rebounding by 229.4 lux over the next 15 minutes.

The temperature and humidity traces are scientifically interesting but much less conclusive. Temperature fell 1.59 °C and relative humidity rose 5.27 percentage points between first contact and maximum, yet the preselected matched evenings in this archive behaved almost the same. The strongest defensible interpretation is:

> EDS unambiguously detected the eclipse in ambient light and captured a possible small thermal response. It did not, by itself, prove that the eclipse caused the full temperature, humidity, pressure, or dew-point changes.

## Visual Asset Pack

All graphics are available as editable SVG and presentation-ready PNG.

| Asset | Best use | Vector | PNG |
|---|---|---|---|
| Light hero | Article hero, presentation, landscape social post | [SVG](../assets/charts/eclipse_2026_08_12_light_hero.svg) | [PNG](../assets/charts/eclipse_2026_08_12_light_hero.png) |
| Control-day comparison | Technical article, study explanation, credibility slide | [SVG](../assets/charts/eclipse_2026_08_12_control_comparison.svg) | [PNG](../assets/charts/eclipse_2026_08_12_control_comparison.png) |
| Environmental response | Technical analysis and secondary carousel slide | [SVG](../assets/charts/eclipse_2026_08_12_environment_response.svg) | [PNG](../assets/charts/eclipse_2026_08_12_environment_response.png) |
| Square social card | Cosmonutz/EDS social post and carousel cover | [SVG](../assets/charts/eclipse_2026_08_12_social_card.svg) | [PNG](../assets/charts/eclipse_2026_08_12_social_card.png) |

Derived public-safe data:

- [111-reading eclipse window CSV](../assets/data/eclipse_2026_08_12_event_window.csv)
- [analysis metrics and five-minute series JSON](../assets/data/eclipse_2026_08_12_summary.json)

## The Primary Signal: Ambient Light

![EDS solar eclipse ambient-light hero](../assets/charts/eclipse_2026_08_12_light_hero.svg)

Three pieces of evidence make the light result unusually strong.

### 1. Timing

The predicted maximum for Espinho was 19:32:14 WEST. EDS sampled 11 lux at 19:32:22—eight seconds later and therefore inside the one-minute sample containing maximum.

The eclipse was partial at the station, with 97.9% of the solar disc covered. Local circumstances are independently documented in the [Espinho eclipse schedule](https://en.tutiempo.net/solar-eclipse/espinho/12-august-2026.html); NASA describes the wider 12 August event and its partial visibility outside the narrow totality path in its [2026 eclipse overview](https://science.nasa.gov/eclipses/future-eclipses/total-solar-eclipse-on-august-12-2026/).

### 2. Depth

The five-minute mean at maximum was 17.8 lux:

- **99.0% below** the first-contact mean of 1,818.4 lux;
- **98.6% below** the control-day median at the same time;
- **92.7% below** even the darkest of the 12 control evenings at maximum.

The sensor therefore captured much more than ordinary late-evening dimming.

### 3. Shape

After maximum, light rose by 229.4 lux in 15 minutes even though the Sun continued descending toward the horizon. Every control evening darkened over the corresponding interval, with changes ranging from −436.8 to −101.6 lux.

That post-maximum reversal is the most useful causal clue in the dataset. A smooth sunset cannot create it. A transient cloud edge could change light quickly, but the minimum's exact alignment with astronomical maximum and the control-day contrast make the eclipse the best-supported explanation in this dataset.

## Why the Control Comparison Matters

![EDS eclipse normalized light compared with control evenings](../assets/charts/eclipse_2026_08_12_control_comparison.svg)

Raw light varies from day to day because of clouds, haze, sensor exposure, and the seasonal solar angle. To make the comparison fair, every evening was normalized to its own light level at first contact.

At maximum:

- eclipse day retained only **1.0%** of its first-contact light;
- the 12 control days retained **45.8–83.7%**;
- the control median retained **64.3%**.

The normalized result is important because it shows that the signal is not an artifact of 12 August simply being a darker day. Its *within-evening trajectory* was categorically different.

## Environmental Response

![EDS temperature humidity dew point and pressure response](../assets/charts/eclipse_2026_08_12_environment_response.svg)

### Temperature

From first contact to maximum:

- eclipse day: **−1.59 °C**;
- 12-control-day median: **−0.91 °C**;
- two preselected matched evenings: **−1.40 °C**;
- eclipse-minus-matched difference: only **−0.19 °C**.

By last contact, the eclipse day had cooled 2.68 °C, versus 2.53 °C on the preselected matched evenings—an excess of only 0.15 °C.

My interpretation: the curve is consistent with a small eclipse-enhanced cooling response, but the additional cooling is too small to isolate confidently from ordinary coastal sunset variability, local airflow, sensor housing effects, and one-minute measurement noise. It is an interesting hypothesis, not a result to advertise as proven.

### Relative humidity

From first contact to maximum:

- eclipse day: **+5.27 percentage points**;
- preselected matched-evening median: **+5.49 percentage points**.

By last contact, humidity had risen 10.21 points, only 0.60 points more than the preselected matched evenings.

This strongly suggests that most of the humidity rise was the expected relative-humidity response to falling evening temperature rather than a unique moisture event caused by the eclipse. No sudden independent moisture pulse is evident.

### Dew point and saturation tendency

The derived dew point changed from 19.44 °C at first contact to 19.09 °C at maximum and 19.04 °C at last contact. It remained far steadier than air temperature.

As a result, the temperature–dew-point spread compressed:

- first contact: approximately **5.87 °C**;
- maximum: approximately **4.64 °C**;
- last contact: approximately **3.59 °C**.

That is a real increase in saturation tendency, but it is also normal when evening air temperature approaches a comparatively stable dew point. The controls do not support claiming a distinct eclipse-driven dew or fog event.

### Pressure

Sea-level pressure rose 0.23 hPa by maximum and 0.44 hPa by last contact. These changes sit inside the nearby-day control range. They are too small and too meteorologically confounded to associate with the eclipse.

## Data Quality

The event record is unusually clean:

- 111 readings between first and last contact;
- exactly one reading per minute;
- mean interval: 60 seconds;
- maximum interval: 60 seconds;
- no gaps or station dropout;
- battery and Wi-Fi telemetry remained present throughout the extracted window.

The comparison uses 18,442 readings from 5–17 August. Twelve non-eclipse evenings form the wider reference set. The 9 and 10 August evenings are bundled as preselected matched controls because their temperature, humidity, dew point, and light during the 30 minutes before first contact were manually chosen as the closest pair to the eclipse day; the current public script documents that choice but does not compute it.

Five-minute centered means use all samples within ±150 seconds of each target time. All stage-to-stage comparisons use changes from first contact rather than only comparing absolute sensor values. The compact 111-row public CSV is enough to reproduce the event-window table, but the wider control metrics still depend on the larger read-only comparison export, which is not bundled here.

## What I Think the Dataset Shows

This is a genuinely strong EDS case study—not because every sensor shows a dramatic eclipse effect, but because the system recorded the event with enough timing precision and continuity to distinguish the obvious signal from the ambiguous ones.

The most valuable findings are:

1. **Event detection:** the illuminance trace identifies maximum eclipse almost to the sample.
2. **Counterfactual evidence:** nearby evenings show that the V-shaped light curve was not normal sunset behavior.
3. **System reliability:** the station captured the entire astronomical window without a missing minute.
4. **Scientific restraint:** the same dataset prevents an exaggerated temperature claim. That honesty strengthens the EDS story.

The best public framing is:

> EDS watched the local environment through a 97.9% solar eclipse. It measured the sky's light collapsing at astronomical maximum, tracked the recovery minute by minute, and captured a possible—but not yet separable—thermal response.

## Limitations

- One station and one eclipse cannot establish a general climatological response.
- EDS did not directly measure cloud fraction or wind during the event.
- The eclipse occurred close to sunset, so the eclipse signal and the normal evening transition overlap.
- Air temperature responds more slowly than light and depends heavily on ventilation, surface heating, station exposure, and local advection.
- The BH1750 illuminance measurement is excellent for relative timing and curve shape here, but the current analysis does not claim laboratory-grade absolute irradiance calibration.
- Camera filenames are treated as nominal first-frame local timestamps; camera-to-EDS synchronization remains provisional rather than certified.

## Recommended Next Study

1. Add independent cloud, wind, and visibility observations for 12 August.
2. Fit an expected sunset curve from the matched controls and analyze residual temperature response with 5–30 minute lags.
3. Compare the time of steepest temperature descent with eclipse obscuration rather than only using the exact maximum.
4. Quantify sensor noise and enclosure thermal lag from stable clear evenings.
5. Preserve the event as a named EDS dataset window for future multi-station or multi-eclipse comparisons.
6. If future stations are available, compare coastal, urban, and inland response simultaneously.

## Reproduction

The dependency-free generator is:

```text
scripts/generate_eclipse_assets.py
```

Run from the `solar_eclipse_2026_site` project root with the read-only comparison export:

```bash
python3 scripts/generate_eclipse_assets.py \
  --input /tmp/eds_eclipse_2026-08-12_telemetry.csv
```

The script writes four SVG files, the derived JSON summary, and the exact event-window CSV. PNG files are raster exports of the SVG masters.