Hidden Signs AI Found to Predict Diving Visibility: Satellite & Weather Data Correlations

March 11, 2026 · 46,000+ real observations · Machine learning model analysis

What AI Learned from 46,000+ Observations

Our dive visibility prediction AI is a LightGBM model trained on 46,000+ real observations combined with weather, marine, and satellite data. Among the features the model values most are 'hidden signs' that humans might not intuitively notice. Here, we analyze which data patterns correlate with high visibility days — essentially, 'when the ocean is clear, here's what the data shows.'

Hidden Sign #1: Satellite kd490 (Diffuse Attenuation Coefficient)

kd490 is the 'diffuse attenuation coefficient' measured by NASA and NOAA ocean color satellites. Lower values mean light penetrates deeper into the water — in other words, clearer water. When matched with real dive logs, days with kd490 < 0.05 showed an overwhelming average visibility of 19.9m.

kd490 Value	Avg Visibility	Obs.	Meaning
Ultra-clear (kd490 < 0.05)	19.9m	7,713	Oceanic clear water; common off Yonaguni/Okinawa
Clear (kd490 0.05–0.10)	13m	23,648	Standard water quality along Japan's coast
Moderate (kd490 0.10–0.20)	10m	10,615	Slightly turbid; common in spring–early summer
Turbid (kd490 > 0.20)	7.7m	1,016	Strongly turbid water

The 12.2m gap between kd490 < 0.05 (clearest) and > 0.20 (turbid) suggests that satellite data alone can approximate visibility even without local dive logs. In our AI prediction model, kd490 is one of the most important features.

Hidden Sign #2: Chlorophyll Concentration

Chlorophyll is the photosynthetic pigment in phytoplankton, measurable by satellite. High-chlorophyll waters have abundant phytoplankton and tend to be less clear. Real data comparison shows 16.9m average for the lowest concentration (< 0.1 mg/m³) vs. 11.2m for the highest (> 0.8 mg/m³) — a 5.7m difference.

Chlorophyll Concentration	Avg Visibility	Obs.	Meaning
Very low (< 0.1 mg/m³)	16.9m	29	Oceanic clear water; minimal plankton
Low (0.1–0.3 mg/m³)	14.7m	3,651	Good water quality
Mid (0.3–0.8 mg/m³)	12.7m	5,132	Typical Japan coastal level
High (> 0.8 mg/m³)	11.2m	700	High plankton → murky tendency

※Chlorophyll from satellite observation; data gaps during cloud cover limit sample sizes.

Hidden Sign #3: 'High Pressure = Clear Water' Is a Myth

The common belief that 'high pressure days are clearer' is contradicted by data. In 46,000+ observations, average visibility during low-pressure conditions (< 1008 hPa) is 14.0m — higher than during strong high-pressure conditions (> 1024 hPa) at 12.7m. High atmospheric pressure does not positively correlate with water clarity.

Air Pressure	Avg Visibility	Obs.	Note
Low pressure (< 1008 hPa)	14m	13,973	Surprisingly high — post-typhoon clearing included?
Average (1008–1016 hPa)	13.8m	16,840	Standard pressure
High pressure (1016–1024 hPa)	13.2m	9,849	High pressure days actually show slightly lower clarity
Strong high (> 1024 hPa)	12.7m	1,541	Prolonged high pressure may drive cold upwelling?

This paradoxical finding may include a 'post-storm clearing effect' — after a typhoon or strong low pressure system passes, oceanic water exchanges and visibility temporarily spikes. The clear sky following a low-pressure system may actually be a prime window. However, this data shows correlation, not causation, and should be interpreted accordingly.

How the AI Uses These Signals

Our LightGBM model considers not just single features like pressure, wave height, chlorophyll, and kd490 individually, but their combinations along with lagged visibility values (3-day, 7-day averages). While 'high pressure = clear' has weak correlation alone, a compound condition like 'low pressure yesterday → high pressure today + previous week avg visibility ≥ 15m + kd490 < 0.08' may strongly suggest high clarity.

The model achieves an R² of 0.824 across all sites, demonstrating that these hidden signs combined into a composite model significantly outperform simple rules of thumb.

Practical Takeaways for Divers

'High pressure = clear' is a false assumption: Data shows good clarity often appears in clear spells after low pressure
Use this site's AI predictions: 8-day forecasts incorporating kd490, chlorophyll and other satellite data updated daily
Last week's visibility is the most reliable sign: The most important model feature is the visibility lag. If it was high recently, it's likely high tomorrow

Summary

Three key insights AI found from 46,000+ observations: ① Lower satellite kd490 = higher visibility (up to 12m difference); ② Lower chlorophyll = higher visibility (5.7m difference); ③ High atmospheric pressure does not directly correlate with visibility (post-low-pressure clear days may actually be better). Some align with human intuition, others contradict it. Data sometimes overturns 'common sense.'

Analysis note: Categories with few observations have wide confidence intervals and should be treated as indicative. The pressure-visibility paradox may vary by region and season.