Shirasaki Visibility Analysis — High Prediction Accuracy from Limited Data

Shirasaki, located in Yura Town, Wakayama Prefecture, is a diving site whose name literally means "white cape." Towering white limestone cliffs plunge into the sea, creating a landscape so striking it has earned the nickname "the Aegean Sea of Japan." These white rock faces, formed from Permian-era limestone approximately 250 million years ago, extend beneath the waves as well, producing an otherworldly underwater landscape found nowhere else in Japan.

Our site has collected 536 days of visibility data for Shirasaki. While this is a relatively modest dataset, our AI model achieves a remarkable AI prediction accuracy: 60.6% from these observations alone. Why does such a small dataset yield such high prediction accuracy? This article explores Shirasaki's visibility patterns while investigating the reasons behind this surprising result.

Monthly Visibility Patterns

Shirasaki's monthly visibility data displays the clear seasonal pattern common to Pacific-coast dive sites: highest visibility in winter, declining through spring and summer. This pronounced seasonality is a key factor in the AI's strong performance.

Situated on the western coast of the Kii Peninsula under the influence of the Kuroshio Current, Shirasaki benefits from the inflow of clear oceanic water during winter, driving high visibility. In spring, phytoplankton blooms create the characteristic "spring turbidity" that reduces visibility. This pattern closely mirrors other Kii Peninsula dive sites such as Kushimoto, reflecting the regional consistency of the marine environment.

How 536 Observations Achieve AI Prediction Accuracy: 60.6%

In AI prediction, limited data typically means lower prediction accuracy. Among sites in our database — some with over 3,000 records — Shirasaki's 536 observations place it near the bottom. Yet the general AI model (trained on data from all sites) achieves an AI prediction accuracy: 60.6% for Shirasaki. This comfortably surpasses Kushimoto (AI prediction accuracy: 41.5%) and Tajiri (AI prediction accuracy: 28.0%).

The key to this high accuracy lies in the general AI model architecture. Our system trains a single AI model on over 46,000+ observations across all 14 sites. This approach allows information gaps in Shirasaki's limited dataset to be filled by similar patterns learned from other sites.

Notably, a site-specific AI model trained only on Shirasaki data achieves an AI prediction accuracy of just 45.7%. The gap of +14.9 points compared to the general AI model demonstrates how effectively knowledge transfer from other sites operates. In particular, the 3,168 records from Kushimoto — located on the same Kii Peninsula — likely contribute significantly to learning Shirasaki's visibility patterns. Shared oceanographic drivers such as Kuroshio influence, seasonal plankton blooms, and coastal topographic effects can be learned from Kushimoto's abundant data and applied to Shirasaki.

Why the General AI Model Works So Well at Shirasaki

The general AI model does not perform equally well at every site. At Tajiri on the Sea of Japan side, AI prediction accuracy drops to 28.0%, indicating that patterns learned from Pacific-side sites do not transfer effectively. So why does it succeed at Shirasaki?

First, clear seasonal patterns. Shirasaki displays the classic Pacific-coast pattern of high winter visibility and low summer visibility, shared with numerous sites including Izu Oceanic Park, Osezaki, and Kushimoto. The AI captures this seasonality by reading seasonal changes in the data.

Second, shared Kuroshio influence. Shirasaki sits on the western coast of the Kii Peninsula while Kushimoto is at its southern tip, but both are directly affected by changes in the Kuroshio Current's path. The mechanism — visibility improves when the Kuroshio approaches and declines when it recedes — is common to both sites, allowing the AI to apply Kuroshio-related patterns learned from Kushimoto to Shirasaki.

Third, effective meteorological and marine data. As an ocean-facing cape, Shirasaki's visibility responds directly to wind, waves, and swell conditions. The AI's 45 types of input data — including wind speed, precipitation, wave height, and swell period — capture the environmental variability at Shirasaki particularly well.

Yearly Visibility Trends

Shirasaki's yearly data covers a limited period, making it premature to draw definitive conclusions about long-term trends. However, several patterns can be discerned from the available data.

Year-to-year variation is relatively large, likely linked to annual fluctuations in the Kuroshio Current's path. During periods of Kuroshio large meander, the pattern of Kuroshio approach to the western Kii Peninsula coast changes, affecting Shirasaki's visibility. As data accumulation continues, it will become possible to quantitatively analyze the relationship between Kuroshio path and Shirasaki visibility with greater precision.

Comparison with Nearby Kushimoto

Shirasaki and Kushimoto both sit on the Kii Peninsula in Wakayama Prefecture and share the Kuroshio Current's influence. Kushimoto is at the peninsula's southern tip while Shirasaki is on the western coast, separated by roughly 60 km in a straight line. This positional difference introduces subtle variations in their visibility patterns.

Kushimoto is the closest point on Honshu to the main Kuroshio stream, and visibility improves directly when the current approaches. Shirasaki is farther from the main Kuroshio stream, but warm water branching from the Kuroshio flows northward through the Kii Channel to reach the Shirasaki area. The Kuroshio's influence is therefore less direct than at Kushimoto, but still substantial.

From the AI model's perspective, Kushimoto's abundant data (3,168 records) effectively serves as "training data" for Shirasaki predictions. Because the oceanographic mechanisms driving visibility at both sites are similar, the general AI model's knowledge transfer operates effectively, enabling Shirasaki's AI prediction accuracy of 60.6% despite its limited local dataset.

Practical Diving Advice

Shirasaki's Appeal: The White Limestone Underwater World

Shirasaki's greatest appeal is the white limestone underwater landscape found at no other dive site in Japan. The white cliffs that earn the "Aegean Sea of Japan" nickname on land continue beneath the surface, where the contrast between white rock and blue water is breathtaking. On high-visibility days, this white-and-blue contrast becomes even more dramatic.

Best Season

As the data shows, Shirasaki's visibility peaks in winter. From December through February, warm Kuroshio-derived water flows into the area, and the clear water showcases the white limestone scenery at its finest. Thanks to the Kii Peninsula's relatively mild climate, winter water temperatures do not drop as low as at sites on the Izu Peninsula.

Planning Recommendations

• Best visibility: December through February. The season when white rock contrasts most vividly with clear blue water.
• Comfort balance: October and November offer a good balance of water temperature and visibility.
• Topographic diving: The unique limestone formations can be enjoyed year-round.
• Periods to avoid: Spring (March through May) brings plankton blooms that reduce visibility. Typhoon season may cause closures.

Access

Shirasaki is approximately 2 hours by car from Osaka, with Kii-Yura Station on the JR Kinokuni Line as the nearest railway stop. Diving is typically based at Shirasaki Marine Park, which also offers camping and barbecue facilities for non-diving activities. Extending the trip to Kushimoto allows for a different style of diving, making a comprehensive Kii Peninsula dive trip.