OCEAN
EMBED
Connecting ocean observations and data to make what lies beneath easier to explore and understand.
SCROLL TO EXPLORE
0m
float descending
Scroll to know
OCEAN
THE OCEAN IS MORE
THAN ITS SURFACE.
Satellites continuously observe the ocean surface and provide valuable information about large-scale physical conditions.
But the surface does not tell the whole story.
WHAT LIES BELOW
REMAINS HIDDEN.
Subsurface temperature reveals how heat is distributed through the water column and helps us understand the changing state of the ocean.
Yet direct measurements become increasingly sparse with depth.
SURFACE DATA IS ABUNDANT.
DEPTH DATA IS NOT.
Satellites provide broad and continuous observations at the surface, while subsurface observations depend on instruments and measurements distributed across the ocean.
The result is an uneven picture of what is happening beneath the surface.
CAN SURFACE SIGNALS
REVEAL THE DEPTHS?
OceanEmbed explores whether observable ocean conditions can be used to reconstruct information about the subsurface ocean.
This is the gap we aim to bridge.
BUILD THE
OCEAN DATASET.
We begin with ocean data that represents conditions across locations and depths.
GLORYS provides the foundation for learning the relationship between surface and subsurface conditions.
TRAIN
THE MODEL.
The model learns patterns connecting observable ocean conditions with variables beneath the surface.
Temperature · Height · Depth
Training across many locations allows the model to learn relationships that are difficult to observe directly everywhere.
RECONSTRUCT
THE HIDDEN OCEAN.
Once trained, the model can estimate subsurface conditions from available ocean observations.
From what we can observe
to what we cannot directly see.
TEST AGAINST
REAL OBSERVATIONS.
Predictions need to be tested against observations that were not used simply as model outputs.
ARGO FLOATS PROVIDE AN IMPORTANT REFERENCE.
Their measurements through the water column allow us to compare reconstructed conditions with observed ocean conditions.
MEASURE
THE ERROR.
We evaluate how closely the reconstructed ocean matches the observations.
RMSE
BIAS
CORRELATION
Together, these metrics help describe the accuracy, systematic error, and ability of the model to capture observed patterns.
LEARN. COMPARE.
IMPROVE.
Model evaluation is not the final step. Prediction errors reveal where the reconstruction can be improved.
PREDICT → COMPARE → MEASURE → IMPROVE
Repeated training and validation help us understand the strengths and limitations of the approach.
FROM TRAINING TO
REAL-WORLD PREDICTION.
After evaluation, the model can be applied to locations where direct subsurface observations are limited.
SURFACE OBSERVATIONS → SUBSURFACE ESTIMATION
The goal is not to replace direct measurements, but to extend our ability to understand the ocean between them.
FROM SCATTERED
OBSERVATIONS TO A
CLEARER VIEW
OF THE OCEAN.
OceanEmbed aims to make subsurface ocean information easier to explore across locations where direct measurements are limited.
Observe the surface.
Learn the hidden patterns.
Reconstruct the depths.