Ambient Occlusion Objective¶
Ambient training learns from sparse EN4/ARGO observations without converting unobserved ocean cells into supervised targets. The model is conditioned on a further-corrupted version of the observed field, while loss is evaluated only where the original sparse field and the paired target are valid over ocean.
Notation¶
x: sparse depth-aligned EN4/ARGO values.m: original sparse observation mask (x_valid_mask).m': further-corrupted mask, withm' ≤ m.y: paired dense target used to establish valid depth/ocean support.v: target-valid mask (y_valid_mask).l: ocean support (land_maskin the batch contract, where ocean is valid).
The model condition contains x ⊙ m', m', the configured dense surface
channels, and coordinate/date context. In the current x0 objective, the
supervision mask is
and the supervised target on that support is the original sparse field x.
Unobserved pixels are not assigned pseudo-labels by this objective.
Corruption policy¶
The local pixel preset uses:
model:
parameterization: x0
clamp_known_pixels: false
ambient_occlusion:
enabled: true
further_drop_prob: 0.25
apply_to_noisy_branch: true
shared_spatial_mask: true
min_kept_observed_pixels: 50
require_x0_parameterization: true
The drop mask is shared spatially across field channels when configured. A
minimum-support guard restores observations when random dropping would leave too
few supervised pixels. With apply_to_noisy_branch=true, the further-corrupted
support also masks the noisy target branch seen by the denoiser.
Constraints¶
- Ambient mode requires
x_valid_maskin each batch. - With
require_x0_parameterization=true, any other parameterization is rejected. - Loss is normalized over valid weighted support, not the full patch area.
clamp_known_pixels=falsemeans sampling does not overwrite predictions with observedxafter each reverse step.- Exporters restore geospatial invalid support and final land nodata after model prediction; that output masking is separate from the training objective.
Relationship to other losses¶
The ambient diffusion term has weight 1.0 in the local preset. Optional sparse observation, profile-increment, GLORYS structure, and spectral terms are disabled. Auxiliary timestep weighting is therefore dormant until an auxiliary term is enabled. See Auxiliary losses.
The HPC synthetic-target and direct-GLORYS presets disable ambient mode and use ordinary dense supervision instead.