This repository documents the methodology used to quantify and model temperature misfits between a physical ocean reanalysis and independent observations, with a specific focus on applications to biologging tag geolocation.
The core objective is to characterize depth-dependent uncertainty in ocean temperature fields using Argo float observations, and to assess whether Gaussian error assumptions are appropriate when comparing model outputs to biologging tag measurements.
Two complementary notebooks are provided:
variance_medit.ipynb: estimation of depth-dependent variance from Argo–model comparisons,test_of_kde.ipynb: evaluation of normalized tag–model errors and non-Gaussian misfit modelling.
Together, these notebooks support the development of a physically grounded and statistically robust likelihood formulation for temperature-based geolocation.
Ocean general circulation models, even when constrained by data assimilation, remain imperfect representations of the true ocean state. Temperature fields produced by such models may exhibit:
- Systematic biases relative to in situ observations,
- Depth-dependent variability and uncertainty,
- Non-Gaussian error structures arising from unresolved processes.
When these model fields are used in inverse problems—such as reconstructing fish trajectories from biologging data—ignoring these characteristics can lead to overconfident likelihoods and unrealistic reconstructed tracks.
This project addresses two key questions:
- How does the variance of model–observation temperature misfits depend on depth?
- Are normalized temperature errors consistent with a Gaussian distribution, or do they exhibit systematic departures requiring a more flexible likelihood model?
The reference temperature fields are extracted from the Copernicus Marine Service physical ocean reanalysis, providing daily three-dimensional temperature fields on a fixed latitude–longitude grid.
The model captures large-scale and mesoscale ocean variability but does not fully resolve fine-scale processes, motivating an explicit characterization of uncertainty.
Argo float temperature profiles are used as an independent observational reference. These profiles provide vertically resolved, in situ measurements spanning a wide range of depths and oceanic conditions.
Argo data are not used to correct the model state itself, but rather to diagnose the statistical properties of model–observation misfits.
Temperature and pressure records from biologging tags are used to:
- Validate the Argo-derived uncertainty estimates,
- Test whether tag–model errors behave consistently with expected variance,
- Diagnose departures from Gaussian error assumptions.
.
├── README.md
├── variance_medit.ipynb
└── test_of_kde.ipynb