Welcome to Mercator Ocean AI competition. You chose this topic as you may be interested in geophysics and environmental field. Do not hesitate to take contact with Mercator Ocean team members to know more about our activities and missions.

You will find below some useful information about your project. 

Oceanic Eddies are  vortices of the order of ~10km in horizontal scale, whose signature is clearly visible in satellite products and well reproduced in ocean models. Sea Level Anomalies (SLA) are used for detecting such object by means of numerical techniques, yet such approach lacks accuracy in detection, namely because Sea Level Anomaly contains errors. Eddy signature is also visible in other variables such as Sea surface Temperature (SST) that do not suffer from such limitation.

The objective here is to develop a Deep Learning approach to identify eddies using SST, SLA and the ocean velocity field. For this purpose you have a dataset consisting of  SLA, SST, and velocity components images which are slightly distorted relative to reality (we call this distorted reality - the reanalysis product), together with labels (the eddy contours, see figure) obtained directly from reality.

Since we can not obtain the real state of the ocean, the so called reality comes from it's virtual emulation. Consequently, we suppose that the ocean is a perfect fluid which can be described by primitive equations (e.g. Navier-Stokes). This virtual reality is then considered as a real time ocean state. 

On the other hand, in reality oceanographers have to model the ocean dynamics driven among other by various sources forcing, such as wind or atmosphere-ocean interactions. Furthermore, thanks to the remote sensing advances, a significant amount of satellite tracked and in-situ observations data are available all around the globe. This data is extremely useful as it is the only source providing the knowledge of the real ocean dynamics. In operational oceanography the process of incorporating the real ocean observations (remote and in-situ) into the modeled ocean an assimilation. However, the assimilation process is in it's core a numerically challenging task, firstly due to the sparse spatial and temporal distribution of observations, and secondly due to the resolution issues. It is worth mentioning, that a number of compromises have to be accepted during the assimilation process so that the modeled ocean contains as much real observations data as possible whilst the overall flui dynamics is not disturbed. Those compromises are source of inaccuracies in modeled ocean representation. Operational oceanography is in constant research of novel ways of data assimialtion, model validation and correction. 

This project presents a big interest for operational oceanography regarding the importance of a reliable ocean dynamics modeling for real time analysis,  forecast, as well as for numerous applications, for instance floating trajectories simulation. being able to detect eddies in the distorted virtual reality, called here and after " NatRun OSSE reanalysis ", would provide a novel way for oceanographers to detect eddies in real modeled ocean, called here and after " Reanalysis product ". Thus, your contribution may open a brand new way of reanalysis product validation through AI eddies detection framework.