Code accompanying the paper:

Optimal sensor placement for the reconstruction of ocean states using differentiable Gumbel-Softmax sampling operator

Overview

This repository provides the implementation of a differentiable framework for optimal sensor placement in ocean-state reconstruction under strict observation-budget constraints.

The method combines:

• a Gumbel-Softmax-based differentiable sampling operator
• a budget-aware probabilistic observation mask
• a joint optimization of sensor placement and reconstruction parameters
• an application to Sea Surface Height (SSH) reconstruction using Optimal Interpolation (OI)

The framework is designed for adaptive observation network design in geophysical systems and is evaluated through Observing System Simulation Experiments (OSSEs) on Gulf Stream SSH data.