Inversion of Remote Sensing Synthetic Data for Essential Climate Variables Characterization

The retrieval of Essential Climate Variables (ECVs) from Earth Observation (EO) data has become increasingly important, as microwave remote sensing has demonstrated unprecedented capabilities for systematic global monitoring of environmental conditions. However, the retrieval remains a highly challenging problem. Over the years, various approaches have been employed, ranging from theoretical and semi-empirical models to statistical methods. The recent development and standardization of machine learning algorithms has opened new possibilities for accurate ECV retrieval. However, these approaches typically require large amounts of data paired with ancillary information, difficult to obtain through purely experimental means. In this context, we consider a theoretical framework for generating synthetic data based on electromagnetic scattering theory, capable of producing large datasets with fully controlled biophysical parameters. These synthetic datasets are then used to train physics-informed neural networks for the retrieval of ECVs from real EO data.