 This paper proposes a novel fully coupled paradigm to improve the retrieval accuracy of soil moisture and land surface temperature from passive microwave remote sensing data. It combines deep learning with physical and statistical methods to solve the ill-posed problem caused by insufficient observational information. The proposed paradigm overcomes the limitations of traditional methods and provides a new way to retrieve geophysical parameters. This article was authored by Kebyao Mao, Han Wang, Zhen Chengshu, and others.