Microwave Remote Sensing of Soil Moisture, Above Ground Biomass and Freeze-Thaw Dynamic: Modeling and Empirical Approaches

TL;DR

This paper addresses the challenge of retrieving key Essential Climate Variables ($SM$, $AGB$, $FT$) from microwave remote sensing data, highlighting inverse-problem approaches and the RETINA project's hybrid framework that combines deterministic analytic inversions with Bayesian methods. It surveys theoretical (RTT, wave theory), semi-empirical (WCM), and empirical models, and discusses how ML and data fusion enhance retrieval accuracy, while introducing the RETINA dataset and multivariate NN operators for robust 2D geophysical estimations. The authors present NN operators $F^d_n$ and $K^d_n$ with IVFS and Bayesian inversion via MCMC/PCA to enable uncertainty-aware, scalable retrievals, and emphasize GPU-enabled parallel computation for large-scale applications. The work underscores the potential of integrating physics-based modeling with probabilistic learning to improve global monitoring of climate-relevant variables and to inform policy and modeling efforts.

Abstract

Human actions have accelerated changes in global temperature, precipitation patterns, and other critical Earth systems. Key markers of these changes can be linked to the dynamic of Essential Climate Variables (ECVs) and related quantities, such as Soil Moisture (SM), Above Ground Biomass (AGB), and Freeze-Thaw (FT) Dynamics. These variables are crucial for understanding global climate changes, hydrological and carbon cycles included. Monitoring these variables helps to validate climate models and inform policy decisions. Technologies like microwave remote sensing provide critical tools for monitoring the effects of human activities on these variables at a global scale. Other than proper tachenological developments, the study of ECVs requires suitable theoretical retrieval tools, which leads to the solutions of inverse problems. In this brief survey, we analyze and summarize the main retrieval techniques available in the literature for SM, AGB, and FT, performed on data collected with microwave remote sensing sensors. Such methods will be some of the fundamental algorithms that can find applications in the research activities of the interdisciplinary, curiosity-driven, project {\it REmote sensing daTa INversion with multivariate functional modeling for essential climAte variables characterization (RETINA)}, recently funded by the European Union under the Italian National Recovery and Resilience Plan of NextGenerationEU, under the Italian Ministry of University and Research. The main goal of RETINA, in which three research units from three different italian universities are involved, is to create innovative techniques for analyzing data generated by the interaction of electromagnetic waves with the Earth's surface, applying theoretical insights to address real-world challenges.

Figures (3)

• Figure 1: The official logo of the RETINA project.
• Figure 2: The screen-shot of the RETINA dataset web page, available at the following link: https://retina.sites.dmi.unipg.it/dataset.html.
• Figure 3: RS images extracted from the RETINA dataset. On the top (from left to right): Berlin (Germany) and Granada (Spain) areas. On the bottom: (from left to right): Rome (Italy) and Lisbon (Portugal) areas.