Survey on AI-Driven Models for Soil Moisture Estimation and Classification

A comprehensive survey has been submitted to arXiv examining data-driven artificial intelligence approaches for soil moisture modelling, a complex spatiotemporal learning problem involving nonlinear environmental interactions, heterogeneous data sources, and limited ground observations. The research, submitted by Ilektra Tsimpidi on 16 June 2026, addresses the limitations of traditional physics-based approaches such as water balance models, which rely on explicit hydrological equations but face significant computational costs and scalability constraints that restrict their deployment at large scales.

The survey organizes existing AI-based methodologies into five distinct categories: statistical time-series models, geostatistical methods, classical machine learning models, deep learning models, and probabilistic or Bayesian methods. These approaches represent a shift toward data-driven alternatives that extract empirical relationships between soil moisture and environmental variables while reducing the need for explicit modelling assumptions. By leveraging diverse input data including historical soil moisture records, meteorological variables, vegetation indices, topography, soil characteristics, and geolocation information, these models perform both regression and classification tasks to estimate soil moisture conditions.

The research highlights how artificial intelligence methods have emerged as flexible alternatives to traditional physics-based models, enabling more efficient large-scale deployment while maintaining predictive accuracy. The systematic categorization of existing approaches provides researchers and practitioners with a structured framework for understanding the landscape of data-driven soil moisture modelling techniques and their relative strengths for different applications and environmental contexts.