Publications

Accurate, privacy-respecting soil information is essential for site-specific nutrient management and carbon accounting, yet the cost of laboratory analyses limits many farms to relatively sparse sampling grids. We propose a collaborative framework that couples a national Sentinel-2 bare-soil composite with FL to produce high-resolution clay and soil organic carbon (SOC) maps while keeping all local...

Soil spectroscopy is a powerful technique for soil monitoring. Hundreds of soil spectral datasets, both open-access and private, are used for various applications, despite potential errors. This study presents a filtering protocol to enhance the quality of soil spectral datasets from laboratory and satellite sources, here termed the Spectral Quality Protocol (SQuaP). At the beginning,...

The advent of artificial intelligence (AI) and machine learning has permeated every aspect of our lives. Water resource management is becoming increasingly important due to the impacts of climate change and human activities. This publication provides examples of how AI can potentially improve the forecasting of water resource variables at both the field and regional...

Soil Organic Carbon (SOC) stocks in croplands play a key role for climate change mitigation and soil sustainability, with proper management techniques enhancing carbon storage to support these goals. This study focuses on the development of a hybrid carbon modeling approach for the simulation of topsoil SOC stocks across the entire agricultural area of Lithuania....

The variability of soil detachment rate (Dr) with rill length is essential to understand the complex process of rill erosion. Moreover, an important variable of rill erosion, such as the ‘flow sediment deficit’ (FSD)–the difference in Dr between the two extreme points of a rill–has been scarcely investigated. Soil disturbances, such as fire and deforestation,...

Soil has garnered global attention for its role in food security and climate change. Fine-scale soil-mapping techniques are urgently needed to support food, water, and biodiversity services. A global soil dataset integrated into an Earth observation system and supported by cloud computing enabled the development of the first global soil grid of six key properties...

Very few studies have explored the effectiveness of polyacrylamide (PAM) application on soil in mitigating rill erosion, especially in deforested environments. This study has measured the soil detachment capacity (Dc) on samples of deforested soil (untreated or treated with PAM). Dc has been estimated by flume experiments under three bed slopes (6.9%, 17.2%, and 18.2%)...

GaiaBot is an AI-driven geospatial platform designed to simplify access to complex soil data for non-expert users, such as farmers and policymakers. Leveraging Earth observation data and global soil models, GaiaBot provides actionable insights into soil properties and field management through an intuitive chat-based interface. The system integrates advanced natural language processing with dynamic geospatial...

Harmonizing spectral data from diverse sensors presents significant challenges due to variability in spectral range and resolution. This work introduces a foundation model based on LSTM autoencoders to reconstruct spectral data from multiple sensors. The model effectively generalizes across diverse datasets, preserving essential features and achieving low reconstruction errors. These results demonstrate the potential of...

This paper explores the application of federated learning in soil science to address the challenges of data privacy and collaboration. Leveraging decentralized machine learning, we demonstrate how to predict soil properties effectively while preserving privacy. Our study focuses on cropland data from Flanders, Belgium, and Central Macedonia, Greece, and utilizes the LUCAS topsoil database and...