As the autumn season unfolds, we're excited to share how data integration methodologies are helping turn diverse agricultural data into actionable intelligence.
In this edition, we spotlight two technological advancements - the DHI Data Integration Framework and VITO's Few-Shot Learning - and explore how they are being applied across our Research and Innovation Labs (RILs) in Grasslands, Crop Management, and Yield Monitoring. Together, these approaches demonstrate how combining satellite and ground data can unlock smarter, more efficient, and more sustainable farming practices.
Thank you for following our journey as we continue building bridges between technology and real-world challenges in agriculture.
Enjoy the read and stay tuned via our website & social media for more updates from the field! Feel free to forward this newsletter or invite colleagues to register.
In this ScaleAgData Newsletter:
Innovation area: Data Integration Methodologies
Recent News Items
Upcoming Events
Data Integration Methodologies
Combining in-situ measurements with satellite observations is at the heart of modern agricultural innovation. While satellites provide consistent, large-scale monitoring of land and crop conditions, in-situ data offer the local accuracy needed to interpret and validate these observations. Integrating both data sources allows us to bridge the gap between field-level insights and regional analyses, resulting in more reliable models, improved decision support, and ultimately, smarter and more sustainable farming practices.
Data Integration Methodologies play a central role.
Let's highlight two complementary approaches developed within ScaleAgData:
The DHI Data Integration Framework, focusing on integrating in-situ and satellite observations for accurate mapping of agri-environmental variables
VITO's capabilities in Few-Shot Learning (FSL), a flexible AI method that learns efficiently from limited labelled data.
Both methods are being tested and applied across our Research and Innovation Labs (RILs) - in Grasslands, Crop Management, and Yield Monitoring - connecting advanced technology with agricultural applications.
The DHI Data Integration Framework: Merging Satellite and Ground Observations
DHI has looked at a number of solutions for combining satellite and ground observations. One of the goals was to do this in such a way that a local observation can improve not only local but also regional or even national maps. The framework that has been developed includes various methods ranging from machine learning data fusion, through using in-situ measurements for bias corrections of models' inputs and outputs, to directly using those observations as model inputs. Those methods are most probably complementary, and some might work better in some circumstances than in others. Whichever method is applied, the overall goal is to combine the abundant data streams of satellite and ground observations to produce improved satellite data products covering large areas. In ScaleAgData, the Data Integration Framework has been used to develop improved maps of evapotranspiration (ET) and root-zone soil moisture.
Why?
Fresh water is one of our most precious resources - yet it is being overexploited. Agriculture alone accounts for about 70% of global freshwater withdrawals. Improving irrigation efficiency is therefore crucial to secure both crop productivity and sustainable water use. Maps of evapotranspiration (ET) and root-zone soil moisture can help estimate irrigation needs more accurately.
However, models estimating these variables rely on diverse inputs - vegetation and soil properties, meteorological data, and even agricultural practices such as irrigation scheduling. Not all of these can be derived reliably from satellite data. However, they can be easily determined using on the ground measurements and observations. Integrating satellite and ground-based measurements enables higher precision of the ET and soil moisture maps and hence more reliable insights.
Impact: Better understanding of soil-water dynamics lead to improved irrigation scheduling and more efficient water use - truly more crop per drop.
Few-Shot Learning for ScaleAgData: Doing More With Less Data
Remote Sensing offers massive amounts of data for agriculture. Satellites continuously capture imagery and signals that can greatly improve our understanding of agriculture and the environment. This data can serve as valuable input for training machine learning and deep learning models for a wide range of applications.
Traditionally, the most effective way to build such models requires large and fully labelled datasets. However, creating these datasets is often challenging due to high costs, privacy restrictions, and the limited availability of annotations. Without sufficient labelled data, it becomes difficult to develop reliable models able to deal with real-world scenarios.
Few-Shot Learning (FSL) provides a solution by enabling models to learn effectively from small, labelled datasets. This is made possible by so-called "Foundation Models" that are trained on extensive collections of unlabeled data. These models capture patterns across diverse sources and create simplified, highly informative representations of the original inputs. Once trained, they can be fine-tuned for a wide range of specific applications such as crop monitoring, yield prediction, or land cover classification.
In the context of ScaleAgData, VITO exploited this approach to offer flexible, user-friendly AI solutions that could be both scalable and efficient. ScaleAgData's FSL Framework, builds on the Presto (Pretrained Remote Sensing Transformer) model. Presto integrates Sentinel-1 and Sentinel-2 time series with meteorological and topographic data, producing compact representations that can be fine-tuned for specific applications such as:
crop yield prediction (regression tasks)
crop type mapping (multi-class classification)
cropland detection (binary classification)
This approach handles missing data, adapts to different temporal resolutions (10-day or monthly), and offers faster, more reliable insights - even with limited training data. It strengthens agricultural research and meets the operational needs across the diverse contexts of the RILs
Impact: The FSL approach accelerates AI deployment in agriculture, offering scalable and accurate predictions with minimal data annotation needs.
Fig. 2. Overview of the Presto-based Few-Shot Learning pipeline fine-tuned for yield estimation. The model is fine-tuned for the regression task on a limited amount of Sentinel 1, Sentinel 2, meteorological and topographic annotated pixel time-series. Once ready, it can be used on a selected Area of Interest to produce a yield map.
In practice: Data Integration in the RILs
Grasslands RIL - Estimating Biomass With Few-Shot Learning
The Grassland RIL focuses on estimating biophysical variables of grasslands in Bolzano (Italy) and Mediterranean savannas in Spain. The team is testing VITO's Few-Shot Learning framework, to estimate above-ground biomass (AGB) over both known and new grassland areas using Sentinel 1/2 imagery, meteorological data and Copernicus DEMs. The model is being trained with ground AGB measurements collected through in-situ campaigns in 2023-2025 across 18 field sites in Northern Italy and Spain.
Impact: More precise AGB estimates to support grassland management.
Crop Management RIL - Integrating EO Data into Decision Support Systems
The Crop Management RIL (Sub-Lab 2b) provides tools for sustainable management of wheat. Earth Observation data (NDVI from Sentinel satellites) have been integrated into HORTA's grano.net® Decision Support System (DSS)*. This allowed the team to improve the wheat yield model and finetune the fertilization recommendation module of the DSS. Evapotranspiration and soil moisture data, providing insights in the water balance, are also being considered to further improve the system. Impact: Improved decision-making for sustainable wheat management.
Further, in the Crop Management RIL (Sub-Lab 2a) satellite-based soil moisture maps were used to identify spatial variability across tomato fields in Greece. This analysis helped determine the optimal number and placement of in-situ soil moisture sensors, which, when combined with satellite data, yield more accurate and spatially distributed soil moisture estimates.
Impact: Efficient planning of the in-situ sensor deployment.
* grano.net® Decision Support System (DSS): an interactive web tool for high quality wheat growers
Yield Monitoring RIL - Predicting Potato Yields With Few-Shot Learning
In the Yield Monitoring RIL, the Few-Shot-Learning Framework was applied to potato yield prediction in Belgium and the Netherlands. Several models were trained using yield data from potato harvesters. The FSL model using Presto embeddings + CatBoost regression clearly outperformed traditional convolutional models incorporating the same raw satellite data inputs.
Impact: More accurate yield forecasts with minimal labelled data.
Recent News
ScaleAgData showcased data-driven innovations at Synergy Days 2025
At Synergy Days, ScaleAgData actively engaged the agrifood innovation community through a project booth, pitch presentation, and collaborative workshops
Would you like to learn about upcoming events focused on the latest innovative data technologies for managing agricultural production and monitoring agricultural environments?
VITO is the project coordinator of the ScaleAgData consortium and will also contribute significantly to the 'product and service development', the RI Lab on Yield monitoring as well as the entire communication segment of the project. As a technology provider, VITO will bring data products and innovative solutions to merge sensor data into the data products. As communication lead, VITO will be managing the ScaleAgData website, social media channels as well as these digital newsletters. Feel free to contact us via scaleagdata@vito.be.
------------------------------------- If you don't want to receive our ScaleAgData newsletter, you can unsubscribe: see the 'unsubscribe' link below.
SHARING IS CARING
Feel free to share this newsletter with colleagues or on social media!
.png?upscale=true&width=1120&upscale=true&name=2025-07-03%20SCALEAGDATA-infographic-lijnendiagram_update%20(1).png)
