FARM-TECH: transforming multi-source data into operational decisions

FARM-TECH is an open-source Decision Support System (DSS) platform for Agriculture 4.0 that integrates satellite, drone, and weather sensor data to transform them into measurable operational decisions. It supports choices such as when to irrigate, where to intervene, how much to treat, and how to estimate yield, using guided workflows, maps, and interpretable dashboards.

The impact of FarmTech on smart agriculture in the Mediterranean was featured by TGCOM24.

Read the deep dive on TGCOM24

What FARM-TECH is and which operational decisions it supports

FARM-TECH (an SDSS platform for smart farming) was created to solve a typical challenge in Mediterranean agriculture and livestock farming: data is available, but it often remains fragmented and fails to translate into repeatable actions. The platform, developed by Smart Shaped S.r.l., integrates Copernicus (the European Earth observation program), drones, and weather stations, converting multi-source signals into operational outputs: priority maps, alerts, and estimations.

According to a summary published by Confagricoltura, investments in Agriculture 4.0 in Italy have reached approximately 2.5 billion euros in recent years, driven by growing expenditures on management software and data platforms (Confagricoltura, social media communication; Instagram).

Agritech technologies include satellite monitoring systems, agricultural drones, field sensors, management software, and predictive analysis platforms that allow farmers to make informed decisions based on real data.

— Smart Agrifood Observatory – Politecnico di Milano, Research center for digital innovation in agrifood

For more details on the project scope, see also FARM-TECH Project: Experimental Development and Data Integration.

How FARM-TECH models the agricultural context without complicating operational work

FARM-TECH maps farms, plots, crops, sensors, and decision outputs through a pragmatic modeling approach: the data model is designed to drive quick decisions, not to introduce unnecessary complexity. In practice, a plot (GIS polygon) is linked to a crop season (time frame and crop type), specific events (irrigation, treatments), and measurements (weather conditions and satellite indices), ensuring that maps and dashboards remain comparable over time.

FARM-TECH adopts a pragmatic approach to predictive modeling. The core philosophy is that the value of a model is determined by its practical utility in field operations, rather than its level of complexity and sophistication: it must function repeatably, using inputs that are genuinely available, and deliver easily interpretable results.

During the project requirements gathering phase, a classic real-world constraint emerged: building an ad-hoc trained supervised model can be unfeasible when (i) the dataset is numerically limited and (ii) the sources are too heterogeneous to guarantee consistent and replicable features during inference. Under these conditions, there is a high risk of producing a model that generalizes poorly and is difficult to maintain in the field.

For these reasons, the platform prioritizes two core operational paths:

1. Starting from traceable and easily verifiable baselines (even when working with partial data);
2. Maintaining an evolutionary trajectory: models and pipelines can be progressively refined and replaced as data quality and availability improve, without disrupting the platform's overarching architecture.

From a geospatial data modeling perspective, leveraging geometries and attributes on PostGIS (the geospatial extension for PostgreSQL) and data exchanges via GeoJSON (a JSON format for geometries) ensures full compatibility with standard GIS workflows and external engineering tools.

What data FARM-TECH integrates and how it normalizes it over time

FARM-TECH integrates weather data, satellite imagery, drone acquisitions, agronomic/livestock datasets, and manual/management inputs into a single versioned model. The goal is not just to import information, but to normalize it and make it accessible long-term for multi-period comparisons and repeatable decisions. Integrations are handled via files (CSV, GeoJSON), REST APIs (application interfaces), and OGC geospatial services (standards such as WMS/WFS) published by GIS components.

Regarding geospatial data, the platform explicitly processes both raster inputs (e.g., GeoTIFF files from satellite/drone or calculated indices) and vectorial/tabular records (plot boundaries, structured datasets, sensor metrics). Two structural requirements make the system field-ready: historization (comparisons over time) and export/download functionalities (such as CSV formats) for offline analysis or external integrations.

The normalization process includes validation checks on data quality and provenance: addressing missing data, matching temporal resolutions, fixing inconsistent georeferencing, and resolving conflicts between data sources. A baseline-first approach allows the system to generate outputs even with partial data, evolving steadily toward more robust pipelines (see the integration of MLOps and LLMOps pipelines for multi-source data).

Which vegetation indices FARM-TECH uses and how they support field decisions

Vegetation indices convert multispectral imagery (e.g., Copernicus Sentinel-2) into georeferenced quantitative indicators, comparable over time as mapping layers. In FARM-TECH, this process makes information more stable and interpretable: the indices summarize crop status into legible numbers and support decisions such as field inspection priorities, zone differentiation, and stress validation.

The initial set of indices is built on consolidated agronomic choices: vegetative vigor, chlorophyll sensitivity, and stress detection. Specifically:

• NDVI: the Normalized Difference Vegetation Index acts as a proxy for vegetative vigor and is calculated using Red (visible red) and NIR (Near InfraRed) bands. Formula: NDVI = (NIR - Red) / (NIR + Red)
• NDRE: the Normalized Difference Red Edge Index, based on the RE (Red Edge) band, is more stable in the presence of high biomass and highly sensitive to chlorophyll content. Formula: NDRE = (NIR - RE) / (NIR + RE)
• GNDVI: the Green Normalized Difference Vegetation Index is used to estimate photosynthetic activity, calculated from Green (visible green) and NIR (Near InfraRed) bands. Formula: GNDVI = (NIR - Green) / (NIR + Green)
• SAVI: the Soil-Adjusted Vegetation Index corrects for soil background influence, which is particularly valuable in early crop growth stages. Formula: SAVI = (1 + L)(NIR - Red) / (NIR + Red + L), where L is a soil correction constant factor, typically set to 0.5.

Operational constraint: indices do not replace field inspections, local calibration, and weather data; they perform best when interpreted alongside historical series and specific agronomic contexts. To explore the role of weather models in field decisions, see artificial intelligence and advanced models for agricultural weather forecasting. An institutional framework on Agriculture 4.0 and DSS is available in the technical document by Agenzia Laore Sardegna (2025): Precision Agriculture and Agriculture 4.0.

How FARM-TECH transforms data analysis into operational decisions step by step

FARM-TECH converts multi-source data into decisions through a 5-phase workflow, designed to be repeatable even when datasets are imperfect. The workflow combines geospatial analysis (GIS), historical series, and inference modules (business rules or machine learning), delivering actionable outputs: cartographic layers, dashboards, and operational alerts.

1. Acquisition: ingestion of GeoTIFFs, CSV files, and API endpoints (e.g., Sentinel-2 via Copernicus). Example: importing a pre-processed drone raster or automated retrieval over a specified time window.
2. Cleaning and Normalization: validation checks on timestamps, units of measurement, and georeferencing coordinates. Example: handling data gaps or anomalous weather sensor spikes.
3. Geospatial and Temporal Fusion: alignment between rasters, vector shapes (plot polygons), and historical series. Example: calculating average NDVI per plot and performing multi-temporal tracking.
4. Inference (Rule or Model): application of deterministic thresholds or predictive models. Example: linear regression for yield estimation or heuristic rules for field inspection priorities.
5. Operational Output and Alerting: rendering of maps, indicators, and exportable data summaries. Example: generating a "critical zones" mapping layer and automated target alerts.

Predictive example (tomato): to ensure the platform remains valuable even under sub-optimal data conditions, FARM-TECH integrates a predictive feature built on a simple and verifiable baseline. In this project use case, tomato yield is estimated starting from the mean NDVI calculated over the target plot, assuming a linear relationship between vegetative status and final production.

Estimated Yield (kg/ha) = a · (Mean NDVI) + b

• a = 63179
• b = 20318 (Note the linear nature of this equation, equivalent to the straight line y = 63179 · x + 20318)

A preliminary validation was performed across 9 fields based on multispectral acquisitions from June 27, 2024. The close fit between estimated mean yield and real yield values fully validates the model for operational deployment as a reliable baseline.

Two other typical use cases: (1) precision irrigation (prioritizing watering windows by combining weather data + soil sensors + NDVI trends); (2) field scouting prioritization (mapping crop anomalies to optimize the time of agronomists and technical personnel). For perspective on advanced predictive setups, see NeuralGCM predictive models for smart agriculture, and for operational traceability, see using digital twins for traceability and compliance in agriculture. An institutional reference on the role of DSS and interoperability can be found in the technical document by Agenzia Laore Sardegna (2025): Precision Agriculture and Agriculture 4.0.

Workflow Limitations: output quality depends directly on satellite cloud cover, sensor noise, and crop/territory calibration; this is why FARM-TECH prioritizes verifiable baselines and an incremental evolution of data pipelines.

What FARM-TECH's architecture is and how it guarantees integration and operational security

The architecture of FARM-TECH is modular, interoperable, and designed to decouple ingestion, analytics, visualization, and access control layers. The platform pairs open-source geospatial components with an application back-end and a dedicated execution layer for analytics/ML pipelines, keeping technical roles distinct (data storage, layer publication, business logic, computing) and simplifying system scaling.

The core components function as an integrated pipeline: GeoNode serves as the central hub for uploading, cataloging, and rendering geospatial datasets; GeoServer manages layer rendering and publication; Django + PostGIS supply back-end handling and spatial data persistence; ML_Runner orchestrates inference tasks; and the Object storage manages massive datasets and raster files. This architecture allows the platform to function natively as a Spatial Decision Support System (SDSS).

Operational security is an embedded asset: in a DSS, trust in data and automated actions relies on Identity & Access Management (IAM), Role-Based Access Control (RBAC), comprehensive logging, and the principle of least privilege. The platform defines strict user categories (guest, registered, uploader, admin) with distinct privileges over sensitive features, managed via a centralized IAM (**Keycloak**) that governs login sessions and security tokens.

Micro-example: an agronomist accesses aggregated maps and multi-temporal metrics; an operator uploads GeoTIFF files or updates data metrics; an administrator configures system policies and user roles, ensuring an immutable audit trail for modifications. To delve into governance and compliance themes, see managing AI governance and compliance under the EU AI Act and, regarding application vulnerabilities, AI security and model poisoning risks. In scenarios requiring data integrity and non-repudiation, it can be useful to assess blockchain technologies for data integration and security.

FARM-TECH vs. other smart farming platforms: differences, costs, limits, and roadmap

FARM-TECH sits at the intersection of general-purpose precision farming platforms and custom-built DSS solutions: it focuses on delivering verifiable workflows, deep geospatial integration, and multi-role governance, while maintaining an open-source, modular layout. In Mediterranean supply chains exposed to severe hydric stress, market leaders frequently highlight the immense value of DSS tools for targeted field actions (Barilla Group, 2025: report).

What drives Total Cost of Ownership (TCO): the number of active ingestion feeds (weather feeds, Copernicus intervals), GIS architecture complexity (rasters vs. vector boundaries), enterprise security requirements (Keycloak configuration, audit trail depth), and the need for customized crop models. There is no flat "standard" price point; the standard path follows a pilot-to-scaling rollout strategy.

Operational Limits: persistent cloud cover minimizes satellite data efficacy during specific windows; raw field data can occasionally be incomplete; without local agronomic calibration, indices can trigger false positives. Agenzia Laore Sardegna (2025) frameworks Agriculture 4.0 as data interconnection and georeferenced DSS tools to optimize irrigation and phytosanitary defenses (technical document).

Expected Roadmap: 0–3 months focused on expanding integrations (new sensors, API connectors, OGC exports); 3–6 months for launching new use cases (irrigation priorities, livestock field scouting); 6–12 months dedicated to tuning models for broader Mediterranean crop varieties and automated alerting. In 2026, digital innovation analyses highlight a structural adoption growth of advanced digital tools in agriculture (CDS Cultura, 2026: article).

Quick Troubleshooting (3 Common Issues)

• Missing Data Feeds: fall back to safe baseline thresholds and wider time frames; explicitly tag data quality and ingestion completeness indicators per source feed.
• Geographical Misalignment: audit CRS projs/datums and plot perimeters; recalculate and re-map polygon statistical boundaries per parcel.
• User Onboarding Blocks: start by isolating 2–3 high-value target decisions (irrigation scheduling, scouting routes, yield metrics) with explicit KPIs and established RBAC structures.

FAQ on FARM-TECH, data integration, and operational adoption

How much does FARM-TECH cost and what licensing does it use?

FARM-TECH is designed as an open-source platform architecture; however, total implementation costs depend on the pilot project's scope, ingestion feeds (weather providers, Copernicus layers), enterprise security setups (Keycloak, logging infrastructure), and custom modules. In practice, the commercial model is available "on request" because the primary value lies in custom provisioning, data onboarding, and operationalizing workflows.

How long does it take to deploy an active operational pilot?

A targeted pilot mapping 1–2 crop variants and a few select data streams can launch in 4–8 weeks, covering user onboarding, plot boundary ingestion, weather/satellite stream connections, and initial dashboards. Timelines scale out when adding complex corporate business rules, deep RBAC role structures, and validation across multiple past growth seasons.

What is the minimum data required to generate valuable outputs?

The minimal actionable dataset consists of: vector plot perimeters, a reliable local weather data stream, and multi-temporal satellite images (e.g., Sentinel-2) to extract vegetation indices like NDVI. While operational management data enhances timing accuracy, FARM-TECH is structurally designed to calculate a verified baseline even with partial or missing data fields.

What is the difference between a descriptive dashboard and a Decision Support System (DSS)?

A descriptive dashboard simply displays raw metrics and static graphs; a proper DSS like FARM-TECH adds rules, predictive layers, and actionable workflows that culminate in an operational decision: intervention priorities, targeted alerts, action recommendations, and traceable logs. Practically, a DSS shortens decision-making cycles and makes the process repeatable, not just the visual data.

What happens if the satellite paths face cloud cover or data fields are incomplete?

During heavy cloud cover or data delivery failures, outputs must be processed carefully and paired with local weather metrics, soil probes, and direct field inspections. FARM-TECH mitigates these issues by leveraging historization patterns, multi-temporal trend comparisons, and simple baseline curves to keep workflows continuous. During critical phases, the optimal path is combining multiple data streams and tracking data quality logs.