Satellite-based crop auditing for agricultural monitoring, insurance validation and fraud detection.
CropAuditing is a technological framework designed to enhance the monitoring of large-scale agricultural areas through the integration of satellite imagery (Sentinel-2) and Machine Learning algorithms. Our solution addresses the limitations of traditional manual surveys by providing an automated, scalable, and precise system for crop species identification and growth stage tracking. By converting complex geospatial data into verifiable analytical reports, the platform optimizes the auditing process for financial institutions and insurance companies. We aim to ensure data transparency, facilitate fraud detection, and provide stakeholders with the necessary tools for informed decision-making in the evolving agricultural landscape.
Using Machine Learning and statistical analysis to accurately identify crop species and calculate their specific land areas.
Leveraging Sentinel-2 satellite imagery to automatically determine and track the growth stages of plantations.
Replacing slow manual surveys with a fast and user-friendly digital process for insurance companies and financial institutions.
Ensuring data transparency to facilitate fraud prevention and simplify compliance with agricultural government aid programs.
CropAuditing connects a clear auditing problem with a scalable technical solution and practical value for institutions that need reliable agricultural verification.
Manual agricultural audits are slow, expensive and difficult to scale across large territories. Field inspections require time, people and repeated verification, making the process inefficient for insurers, banks and public-sector entities.
CropAuditing combines Sentinel-2 satellite imagery, NDVI time-series analysis and Machine Learning models to classify crop species, monitor growth stages and compare declared agricultural information with satellite-derived evidence.
The platform enables faster verification, scalable monitoring and stronger fraud or inconsistency detection, supporting more transparent decisions in insurance validation, agricultural subsidies and financial risk analysis.
Developed with support and feedback from academic, agricultural and technology stakeholders.
CropAuditing transforms satellite imagery and declared agricultural parcel data into automated crop auditing outputs, combining database management, statistical analysis, Machine Learning and Google Earth Engine visualization.
Satellite and declared parcel data
DuckDB stores and queries large volumes of parcel and crop data.
Spectral descriptors, temporal curves and vegetation patterns.
Data split for model development, testing and performance evaluation.
Classification and phenology monitoring
User interface for selecting agricultural terrain, visualizing classified parcels and supporting automated crop audit decisions.
Open Web AppWe collect and organize agricultural parcel data using EuroCropsV2 and Sentinel-2 satellite imagery. For the current study, the database focuses on Portugal and selected crops such as corn, rye and oats, supporting both model training and validation.
The processing layer uses DuckDB and Python-based analysis to handle large volumes of declared parcel data, extract satellite-derived features, and prepare the datasets used by the Machine Learning models.
The intelligence layer combines Random Forest classification for crop species identification with NDVI time-series analysis for growth stage monitoring. This allows the system to compare declared agricultural information with satellite-derived evidence.
Results are presented through a Google Earth Engine web application, allowing users to select agricultural terrain, visualize classified areas and access practical crop auditing outputs.
Evaluation of crop type and growth-stage classification performance.
Comparison between predicted crop maps and validated parcel data.
MAE and RMSE used to compare predicted and observed vegetation curves.
Evaluation of growth-stage timing from NDVI time-series behaviour.
The people behind CropAuditing, working across data collection, machine learning, web development and project communication.
A simplified overview of the CropAuditing development process, from stakeholder validation and data preparation to model development, prototype integration and final presentation.
Initial report, interview report and stakeholder interviews.
Data collection, crop species, growth stages and validation methods.
Website, blog, web app development and model integration.
Testing, pitch deck, report, poster, video and ElectroDay.
Access our project documentation, reports, and deliverables.
Stay up to date with the latest news, projects, and articles from our group.
Week 1
Week 2
Week 3
Week 4
Week 5
Week 6
Week 7
Week 8
Week 9
Week 10
We are available to answer any questions you may have about us. Leave us your message and email address.
