Methodology summary

The coffee model uses supervised deep learning techniques, particularly convolutional neural networks, to automatically identify coffee crops in images from Sentinel-2 and PlanetScope satellites. As it is a supervised method, training the model required reference annotations, obtained from data provided by EMATER/MG and CONAB. Considering that coffee is a perennial crop, it was possible to explore a temporal inference approach using images from 2020 to 2024, allowing for the analysis of its behavior over time, including management and pruning practices. This strategy enhanced the model’s ability to distinguish coffee areas from other land uses with similar spectral signatures but distinct phenology.

The platform preferentially integrates data from government organizations that, in partnership with CONAB, are responsible for different key components:

1. High-resolution mapping of coffee crops in the main producing regions of the national coffee growing area: UFMG is responsible for developing the methodology that uses multi-year mosaics of remote sensing images as input for the Artificial Intelligence algorithm (U-Net), allowing for systematic updates of coffee crops.
2. Deforestation – Legal Benchmark: The National Institute for Space Research (INPE), via the PRODES satellite monitoring system, is responsible for providing and annually updating post-2020 deforestation data.
3. The Rural Environmental Registry (CAR) is an electronic register of all rural properties. The Ministry of Management and Innovation in Public Services (MGI) manages the system, which is accessed via a gov.br account for registration and management of the CAR. It is essential for environmental regularization and access to public policies.
4. National Rural Cadastre System (SNCR): The National Institute for Colonization and Agrarian Reform (INCRA) is responsible for updating the SNCR, which includes the registry of rural property owners and holders, the registry of rural tenants and partners, the registry of public lands, and the national registry of public forests. The service provided via the government API allows for the consultation of rural property information and verification of ownership/possession for individuals or legal entities.
5. CONAB will evaluate the opportunity to integrate other databases into the platform.

The platform allows producers and their representatives, when logged in via gov.br, to analyze their areas and request suggestions and change requests for the data layers, especially in cases related to deforestation and coffee mapping.

The Mappia interface enables the construction of a client-server architecture that handles a large volume of spatial data to quickly visualize a series of map algebra operations and other spatial queries. All geospatial analyses are integrated relationally using open-source tools capable of handling “spatial big data,” such as PostgreSQL and Geoserver.

The platform will provide maps at different scales, including property (CAR), municipality, intermediate regions, and states, using interactive map services. It will also include interactive dashboards, thereby ensuring accessibility and usability for its users.

CONAB, in partnership with public and private institutions as needed, will conduct field validation.

The validation assesses areas identified through the intersection between the coffee crop mapping and the deforestation polygons recorded by the PRODES/INPE system. To implement the field validation, proximity criteria or other criteria to be aligned between CONAB and its institutional partners will be defined.