Agro-informatics Tech Talks ‘Early Season Crop Type Mapping: An Essential Product for Parcel Level Crop Monitoring. Followed by an update on the most recent progress on FAO - Google Memorandum of Understanding (MoU)’
Virtual Event, 28/06/2024
.png?sfvrsn=2ec26b40_1 "Agro-informatics Tech Talks ‘Early Season Crop Type Mapping: An Essential Product for Parcel Level Crop Monitoring. Followed by an update on the most recent progress on FAO - Google Memorandum of Understanding (MoU)’")
Date: 28/06/2024
Time: 15:00-16:00 CEST
The Agro-informatics TechTalks organized by the Digital FAO and Agro-informatics Division (CSI) aim to spark peer-to-peer discussions on the central role that technology, science and data play in sustainable digital agriculture for targeted impact and to strengthen links within the agro-informatics community and beyond. The series serve as a dedicated space to showcase experts' knowledge and how their research, initiatives or products contribute to agricultural transformation and sustainable development, in a globally interconnected ecosystem, leaving no one behind.
Join us on Friday, June 28, at 15:00 CEST for a virtual presentation on the “Early Season Crop Type Mapping: An Essential Product for Parcel Level Crop Monitoring” by Dr. Martin Claverie, Data Scientist and Researcher, Joint Research Center (JRC), European Commission, followed by an update on the most recent progress on FAO - Google Memorandum of Understanding (MoU) by Julian Fox, Senior Forestry Officer, FAO.
Early-season crop type classification is vital for predicting crop yields and monitoring agricultural parcels, yet it poses challenges due to complex growth patterns and variability. Traditional deep learning methods often rely on single data sources like satellite imagery. To address this, a novel multimodal fusion model will be introduced during this session. This model integrates satellite time series data, crop rotation, and local crop distribution, significantly enhancing accuracy and robustness. The research introduces a dataset comprising 7.4 million parcels from France and the Netherlands, annotated with surface reflectance and biophysical variables. The model features a hierarchical crop classification system and a data augmentation technique specifically designed for early-season identification. With an accuracy rate of 85-95% across various crop type groupings, this approach surpasses existing methods, leveraging spatio-temporal context for better transferability and generalization across different countries.
Join the conversation to contribute to building a more resilient and sustainable digital agricultural future.
Register here