Open-Earth-Monitor Global Workshop 2025

Transfer Learning as a Solution for the Large Areas Classification Dilemma
2025-09-17, 14:10–14:30 (UTC), Aula Magna

This study explores transfer learning in the Brazilian Amazon over the period from 2015 to 2022.

We use a time-series Random Forest model taking samples from 2022 to classify yearly date cubes up to 2015. The classification achieved an agreement of 89.20% with the reference map for 2022. Over the years, the agreement showed a cumulative decline of 2%. Our results suggest that the transferability of a machine-learning model is highly correlated with a set of highly representative training samples.

What is your current associations to EU Horizon projects (if any)? –

Open-Earth-Monitor Cyberinfrastructure (Grant agreement ID: 101059548)
Gilberto Camara
This speaker also appears in:
Felipe Carvalho
This speaker also appears in: