Open Earth Monitor — Global Workshop 2023

Land use and land cover maps are important indicators of human footprint and environmental dynamics. These maps are crucial data sources for measuring global indicators related to the United Nations’ Sustainable Development Goals (SDGs), including land degradation, freshwater ecosystems, and urban development.

Recently, satellite image time series and machine learning methods have been widely used to produce land use and land cover maps from big Earth observation data with promising results. Most of these methods are based on supervised learning, and thus they require a training phase using a significant number of samples labeled a priori. The quality of land use and land cover training samples is crucial in the classification process. Good samples lead to classification results with better accuracy. It is essential to advance in new software tools that help specialists to collect good-quality land use and land cover training samples.

This abstract presents an open-source web platform, called TerraCollect, to collect land use and land cover samples based on image time series mainly for machine learning method training. This platform has components to visualize remote sensing images over time; to extract image time series from Earth observation data cubes; to access distinct land use and land cover data sets and their trajectories; and to edit and label points. Besides that, it has a component for sample analysis where specialists can examine the balance of samples by class, the image time series pattern that represents each class and possible occurrences of class noise. The TerraCollect platform is being developed by the Brazil Data Cube project team (http://www.brazildatacube.org/).