As part of this cooperation, different use cases are analysed systematically, and methodological building blocks for a scalable AI pipeline are developed and validated. The pipeline processes multispectral data from the Sentinel satellites of the EU Copernicus programme.
A particular focus lies on designing and testing suitable machine learning methods for integrating, modelling and interpreting the heterogeneous data streams.
Research question
How can satellite data, measurement data and expert knowledge be combined so that changes in land, vegetation and soils are reliably detected and presented in a way people can follow?
Approach
Sentinel-2 delivers images every few days in thirteen spectral bands – from visible light to short-wave infrared. We systematically examine different use cases, prepare the data as time series and test machine learning methods that bring together, model and interpret the heterogeneous data streams. The video shows such a time series over Chemnitz-Adelsberg.
Why explainability matters here
Anyone who makes decisions based on satellite data – about land management, protective measures or planning – needs to know what a result is based on. That is why we combine the models with expert knowledge of landscape and vegetation and make visible which data leads to which statement.
Research topic: explainable AI →Status and outlook
Methodological building blocks are being developed and validated in selected areas. The pipeline is designed to be scalable and is intended to be transferable to other regions in Saxony and to other questions.