Deep learning models based on diffusion processes have shown great potential in a range of generative tasks, such as image generation. For remote sensing applications, generative models are not that common. The question that we tried to answer is whether diffusion processes can be used to efficiently pretrain models for discriminative tasks?

Multimodal datasets for remote sensing are oftentimes limited to two data modalities, such as multispectral and SAR polarization data. In order to experiment with a much wider range of data modalities, we extended the well-known BigEarthNet dataset to includes a wide range of data modalities.

Road traffic noise is a global issue that can lead to severe health effects. Despite the ubiquity of traffic noise, its quantification or estimation is complicated and detailed road traffic maps are only available for select countries or areas. We investigate whether it is possible to train a segmentation model to esimate road traffic noise from satellite imagery.

Self-supervised learning provides a powerful means to pretrain models based on un-labeled data. Un-labeled Earth observation data are abundant: this circumstance combined with the availability of multi modal data makes Earth observation a perfect playground for self-supervised learning. Our early results are very promising…

Tracking Greenhouse Gas emissions will be increasingly important in the future, as power generation from fossil fuels is supposed to fade out. Independent tools to monitor power plants are required for this task. We present a method that uses freely available satellite imagery to estimate power generation and CO2 emission rates on a global scale.

Air pollution is a major health issue and often also contributes to climate change. Measuring air pollution is costly and therefore only available in some countries. We investigated whether freely available Earth observation data can be utilized to estimate air pollution on the surface level.

Can we identify trucks from space and estimate truck traffic rates anywhere on the planet? Yes, we can!

We developed a deep learning model that is able to distinguish between 10 different types of power plants in an effort to automatically identify and characterize industrial sites in satellite imagery. This work will help us to estimate greenhouse gas emission rates for individual industrial sites in the future.

Greenhouse gas emissions from the industrial economic sector are a major driver of the currently observed climate change. We developed a deep learning approach to identify and characterize industrial smoke plumes. In the future, we will utilize this approach to estimate greenhouse gas emissions from remote sensing data on a global scale.