MULTI-MODAL DEEP LEARNING WITH SENTINEL-3 OBSERVATIONS FOR THE DETECTION OF OCEANIC INTERNAL WAVES

Shrinkage of the Antarctic ice sheet (AIS) leads to sea level rise. Satellite gravimetry measures AIS mass changes. We apply a new method that overcomes two limitations: low spatial resolution and large uncertainties due to the Earth's interior mass changes. To do so, we additionally include data from satellite altimetry and climate and firn modelling, which are evaluated in a globally consistent way with thoroughly characterized errors. The results are in better agreement with independent data.

For the pilot region Mississippi River basin, we explored three model calibration approaches for using both streamflow and total water storage anomaly (TWSA) observations to adjust parameter sets of a global hydrological model. We developed a method for considering the observation uncertainties to determine not only optimal but also behavioral parameter sets and model output uncertainty. Model performance was improved but trade-offs between optimal fits to streamflow and TWSA may be large.

Ozone is a toxic greenhouse gas with high spatial variability. We present a machine-learning-based ozone-mapping workflow generating a transparent and reliable product. Going beyond standard mapping methods, this work combines explainable machine learning with uncertainty assessment to increase the integrity of the produced map.

With the AQ-Bench dataset, we contribute to shared data usage and machine learning methods in the field of environmental science. The AQ-Bench dataset contains air quality data and metadata from more than 5500 air quality observation stations all over the world. The dataset offers a low-threshold entrance to machine learning on a real-world environmental dataset. AQ-Bench thus provides a blueprint for environmental benchmark datasets.