Navigation and service

Scientific interests

of Martin Schultz

Photo of Martin Schultz in front of the supercomputing JUQUEENDr. Martin Schultz

  • Apply deep machine learning and modern statistical methods to enable novel analyses of Earth system data with a focus on air quality and weather data
  • Develop performant interoperable workflows and large-scale data architectures to bridge the gap between Earth system observations and modelling
  • Analysis of global air pollution and its trends over time

My research objective is to exploit the rapid technical and conceptual development of web services and computing infrastructure to open new ways for analyzing complex Earth system data from observations and models, in particular through deep learning and big data analytics methods. While my background and focus is on global air quality and its interactions with climate and ecosystems, I am also interested in new possibilities for weather forecasting. As co-chair of the Tropospheric Ozone Assessment Report (TOAR), phase II, I am responsible for the development and maintenance of one of the world’s largest data collections on air quality measurements. Within TOAR, my team and I explore novel ways for online analyses of atmospheric data.

Ongoing grants

IntelliAQ – ERC Advanced Grant, 2018-2023
The IntelliAQ project (Schultz, 2020) develops novel approaches for the analysis and synthesis of global air quality data based on deep neural networks and on the TOAR database and web services. The information from air quality measurements is enhanced through several geospatial datasets and sampling of output from multi-year meteorological reanalyses. State of the art deep learning methods are applied to this unprecedented dataset in order to 1) fill observation gaps in space and time, 2) provide short-term forecasts of air quality, and 3) assess the quality of air pollutant information from diverse measurements. Specifically, we explore techniques for time series analysis, video prediction and unsupervised learning. The combination of diverse data sources is unique, and the project is the first to apply the full potential of deep neural networks on global air quality data. IntelliAQ will shift the analysis of global air pollutant observations to a new level and provide a basis for the future development of innovative air quality services with robust scientific underpinning. Due to the heterogeneity of the multivariate data, lack of structure, and generally unknown uncertainty of the input data, the project poses challenges for existing deep learning methods, and will thus lead to new developments in this field. Direct outcomes of the project will be a substantial improvement of global air quality information including methods to assess the quality of air pollution measurements, and a new data-driven method for forecasting air quality at local scales.

DeepRain – BMBF project (2018-2021)
The DeepRain project combines modern methods of machine learning with high-performance data provisioning and processing systems to generate spatially and temporally high-resolution maps with improved and validated precipitation predictions including their uncertainties based on regional weather models. It is a collaboration between Forschungszentrum Jülich, the University of Osnabrück, the University of Bonn, Jacobs University in Bremen, and the German Weather Service DWD). Our role at JSC is to set-up and maintain the DeepRain data infrastructure and manage about ½ Petabyte of weather model data for analyses with various machine learning tools. We also operate a rasdaman server with a portion of these data as a node in the EarthServer federation.

Pilot Lab Exascale Earth System Modelling – Helmholtz project (2019-2021)

In the Pilot Lab Exascale Earth System Modelling several Helmholtz centres from three different research fields cooperate to explore novel ways for climate modelling on next generation high performance computer systems. New hardware developments and urgent requirements to enhance the resolution of current climate simulations make it necessary to find new solutions for scalability and data handling in Earth system models. We investigate the use of model dwarfs and domain specific languages, look into asynchronous workflows and the use of hierarchical storage infrastructures and test the potential of machine learning to replace empirical model parametrisations. The lessons learned from these activities are fed back to the system design of the next generation HPC hardware.

Aside from these 3 grants where I act as principal investigator, I am also contributing to the Helmholtz Climate Initiative, the data infrastructure of the “Bioökonomie Revier”, and the KI:STE project.