Test Main Research Field

Computational Environmental Science (CompEnvSci) at IWR brings together researchers who strive for a deeper understanding of the Earth's environmental systems using computational methods and rich digital databases. Thereby, the IWR forms the computational hub for multidisciplinary research teams in the environmental sciences at Heidelberg University and at the Heidelberg Center for the Environment (HCE) in particular. The core groups are formed by the disciplines of Physics, Geography, Computer Science, Chemistry and Medicine. Our scientific network covers the full range of disciplines in Heidelberg including natural, social sciences and humanities. Through that network advances in IWR are transferred rapidly and contribute importantly to environmental sciences at large.

One of humankind's immediate challenges is the understanding of the growing coupling between our socio-cultural and physical environment, and the control of the entire system within acceptable bounds. There is the urgent need of our society for comprehensive, trustworthy, and quantitative predictions – including uncertainties. Modeling and numerical simulations as well as analysis of rich observational geographic data have been key in environmental sciences for the past decades. This is bound to deepen in the years to come. This is demanded by our environment's very nature: highly nonlinear, non-separable with often strong coupling across disparate fields, multi-scale heterogeneous, fundamentally chaotic, self-organizing, and rapidly evolving, the latter from Earth's biodiversity to humankind's culture with its technology. There are opportunities from the massive growth of computing power, communication capacity, observational capabilities (technical and human sensors), and methodological developments, the latter from high-performance numerical solvers to machine learning and artificial intelligence.

These big challenges reflect themselves in the IWR members together working in the main research field of CompEnvSci. The groups of Butz, Platt, Gutheil, Köthe aim at deciphering the processes that drive today's climate change with a particular focus on the biogeochemical cycles that control greenhouse gas (GHG) concentrations in the Earth's atmosphere. We develop and use computational tools for data reduction of satellite measurements and to simulate the composition of the Earth atmosphere. For example, we study the role of chemical instabilities in the troposphere (Platt, Gutheil). Furthermore, we push forward submitting Earth-observation satellite data to a machine learning technique with the goal to determine atmospheric abundances of carbon dioxide (Butz, Köthe). Besides being computationally extremely efficient compared to traditional physics modelling, the technique in particular delivers uncertainty estimates which are crucial to make the data actionable. We have just embarked on an emerging collaboration in the field of Climate Action Science within a multidisciplinary team of the HCE (Butz, Gutheil, Höfle, Zipf). We want to collect atmospheric GHG data and simulations, geographic boundary conditions and geostatistical data for a community-scale GHG information system to enable efficient community climate action. In addition to technical sensors, citizen science and crowdsourcing approaches are complementary means for environmental monitoring, esp. from a human in situ perspective. As this data is not always provided by experts we need to develop methods for quality analysis, assurance and enhancement (Zipf).

The Humboldt Professorship and IWR group of Till Bärnighausen at the Heidelberg Institute of Global Health drives global health research and examines the group of carriers and sufferers, he also makes connections to society as a whole. For this digital health and geographic data and computational analysis approaches are increasingly applied. Furthermore, disease outbreaks, natural and man-made hazards require spatial information systems to aid governmental and non-governmental agents to plan and organize activities. The Zipf group employs a toolchain that involves crowdsourcing (e.g. MapSwipe App), intrinsic data quality assessments for crowdsourced information and object detection based on deep learning to estimate population and critical infrastructure. Novel routing approaches are used to identify the accessibility of critical infrastructure in case of disaster. Transfer of developed methods in practice is provided by the spin-off company HeiGIT. Furthermore, we develop methods to include geographic knowledge into the computational analysis based on highly detailed 3D/4D Earth observation data (Höfle group). Solutions include the open source codes of HELIOS++ (Virtual Laser Scanning) and several web-based 3D-crowdsourcing tools. Particularly interesting is the research frontier of autonomous and responsive 3D Earth observation, which demands for computational methods making use of the full data stream to adapt data acquisition in near real-time.

The Roth group follows two major lines: (i) With a narrow focus on flow and transport in unsaturated porous media, process-based simulation software is developed (DORiE, based on DUNE/PDElab) and integrated into a general data assimilation framework (KnoFu). Exploration of describing active transport experiments using convolutional neural networks (CNN) was done successfully. (ii) With a very wide and fundamental perspective, chaotic, complex, and evolving systems are explored. The main tools are numerical simulations of cellular automata (CA), agent-based model (ABM), and evolution. They were recently aggregated into an uniform framework (Utopia, open source).

The Emmy Noether group of Kira Rehfeld tests the validity of projected changes in climate variability by integrating palaeoclimate data, climate simulation and statistics. Within IWR we test the efficacy of machine learning methods to enhance model simulations for water isotopes (collaboration U. Köthe).

The atmospheric physics group develops radiative transfer and inverse estimator codes for remote sensing of atmospheric composition from satellites and other platforms and, since recently, they explore machine learning techniques to complement the computationally costly physics codes (groups of Butz, Köthe). For the 3D simulation, large eddy simulations are performed using the open-source software WRF-Chem where various modules such as the detailed chemistry model are modified and included (groups of Platt, Gutheil). Meteorological data are nudged to account for realistic weather conditions in the atmosphere. Inhouse codes for 0D (box model) and 1D simulations have been developed to study the detailed chemical reactions and their sensitivity to the ozone depletion events. 

CompEnvSci research projects are supported by a diverse set of funding bodies such as DFG, EU, Humboldt Foundation, Federal State of Baden-Wuerttemberg, etc., and partly by the industry.