Physics-based modeling in the age of open science: one modeler's perspective, concerns, and challenges

Physics-based modeling in the age of open science: one modeler’s perspective, concerns, and challenges. 1 Dr. Jared Bell, PhD [email protected] Science Data Officer for Heliophysics NASA HQ 2 Physics-based modeling in the age of open science: one modeler’s perspective, concerns, and challenges. Modeler Guy from headquarters wants to learn about open science and modeling. Dr. Jared Bell, PhD [email protected] Science Data Officer for Heliophysics NASA HQ Quick Introduction – I’m a modeler now at HQ ●Develop global circulation models for planetary atmospheres. ○Global Ionosphere-Thermosphere Model (Titan, Jupiter, Saturn, Mars, and ExoPlanets) ■University of Michigan, Ridley et al. [2006]. ○Thermosphere-Ionosphere-Electrodynamics General Circulation Model (TIE-GCM) for Mars ■Bougher et al. [1999]. Titan GITM (Bell et al. [2014]) Mars GITM (Bougher et al) Quick Introduction – I’m a modeler now at HQ ●Currently the Chief Science Data Officer for Heliophysics at NASA HQ. ○Develop coherent policy for open science data approaches for research, missions, and repositories. ○Establishing common formats, meta data, etc: ■Apply FAIR and FAIRUST principles ■FAIR = Findable, Accessible, Interoperable, and Reusable. ■FAIRUST = FAIR + Understandable, Secure, and Trusted. ■Can involve the repositories (SDAC/SPDF/HDRL) ■Definitely involves the CCMC! My primary goal is to listen and learn ●Learn about open science approaches to modeling: ○Approaches for highly complex physics-based codes running on supercomputers ○Or, approaches for more simplified models that can run on a laptop. ●How do current NASA resources play into the community’s movement toward more open modeling? ○For example, CCMC is a key resource for making models, software available. ○CCMC should and can play a role moving forward. ●How can policy at HQ support an open science approach for modeling? ○Should policy incentivize open science? ○Want to avoid unintended consequences that stifle innovation.. General modeling concerns (The Decadal) ●As modelers, the codes and software that we produce represent the source of innovation, creation, and the foundation for funding ourselves and our teams. ●Modeling -focused proposals can struggle in the ROSES funding system and in mission support roles. ○Chicken and the egg. ○Everyone wants the products from models, but getting investment into new modeling paradigms is tricky at best. ●Real innovation in modeling requires multi-year investments prior to significant deliverables. ○Long-term vision. ○Current short-term approaches can be severely limiting if not completely antithetical to the needs of modeling. Possibly irrational fears of open modeling ●Could open science force modeling teams and individuals to give up potential “competitive advantages?” ●Could an open science approach to modeling expose innovators’ work in a way that shuts off funding. ○Other groups get the code and submit competing proposals or cut the original developer out of work? ○Other groups mis-use the model and give it a poor reputation with funding agencies? ●Could an open modeling approach lead to being “scooped?” ○Someone uses the model to study a phenomenon without proper attribution or mis-uses a model and poorly represents the original developer. Open science will drive change ●If we (NASA and the heliophysics community) expect the modeling community to adopt a more open science approach, then the rest of the science ecosystem most likely needs to evolve with them. ○How do we deal with potential bad actors? ○What can the agency do to protect our researchers while also fostering a free and open science environment. ●Many proponents of open science talk of educating the community about open science practices. ●Agencies will also need to be educated about the best policities to support this shift. Some final thoughts ●As we move forward, how do we define open science for models? ●Do we first focus on the DOI/identification of model versions for citations and referencing? ●Should we focus first on establishing (if this isn’t already established) common formats or community-wide supported formats? ○Tied to existing open software? ○E.g. Spacepy, SolarPy, PyHC, Komodo, etc. ●Should we classify the “openness” of models or have checkpoint versions like in LLM distributions? ○Akin to the LLMs that one can get from sites like HuggingFace and/or Ollama? ■Different size models (from heavy to light hardware needs?)