The Science Data Understanding Group carries out research and technology development in computational techniques to extract knowledge from science data.

What we do: Model the statistical variability of physical systems like oceans and atmospheres; infer parameters, recognize patterns, and quantify uncertainty in massive data sets taken from such systems.

Uncertainty Quantification

  • Uncertainty quantification (UQ) for retrievals from Earth-observing instruments including OCO-2, MLS, ECOSTRESS, and MAIA
  • Uncertainty assessment and propagation for a large-scale hydrological routing model, and uncertainty assessment for groundwater using models and gravimetric observations

Statistical Data Fusion and Science Data Analysis

  • Spatial and spatio-temporal data fusion for atmospheric fields in a Bayesian context, with applications to near surface temperature products from AIRS
  • Developed a leading system for recognizing, grouping, and tracking solar active regions used for SOHO and distributed as a data product for SDO
  • Proprioceptive and appearance-based terrain classification for autonomous rover navigation (DARPA, ARO)
  • Developed multi-observation compressed summaries for MISR and AIRS, which have been distributed as a Level 3 data product
  • Developing new statistical methods for assessing agreement between climate model simulations and observations in a distributed data context

Physical Modeling and Advanced Inversion Algorithms

  • Inversion of cosmic microwave background to recover spherical harmonic power, and structural parameters of cosmological models (Wilkinson Microwave Anisotropy Probe and Planck)
  • Data assimilation and Bayesian risk assessment for highly nonlinear weather systems



A 2019 seminar highlighted recent activities in data fusion and uncertainty quantification:

Earlier seminars: