Software infrastructure and algorithms to facilitate co-location of observation and model data

Observation data (e.g., from the A-Train satellite constellation) and model data (e.g., from ECMWF analysis) live in their own respective grid spaces, have different sampling characteristics, and use different formats and structures in archiving. This deters synergistic usage of these datasets, e.g., to use one instrument to calibrate another, or to conduct model data assimilation using observation. In response to the challenge, we have built a software tool to facilitate the co-location of A-Train and ECMWF data. That is, to interpolate from a source data grid onto that of a target. This tool is written in Python with C extension, has RESTful web interfaces in the frontend, and uses parallel computing for high performance in the backend. In this presentation, we will focus on the computer science aspects of this tool, namely the architectural design, the infrastructure of the web services, the approach to parallelization, and the key algorithms being used for, e.g., linearly-scaling nearest neighbor search among different grids. Some use cases and their scientific significance will also be discussed briefly.