Model Insight Patterns
Jataware recently open sourced Model Insight Patterns (https://github.com/worldmodelers/insight-patterns). This is a scientific computing transformation menagerie that contains various Python modules for post-processing model output to improve explainability, understanding, and the portability of results. These patterns are designed to improve modeler's and analyst's ability to derive insights from their models and to make it easier for them to share their results.
It is organized into 4 main categories of transforms, each of which contain examples for executing the pattern against NetCDF, GeoTIFF, and CSV model outputs. These are:
- regrid
- time-scale
- clipping
- analytics
Model regridding
Regridding a model output involves either upscaling or downscaling the model output to a desired resolution. This is typically done to decrease the overall size of the model output on disk.
Time Scaling
The goal of adjusting a model's time scale is to ensure that the time resolution is sufficiently coarse grained to readily facilitate end-user insight generation.
Clipping
Often, a model produces output for a large area or even the entire globe. These outputs can be at a massive scale for high
resolution satellite imagery based models and analysts often only care about a specific region. Clipping these outputs to a
region of interest can provide extreme efficiencies downstream and enable modelers to more readily share their results.
Analytics
We primarily focus on two categories of analytics: those derived by leveraging population data and those derived by leveraging land cover masks. Here we demonstrate how to join model output with population and land cover masks in order to generate insights at the human scale.
