Speaker
Description
In the community of environmental modelling, the advent of hyper-resolution Earth observations and datasets in conjuncture with growing computational resources lead to an increase in model resolution.
The mathematical representations of biogeophysical processes need to be solved for billions of grid cells and thousands of time points.
Each process requires parameters that cannot be easily and sensibly set fix nor calibrated.
Instead they need to be inferred directly from the land surface properties through transfer functions.
In a simple form, they follow a context free grammar which follows the Fortran language.
The transferred effective parameters have a hierarchical interdependency forming a tree structure.
Yet finally, the shape of the arrays containing the land surface properties does usually not conform with the shape of the array of process parameters of the model.
This necessitates multiple array broadcasting, slicing and remapping steps.
The scientific approach - the Multiscale Parameter Regionalization (MPR) concept - is now available as an object-oriented and flexible Fortran library (https://git.ufz.de/chs/MPR).
In this presentation, we discuss the design of the MPR library, show implementation details and highlight major difficulties. We are strongly interested in community feedback on the implementation.
