Optimizing NESTOR, a tool for preparing meteorological and surface data for the MAR model

Abstract

Abstract - Simulating the Earth's climate is a complex and challenging task that the scientific community is actively researching with various approaches. Besides the computer models that simulate the climate for the entire Earth's surface, also called Global Climate Models (GCMs), a significant part of the ongoing research focuses itself on Regional Climate Models (RCMs). These computer models simulate the climate over high resolution grids modelling specific regions of the globe in order to produce finer results w.r.t. GCMs, notably regarding precipitations (i.e., rainfall and snowfall). The MAR model is a good example of RCM, and is particularily effective at simulating precipitations as well as snow and ice. As such, several research groups actively use it to study the evolution of polar regions or to predict future hydroclimatic conditions in specific regions of the world.

In order to be as realistic as possible, a RCM needs to feed the borders of its grid with pre-existing meteorological data in order to take account of how the climate evolves outside of the region of interest. This data, or forcing fields, can come either from real-life measurements or from predictions computed by a GCM. In the case of the MAR model, the large scale grids must first be processed by NESTOR, a companion software that is responsible for downscaling said grids, i.e., inferring high resolution data from the input domain to initialize the regional grids and prepare the forcing fields.

While the MAR model has been maintained and updated on a regular basis since its creation, NESTOR has not received a major update since 2004, though some components have been added much more recently to meet the needs of MAR users. While it is still doing its intended task, NESTOR requires a significant amount of time to process most of its typical use cases, a problem which constitutes an additional constraint to MAR users.

This master thesis thoroughly reviews the source code of NESTOR to identify its main issues and subsequently introduces a few simple changes that significantly improve its performance. When all suggested changes are applied, typical use cases can be computed in (much) less than one minute, with some of these use cases having been accelerated up to 40 times with respects to the unedited NESTOR. A comparison of the results of a MAR simulation started with the old NESTOR with those of a second simulation kickstarted with the output files produced by the updated NESTOR demonstrates that the suggested changes did not alter the results of the MAR model, despite the well-known chaotic nature of some atmospheric processes. Finally, the updated NESTOR also drops obsolete features with respects to the previous version and slightly improves the readability of its source code.