Contents

Python is a great language for writing programming frameworks. Python frameworks are normally addressed to software developers who are Python professionals. I developed a software package in a scientific institution, designed to be used by non-programmers, but also designed to enable customisation through programming by some users. I finally designed a three level package:

• A Python programming framework, addressed whoever wanted to invest in Python and had plans to develop applications;
• An application built on that framework with an interface designed for non-programmers;
• Ways of customising the behaviour of the application by providing some basic Python functions or more elaborated code.

One of the challenges was to offer an application with an easy to use interface, not graphical, not web-based and not requiring Python programming. This interface was necessary for batch processing.

This talk addresses how this project was carried out, the technical solutions adopted and how Python was introduced in an operational scientific institution (http://www.ecmwf.int) where most users were Fortran programmers. Python was introduced as early as 2004 and it was a challenge to gain acceptance. I will also make a parallel with a project I am currently working on for NASA (http://gmao.gsfc.nasa.gov/). Desperately trying to forget technical details summarises how I tried, using Python, to help Meteorology scientists to focus on their domain of expertise instead of constantly solving technical problems.

Background:

The disciplines of Meteorology and Climate involve numerical modelling of physical phenomena. The amount of data going in and out of the model is considerable. The organisation and the storage of data is complicated, their post-processing is a challenge. Scientists need to access and process input and output data to monitor the trends of the input data and to evaluate the performance of their models. Those statistics, diagnostics, plots and verifications are crucial to the improvement of the quality of the models. Finding the right data, decoding it, transforming it to be ready for use are necessary steps to initiate the pre-processing. All these actions are fundamentally the same between different prediction centres, but the data organisation and file formats can differ.