Quality and sustainability of machine learning software for climate change

Climate change impacts human health through extreme weather events, higher risks of illnesses and resource displacement.To battle these challenges, it is imperative to combine and correlate data belonging to different domains, such as health, weather, supply chains etc.. Furthermore, the application of cutting-edge technologies such as powerful machine learning will rely on the collaboration of organizations across international borders. This entails creating integrated systems able to support a constellation of local initiatives including data resources, complex prediction models and custom-developed software applications. From a software quality and software sustainability perspective, this goal poses several novel challenges, especially related to the long-term viability of machine learning applications, for which engineering practices are considered an emerging domain. Maintaining and evolving such a heterogeneous system in the long term is a crucial software engineering research goal that is essential to provide a powerful tool for humanity to deal with climate change.    

Machine learning is an emerging domain where non-functional qualities such as the maintainability and evolvability of developed software systems and complex data management approaches are not well understood from a software engineering perspective, as we have  recently investigated with one of our PhD students [1].This challenge is amplified by the necessity of integrating such ML applications, often developed for a limited local usage, with traditionally-developed and legacy applications into a well architected ecosystem of independent services, which needs updated development strategies such Agile and DevOps practices as well as solid APIs to be exploited by international and diverse stakeholders.