Open science is built on the principles of making scientific research accessible to all, benefiting the whole of society. One notable facet of EXPECT, underscored by Theme 4, is its commitment to open science. The provision of open-source code fosters reproducibility and enables researchers to build upon the project’s work.
Machine learning applied to Earth Sciences
The EXPECT project aims to uncover deeper insights into the drivers of climate change by leveraging underutilised data and advanced machine learning models. By focusing on accessible tools and streamlined processes, the project paves the way for broader engagement and impactful discoveries in climate science.
Now, complex codes and datasets developed for the EXPECT project are being transformed into accessible scripts, enabling non-experts to run sophisticated analyses with ease.
Amanda Duarte, a Senior Researcher at BSC, is at the forefront of this work. With a PhD in machine learning focused on natural language processing and computer vision, Amanda brings a unique perspective to the Earth Sciences Department. Her mission is clear: to simplify the complexities of climate research and make advanced tools more practical and impactful for researchers worldwide.
”The main motivation to create an open science platform is that, usually, working with climate data is not so easy. Even though some of this data is public, the process to access, preprocess and adjust it can be long and complex.”
How did you become involved in EXPECT?
Climate change stands as one of the most pressing issues of our time, a reality that deeply influenced Dr Duarte’s decision to join EXPECT. ‘I was interested in how I could apply what I had learnt in my background studies and give back to society,’ she says.
‘The main motivation to create an open science platform is that, usually, working with climate data is not so easy. Even though some of this data is public, the process to access, preprocess and adjust it can be long and complex.’
What is your role within EXPECT?
Besides overseeing the rollout of the open science platform, Dr Duarte coordinates a team that is working on machine learning applied to Earth sensors inside the department at BSC. Her work emphasizes making advanced tools practical and impactful for a wider research community.
‘Our goal is to create a better way of reproducibility and give more visibility to the discoveries made in EXPECT,’ Duarte says.
The team refines complex code and standardises formats and grids in datasets, turning them into plug-and-play scripts that streamline usability in different models. ‘We clean up the code, create scripts, and make it as simple as pressing a button to train a model,’ she says.
By reducing the technical burden, the platform developed at BSC seeks to ‘democratise the use of the data and machine learning models we generate’, adds Dr Duarte. Eventually, this should enable a broader range of researchers to engage with and build on critical climate insights with minimal effort.
A dual approach to benchmark progress
‘One way is to compare the original code created by the scientists with the scripts we produce and then give both to someone unfamiliar with the project,’ says Dr Duarte. By testing scripts to assess ease of use, ‘we observe which version is simpler to work with and whether our scripts can be used without significant difficulties.’
Another key metric involves adoption rates over time. ‘We measure how many people have downloaded the datasets we’ve created or used the codes we developed, Duarte adds. However, the true impact of an open science platform requires patience. ‘These results take time to materialize, which we may only fully understand once the project is complete,’ she concludes.
Data validation: the cornerstone of effective climate modelling
Working with climate data presents unique challenges. The sheer volume of data makes it difficult to interpret and extract meaningful insights. Machine learning deepens complexity. ‘These models excel at learning from data, but only if the data provided is accurate and relevant,’ she adds.
Ensuring data quality and having the computational power to process it are significant hurdles. ‘Supercomputing centers like ours are essential, but without access to such resources, conducting this kind of research isn’t always possible,’ Duarte notes.
Inspired by all things open science? Pass this knowledge on and venture into more project explainers on the EXPECT website.
We’re on LinkedIn, X, and Bluesky. Follow us there if you haven’t already!