New article from the ARIES team at BC3, now published in Plos One, shows how scientific modelling applied to agricultural systems can be accessible, interoperable and user-friendly by integrating a set of complex ecological models coming from various disciplines, backgrounds and spatiotemporal scales in the ARIES platform.
The study is led by doctoral researcher Alba Márquez alongside with Ikerbasque professor and ARIES’ main lead researcher Ferdinando Villa and Ikerbasque research fellow and ARIES managing director Stefano Balbi.
New technology, such as cloud computing, remote sensing and artificial intelligence (AI) can play an important role in sustainable agriculture, optimizing resources, providing key spatial-temporal information and identifying the most appropriate and effective practices for better management.
However, one of the main issues preventing the full use of these new technologies in agricultural modelling arises from the multidisciplinary nature of the problem and the variety of data and models needed, produced by different scientific domains from climatology to ecology and social sciences and their lack of interaction.
The challenge: Scaling up the ‘PUERTO’ model
To create the model set, BC3 researchers collaborated with Dr Busqué (Cantabrian Agricultural Research and Training Centre) to adapt his so-called ‘PUERTO’ model to the ARIES platform.
‘PUERTO’ is a key model for the Cantabrian rangeland that was developed around the structure, growth and utilization of pastures in the Spanish region. However, it had only been used in regional projects, never contributing to more comprehensive computational workflows. That was BC3 researchers’ challenge for the past year.
The model was written using R software, and although it was divided into different files, it had been built as a single model. The main code was monolithic, meaning that it had been written continuously, and the source lines of code could not be run independently. An interface for expert users in R software wasn’t of much help either to understand the code.
Turning ‘PUERTO’ FAIR
Internet and online data repositories have made it easier for both humans and computers to retrieve digital resources, making these both findable and accessible. When resources are related to the same topic and can interact among themselves it means that they are interoperable. Besides, we speak about reusability when resources and models can be used in other contexts.
These four principles – FAIR (Findable, Accessible, Reusable and Interoperable) – are the main objectives to be accomplished when building a model. Unfortunately, ‘PUERTO’ did not fulfil any of these, as the model worked through associated tables at a local level that related to a specific study area.
One of ARIES’ main goals is to have different models talking and interacting among each other and, to achieve this, we need to make them interoperable.
In order to adapt ‘PUERTO’ to k.IM, the semantic language of the ARIES technology – k.LAB –, Márquez divided the code of ‘PUERTO’ into 10 thematic script files, also called ‘modules’. Then, they transformed the homogeneous R code into small semantic models associated with each of these modules in order to execute them independently. As models were logically consistent and self-contained, only those models that were needed were executed, rather than the whole script block.
Results speak for themselves: 246 independent models could interact with any model in the k.LAB software, whether being in the same module or not.
Besides being more understandable to end-users thanks to the semantic modelling, models could be executed in any spatial or temporal context. And, even more importantly, the whole process was much faster because it was powered by AI and Machine Reasoning.
Likewise, models were constantly improved due to scientific and stakeholder teamwork, since ARIES consists of a collaborative and open-source platform for everyone.
Integrating knowledge through semantic modelling
In Alba’s words, sometimes, the complexity of modelling a problem involving different disciplines, identities and scales, can generate unsatisfactory outcomes not reflecting reality as it is. This can be due to different reasons: the limited access to knowing all sides of the issue, or the challenging reach to resources.
Thus, it is key to articulate all the various components properly, as well as to boost collaboration among all stakeholders – from governments to individuals – to share their knowledge. It is equally important to adapt any model and resource from a particular context to a different one, by having means to reuse them.
Integration is about defining every model individually, in order to link them without any human intervention respecting their own granularity. Integration consists of adapting models to a changing problem in a changing world.
Márquez’s work focuses on modelling forest ecosystem services through remote sensing and AI. She is involved in the ALICE Project, a European initiative aiming to promote sustainable investments in Blue-Green Infrastructure Networks (BGINs) through the identification of nature’s benefits in the Atlantic Region – Portugal, Spain, France and North Ireland –.