Scientific workflows designed to handle massive datasets through distributed high-performance computing (HPC) infrastructures or elastic on-demand computational services have established themselves as a robust and mature paradigm within data science. Within this context, one of the most consolidated production applications is the orchestration of environmental models for simulation and forecasting tasks.

This presentation illustrates our perspective on workflows as essential building blocks for environmental systems, where numerical modeling is combined with artificial intelligence to strengthen forecasting and predictive capabilities. At the HPSC SmartLab of the University of Naples "Parthenope," we developed DAGonStar, a workflow engine designed to orchestrate environmental models used by the Center for Monitoring and Modeling Marine and Atmosphere (CMMMA) to produce weather and marine predictions.

Among the laboratory's operational applications is MytilEx, a project funded by the Campania Regional Government, which aims to forecast E. coli contamination in cultivated mussels. The system improves pollutant transport and dispersion simulations (carried out with the WaComM++ model) by integrating an artificial intelligence module (AIQUAM++), trained on microbiological observations. Initial system evaluations reveal prediction accuracies above 90% for E. coli presence, a substantial step forward in applying computational intelligence to environmental and food safety domains.

The same workflow building blocks that supported MytilEx have also enabled the development of two further projects. The first, MytilX—currently underway and funded by the Istituto Zooprofilattico Sperimentale dell'Umbria e delle Marche "Togo Rosati"—has shown through preliminary experiments that the MytilEx success case can be replicated at other sites. The second, SmokeTracer, funded by the Campania Regional Government, is an on-demand HPC workflow aimed at estimating the potential soil contamination footprint caused by wildfires or arson. SmokeTracer has been implemented partly by reusing modules already available in the DAGonStar framework.