WP1 aims to ensure the execution of the work plan, quality control, knowledge management, and effective communication between stakeholders. This work package also includes procurement activities for the necessary infrastructure (hardware/software). Furthermore, WP1 manages the project as a whole, including reporting activities, supervision of data management, and financial, legal, administrative, and technical matters. It also includes technical coordination of activities between the various WPs of the project, regular communication between project members (video conferences, meetings), monitoring of project performance, risk analysis, and contingency management, as well as planning, monitoring, and justification of resource allocation.

WP2 concerns analyzing the requirements to be carried out after collecting specifications to define Hi-WeFAI.
Nowcasting consists of predicting meteorological behaviors with an advance of a few minutes to hours, allowing local authorities to be warned of dangerous situations. Nowcasting is based on data detected by weather radars, which allow meteorologists to produce highly detailed forecasts specific to the location. Adaptable solutions for federated learning models for XWR and WS and the generative model for synthesizing the dataset to be used with input for the hydrometeorological model are analyzed in this WP.

WP3 aims to design and develop a nowcasting model based on AI techniques for XWR and WS instruments. In this phase, the design of the software infrastructure to implement the FAI, the individual models on the edge computing resources for XWR and WS, the generative model for the synthesis of the dataset to be used with the hydrometeorological model, the decisional component for the activation of said model, and the system for the dissemination of the results are carried out.

A large sample of reflectivity data collected between 2018 and 2023 will be pre-processed to train the nowcasting model by suitably simulating the federated learning through the two different XWRs available. Once the federated learning procedure has been implemented and the ad-hoc protocol based on the DTF has been developed, the generative model for the synthesis of the precipitation prediction dataset as input for the hydrometeorological model is implemented and the decisional component for the activation of the hydrometeorological model is implemented. In this WP all the static data necessary for the configuration of the hydrometeorological model are prepared. The ParFlow implementation of the Hi-WeFAI project will be validated in WP5.
The DAGOnStar workflow engine will be used and extended with a new software component for generating automatic alerts provided by the direct interface with different notification technologies.

WP5 is dedicated to hydrometeorological nowcasting, evaluating each component of the workflow. The predictions of the precipitation field evolution provided by the deep learning model in the next 180 minutes will be compared with the precipitation data observed on the ground (XWR, WS) using different statistical indices.

Within WP6, the use case will be tested on on-premises, cloud, and hybrid computing resources thanks to the experience of industrial partners.

WeFAI releases, and the public dissemination of the project. A dedicated web portal for the project will be created. Workshops and technical presentations will be held to foster knowledge exchange with other research groups, companies, and associations, as well as training events, discussion forums, and mailing lists. Promotional material will be prepared to present the project results and encourage their use.