The module adddresses wildfires analysis and prediction providing tools allowing users to pre-processdata, choose model architecture and train the model, post-process results (basic visualisation)
The development of the Python package is currently underway. In this first version, a Machine Learning pipeline has been developed for running the different thematic modules. The capabilities include subsetting SeasfireCube data in Training, Validation and Testing, after a preliminary selection of drivers and target variables. The ML model used for this version is a standard UNet++ network; users can customise the network depth. The module includes model training and data visualisation utilities.
Train, validate and test a ML model for the fire ignition prediction by selecting the best hyperparameter configuration, including potentially different ML architectures. Validate the resulting model on CMIP6 reanalysis data and build inference pipelines with different CMIP6 scenarios. Support for multi-model (on CMIP6 data) ensembles will also be evaluated.