Data driven processes (models) in FABM?

[Fenjuan Hu]

Dear FABM developers, att. Jorn &Karsten

For the spirit of  community based model framework, I embrace openness of FABM. So I am trying this wild idea with FABMers. 

With current development of FABM, that parsac has some of the Bayesian models perforME on parameter estimation and data, and future features of data assimilation in FABM, I can see the light of some data driven approach in FABM framework. 

I am currently working on a concept of process driven model combined with data driven model, where we would like to combine Machine learning, Bayesian Networks, Bayesian hiearchical model, with some of the FABM based model (both hydroydanmic and biogeochemical, fx. GOTM-FABM-WET). A typical useful case would be predicting algal blooms. The workflow(for example) could be 1)using GOTM-FABM-WET to simulate the lake process calibrated with long term monitoring data; then 2) using Machine learning model to simulate the growth season, where algal blooms are more of interests. The ML model will use high frequency algal data( some good sites avaialble) and process driven model generated high frequency data to train the ML model. Finally 3)process guided ML model will be able to predict future bloom events based on future scenarios and avaialble data. 

With the current development of parsac and data assimilation approach, I could imagine combining certian ML model with FABM would be possible? 

I have also in mind of using Bayesian Hierarchical Model  (Python package), and combined with FABM models, think the workflow can also be realizable?

Of course this still could be done independtly without FABM support, how ever, a nice package or workflow enable smart integration of Data driven models in to FABM models would be optimal. 

What do you think?

Best 

Fen