A novel hybrid framework for efficient distributed model spin-up

Initializing spatially distributed ecohydrological models with soil biogeochemistry is computationally expensive, particularly when lateral fluxes of water, carbon, and nutrients need to be explicitly resolved across complex terrain.

In a recent study published in Geoscientific Model Development (GMD) and led by PhD student Taiqi Lian, we developed a hybrid initialization framework for distributed ecohydrological models with soil biogeochemistry that combines physically based 1D flux-tracking spin-up simulations with a random forest–based machine learning component.

The proposed framework addresses this challenge by first running detailed 1D simulations on a subset of representative grid cells to reach steady-state conditions. These results are then used to train a random forest model that extrapolates spatially heterogeneous, topography-informed initial conditions across the full model domain.

By integrating process-based simulations with data-driven learning, the framework preserves the influence of topography and lateral transport on ecosystem states while substantially reducing computational costs compared to fully distributed spin-up approaches. This makes high-resolution ecohydrological-biogeochemical modeling more tractable for large and complex catchments without compromising the realism of spatial patterns.

If you are interested, read the full publication at this link!