iEMSs 2022 Conference - Brussels, Belgium

iEMSs 2022 - Session C.1

Advances in calibration of process models and consequent training of surrogate models (machine learning)

Stream    : C - Computational Methods, Workflows, Informatics and Integrated Systems in Environmental Modelling

Session Leader: Francesco Serafin (Colorado State University), Timothy Green (USDA-ARS), Olaf David (Colorado State University), Jack Carlson (Colorado State University)

Emerging surrogate models from process models require a progression of operations that cover the entire spectrum from process model calibration to surrogate model validation. Each step is an essential part of this pipeline that can make or break the final product. Since each step can be implemented in numerous ways, this session focuses on sharing successful applications and innovative approaches to build the pipeline.


Potential topics for discussion within the session include, but are not limited to:


  • Selection of objective functions;

  • Evaluation of model metrics;

  • Data splitting for cross-validation;

  • Geospatial sampling and clustering;

  • Parameter space dimensionality/complexity;

  • Parameter non-uniqueness;

  • A priori & a posteriori sensitivities;

  • Error and uncertainty propagation from process models to surrogate models.


Specific case studies that can be targeted are simulation of crop growth, water quality and quantity, carbon sequestration and greenhouse gas emissions, as well as other environmental modeling studies.