Developing and applying a vetting toolkit to assess the state-level input calibration in GCAM-USA

GCAM-USA extends the Global Change Analysis Model to provide state-level resolution for energy transformation, end-use demands, and inter-state energy trade across the United States. While the model relies on extensive USA-specific data for calibration, capacity constraints prevent full capture of state-level variability, producing systematic discrepancies that limit the model's utility for state policy applications. To address this, we developed a diagnostic vetting toolkit that compares and harmonizes historical data with modeled base-year outputs. 

The toolkit comprises four components: automated EIA API data retrieval with cloud-based cataloging, processing scripts that query GCAM outputs and harmonize them with recategorized historical data, visualization tools that generate diagnostic reports and algorithmically rank calibration priorities, and calibrated input files that resolve identified discrepancies and remain continuously updatable.

We demonstrate the toolkit through three use cases. First, applying it to electricity generation efficiency revealed that GCAM-USA's downscaling methodology obscured significant state-level variation in generation efficiency even when national aggregates matched historical data closely. Deriving state-specific efficiency values directly from EIA fuel consumption and net generation data substantially reduced these discrepancies. Second, in a Maryland policy modeling context, the toolkit enabled targeted calibration of industrial cement production yielding a more accurate baseline for evaluating decarbonization scenarios. Third, we set solar generation floors based on EIA observations for historical years that the model still treats as projections in order to match the scale of 2022 to 2025 solar deployment in states like California and Texas.

These cases illustrate how the same framework scales from broad structural corrections to facility-level adjustments. The toolkit's modular, automated design lowers the barrier to ongoing calibration and offers a transferable template for improving state- and regional-level accuracy for major model outputs in other integrated assessment models.