Climate projection pipeline for net load forecasting and BTM attribution

The operational picture

Net load forecasting on CAISO is an attribution problem: the grid operator sees aggregate load at the substation but can't directly meter rooftop solar, home batteries, or EV chargers. Gramm's approach is a two-stage decomposition — a gross-demand model driven by temperature and calendar variables, and a physics-informed BTM solar model driven by spatial Global Horizontal Irradiance (HRRR GHI) over the ~17 GW installed CAISO rooftop-PV fleet with a calibrated system-efficiency term. Net load is reconstructed as gross demand − BTM solar rather than predicted directly, so the two error sources stay separable.

Identifiability of the two components rests on two priors: installed CAISO rooftop-PV capacity is known to within a few percent from the CEC Distributed Generation Interconnection Dataset, which constrains the BTM-solar scale; and AMI-metered subsets from participating LSEs provide per-customer ground-truth that calibrates the GHI-to-PV coefficient where available. Gross demand and BTM solar are not separately observable from aggregate substation load, but under these priors they are separately recoverable.

Climate projections extend the same decomposition out over multi-year planning horizons for IRP filings, resource adequacy analyses, and procurement cycles where a short-term operational forecast isn't the right horizon. Built in collaboration with Fairbuild.