Skip to content

Downscaling demand

The IAM reports demand per sector, per year, and per multi-country region. PyPSA models need country-level demand with hourly profiles. This requires a downscaling approach for both the spatial and the temporal dimensions.

Scope: demand only

Downscaling in iampypsa today covers electricity demand only — not capacities. Capacity targets stay at whatever region resolution the IAM reports them at; they're reconciled with the model's existing plant database directly instead, a different process covered in Harmonising capacities.

Spatial downscaling: region → country

Work in progress

The spatial downscaling is still work in progress. It currently does not match historical patterns.

This part is implemented in iampypsa. A model region like REMIND's ECE typically covers many countries, whereas a PyPSA model requires country-level demand.

  • The region → country map — read_region_map() reads the packaged region-mapping CSV (data/remind/regions.csv) and returns {region: [country, ...]} (or the reverse, depending on source/target).
  • Proxy shares — each sector's regional demand is weighted by proxies: normalised population and GDP projections (downscale/proxy.py) for most sectors, and degree-day-weighted demand proxies for heating (heatpump, resistive) and cooling (space_cooling). A sector's sector_weights entry names which proxies to blend and in what proportion (e.g. {gdp: 0.6, population: 0.4}).
  • Applying the split — disaggregate_demand_to_country() applies those shares to every (year, region, sector) row. Most integrations don't call this directly; they call Coupler.downscale_country_demand(), which reads the regional demand and disaggregates it in one step.

Temporal downscaling: annual → hourly

This part is not implemented in iampypsa — it happens on the PyPSA-model side, consistent with the general split: IAM-side logic lives in the package, PyPSA-side logic lives in each regional PyPSA model.

In each PyPSA model, the spatially downscaled total demand per (country, year, sector) then needs to be downscaled temporally into an hourly timeseries by taking a demand shape from another source and scaling it such that the profile's annual sum matches the country-level annual total.

Next