`Capacity|Electricity` is over-estimated when `desired_regions` is a subset of regions

[jiheunha]

When generate_report() is run with a subset of regions via desired_regions, the resulting Capacity|Electricity|* values for the selected regions are inflated by roughly 7x–60x compared to the same region in an all-regions run. Running with desired_regions = "All" gives correct, realistic capacities.

Observed behavior

Same single scenario, same region (South Korea), same GCAM_version = "v7.0", same database. The only difference is desired_regions. Values in GW, year 2020:

Variable	desired_regions = "All"	desired_regions = c("South Korea","Russia")	ratio
Capacity|Electricity (total)	153.0	3091.3	20.2x
Capacity|Electricity|Coal	39.8	1139.7	28.6x
Capacity|Electricity|Gas	65.6	595.2	9.1x
Capacity|Electricity|Nuclear	22.9	162.7	7.1x
Capacity|Electricity|Solar	14.5	655.1	45.2x
Capacity|Electricity|Wind	3.4	199.3	58.2x

The "All" column matches real South Korea installed capacity well (≈150 GW total, ≈40 GW coal, ≈23 GW nuclear, ≈15 GW solar in 2020). The subset column is physically impossible. The inflation grows as fewer regions are selected.

Root cause

In get_cf_iea_tmp() (R/functions.R), the IEA-consistent capacity factor is

cf = EJ / (value * hr_per_yr * EJ_to_GWh)

where the numerator and denominator are computed at different region scopes:

• Numerator EJ comes from secondary_energy_clean, summed by technology:

  tmp1 <- secondary_energy_clean %>%
    dplyr::filter(year == base_year_p, scenario == unique(secondary_energy_clean$scenario)[1]) %>%
    dplyr::group_by(var) %>%
    dplyr::summarise(EJ = sum(value, na.rm = T))

  secondary_energy_clean only contains the regions present in the project, so when desired_regions is a subset this sum is the generation of only those regions.

• Denominator value comes from iea_capacity, which is provided only for region "World" (global installed capacity, region-independent and fixed):

  tmp2 <- iea_capacity %>%
    dplyr::filter(period == base_year_p, scenario == "Current Policies Scenario")

So selecting fewer regions shrinks the numerator (less generation) while the denominator stays at the global World capacity. cf_iea therefore comes out far too small, and since downstream capacity is generation / (cf * ...) (get_elec_capacity_tot()), the reported capacity explodes. The approximate inflation factor is (world generation) / (selected-region generation), which is consistent with the 7x–60x range above.

cf_iea is intended to be a single global (world-average) capacity factor, so the numerator must be world generation. With a region-filtered project, the world generation is simply not available, which is what breaks the calculation.

How to reproduce

# Correct
generate_report(db_path, db_name, scenarios = <scenario>,
                GCAM_version = "v7.0", desired_regions = "All",
                desired_variables = "Capacity*")

# Inflated — compare South Korea Capacity|Electricity between the two
generate_report(db_path, db_name, scenarios = <scenario>,
                GCAM_version = "v7.0", desired_regions = c("South Korea", "Russia"),
                desired_variables = "Capacity*")

Confirmed on v7.0. Because iea_capacity is World-only for every version, the
same scope mismatch should affect all GCAM versions.

Possible fixes

The proper fix is a design decision, so we would like your input before sending a PR:

1. Compute the cf_iea numerator from global base-year generation regardless of desired_regions (e.g. always query/keep world secondary energy, or bundle a precomputed global base-year generation so cf_iea is region-independent by construction), or
2. Restrict Capacity|Electricity to desired_regions = "All" and clearly document (and warn) that capacity is not valid for region subsets.

Happy to help with a PR if useful.