Grow waste on the WPP2024 year-by-year population curve, not a frozen CAGR

[HughRunyan]

Output

Waste generation grows along the UN WPP2024 medium-variant population series year by year — factor(year) = P(year) / P(pivot) — instead of compounding a single frozen CAGR (growth_rate_*) indefinitely. Within the 1990–2050 horizon this is pure interpolation of the UN's own decelerating trajectory (no extrapolation), so long-run projections are far less dramatic and are citable.

This is the growth-math half of the change; the pipeline half is RMI_Climate_TRACE_Waste_Methane#wpp-population-growth-model. Merge this first — it's the editable dependency.

User story

As a modeller, I want landfill waste to follow the UN's projected population curve so that emission estimates past ~2035 are realistic and defensible instead of exploding on a compounded fixed rate.

What changed

• growth_factors_for_years() (new, class_defs.py): returns P(year)/P(pivot) when a population_series is supplied, and reproduces the legacy piecewise frozen-CAGR exactly when it is None — so every existing caller is unchanged by default.
• Threaded an optional population_series through WasteGeneratedDF.create and DivsDF.create / _create_dataframe (default None → old behaviour).
• _population_series_from_pop_data() (new, city_params.py): pulls pop_1990..pop_2050 for the site's iso3 out of pop_data; returns None (scalar fallback) when the yearly columns or the country are absent. Wired into the TRACE site path.
• Bug fix — double-applied growth (multi-row TRACE path): growth was applied once while building waste_time_series and again in WasteGeneratedDF.create, turning a 2 %/yr country into ~4 %/yr. For a KEN-like site the 2050/2023 ratio was 3.09; it is now 1.52 (matches the single-row path and the pure WPP curve).
• Bug fix — anchor mismatch: the multi-row projection pivot was hardcoded to 2020 while the single-row path pivots at the data year; both now pivot at the data year, eliminating a ~6 % offset. Observed values in the reported years are still respected (the reason the time-series path exists).

Acceptance criteria

• With a population_series, generated waste equals the WPP curve P(year)/P(pivot) exactly.
• With population_series=None, output is byte-for-byte the legacy frozen-CAGR (backward-compatible default).
• Growth is applied once, not twice, on the multi-row path.
• Single- and multi-row paths agree on projected (non-observed) years.
• No mass-balance regression: scenario diversion stays proportional to generated waste.

Definition of Done

• Acceptance criteria met
• Tests pass — 6 WPP regression tests + 113 mitigation-integration tests green (tests/manual in the paired repo, full-dep lane)
• Reviewed & merged (merge before the paired pipeline PR)

🤖 Generated with Claude Code