fix(ciceroscmpy2): hold aerosols at PI for idealised conc-driven runs

changelog/108.fix.md

@@ -0,0 +1 @@
+Fixed CICEROSCMPY2 holding aerosol / ozone-precursor emissions at the rising historical baseline on the concentration-driven idealised path (`abrupt-4xCO2`, `1pctCO2`). Because `emstart` is `nyend` there (to keep CO2 concentration-driven), the supplied PI-flat emissions were applied only at the final year and every prior year fell back to the `historical_em` baseline, producing a spurious growing aerosol cooling (~0.6 W/m² ERF drift across a 4xCO2 step). `_build_hybrid_emissions_data` now takes an `overlay_from` argument and applies the supplied emissions from `nystart` for idealised runs, so non-CO2 forcers stay at preindustrial; CO2 is unaffected (concentration-driven via `conc_run`).

src/openscm_runner/adapters/ciceroscm_py2_adapter/ciceroscmpy2_adapter.py

@@ -685,6 +685,13 @@ def _build_scendata_list(
             nyend=nyend,
             emstart=scen_emstart,
             zero_unsupplied=idealised,
+            # Idealised runs hold every non-CO2 forcer at PI. On the
+            # conc-driven path scen_emstart is nyend (keeping CO2
+            # concentration-driven), but the supplied PI-flat aerosol /
+            # ozone-precursor emissions must apply across the whole
+            # window rather than fall back to the rising historical
+            # baseline. Pin the overlay to nystart for idealised runs.
+            overlay_from=nystart if idealised else None,
         )
 
         conc_data = None
@@ -839,6 +846,7 @@ def _build_hybrid_emissions_data(
     nyend: int,
     emstart: int,
     zero_unsupplied: bool = False,
+    overlay_from: int | None = None,
 ):
     """
     Build an RCMIP-format emissions DataFrame for one scenario.
@@ -859,6 +867,18 @@ def _build_hybrid_emissions_data(
 
     Parameters
     ----------
+    overlay_from
+        Year from which the user's supplied emissions overwrite the
+        baseline trajectory. Defaults to ``emstart``. For idealised
+        concentration-driven runs the model ``emstart`` is ``nyend``
+        (so CO2 stays concentration-driven), but the supplied PI-flat
+        aerosol / ozone-precursor emissions must still apply across the
+        whole window -- otherwise the species fall back to the rising
+        historical baseline (a spurious growing aerosol cooling that
+        drifts ERF down through abrupt-4xCO2 / 1pctCO2). Pass
+        ``nystart`` in that case to pin the supplied emissions from the
+        run start; CO2 is unaffected because it is not supplied on the
+        conc-driven path (it stays concentration-driven via conc_run).
     zero_unsupplied
         When ``True``, every species column the user does NOT supply
         is overwritten with ``0.0`` across the whole ``[nystart, nyend]``
@@ -881,6 +901,8 @@ def _build_hybrid_emissions_data(
 
     from ..utils.cicero_utils.make_scenario_common import cicero_comp_dict
 
+    apply_from = emstart if overlay_from is None else overlay_from
+
     df = (
         pd.read_csv(
             baseline_em_file, delimiter="\t", index_col=0, skiprows=[1, 2, 3]
@@ -965,7 +987,7 @@ def _build_hybrid_emissions_data(
             )
             continue
         for year, val in user_row.items():
-            if year in df.index and year >= emstart and not pd.isna(val):
+            if year in df.index and year >= apply_from and not pd.isna(val):
                 df.at[year, col] = val * convfactor
         overlaid.append(cicero_species)
         overlaid_df_cols.add(col)