PERSEUS multi-model vs the American Forests state CBM reports

✓ 2026-06-04 corrected CONUS v2 sweep posted

Full 48-state corrected sweep ran on Cardinal SLURM array 11295550 in ~10 minutes wall. CONUS-wide year-50 delta under LCMS-anchored disturbance rates (--from-bundle mode):

Scenario uncertainty is ~25 TgC at CONUS scale. Inventory stratification uncertainty is ~6,760 TgC. The methodological uncertainty axis (B1.1 vs B1.3 EXPNS) is roughly 270× larger than the scenario uncertainty axis under empirical disturbance rates. This is the primary finding. The qualitative story is unchanged from the v1.0 methods note: inventory stratification dominates cross-model uncertainty for this generation of CBM-CFS3 carbon models.

⚠ Earlier 2026-06-04 literature-rate correction (archived)

The CONUS scenario sweep originally published below (BAU sink to HIST source, 15,000 TgC swing) used literature-calibrated regional disturbance rates that overestimated the actual schedule by approximately 5.6× in MN. The previously published BAU was either the LCMS-calibrated schedule (MN, WA) or a true no-disturbance counterfactual (OR, CA, MD, PA — whose bundles ship with zero LCMS events). The scenario tool now has a --from-bundle mode that scales the bundle's empirical disturbance events directly.

Corrected 6 AF state result (year 50 delta Mg/ha under LCMS calibrated rates):

The corrected finding is methodologically stronger. Under realistic (LCMS calibrated) disturbance rates, policy scenario effects on 50 year ecosystem stock are within simulation noise (<1 Mg/ha for MN, <2 Mg/ha for WA). This means methodological uncertainty axes (inventory stratification, HWP storage, engine choice) dominate the scenario uncertainty axis at the 50 year horizon under empirically observed disturbance regimes. The qualitative methods finding (inventory stratification dominates cross model uncertainty, PNW source signature) is unchanged. The larger scenario effects in the v1.2.0 deposit reflect literature rates that are roughly 5 to 6× higher than LCMS observed. The original v1.2.0 outputs remain archived for reproducibility; the corrected v1.3.0 sweep with LCMS calibrated rates is in preparation.

Why this comparison matters

The American Forests state CBM reports are single model CBM-CFS3 outputs with one disturbance schedule and one set of scenario parameters per state, published one state at a time. PERSEUS runs both libcbm and GCBM under matched conventions across the full 48 state CONUS, identifies five refinement axes where the single model approach can be improved (inventory stratification, HWP long term storage, disturbance risk, landowner dynamics, economics plus uncertainty quantification), and projects 50 year ecosystem stock trajectories that confirm AF's directional findings while widening the reported uncertainty band. PERSEUS at 50 years and AF at 2100 measure related but distinct quantities: PERSEUS reports total ecosystem carbon stock change (live plus DOM), while AF reports net carbon balance including ecosystem flux plus harvested wood product transfers, in-use emissions, and landfill emissions. The two are directionally comparable but operate on different scopes.

Three headlines from the CONUS sweep

1. The CONUS sign flips from sink to source between BAU and HIST. A 15,000 TgC swing exposes how much of PERSEUS's previously published BAU sink was an artifact of the no-disturbance assumption. Once realistic disturbance is replayed, CONUS becomes a -10,100 TgC ecosystem source over 50 years (-36.8 Mg/ha mean). 36 of 48 states are sources under HIST, vs 7 under BAU. The remaining sink is concentrated in NE Atlantic Maritime and Lake States.

2. Reduced harvest is the highest leverage policy lever. RH (-30% harvest probability) recovers +3,001 TgC vs HIST over 50 years. RH_CC (-50% clearcut, partial harvest unchanged) recovers +5,042 TgC, about 1.7× more effective despite a smaller area footprint. The result underscores that clearcut probability is the most efficient knob to dial down per unit hectare in CBM-CFS3 family carbon accounting.

3. Climate-amplified wildfire adds another -1,000 TgC at CONUS scale. WARM_HIST applies region-specific wildfire amplifiers (PNW × 1.5, Mountain West × 1.5, Lake States × 1.2, NE × 1.2, South + Southeast × 1.0) on top of HIST. PNW takes the largest additional hit (-10 Mg/ha vs HIST), Mountain West -10 Mg/ha, Lake States and NE near zero. The asymmetry matches the AF Climate Change Impacts scenario but quantifies it across every CONUS state under uniform conventions.

Three headlines from the cross state sweep

1. HIST reframes everything. The previously published PERSEUS BAU was a no-disturbance counterfactual that placed OR at +2.3 Mg/ha and WA at -2.6. Once realistic disturbance is replayed (HIST), OR drops to -86.9 Mg/ha and WA to -77.2. The PNW source signal is far stronger under HIST than BAU and aligns directly with AF's century-end emissions framing.

2. Reduced harvest works in every state, but the magnitudes vary. RH_CC (clearcut down 50%, partial unchanged) recovers between +12 and +23 Mg/ha over the 50 year horizon across the six states, with the largest gain in MD (+23) and the smallest in CA (+12). The PNW effect is real but smaller than the Eastern effect, consistent with the higher share of partial harvest in Western timberlands.

3. WARM_HIST signal is regionally diagnostic. A 50% wildfire amplifier produces -8 to -12 Mg/ha additional loss in PNW (OR, CA, WA) over 50 years, while Lake States (MN) and NE (PA, MD) show -1 to 0 Mg/ha. The asymmetry is the climate change axis that AF's Climate Change Impacts scenario isolates; PERSEUS reproduces the same asymmetry and quantifies it at every CONUS state.