# GLAMBIE v17a_TIModel — Comparison Report

Generated by `plot_v17a_comparisons.py`. Two comparisons, both over the full submission period
(1975–2025, monthly, all 19 RGI regions):

- **A.** `glambie_submission_fixed_geometry_v17a_TIModel` vs `glambie_submission_fixed_geometry_v163_TIModel`
  — same method (fixed geometry), old (v163) vs new (v17a) OGGM source/calibration.
- **B.** `glambie_submission_spinup_geometry_v17a_TIModel` vs `glambie_submission_fixed_geometry_v17a_TIModel`
  — same source (v17a), fixed vs dynamically-evolving geometry.

Underlying numbers: [comparison_A_v17a_vs_v163.csv](comparison_A_v17a_vs_v163.csv),
[comparison_B_spinup_vs_fixed.csv](comparison_B_spinup_vs_fixed.csv).

## A. v17a vs v163 (fixed geometry): the source update didn't move the needle

[plot_A1_global_v17a_vs_v163.png](plot_A1_global_v17a_vs_v163.png) —
the two global cumulative curves are visually indistinguishable (v163's grey line is fully hidden under v17a's red).
Global cumulative mass change 1975–2025: **−9115.7 Gt (v163) vs −9124.0 Gt (v17a), a −8.3 Gt difference — 0.09%**.

[plot_A2_by_region_v17a_vs_v163.png](plot_A2_by_region_v17a_vs_v163.png) and the
own-scale [plot_A2b_diff_only_v17a_vs_v163.png](plot_A2b_diff_only_v17a_vs_v163.png) — per-region differences are
all small (largest is RGI19 at −7.6 Gt on a −978 Gt base, 0.8%). No region flips sign or changes materially.

[plot_A3_annual_diff_v17a_vs_v163.png](plot_A3_annual_diff_v17a_vs_v163.png) — the annual specific-MB difference
per region oscillates around zero with no trend or step-change across the 50-year record (±0.01–0.03 m w.e./yr,
noise-level). This is exactly what we'd hope for: the v1.7a source (new prepro pipeline, `informed_threestep` +
dynamic-spinup calibration) reproduces the same fixed-geometry climate-driven signal as the older v1.6/v163
source, with no systematic bias — a good consistency check on the pipeline update itself, independent of the
unit-conversion bug that was fixed separately (see CLAUDE.md).

**Bottom line: A is a clean validation, not really a "difference to explain."** Use v17a_TIModel going forward
with confidence that it isn't silently changing the fixed-geometry answer.

## B. Spinup (varying) vs fixed geometry: small globally, substantial regionally

[plot_B1_global_spinup_vs_fixed.png](plot_B1_global_spinup_vs_fixed.png) — global curves track closely, with a
mild, physically-sensible pattern (spinup slightly less negative than fixed through the 1990s–2000s, converging
again by 2025). Global cumulative: **−9124.0 Gt (fixed) vs −9114.6 Gt (spinup), a +9.4 Gt difference — 0.1%.**

That global agreement is misleading on its own — it comes from **regional differences that are an order of
magnitude larger than the global difference and don't all point the same way, so they largely cancel out in
the sum.** This is visible in [plot_B2_by_region_spinup_vs_fixed.png](plot_B2_by_region_spinup_vs_fixed.png)
(absolute Gt, dominated visually by the largest regions) and much more clearly in the two difference-only
charts on their own scale:
[plot_B2b_diff_only_spinup_vs_fixed.png](plot_B2b_diff_only_spinup_vs_fixed.png) (Gt) and
[plot_B2c_pct_diff_spinup_vs_fixed.png](plot_B2c_pct_diff_spinup_vs_fixed.png) (% of the fixed-geometry total).

| Region | Fixed (Gt) | Spinup (Gt) | Diff (Gt) | Diff (%) | Area change 1975→2025 |
|---|---:|---:|---:|---:|---:|
| RGI01 Alaska | −2251.2 | −2038.7 | **+212.5** | +9.4% | −2.2% |
| RGI17 S. Andes | −1095.4 | −1203.4 | **−107.9** | −9.9% | −11.8% |
| RGI19 Antarctic/Subantarctic | −977.9 | −1086.3 | **−108.4** | −11.1% | −0.9% |
| RGI16 Low Latitudes | +2.6 | −12.7 | −15.3 | −592%* | +5.3% |
| RGI18 New Zealand | −26.5 | −22.4 | +4.0 | +15.2% | −9.3% |
| all other regions | | | ≤ ±5.8 Gt | ≤ ±10% | −0.8% to −7.5% |

\* RGI16's fixed-geometry baseline is close to zero (+2.6 Gt), so a modest absolute swing produces a huge,
not-very-meaningful percentage — flagged, not a typo. See caveat below.

**RGI01, RGI17, RGI19 — the calving story.** These three account for essentially all of the large regional
differences, and all three are dominated by tidewater/calving glaciers (Alaska, Patagonia, peripheral
Antarctica). Fixed-geometry MB is a pure climate-driven elevation-band calculation with no calving or dynamic
retreat; spinup-geometry runs the actual flowline model, including calving flux and the geometry/elevation
feedback that comes with real retreat and thinning. RGI01's area barely changes (−2.2%) yet its mass-change
gap is the largest in absolute terms (+212 Gt, spinup *less* negative) — consistent with calving/dynamics
redistributing mass loss rather than simple area shrinkage explaining it. RGI17 and RGI19 show the opposite
sign (spinup *more* negative) combined with much larger area retreat (−11.8%, −0.9%), suggesting a real
geometry-drawdown feedback dominates there instead. This is the expected, useful signal that motivates having
a spinup-geometry contribution at all — worth highlighting in the methods writeup as the headline finding.

[plot_B3_annual_diff_spinup_vs_fixed.png](plot_B3_annual_diff_spinup_vs_fixed.png) shows *when* this happens:
RGI17 (olive) diverges gradually and smoothly from the mid-1990s onward — a genuine secular trend, consistent
with progressive dynamic retreat. RGI16 (grey) is different: it shows large, sign-alternating **transient
spikes in 1975–2000** (up to ±0.8 m w.e./yr) rather than a smooth trend, then only develops a real trend after
~2015.

**Caveat on RGI16.** [plot_B4_area_evolution.png](plot_B4_area_evolution.png) shows RGI16's dynamic area
overshooting to 130–140% of its 1975 value in the first ~25 years before slowly declining — a much larger and
longer-lived transient than any other region (most regions settle within a few years). RGI16 (tropical Andes /
Africa / Indonesia) has very few, very small glaciers, so it's the region most exposed to individual-glacier
calibration noise or dynamic-spinup transients dominating the regional aggregate. The sign flip (fixed: +2.6 Gt
net gain vs spinup: −12.7 Gt net loss) is real in the data but I'd treat it as **the one result in this
comparison worth a follow-up per-glacier look** before leaning on it in the methods writeup — it isn't
implausible, but the mechanism (a 25-year area overshoot) looks more like a dynamic-spinup calibration artifact
for a handful of glaciers than a robust climate signal, unlike RGI01/17/19 where the calving/retreat story is
straightforward.

## Suggested next step

If RGI16 matters for the submission narrative, worth pulling `per_glacier_specific_mb_spinup_RGI16_v17a_TIModel.nc`
and checking which individual glaciers drive the early-period area overshoot (likely a small number, given how
few glaciers the region has) — same kind of diagnosis already done for region 6's spinup-coverage question.