Snowpack-departure methodology lit review: rag-build + 11 papers + citation map

[NewGraphEnvironment]

Summary

Closes the #53 lit-review companion to #48. Mines 11 peer-reviewed papers (10 newly added to the NewGraphEnvironment/hydrology Zotero collection plus 1 already-in-library) for snow-departure methodology citations, builds a local ragnar DuckDB store, runs 23 queries across 8 topic areas, and produces a citation-keyed findings.md ready for #48 Phase 5 to consume.

Decoupled boundary: this PR produces methodology notes + a rag store. The actual [@key] citation insertions in vignettes/peace-fwcp.Rmd happen in #48 Phase 5, sourced from this branch's findings.md.

What's in this PR

• scripts/rag_build_snow_methodology.R — clones rag_build_departure_framing.R pattern; reads PDFs from data/rag/snow_methodology_pdfs/ (gitignored, populated via Zotero Web API); builds DuckDB with Ollama nomic-embed-text embeddings.
• scripts/rag_query_snow_methodology.R — runs 23 focused queries (per Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 metric + cross-cutting) and dumps top-5 chunks per query.
• planning/active/snow_methodology_quotes.md — raw retrieval (727 lines, archived as part of /planning-archive on merge).
• planning/active/findings.md — synthesis: methodology quotes by Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 metric, cross-cutting (baseline window, MK + autocorrelation, ERA5-Land bias), deviations from consensus, and a 15-row "cite this for that" citation map.

Headline findings

1. cd_trend()'s raw Mann-Kendall + Theil-Sen (no prewhitening) is methodologically correct per Yue & Wang (2002). Their Monte Carlo result: prewhitening fails when a real trend exists in the series — it dramatically underestimates slope. For our 76-year series with strong climate trends, raw MK is the right call. We were lucky.

2. snowmelt_rate_peak (annual max of 7-day rolling daily snowmelt) is our invention. The literature canon for freshet flashiness is streamflow-based (Stewart 2005 center timing, Kang 2016 Fraser timing). Our metric is upstream of that — direct on the snowmelt flux. Diagnostic of snowpack-side intensity but no close methodological precedent. Document as such in the vignette interp.

3. ERA5-Land overestimates SWE in mountains by 150–200% NH-wide (Kouki et al. 2023). Bias is approximately stable over time — trends remain valid even though absolute values are biased. The ASWS QA cross-check planned for Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 Phase 3 will document our specific BC bias structure.

4. swe_max (true annual max of daily SWE) is a slight deviation from the April-1-SWE canon (Pederson 2011, Mote 2005/2018). Equivalent in effect for BC pixels (peak is at or near April 1) but date-insensitive. Worth a brief methodology note in the vignette.

5. BC-specific anchors are strong. Najafi et al. (2017) attributes BC spring SWE decline to anthropogenic forcing using VIC + CMIP5 fingerprinting; Kang et al. (2016) documents the 10-day advance of the Fraser freshet during the salmon migration window 1949–2006. Both directly map to our fish-passage reporting context.

Test plan

• All 11 papers added to top-level NewGraphEnvironment/hydrology Zotero collection (key X29BX4U8) with PDFs attached
• DuckDB store builds clean: 11 sources, 1006 chunks
• 23 queries return sensible top-5 chunks (sample inspected manually)
• Synthesis in findings.md covers all four Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 metrics + 3 cross-cutting topics + deviations + citation map
• Post-merge: /planning-archive. Archived findings.md becomes the long-lived methodology reference.
• Downstream — Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 Phase 5 inserts [@key] markers from the citation map into vignettes/peace-fwcp.Rmd

Out of scope

• Vignette citation insertion — happens in Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 Phase 5.
• Setting up bibliography: YAML field in peace-fwcp.Rmd — also Add snow-related variables (SWE, snowfall fraction, melt timing) for hydrology departure #48 Phase 5.
• Original methodology research — picked from established literature.
• Refactoring cd_anomaly / cd_trend — Yue & Wang 2002 supports current behavior; no refactor needed. Could file an autocorrelation-aware variant against cd_compare(): add p-value for window-vs-window comparison #43 if motivated.
• Glacier dynamics, streamflow modelling.

Notes

The findings.md "cite this for that" table is the primary deliverable. It maps 15 vignette claim types to specific citation keys. #48 Phase 5 should be a copy-paste exercise rather than a literature search.

BBT will generate citation keys when the user restarts Zotero. The keys below match my proposed labels (firstauthor[_etal]year) and should align with BBT's auto-generation, but if they diverge the vignette's [@key] references will need a sweep.

Fixes #53
Relates to NewGraphEnvironment/sred-2025-2026#23

🤖 Generated with Claude Code