Skip to contents

Changelog

Source: NEWS.md

cd 0.5.9 (2026-10-02)

  • The installed package is 4.7 MB smaller, about 38%. It no longer ships inst/extdata/context_kotl.gpkg, a 4.9 MB layer set that nothing has read since the KOTL vignette was removed in 0.1.4. The Kootenay Lake vignette’s context_kootenay_lake.gpkg covers its layers and area. example_aoi_kotl.gpkg, which the README uses, stays. (#112, #115)

cd 0.5.8 (2026-10-01)

  • sf moves from Imports to Suggests, so installing cd no longer installs sf. No cd function calls sf: cd_crop() and cd_extract() still accept an sf AOI, which terra::vect() converts, loading sf itself. The examples, README and vignettes read AOIs with sf::st_read(), so install sf to follow them. The .data and .env pronouns are now imported from rlang, and R CMD check reports no NOTEs. (#111, #114)

cd 0.5.7 (2026-10-01)

  • The package tarball now ships only the package. .Rbuildignore had held its six scaffold lines, so every build and every install from GitHub carried planning/, scripts/, data-raw/, dev/, logs/, .claude/, CLAUDE.md, CITATION.cff and .lintr: 307 files where 92 belong. R CMD check drops the “hidden files”, “portable file names” and “CITATION file in a non-standard place” NOTEs. (#100, #113)

cd 0.5.6 (2026-09-30)

  • cd_trend() returns a typed empty table when no series is long enough. A trend in which every variable and period had fewer than 3 years in its window came back as a 0 x 0 tibble, so cd_summary() on it failed with Column 'period' not found and cd_plot_timeseries(trend = ) warned about uninitialised columns. It is now a zero-row tibble with every column cd_trend() documents, and cd_summary() returns an empty table.

    Behaviour changes for existing callers:

    • slope and intercept no longer carry the names "yr" and "Intercept" that the Theil-Sen fit attached.

    (#101, #110)

cd 0.5.5 (2026-09-30)

  • cd_summary() names the trend window. A trend table holding several trend_start values, such as cd_trend(x, trend_start = c(1951, 1981)), gave two rows per variable and period that differed only by Years. It now gains a Start column, holding the start year asked of cd_trend(), after Period (or after Trend on). Both vignettes’ trend tables had 59 such row pairs. They now show Start, and each 1951/1981 pair sits on adjacent rows.

    Behaviour changes for existing callers:

    • cd_summary() gains a Start column only when the table holds more than one trend_start, counted over the whole table, with an NA counting as one. A single-window table keeps its shape. As with Trend on, per-region summaries bound together can differ in having it, and it is NA for those that lack it.

    (#106, #109)

cd 0.5.4 (2026-09-30)

  • cd_summary() keeps rows distinguishable. A long_name shared by several variables, such as one “Mean discharge” on many stations, now gets the variable appended: Mean discharge (q_site1), Mean discharge (q_site2). This is the rule cd_plot_comparison() already used, now in one internal helper that both functions call. A trend table holding both raw-value and anomaly trends, such as bind_rows(cd_trend(x), cd_trend(ano)), gains a Trend on column (Value or Anomaly) after Period. It uses the same rule as Unit, so a missing or NA trend_on reads as Anomaly.

    Behaviour changes for existing callers:

    • cd_summary() Parameter gains (variable) wherever several variables share a label. Registered ERA5 variables are unaffected, because their long_names are unique.
    • cd_summary() gains a Trend on column only when the input mixes scales. A table on one scale keeps its shape.
    • cd_plot_comparison(): where a suffixed label equals another variable’s own label (long_name = c("Q", "Q", "Q (a)")), the facets now read Q (a) (a), Q (b), Q (a) (c). Previously two of them both read Q (a).
    • A variable name built to collide (a) (a beside a) gives labels that never settle. cd_summary() and cd_plot_comparison() now abort on it and ask for the variables to be renamed. cd_plot_comparison() used to plot it.

    Follow-up filed: #106. (#98, #108)

cd 0.5.3 (2026-09-29)

  • cd_plot_timeseries(trend =) now draws only trend lines on the plotted scale. A table holding both raw-value and anomaly trends of a series, such as bind_rows(cd_trend(x), cd_trend(ano)), drew the raw line (in mm, say) over the anomaly bars. The overlay now keeps rows whose trend_on names the plotted column. It warns when rows match the series but none is on that scale. It orders rows by trend_start, so the earliest start is always the dashed line, as the vignette captions say.

    Behaviour changes for existing callers:

    • A trend on the other scale is skipped with a warning. For example, cd_plot_timeseries(ano, trend = cd_trend(ts)) now draws nothing, where it drew the raw line on the anomaly axis.
    • A trend table without trend_on, whether hand-built or saved before 0.5.2, still draws, on either scale. cd_summary() instead reads such rows as anomaly trends.
    • Trend rows whose variable or period is NA are no longer drawn as an empty line. They no longer take the dashed style either.
    • The earliest trend_start is dashed whatever the row order.

    (#103, #105)

cd 0.5.2 (2026-09-29)

  • cd_summary() no longer labels a trend of raw values with the anomaly unit from cd_variables(). A raw precipitation slope in mm read %, and a unit carried on an absolute series (m3/s, say) was dropped. cd_trend() now records what it ran on in a trend_on column ("value" or "anomaly"). On raw input it also carries anomaly_type and unit, keeping the unit only where it is also the unit of the values (absolute, pct_point_diff). cd_summary() applies that rule row by row, so its Unit column now agrees with the cd_plot_timeseries() axis label for the same series. This replaces the 0.5.0 rule that anomaly_type and unit are carried only when trending anomalies.

    Behaviour changes for existing callers:

    • Every cd_trend() result gains a trend_on column.
    • cd_summary() Unit on raw-value trends: pct_normal series (prcp, soil_moisture, swe, snowfall, snowmelt) go from % to NA, and a unit carried on an absolute series is now shown.
    • cd_trend() on raw input now errors when anomaly_type or unit varies within one series, as cd_plot_timeseries() already did.
    • A trend table without trend_on, such as one saved before this release, is read as an anomaly trend, as before.

    Follow-ups filed from review: #101, #102, #103. (#104)

cd 0.5.1 (2026-09-29)

  • cd_plot_timeseries() and cd_plot_comparison() now label a series from the long_name and unit it carries, falling back to cd_variables() by the rules of the 0.5.0 input contract, so a series from another package (streamflow, say) plots under the label cd_summary() prints for it instead of anomaly. cd_compare() carries a long_name column through when its input has one (no unit: it works on raw values). A long_name shared by several variables gets the variable name appended in the comparison plot, and each variable always gets its own facet.

    Behaviour changes for existing callers:

    • cd_plot_timeseries() on raw values no longer shows an anomaly unit that does not describe them: raw precipitation, which read “Precipitation (%)”, now reads “Precipitation”. Anomaly plots are unchanged.
    • cd_plot_timeseries() errors on duplicate years in the plotted series (they were stacked into one bar) and on conflicting metadata within it, and ignores rows whose variable or period is NA.
    • cd_compare() gains a long_name column when the input carries one.

    Follow-ups filed from review: #96, #97, #98. (#99)

cd 0.5.0 (2026-09-29)

  • The consumer chain now works on any annual series in cd’s long format, not only the ERA5-Land variables in cd_variables(). Streamflow and stream temperature from other packages are included. cd_anomaly() used to return a silent NA for any variable outside the registry. It now takes anomaly_type, unit and long_name from the input where present, falls back to cd_variables() row by row, and raises an error naming any variable whose type it cannot resolve. unit means the anomaly’s unit, as it always has in cd_variables(). cd_trend() carries the metadata through (anomaly_type and unit only when trending anomalies) and cd_summary() uses it. The contract is documented under “Input contract” in ?cd_anomaly.

    Behaviour changes for existing callers:

    ERA5 results are unchanged: the regional trend tables in both vignettes reproduce exactly. (#94)

cd 0.4.3 (2026-09-07)

  • Producer-side only — no change to any exported function. Fixes what four independent review rounds found in the 0.4.1 and 0.4.2 changes, which had shipped on self-review alone.

    The offline test suites added in those releases were decorative for the distinction they existed to make: months_available() stubbed to return the index of the last month instead of the count of distinct months passed 7/7, because every fixture’s year began on 1 January and was contiguous, and for that shape the two are the same number. A guard that over-counts writes a partial year. The R suite never asserted the connection-failure diagnosis at all, so a message colliding with the 401 text would have sent an operator to rotate a healthy secret on a DNS timeout. Both suites now carry fixtures that reach those cases, and every previously surviving mutant is red.

    In the probe itself: edh_err was not reset between retry attempts, so one attempt’s connection error printed beneath another’s rotate the secret diagnosis; the error-text fetch never checked its own status, so a recovered server’s 200 body could be quoted as EDH’s explanation; and neither curl handle bounded the transfer, so a server that completed the handshake and stalled hung until the job was cancelled — which if: failure() skips, silencing the alarm in exactly the hang the probe exists to catch. The workflow now also fires on cancelled().

    Three pre-existing defects the review surfaced are filed rather than fixed here: #88, #89, #90. (#91)

cd 0.4.2 (2026-09-07)

  • Producer-side only — no change to any exported function. The EDH credential probe in scripts/pipeline_update_edh.R STEP 0 collapsed every HTTP status >= 400 into EDH rejected the token, so a 403 sent the reader to rotate a secret that was in perfect health — which is exactly what happened on 2026-09-07. A 401, 403 and 404 now give three distinct diagnoses, and EDH’s own error text is quoted on the failure path. The probe also gains a bounded retry over the statuses that can clear on their own (connection failure, 408, 429, 5xx, and 403 — observed transient), while 401 still fails immediately, since retrying a rejected credential only delays the report. Both decisions are pure functions in a new scripts/_lib.R, asserted offline by scripts/test_lib.R. (#87)

cd 0.4.1 (2026-09-07)

  • Producer-side only — no change to any exported function. The three EDH backfillers now establish whether a year is complete before fetching it. .compute(), where the lazy xarray graph actually pulls from the Zarr store, previously ran ahead of the 12-month guard, so every monthly climate-update run downloaded a full partial year across 15 variables and discarded it — against a metered EDH free tier, roughly eleven times a year. A new months_available() helper answers the same question from the store’s time coordinate, which Zarr materialises on open and so costs no data transfer. The check is per store, since the hourly and daily stores advance independently (measured two months apart). Also closes an unguarded write: the four annual-derived snow variables had no completeness check at all, so a partial year produced rasters that the per-output idempotency check then preserved permanently. (#85)

cd 0.4.0 (2026-06-25)

  • On-disk caching wired into the consumer read path, so repeated extractions, report renders, and vignette rebuilds pull each COG from S3 once and read locally thereafter — turning the dominant recurring S3 egress driver into a one-time cost. New exported cd_cache_fetch() downloads a remote http(s) COG to the cd cache (keyed by URL hash, with a sidecar .meta recording the S3 ETag and size), validates freshness with a cheap HTTP HEAD (ETag, falling back to Content-Length), and serves the local copy on a hit. Downloads are size-validated and atomically renamed so a truncated file is never served; a failed HEAD with a cached copy present serves the cache, and options(cd.cache_revalidate = FALSE) skips revalidation entirely for offline work. cd_crop() and cd_extract() gain cache = TRUE (default), threading remote reads through the cache while local paths pass through unchanged. Live S3 confirmation: a repeat read drops from a full-COG download (megabytes) to a ~1 KB HEAD (or zero network with revalidation off). Adds curl to Imports. See the new README “Caching” section, which also documents the GDAL /vsicurl/ env-var stopgap. (#76)

cd 0.3.2 (2026-06-06)

  • Both regional vignettes (kootenay-lake, peace-fwcp) rewritten for new readers: plainer-language opener for the snowpack section (“In BC, most of the year’s runoff starts as winter snow…” instead of the “hinge of BC hydrology” metaphor), Trends / Recent-Decade / bias-notes preambles compressed and de-jargoned, Annual snowpack signals intro reduced to a 3-bullet plain-language list, salmonid Interpretation closer tightened to one paragraph with three bold knock-on effects. Figure trim: cut plot-tmean (covered by facet-tmean), plot-dtr (asymmetry numbers already in prose), and snow-rate-peak (not load-bearing); fold plot-tmax + plot-tmin into one 2-panel faceted plot-tmaxmin, and snow-swe-max + snow-doy-50 + snow-fraction into one 3-panel faceted snow-annual (free y-scales). Net per vignette: 3 fewer standalone figures, same coverage. Bibliography: dropped kouki_etal2023 and yue_wang2002 (no longer cited); union now 15/15. (#75)

cd 0.3.1 (2026-06-06)

  • Kootenay Lake vignette: drops cross-references to the Peace vignette so the regional narrative stands on its own — recent-decade table preamble, snowpack section header, ASWS QA cross-check, snowpack-meaning section, cordillera-wide background, and the precipitation / snow Interpretation bullets all rewritten to focus on Kootenay-only findings (what is happening and what it means). Institutional FWCP reference at the watershed-group section and maintainer-facing peace-fwcp.Rmd build comment retained. (#74)

cd 0.3.0 (2026-05-10)

  • cd_compare() API: adds defaults window_a = 2015:2025, window_b = 1951:1980 (the WMO-style standard normal vs recent decade — the framing both regional vignettes had settled on) and a new test = "t" argument that adds a p_value column to the output. test = "t" runs Welch’s two-sample t-test on the annual values within each window; test = "wilcox" runs Mann-Whitney U; test = NULL skips and drops the column. Rows where either window has < 8 non-NA years get p_value = NA and a single batched warning naming affected variable/period rows. The window-vs-window p-value answers a different question than cd_trend()’s Mann-Kendall test — “do the two windows differ” vs “is there a monotonic trend” — and step changes / U-shapes can produce significant Δ p with non-significant trend p (and vice versa). Both regional vignettes now report both p-values side-by-side (Δ p (windows) + Trend p (75-yr)), and the visible compare-recipe chunk drops to cd::cd_compare(ts) letting defaults carry the call. Vignette structure tweak: the Recent Decade vs Pre-Warming Reference section (renamed from “Recent vs Pre-warming”) moves up directly after Trends so the headline cumulative-impact number lands before DTR / Snowpack / Spatial Pattern drill-down. Minor bump (output schema gains p_value column by default; new test argument). (#73)

cd 0.2.8 (2026-05-07)

  • Vignette polish bundle covering both regional vignettes (kootenay-lake, peace-fwcp): (1) citation prose tightening — surfaces source-specific findings (Karl et al. 1993’s 3:1 min:max ratio, Mantua et al. 2010’s Washington-State scope + Eaton & Scheller 1996’s 57-species comparative finding, Kang et al. 2016’s ~10-day Fraser advance) instead of name-drop end-tags; drops arguez_vose2011 as decorative-only attribution; references.bib regen 18 → 17 entries; (2) km² superscript fix + highway symbology lifted from gq/inst/registry/reg_qgis_restoration.json (grey casing under warm-yellow fill); (3) kootenay town filter to Rossland, Castlegar, Nelson, Cranbrook, Kaslo, Nakusp; (4) chunk restructure (Option B) — hide ~13 visualization/styling chunks per vignette via echo = FALSE, lift cd_*() recipes (cd_extract, cd_baseline + cd_anomaly + cd_trend + cd_summary, cd_compare, per-ecoregion + per-WSG cd loops) out of # Equivalent to: comment blocks into shown eval = FALSE companion chunks. Reader’s flow becomes: cd recipe (visible, copy-pasteable) → output (table/figure, source hidden) → narrative; (5) cd:: namespace prefix on every cd function call in both vignettes (visible + hidden) for explicit package attribution. Independent Explore subagent re-verified each tightened citation against source archives + PDFs; both vignettes render cleanly post-restructure. (#72)

cd 0.2.7 (2026-05-06)

  • Tooling. Adds data-raw/regenerate_bib.R — a one-line helper that regenerates vignettes/references.bib from the union of pandoc citation markers across both regional vignettes by pulling source records from Zotero via Better BibTeX (rbbt::bbt_bib). Run after editing cites: Rscript data-raw/regenerate_bib.R. Each vignette gets a top-of-file HTML comment pointing to the helper; CLAUDE.md Vignettes section documents the prerequisite (BBT 9.x for Zotero 8/9). The conservative option vs embedding the rbbt call in YAML at render time, which would break pkgdown CI (no Zotero on the runner). (#70)

cd 0.2.6 (2026-05-05)

  • Wires up vignettes/peace-fwcp.Rmd interpretation paragraphs with citations from the climate-departure 3-split lit reviews — companion to v0.2.5’s kootenay-lake wire-up. Same 7-insertion pattern (8 unique new keys + 1 reuse), sparingly applied per the plain-language vignette philosophy. Two AOI-specific tweaks vs the kootenay PR: the [@pepin_etal2015Elevationdependentwarming; @rangwala_miller2012Climatechange] cite lands at the Interpretation paragraph rather than the Spatial Pattern section (Peace’s dominant warming gradient is east-west / windward-of-Rockies, not pure elevation), and the [@ficklin_novick2017Historicprojected] VPD-drying cite is stronger here than in Kootenay because Peace’s precipitation is up 3-4% in 2 ecoregions yet soil moisture is flat — pure VPD-driven evaporative-demand effect. Audit log at planning/archive/2026-05-issue-67-peace-vignette-wireup/citation_audit.md records the per-cite source quote, paraphrase, and visible-in-vignette warrant. An independent Explore subagent verified each row against source quote archives + PDFs and signed off all 7 cites with zero edits or removals required (stronger pass than the v0.2.5 review, which caught one minor scope nit). vignettes/references.bib regen produced no diff — both vignettes draw from the same 18-key union. Snow-section citations from #54 untouched. (#68)

cd 0.2.5 (2026-05-05)

  • Wires up vignettes/kootenay-lake.Rmd interpretation paragraphs with citations from the climate-departure 3-split lit reviews (#53/#58/#61/#63). 8 new BBT-keyed citations + 1 reuse, sparingly applied per the plain-language vignette philosophy. Insertions span all four non-snow interp themes: Trends/baseline-window (Hansen et al. 2012, Arguez & Vose 2011), DTR asymmetry (Karl et al. 1993), elevation-dependent warming (Pepin et al. 2015; Rangwala & Miller 2012), continental-scale VPD drying (Ficklin & Novick 2017), and the climate→stream-temperature→fish thermal-habitat bridge (Mantua et al. 2010; Eaton & Scheller 1996). Audit log at planning/archive/2026-05-issue-65-kootenay-vignette-wireup/citation_audit.md records, per cite, the vignette excerpt, source quote, rag store + topic where retrieved, and paraphrase as written — the “what is where by who said what where” trail. An independent Explore subagent verified each row against source quote archives and PDFs before merge; one minor scope concern surfaced (Ficklin & Novick paraphrase narrowed continental-US scope to “western US”) and was fixed inline. vignettes/references.bib regenerated via rbbt::bbt_update_bib() — now 18 entries (was 11). Snow-section citations from #54 untouched. (#66)

cd 0.2.4 (2026-05-05)

  • Interpretation framing methodology literature review — wraps the climate-departure 3-split (snow done in v0.1.7, temperature in v0.2.2, precip+drying in v0.2.3, interpretation framing here). Adds scripts/rag_interpretation_framing_build.R and scripts/rag_interpretation_framing_query.R, builds a local ragnar DuckDB from 4 peer-reviewed papers (now in the NewGraphEnvironment/climate Zotero collection), and ships the methodology synthesis + 11-row “cite this for that” citation map at planning/archive/2026-05-issue-63-interpretation-framing-lit-review/findings.md. Coverage spans WMO climate normal definition + alternatives (Arguez & Vose 2011), estimating normals when trends exist (Livezey et al. 2007), time of emergence / signal-to-noise framing (Hawkins & Sutton 2012), and cumulative-impact / “loaded dice” framing (Hansen et al. 2012). Headline finding: Hansen et al. (2012) explicitly use the same 1951–1980 base period that cd uses, providing the strongest direct precedent for cd’s baseline window choice across all three lit reviews; combined with Arguez & Vose (2011)’s framework, cd’s choice is a defensible “alternative climate normal” with cumulative-impact framing for FWCP fish-passage planner reporting. After this release, the citation backbone for the climate-departure 3-split is complete: 32 peer-reviewed papers across 4 ragnar stores, with cite-this-for-that maps in 4 archived findings.md files ready for the downstream vignette wire-up branch. (#64)

cd 0.2.3 (2026-05-05)

  • Precipitation + drying methodology literature review (2/3 of the climate-departure 3-split; temperature done in v0.2.2, interpretation framing forthcoming). Adds scripts/rag_precip_drying_methodology_build.R and scripts/rag_precip_drying_methodology_query.R, builds a local ragnar DuckDB from 7 peer-reviewed papers (now in the NewGraphEnvironment/climate Zotero collection), and ships the methodology synthesis + 15-row “cite this for that” citation map at planning/archive/2026-05-issue-61-precip-drying-lit-review/findings.md. Coverage spans anthropogenic precip-extremes attribution (Min et al. 2011), Canadian / BC adjusted precip dataset methodology (Mekis & Vincent 2011), VPD continental-scale drying (Ficklin & Novick 2017), VPD ecosystem responses (Grossiord et al. 2020), drought attribution (Williams et al. 2020), drought framework (Trenberth et al. 2014), and 20th-century hydroclimate signal (Marvel et al. 2019). Headline finding: the v0.1.1 vignette claim that “soils dry from both ↓P and ↑ET” is now backed by Ficklin & Novick (2017) rising-VPD continental drying + Williams (2020) anthropogenic-warming attribution of NA megadrought + Trenberth (2014) Penman-Monteith drought framework. Macos Zotero auto-restart pattern (osascript quit + open -a + 30 s wait) automated the BBT key generation step end-to-end (added to soul#43 for /lit-search + /zotero-api). (#62)

cd 0.2.2 (2026-05-05)

  • Temperature-departure methodology literature review (1/3 of the climate-departure 3-split; precip+drying and interpretation framing follow as separate issues). Adds scripts/rag_temp_methodology_build.R and scripts/rag_temp_methodology_query.R, builds a local ragnar DuckDB from 10 peer-reviewed papers (now in the NewGraphEnvironment/climate Zotero collection), and ships the methodology synthesis + 18-row “cite this for that” citation map at planning/archive/2026-05-issue-58-temperature-lit-review/findings.md. Coverage spans DTR asymmetry methodology (Karl et al. 1993, Easterling et al. 1997, Vose et al. 2005), Canadian/BC temperature trends (Vincent et al. 2018), elevation-dependent warming (Pepin et al. 2015, Rangwala & Miller 2012), BC downscaling (Wang et al. 2012), the climate-fish thermal-stress bridge (Mantua et al. 2010), and salmonid thermal envelope (Eaton & Scheller 1996, Richter & Kolmes 2005). Headline finding: cd_trend()’s raw Mann-Kendall + Theil-Sen aligns with Vincent et al. 2018’s Canadian-temperature-trend methodology (Sen slope + Kendall’s τ); raw MK is the right call for our 76-year strong-trend series per Yue & Wang 2002 (cross-rag from snow methodology store). Also renames existing rag_build_*.R / rag_query_*.R scripts to rag_*_build.R / rag_*_query.R (noun_verb per cd convention). (#60)

cd 0.2.1 (2026-05-04)

  • Adds a new kootenay-lake vignette covering the southern Kootenays — KOTL + LARL + DUNC + SLOC, ~24,200 km². Reads stand-alone alongside peace-fwcp and uses the v0.2.0 snow variables to tell a snow-pack story for a region with a sharp east-west precipitation gradient (Selkirks vs Purcells). Headline regional findings (2015-2025 vs 1951-1980): annual SWE -23%, annual snowfall -15%, freshet shift -12.6 days earlier, annual precipitation -7% (p = 0.02). Total annual snowfall is dropping, not just melt timing shifting — consistent with Knowles et al. 2006’s threshold finding that significant snowfall-fraction declines occur where winter wet-day Tmin > -5 °C. Adds a per-watershed-group facet view and data-raw/qa_snow_validation_kootenay_lake.R ASWS cross-check (5 sites, 74 paired station-years, pooled r = 0.90, mean bias -54%, bias-stable). New bundled assets: inst/extdata/example_aoi_kootenay_lake.gpkg, inst/extdata/context_kootenay_lake.gpkg, inst/vignette-data/kootenay_lake.rds. (#57)

cd 0.2.0 (2026-05-04)

  • Adds 8 snow-related variables for hydrology departure analysis. Four monthly natives (swe, snowfall, snowmelt, snow_cover) ship as 12-band/year COGs alongside the existing 7 climate variables; four annual derived scalars (swe_max, snowfall_fraction, snowmelt_doy_50, snowmelt_rate_peak) ship as 1-band/year COGs. New pct_point_diff anomaly type in cd_anomaly() for snow_cover and snowfall_fraction (already-percentage variables). New “Snowpack” section in the FWCP Peace vignette with seasonal-curve table, four annual time-series plots, per-ecoregion view, and citation-grounded interpretation; ASWS QA cross-check at 4 BC stations confirms bias is stable over time even where absolute values are biased. Producer pipeline extends with scripts/backfill_edh_snow.py (first place ECMWF’s accumulation-reset trick lands in code). Also bug-fix in cd_stac_item() filename parsing (substring match was mis-routing variables whose names contain other variable names — swe_max was being filed under swe). Headline finding: regional summer SWE has collapsed by 75% over the record, freshet-timing shift is uniform across all 5 ecoregions at ~1 day/decade earlier melt. (#55)

cd 0.1.7 (2026-05-04)

  • Snow-methodology literature review for the upcoming “Snowpack” vignette section. Adds scripts/rag_build_snow_methodology.R and scripts/rag_query_snow_methodology.R, builds a local ragnar DuckDB from 11 peer-reviewed papers (now in the NewGraphEnvironment/hydrology Zotero collection), and ships the methodology synthesis + 15-row “cite this for that” citation map at planning/archive/2026-05-issue-53-snow-lit-review/findings.md. Headline finding: cd_trend()’s raw Mann-Kendall + Theil-Sen (no prewhitening) is methodologically correct for our 76-year series with strong trends per Yue & Wang 2002 — prewhitening would underestimate slope when a real trend exists. (#54)

cd 0.1.6 (2026-05-03)

  • Producer-side refactor. Extracted shared safeguards (single-instance pgrep guard, exponential-backoff retry, atomic GeoTIFF write, timestamped logging, EDH token loader) from scripts/backfill_edh_all.py into a new scripts/_lib.py, and ported them to the sibling scripts/backfill_edh_tmax_tmin.py (which previously had none). Adds a backup_before_delete() helper codifying the on-disk pattern at data/backfill/monthly/_cds_backup/. No consumer-side changes; sets up the planned snow-variables backfill (#48) to inherit the safeguards via a single import. (#52)

cd 0.1.5 (2026-05-02)

  • Adds a “Watershed Groups Across Ecoregions” section to the peace-fwcp vignette: a map of the 16 canonical FWCP Peace watershed groups labelled with codes on top of ecoregion fills, plus a table showing the percent of each watershed group’s area falling in each of the five ecoregions. Lets readers map per-ecoregion climate departure findings (precipitation up only in BMP and NRM) onto the watershed-group reporting unit. Canonical 16-WSG list (CARP, CRKD, FINA, FINL, FIRE, FOXR, INGR, LOMI, MESI, NATR, OSPK, PARA, PARS, PCEA, TOOD, UOMI) hardcoded in data-raw/example_context_fwcp_peace.R for reuse — UPCE and MURR dropped because they sit mostly outside the FWCP boundary. Recent vs Pre-warming consolidated into one table. All six vignette tables now render with single, clean bookdown captions (label = NA + caption = "..." pattern). (#47)

cd 0.1.4 (2026-05-01)

KOTL vignette removed. peace-fwcp is now the canonical worked example — it covers everything KOTL did at higher fidelity (regional + per-ecoregion + day-night asymmetry + plain-language explainers + interpretation), and dropping the second live-S3 vignette saves ~60 s per pkgdown render and removes the last /vsicurl/ flake surface from the doc build. KOTL polygon assets stay in inst/extdata/ because the README quick-start still uses them. Future direction (single-vignette snowpack story or split vignettes by AOI scale) deferred until snow-pack variables land — see #49. (#50)

cd 0.1.3 (2026-05-01)

CI fragility patch. The peace-fwcp vignette previously re-fetched ~144 /vsicurl/ COG range requests on every pkgdown render; one transient flake failed the whole build. Heavy data is now pre-computed by data-raw/peace_fwcp_vignette_data.R and shipped under inst/vignette-data/ (160 KB rds + 6 KB tif). Vignette loads via system.file() and renders in ~10 s instead of ~6 min. Live cd_catalog() read kept as the consumer entry-point demonstration. (#45)

cd 0.1.2 (2026-04-30)

Vignette and docs patch. New peace-fwcp vignette runs the consumer pipeline on a regional administrative AOI (FWCP Peace Region, ~73,000 km², ~11x KOTL) — catalog → extract → trends → recent vs pre-warming → spatial map → per-ecoregion breakdown across the five BC ecoregions intersecting the region, with faceted time-series carrying both 75-yr and 45-yr Theil-Sen trend lines, a wide roll-up table, day-night asymmetry section (textbook signal does show up here, unlike KOTL), and three-finding interpretation. Plain-language explainers for trend windows, WMO climate normal, and “warming has accelerated/slowed” framing. README gains a Data section with the catalog URL and the /vsicurl/ direct-read pattern so the COGs are usable outside R (QGIS, gdalcubes, rasterio). Issue #43 filed for cd_compare() to gain a proper window-vs-window p-value. (#42)

cd 0.1.1 (2026-04-15)

Vignette and docs patch. The climate-departure vignette gained a “Daytime Highs and Overnight Lows” section using the tmax/tmin variables now on STAC, with honest framing for the example watershed (the textbook day-night asymmetry doesn’t show at Kootenay Lake — the dominant signal is summer daytime maximum, the temperature envelope for salmonid thermal stress in tributaries). Existing maps now clip context layers and mask departure rasters to the watershed group polygon for tighter framing. Interpretation section corrected: precipitation has declined ~10% (statistically significant) and soils are drying due to both falling precipitation and rising evapotranspiration. README quick-start fixed (was referencing files that don’t exist) and now links to the live pkgdown vignette. (#39)

cd 0.1.0 (2026-04-14)

CRAN release: 2020-10-22

First minor release. Producer pipeline migrated from Copernicus CDS to DestinE Earth Data Hub (Zarr). Same ERA5-Land data at the same 9 km native grid, no rate limiting, ~5x faster fetches. All 7 cd variables (tmax, tmin, tmean, prcp, vpd, rh, soil_moisture) regenerated on a single internally-consistent EPSG:4326 BC grid. Monthly GitHub Action rewired to use EDH. Consumer API unchanged — cd_catalog() and friends work exactly as before against the refreshed STAC catalog on s3://stac-era5-land. See pkgdown reference for the current function list. (#36)

cd 0.0.0.9000

Initial development version. Consumer and producer pipelines for ERA5-Land climate departure analysis.