Skip to contents

Changelog

Source: NEWS.md

wet 0.4.7

  • Scripts, data scripts and live tests now read a named HYDAT release, from one per-user cache (tools::R_user_dir("wet", "cache")/hydat/<release>/), instead of whichever release a machine last downloaded (#46). The internal wet_hydat() downloads ECCC’s current release into the cache, and checks every file is whole and holds the release it is named for. WET_HYDAT_RELEASE names the release, by default the shipped one; WET_HYDAT is retired and refused. Station-flow outputs record their release. wet_station_*() still default to tidyhydat’s own directory for interactive use.

wet 0.4.6

  • The published discharge at s3://fresh-bc/discharge/fwa/v1/ now carries the snow-informed monthly balance (build acb820f, #77). Annual flow and each segment’s pick are unchanged. The monthly-flow article’s data were rebuilt from it, and the article now describes the snow terms. The balance reads high only in April and May, by a median of 7 % to 11 %, and is within 5 % in the other months. The freshet is still flattened. data-raw/monthly_vignette_data.R reads the held-out months each fit stores, instead of recomputing them (#68).

wet 0.4.5

  • The water balance’s monthly shares now also use snow: ERA5-Land’s 1981-2010 day by which half the snowpack has melted and its peak snow water equivalent (capped at 3,000 mm for glacier points), from cd through the new wet_snow_normals(), averaged upstream like the other predictors. A rule fixed before scoring kept them: the held-out median share NSE rose from 0.856 to 0.874 on the shipped fit. Winter and early-spring flow is much less overpredicted, the spring peak is a little less flattened, and zone 27 is worse. The annual balance is unchanged (#16). Each fit now stores its held-out monthly balance (#68). The published discharge still carries the earlier monthly values until #77.

wet 0.4.4

  • wet_segment_monthly() reads flow per FWA segment and source in the shape of wet_station_monthly() (linear_feature_id, month, q_m3s, source), so a gauge comparison is a join on month (#59). It reads the published product, from Postgres after wet_db_load() or from the parquet parts through duckdb, or with from = "fwapg" fwapg’s routed table, which scripts/pcic_routed_compare.R scores before a publish. A segment with no value comes back as 13 NA rows. wet_flow_window() turns monthly means into mean flow over month-day windows, labelled basis = "monthly"; the monthly article’s read and its Chinook windows now go through both, with its numbers unchanged.

wet 0.4.3

  • The segment-discharge article is simpler (#71). It now says what the published product is, which value to use, how to read it and how good it is, in that order. pick has its own early section with its rule and the province counts: routed PCIC on 3,152,838 segments, the balance on 1,380,005, neither on 374,598. One province map of the source pick takes on main rivers replaces the Salmon River maps and the scatter. The map’s picks are read from the live published parts by data-raw/segment_vignette_map.R, under the guards the values script uses. The rename of pick to is_default at v2 is recorded in research/discharge_publish.md.

wet 0.4.2

  • A new article covers months 1–12 of the published product (#65): how to read one stream’s twelve months, both sources’ hydrographs at three Bulkley gauges with the Chinook windows, the flow in each window beside the gauge’s daily record, and month-by-month scores at the calibration gauges. Held out at all 315, the water balance reads 8–16 % high from January to May and within ±5 % from June to December, and 67–96 % low in January–March at the 25 northern gauges routed flow does not reach (#16). Routed PCIC, at 279 gauges, reads 20 % low in March–April and 19 % high in July–August, and its 1951–2012 period explains only part of that (#67). The values are read from the published parts and checked against their sources before they are bundled (data-raw/monthly_vignette_data.R), under guards now shared with the segment article (scripts/vignette_data_lib.R).

wet 0.4.1

  • The segment-discharge article now introduces the published product (#64): where it is, what a row holds, how to read it with wet_catalog(), wet_db_load() or duckdb, what pick selects, how each source was made, and how both score at the 315 calibration gauges, with the Salmon River as the one worked group. The values it shows are read from the published parts and checked against their sources before they are bundled. scripts/discharge_publish.R now reads each part it writes back and compares the flows, and the segment rows’ runoff, coverage and pick, value by value with what they were built from: on build 7c62b9e, 555,857,796 values with none differing, and all 67 parts identical to the live upload. The fwapg parity and sampling results moved to research/fwapg_mad_method.md.

wet 0.4.0

  • Per-segment and per-watershed discharge have a published layout at s3://fresh-bc/discharge/fwa/v1/ (#7): Hive-partitioned parquet keyed scenario / forcing / source, one part per FWA top-level drainage with one row group per watershed group, and a manifest.json describing every part, because the bucket allows reads but no listing. wet_catalog() reads the manifest. wet_db_load() loads one scenario of a dataset into Postgres (discharge.fwa_v1_segment_monthly or discharge.fwa_v1_watershed_monthly), checking each part’s size and md5 first, so fresh can join month = 0 AND pick. scripts/discharge_publish.R conforms each part to its crate schema (crate 0.7.0), builds the tree and writes a tracked report. Uploading is a separate, explicit step and has not been run yet. One dataset ID, <variable>/<network>/<version>/<unit>_<timebase>, names the S3 prefix, the crate schema and the table (research/discharge_publish.md).

wet 0.3.2

  • FWA segments now carry both estimates where each exists, PCIC’s routed flow (NewGraphEnvironment/fwapg#6) and the open water balance, labelled by source and period, with a logical pick marking the default (#57). scripts/segment_output.R writes them per basin as linear_feature_id, month, runoff_mm, q_m3s, coverage, source, period, pick. The default is routed unless its mean annual flow is under 0.2 of the balance’s, and only a segment on its watershed’s own line can fall to the balance, because the balance reaches a segment through fwa_streams_watersheds_lut and off that line it is another line’s flow. 224,199 of 3,374,133 segments with both sources fall to the balance, 26.7 % of them in the Nechako Reservoir groups, where PCIC’s outlets are almost all unplaced (NewGraphEnvironment/fwapg#24). At the 290 calibration gauges routed flow covers, the pick’s mean absolute error is 22.9 % against routed’s 24.1 % and the balance’s 27.6 %. The 369,217 segments with no lut row get neither. The balance stays the independent estimate both are scored against.

wet 0.3.1

  • The segment vignette compares the open water balance with PCIC’s routed flow (NewGraphEnvironment/fwapg#6) instead of fwapg’s older annual table (#58): the table of strengths and limits, the scatter, the province coverage outline, the Salmon River pair and the Bulkley gauges. At the 290 of 315 calibration gauges routed flow covers, its mean absolute error is 24.1 % against the held-out balance’s 27.6 % (median 13.0 against 18.0); by mean error the balance is closer in the Peace and the Fraser. Where the older annual table also had values, the two PCIC products tie once 7 far-low gauges are set aside, so what routed flow adds is coverage: the coast and the largest rivers. The scores come from scripts/pcic_routed_compare.R, whose report records the fwapg commit and the table’s fingerprint, and the vignette’s data script refuses any other build.

wet 0.3.0

  • wet_temp_fill() fills gaps in daily water temperature at stations, loggers included (#40). Air2stream is run open-loop on daily air temperature from cd::cd_extract_daily(), and a Kalman smoother carries each station’s departure from it across a gap, from both ends, adding the error its peers in the same pool saw that day. Held out at 999 station-years, it beats air2stream alone on every gap: 0.48 against 1.31 °C on 7 days, 0.76 against 1.43 °C on 30 days, 0.84 against 1.14 °C on a seasonal logger’s spring and autumn. Over a whole missing season its GSDD is no better than air2stream alone (130 against 129 °C-days), the one criterion of a rule fixed beforehand that it fails. wet_temp_gsdd() gives growing season degree days per station-year in wet_window_stats()’s long format, with the share of each season that was filled. At the 17 Skeena stations, every site’s GSDD lies inside the network model’s 95 % interval (hill_etal2025Spatialstream).

wet 0.2.5

  • The dry-interior gauges are scored against licensed water use above them, under a rule fixed before any licence was matched to a gauge (#53). BC water rights licences, placed upstream of each calibration gauge on the FWA network, account for 0–2 mm/yr above the zone-24 gauges against overshoots of 26–245 mm; returning all of it moves zone 24’s held-out error by half a point. Diversions above the gauges do not explain the overshoot. The shipped estimate does not change.

wet 0.2.4

  • Climr precipitation is scored against the River Forecast Centre’s plateau-elevation precipitation gauges and snow courses, under a rule fixed before any product value was computed (#50). climr is at most modestly high on the dry plateaus relative to the rest of the interior (dry over contrast 1.09, not decided), and part of #45’s gap between climr and PNWNAmet is PNWNAmet’s. Precipitation alone does not explain zone 24’s overshoot; the next check is gauges with water taken out upstream (#53). The shipped estimate does not change.

wet 0.2.3

  • The open water balance’s overshoot in the dry interior (hydrologic zones 15, 17, 23 and 24) is diagnosed against PCIC VIC-GL’s split of precipitation and evapotranspiration, under a rule fixed before any PCIC value was computed (#45). Neither term is named: our precipitation runs 1.4–1.6 times PCIC’s there, and our AET offsets most of the difference. The dry zones differ from the rest in precipitation, not AET, so the next check is precipitation observed at plateau elevation. The shipped estimate does not change.

  • The README and package description now cover the water-temperature path, and no longer claim seasonal or climate-scenario output, which no script builds.

wet 0.2.2

  • The open water balance is refit on HYDAT 2026-07-17 (#43). The pipeline now keeps one fit per HYDAT release (WET_HYDAT names the file), and the 2025-10-14 fit is reproduced exactly beside it. The new fit has 315 calibration gauges and a held-out mean absolute error of 27.5 % (headwater 30.7 %, nested 17.6 %). Pooled zones get no adjustment, settled on short-record gauges no fit uses, which retires the “mildly optimistic” caveat. The zone adjustment ties raw P − AET on headwater gauges and is kept by a recorded decision because it corrects the major rivers (#47 revisits the gate).

  • The segment discharge article compares the two estimates in a table of each one’s strengths and limits, with guidance on which suits which job, in place of a single recommendation. PCIC’s model fits gauges more closely where fwapg has a value; the open water balance covers all of BC and ships its code.

wet 0.2.1

  • The segment discharge article is reordered for readers: what mean annual discharge is, which estimate to use and why, how close the open water balance comes at 290 gauges, the Salmon River and Bulkley River as maps that hold every gauge they name, and limits in order of what affects a user. fwapg (PCIC) is now scored at the 168 gauges it covers: it is closer there (25.5 % against 30.5 % mean absolute error), and the article recommends the balance everywhere on the strength of its coverage and open code, stating that gap (#39).

wet 0.2.0

  • wet_temp_daily() starts a wet_temp_* family: daily mean, minimum and maximum water temperature at about 300 hydrometric stations from the water-temp-bc archive, in each station’s local standard time, in the shape wet_window_stats() takes. Departures are against 2016-2025, since few stations have year-round records before the 2010s (#36).

wet 0.1.1

  • Hex sticker, and a README that follows fresh, link and drift: install, prerequisites, a station example, vignettes and the package ecosystem.

wet 0.1.0

  • Package scaffold, PCIC VIC-GL OPeNDAP subsetting, annual/monthly runoff aggregation, upstream area weighting and an fwapg MAD parity check (#1).

  • Join-free upstream accumulation over FWA codes (wet_upstream_sums(), wet_upstream_irregular(), wet_ws_fetch()): whole basins, no order-8 skip. wet_upstream_mean() now takes (ws, values, denom, irregular_pairs, upstream_area) instead of a pairs table. wet_pcic_annual() fetches PCIC one year at a time; scripts/mad_basin.R builds the Fraser (#2).

  • An open, province-wide water balance after Chapman et al. (2018), the BC Water Tools method, validated out of sample at 290 HYDAT stations (#11).

  • The water balance’s annual AET is now CGIAR floored by a Fu-Budyko AET, chosen from four candidates by a rule fixed before scoring (#15). Blocked-CV mean absolute error on annual runoff drops from 33.1 % to 27.7 % (headwater basins 38.1 % to 31.2 %; 32.3 % under a fully nested selection), mostly on the dry interior plateaus.

  • MOD16A3GF v061 annual ET (wet_mod16_aet()) was scored as a challenger to the shipped AET under a rule fixed before scoring (#18). It improves on CGIAR alone but not on CGIAR floored by Fu-Budyko, which stays: headwater MAE is 36.3 % for MOD16 and 34.6 % for max(CGIAR, MOD16), against 31.2 %. scripts/wb_aet_compare.R now re-runs #15’s comparison, which must reproduce, before the #18 one. See research/water_balance_method.md.

  • The input builders wet_cgiar_aet(), wet_dem_glo90() and wet_climr_normals(), and the hydrologic-zones grid in scripts/wb_inputs.R, now key their caches on a method version (and the normals on climr’s version too), so a builder change never reuses an old file (#19). Existing caches rebuild once under the new names; re-run under them, the water balance reproduces every shipped number.

  • Flow departure at hydrometric stations (#25).

    • wet_station_daily() gives one daily flow series per station, from approved HYDAT through ECCC provisional daily means (the water-temp-bc archive) to real-time. Each source continues the one before it, from its whole record. Ported from ngr::ngr_hyd_q_daily().
    • wet_window_stats() summarises any daily series over month-day windows once per year, in cd’s long format, and wet_windows_calendar() supplies the calendar windows. Departure and trend come from cd.
    • scripts/station_departure.R runs it for real-time stations. At Buck Creek and Bulkley nr Houston, 2023-2026 summer and early-fall flow ran 54-87 % below the 1981-2010 mean. Provisional winter flows are uncorrected ice readings; see research/station_flow_departure.md.

  • A pkgdown site and the first vignette, Chinook flow at two stations (#27). It maps Buck Creek and Bulkley nr Houston and their nested catchments, then compares the last five years with 1981-2010 in the three open-water Chinook windows (migration, spawning, fry migration), from bundled data (data-raw/station_vignette_data.R, data-raw/station_vignette_map.R). Every window at both stations was below its 1981-2010 mean in 2023-2026, and 21 of the 24 station-window values from those years were drier than nine baseline years in ten.

  • A second vignette, Mean annual discharge per segment (#28). It reproduces fwapg’s discharge on the 9,000 Salmon River (SALR) segments it gives a value, and shows how area-weighted sampling moves small watersheds. It maps discharge per segment for SALR and for the Bulkley (BULK), where fwapg has none and the open water balance is the only estimate, and plots the balance’s blocked-CV skill at 290 HYDAT gauges by zone. On SALR the balance runs a median 1.47 times PCIC. At the four gauges within 30 km of SALR and its outlet gauge, the balance is high at all five. PCIC is within 5 % at the two small basins and low at the three large ones. Data from data-raw/segment_vignette_data.R and data-raw/segment_vignette_map.R.