Lake km: should 1450 connection lines count as lake rearing? #317

If done: default's lake km count only lines that trace a lake's flow, so rearing km stop scaling with how many tributaries a lake has. If never: a big lake's km stay dominated by FWA connection lines. Adams Lake alone puts ADMS CH at +91 % rearing against bcfishpass.

Problem (link@4580717, v0.61.0)

• Default lake and wetland rearing: connected to spawning, centrelines kept, km and ha broken out #310 admits lake and reservoir lines to the default lake rear rule: 1000/1100/1200/1250/1300/1350/1400/1450/1475.
• Edge type 1450 ("Construction line, connection") joins each tributary mouth to the main-flow line (1200) across the lake. It is not a flow path through the lake.
• Measured (data-raw/logs/lake_connected_310/):
  • Adams Lake (13,229 ha) holds 211.6 km of ADMS CH / CO rearing lines: 62.8 km of 1200 and 148.8 km of 1450.
  • On NATR, BT's lake km are 275.0 km of 1450 and 227.6 km of 1200.
  • Province-wide, lakes hold 54,402 km of 1200 and 44,528 km of 1450.
• Against bcfishpass, ADMS CH rearing goes from +14 % to +91 %, and CO from +3 % to +75 %. Without Adams Lake those would be about +22 % / +15 %.

The issue named 1450 among the lake edge types, so #310 kept it.

Decision and result (2026-10-08)

Lake km are flow lines. Connection lines stay in rearing, and the rollups report them apart. Measured in data-raw/logs/lake_connection_317/; write-up in research/habitat_thresholds.md, "Lake connection lines".

• Connectors only: 1450. 1400 is "other flow / inferred connection", a construction flow line, so it stays in the km. The first cut also excluded 1400 and was narrowed before merge (operator, 2026-10-08: "construction and connector are totally different"). Lake 1400 is at most 1.4 km of rearing.
• The rules are unchanged. 1400 / 1450 stay in the L rule, so cluster_rearing still joins inlet rearing to the lake, and nothing re-classifies.
• The compare rollups split them out. lnk_compare_rollup() / lnk_compare_wsg() report rearing_lake_connection (km) and leave it out of rearing and rearing_lake km.
• The rule runs on both sides. bcfishpass counts these lines in BT and SK rearing too, so the bcfishpass side gets the same rule.
• lnk_rollup_wsg()'s default rearing_km stays the flag total, for the validator and the parity scripts.
• ADMS against bcfishpass: CH +91.1 % → +38.9 %, CO +75.0 % → +28.7 %, BT +16.4 % → +2.0 %. CH / CO stay high because bcfishpass has no CH / CO lake rearing on ADMS. The residual is lake flow lines, Default lake and wetland rearing: connected to spawning, centrelines kept, km and ha broken out #310's call.
• SK rearing km fall 55–95 % (seven WSGs) on both sides, so SK parity stays 0.0 %. KO falls 62 % on NATR, with no bcfishpass reference.

The proposal below, dropping 1400 / 1450 from the rule, was not taken.

Proposal

Decide whether lake km are "centrelines" (1200 / 1250 / 1300 / 1350 / 1475, the lines that trace flow) or "every line in the polygon".

If centrelines, drop 1400 / 1450 from the lake edge set in R/lnk_rules_build.R. That set is one list, used in both edge-type modes. Then rebuild the rules and re-measure with data-raw/logs/lake_connected_310/run.R.

1450 lines may still matter for connectivity: they join inlet clusters to the lake. Dropping them from rearing could disconnect inlet rearing from outlet spawning. Measure that before deciding.

Relates to #310.