Scoring crossings for fish passage prioritization
Source:vignettes/scoring-crossings.Rmd
scoring-crossings.RmdThis vignette walks through the crossing interpretation pipeline using the Morice watershed group (MORR) as an example. The same steps apply to any watershed group in the province.
Data generated by data-raw/vignette_morr.R. Sources:
crossings from fresh CSV, PSCIS
assessments from BC Data Catalogue via bcdata, override CSVs from bcfishpass. No
database required.
How link fits with fresh
fresh is a generic stream network engine. It segments streams at break points, classifies habitat by species thresholds, and rolls up results. But fresh doesn’t know about crossings, severity, or field assessments — that’s link’s job.
link (interpret, score crossings) --> break source spec --> fresh (segment, classify habitat)link reads raw crossing data, applies field corrections, scores
severity, and produces a break_sources spec that plugs
directly into frs_habitat(). Confirmed barriers break the
network geometry. Unassessed crossings are features registered on the
network for downstream indexing (see fresh #92,
#93).
The problem with binary classification
Provincial crossing assessments produce BARRIER or PASSABLE. This tells you nothing about how bad a barrier is or what habitat it blocks.
morr_raw <- readRDS(system.file(
"testdata", "morr_crossings_raw.rds", package = "link"))
nrow(morr_raw)
#> [1] 1599
table(morr_raw$crossing_source)
#>
#> CABD MODELLED CROSSINGS PSCIS
#> 3 1424 172
table(morr_raw$barrier_status)
#>
#> BARRIER PASSABLE POTENTIAL UNKNOWN
#> 1 86 116 1383 13MORR has 1599 crossings — 1424 modelled, 172 with PSCIS field assessments. Most modelled crossings have no field data.
Override corrections
bcfishpass maintains correction CSVs — tens of thousands of hand-reviewed crossings accumulated across field seasons and imagery review.
morr_fixes <- readRDS(system.file(
"testdata", "morr_modelled_fixes.rds", package = "link"))
nrow(morr_fixes)
#> [1] 947
table(morr_fixes$structure)
#>
#> FORD NONE OBS
#> 718 4 162 63For MORR: 947 modelled crossing fixes. Most reclassify crossings as empty (no structure found on imagery) or open-bottom structures (passable).
head(morr_fixes[, c("modelled_crossing_id", "structure",
"reviewer_name", "source")], 8)
#> modelled_crossing_id structure reviewer_name source
#> 827 14001728 NONE OB Bing/Google/ESRI imagery
#> 829 14001727 NONE OB Bing/Google/ESRI imagery
#> 830 14001726 NONE OB Bing/Google/ESRI imagery
#> 831 14001724 NONE OB Bing/Google/ESRI imagery
#> 833 14001721 OBS OB Bing/Google/ESRI imagery
#> 834 14001714 NONE OB Bing/Google/ESRI imagery
#> 835 14001698 NONE OB Bing/Google/ESRI imagery
#> 836 14001697 NONE OB Bing/Google/ESRI imageryThe link override pipeline loads, validates, and applies these corrections:
conn <- lnk_db_conn()
# Load --- validates CSV structure before writing to DB
lnk_override_load(conn,
csv = "user_modelled_crossing_fixes_morr.csv",
to = "working.morr_modelled_fixes",
cols_id = "modelled_crossing_id",
cols_required = c("structure"),
cols_provenance = c("reviewer_name", "review_date", "source"))
# Validate --- find orphans and duplicates
lnk_override_validate(conn,
overrides = "working.morr_modelled_fixes",
crossings = "working.morr_crossings")
# Apply --- auto-detects which columns to update
lnk_override_apply(conn,
crossings = "working.morr_crossings",
overrides = "working.morr_modelled_fixes")PSCIS matching
PSCIS assessments have field measurements — outlet drop, culvert slope, length, channel width. The crossings table from fresh has network position. Matching links the measurements to the network so severity scoring uses real data.
morr_pscis <- readRDS(system.file(
"testdata", "morr_pscis.rds", package = "link"))
morr_xref <- readRDS(system.file(
"testdata", "morr_xref.rds", package = "link"))
nrow(morr_pscis)
#> [1] 169
nrow(morr_xref)
#> [1] 8The 8 xref corrections are hand-curated GPS error fixes from field work. These take priority over spatial matching.
lnk_match_pscis(conn,
crossings = "working.morr_crossings",
pscis = "working.morr_pscis",
xref_csv = "pscis_modelledcrossings_streams_xref_morr.csv",
to = "working.morr_matched")PSCIS barrier status overrides
morr_pscis_fixes <- readRDS(system.file(
"testdata", "morr_pscis_status_fixes.rds", package = "link"))
if (nrow(morr_pscis_fixes) > 0) {
morr_pscis_fixes[, c("stream_crossing_id",
"user_barrier_status", "notes")]
}
#> stream_crossing_id user_barrier_status notes
#> 1295 197962 PASSABLE Bridge; Imagery ReviewOne MORR crossing has a barrier status override — confirmed PASSABLE by imagery review.
Severity scoring
Default thresholds classify crossings by biological impact:
th <- lnk_thresholds()| Severity | Criteria | Interpretation |
|---|---|---|
| High | outlet_drop >= 0.6m OR slope x length >= 120 | Impassable at most flows |
| Moderate | outlet_drop >= 0.3m OR slope x length >= 60 | Flow-dependent |
| Low | everything else | Likely passable |
lnk_score_severity(conn, "working.morr_crossings")
morr_dist <- readRDS(system.file(
"testdata", "morr_severity_dist.rds", package = "link"))
morr_dist
#>
#> high low moderate <NA>
#> 11 73 19 66
morr_scored <- readRDS(system.file(
"testdata", "morr_pscis_scored.rds", package = "link"))
# The reveal: crossings with PSCIS measurements scored by severity
scored_with_data <- morr_scored[!is.na(morr_scored$severity), ]
head(scored_with_data[order(-scored_with_data$outlet_drop),
c("stream_crossing_id", "barrier_result_code",
"outlet_drop", "culvert_slope",
"severity")], 10)
#> stream_crossing_id barrier_result_code outlet_drop culvert_slope severity
#> 122 198039 BARRIER 1.60 5.0 high
#> 157 198085 BARRIER 1.60 5.0 high
#> 9 197367 BARRIER 1.57 5.0 high
#> 130 198056 BARRIER 1.20 4.0 high
#> 58 197951 BARRIER 1.10 4.5 high
#> 119 198036 BARRIER 1.10 6.0 high
#> 94 198009 BARRIER 0.80 5.0 high
#> 163 198936 BARRIER 0.75 2.0 high
#> 51 197944 BARRIER 0.70 4.5 high
#> 161 198934 BARRIER 0.70 4.0 highThe binary barrier_result_code treats these crossings
identically. The severity column reveals the gradient —
from impassable drops to likely passable culverts.
Handing off to fresh
lnk_break_source() produces the spec that plugs directly
into fresh’s
frs_habitat(). Link’s severity labels translate to fresh’s
access labels via label_map:
-
highseverity –>"blocked"(confirmed barrier, breaks geometry) -
moderateseverity –>"potential"(unconfirmed, may break geometry)
morr_spec <- readRDS(system.file(
"testdata", "morr_break_source.rds", package = "link"))
str(morr_spec)
#> List of 3
#> $ table : chr "working.morr_crossings"
#> $ label_col: chr "severity"
#> $ label_map: Named chr [1:2] "blocked" "potential"
#> ..- attr(*, "names")= chr [1:2] "high" "moderate"
src <- lnk_break_source(conn, "working.morr_crossings")
# The handoff --- fresh takes it from here
fresh::frs_habitat(conn, "MORR", break_sources = list(src))
# After fresh classifies habitat, link can roll up per crossing
lnk_habitat_upstream(conn, "working.morr_crossings",
"fresh.streams_habitat")Scaling to any watershed group
The same steps work for any watershed_group_code. The
override CSVs have watershed_group_code columns for
filtering. The crossings CSV from fresh covers the entire province.
crossings <- read.csv(system.file("extdata", "crossings.csv",
package = "fresh"))
wsgs <- unique(crossings$watershed_group_code)
for (wsg in wsgs) {
# 1. Filter crossings to WSG
# 2. Load and apply overrides (filtered to WSG)
# 3. Fetch PSCIS from bcdata, match with lnk_match_pscis()
# 4. lnk_score_severity()
# 5. lnk_break_source() -> frs_habitat()
}