Access model
bcfishpass is based on previous BC Fish Passage Technical Working Group (FPTWG) Fish Passage models - the basic logic for evaluating connectivity is much the same as in previous versions. Using BC Freshwater Atlas (FWA) streams as the mapping base, barriers to a given species are identified and any watercourse downstream of all barriers to that species is considered ‘potentially accessible’ to the species in the absence of anthropogenic (dams, road-stream crossings) or transient natural barriers (debris flows, log jams, etc) - with an assumption that all mapped streams have sufficient flow for migration.
The processing steps involved are:
1. Collect known natural barriers
Collect known natural barriers:
barrier type |
source |
|---|---|
waterfalls (of height >=5m) |
|
subsurface flow |
|
expert/stakeholder identified barriers |
Natural barriers identified by stakeholders include:
waterfalls or cascades not identified in existing inventories / CABD
steep gradients not captured by modelling
channels known to be dry year-round
other similar features
2. Generate gradient barriers
FWA stream network lines hold standardized Z values; each vertex of a stream line holds an associated elevation value derived from the BC Digital Elevation Model. Absolute elevation accuracy is subject to error in the DEM, but all elevations have been processed to ensure relative elevation is clean - all streams flow downhill (see the Technical Specification for details). With these clean Z values, we can confidently calculate a reasonable estimate of the gradient of a stream at any point.
To identify locations where a stream’s slope exceeds a given threshold, the model iterates through each vertex of the stream’s flow line (identified by the blue_line_key) - starting at the mouth and working up to the headwaters. At each vertex, the slope is calculated from the vertex to a point 100m upstream. If this 100m slope exceeds a given threshold (eg 25% for salmon), the point location is recorded as a ‘gradient barrier’.
100m was chosen as the interval because:
100m is the traditional minimum length used for measuring slope and defining reaches in BC, as per the Fish Stream Guidebook
smoothing the FWA stream profile over 100m removes small irregularities in the data (spikes)
The gradient threshold used for a given model is species dependent. Thresholds applied to existing models are:
Species |
max gradient |
|---|---|
Pacific salmon (all) |
15% |
Steelhead |
20% |
Bull trout |
25% |
3. Filter natural barriers
For anadromous species, natural barriers downstream of Known Fish Observations are removed from the model - we presume that if there are observations of the given species upstream of a natural feature, it cannot be a barrier to that species. To correct for species misidentification or other issues in the observations dataset a count and or date threshold can be set. For example, current access models for Pacific Salmon and Steelhead assume all natural barrier features with 5 or more observations upstream since 1990 are not currently barriers to fish passage.
For potamodromous (freshwater migration only) species such as Bull Trout and Westslope Cutthroat Trout, models remove any natural barrier below an observation.
4. Identify stream downstream of all natural barriers
Watercourses downstream of all natural barriers to a given species can be considered “potentially accessible” to that species. In other words, a migratory fish of the given swimming ability could potentially access all these streams if no anthropogenic barriers are present (presuming all else is equal and adequate flow is present in the stream).
5. Collect known anthropogenic barriers
Dams, assessed PSCIS barriers and expert/user identified barriers are collected from these sources and loaded to the database:
barrier type |
source |
|---|---|
dams |
|
PSCIS assessed barriers |
|
expert/user identified barriers |
6. Model potential anthropogenic barriers
Potential anthropogenic barriers (closed bottom structures, ie culverts) are identified by mapping intersections of FWA streams and linear infrastructure (roads, rail lines, major trails). This set of potential barriers is filtered, removing locations of known bridge structures and locations where a open bottom/bridge structure is presumed to exist.
a. Download linear infrastructure
Road and railway features are downloaded from DataBC. Features used to generate stream crossings are defined by the queries below. The queries attempt to extract road and railway features only where the built feature type is likely to include a stream crossing structure.
Source |
Query |
|---|---|
|
|
|
|
|
|
|
|
|
b. Overlay and de-duplicate
Intersection points of the stream and road features are created. Because intersections can occur at road intersections / stream confluences where only one structure is likely to exist, crossings (on the same stream) are de-duplicated using several data source specific tolerances:
merge DRA crossings on freeways/highways with a 30m tolerance
merge DRA crossings on arterial/collector road with a 20m tolerance
merge other types of DRA crossings within a 12.5m tolerance
merge FTEN crossings within a 12.5m tolerance
merge OGC crossings within a 12.5m tolerance
merge railway crossings within a 20m tolerance
DRA crossings are also merged across streams at a tolerance of 10m.
After same-source data crossings are merged, all crossings are de-duplicated using a 10m tolerance across all (road) data sources (railway crossings are not merged with the road sources). The location of a given output crossing corresponds to the location from the highest priority dataset, in this order of decreasing priority: DRA, FTEN, OGC permits, OGC permits pre2006. Despite the removal of duplicates, the unique identifier for each source road within 10m of a crossing is retained, allowing all crossings to be linked back to all source road datasets that apply.
c. Identify open bottom structures
Modelled crossings are created at all intersections, but open bottom structures (bridges) are presumed to be passable. Open bottom structures are identified via these data sources and properties:
Source |
Query |
|---|---|
|
|
|
|
|
|
|
|
|
|
|
d. Manually QA potential barriers
While the scripts attempt to identify crossings with open bottom structures based on above rules, many more exist on the landscape. Manual review of commonly available satellite imagery (Google/Bing/ESRI) at crossing sites can often reveal if a bridge is present (and thus presumed to be passable). Where review has been done, the crossing is removed from the potential barriers dataset only if the imagery clearly shows a bridge or that the crossing does not exist (road or stream not present). Where the imagery is not definitive, the site is retained as a potential barrier.
Note that manual QA/QC of the potential barriers dataset has been project based - some areas have been reviewed, many have not. To date, review has primarily been along rail corridors and at very high impact sites. Before planning field visits, further review of potential barriers is always advised.
7. Report
For a given area of interest, modelling reports on:
total length of stream potentially accessible to species of interest
length of potentially accessible stream with no known or potential anthropogenic barriers downstream
amount of potentially accessible stream upstream of each known/potential barrier
amount of potentially accessible stream upstream of each known/potential barrier, and downstream of any other known/potential barrier
8. Known limitations
The model represents streams potentially passable/accessible to fish based only on known barriers, modelled gradient barriers, the species theoretical swimming ability, and known fish observations. Fish inventories are incomplete, the modelled swimming ability of various species is an approximation, and the model does not account for partial barriers.
Field testing / validation / ground truthing of this model and the assumptions that went into creating it have not been carried out. To date, peer review of the methodology has been limited.
There are also many limitations associated with the source inventories used to create the model. Source errors and omissions from model inputs are present in model outputs, and source errors may be compounded when combined with other inputs.
FWA stream network
The FWA stream network is based on TRIM I stream linework. Streams tend to be under-represented in wet, coastal areas and over-represented in dry interior areas.
TRIM features were delineated through air photo interpretation and have varying degrees of accuracy – particularly when it comes to smaller streams. TRIM commonly under-represents the number of streams in the wetter, coastal areas of the province and field surveyors may regularly find small streams which were not visible on the original air photos due to canopy, terrain, shadows, etc. As a result, these streams do not exist in the FWA and as a result, the model. In wet areas, field crews will tend to find additional streams and road-stream crossings.
Conversely, in the drier, interior portions of the province, TRIM may over-represent the number or magnitude of streams. These may be ephemeral or intermittent streams which only have water in them at the wettest times of the year. Field crews often report finding only a ‘dry draw’ at locations where a stream has been shown on the mapping.
Also, while Z values of FWA geometries have been smoothed to ensure that all streams flow downhill, elevation errors are still present:
error present in the DEM will also be present in the FWA
resolution of elevation is limited by the resolution of the DEM
resolution of elevation is further limited by the smoothing/cleaning processing
despite the smoothing to FWA elevations, some small elevations spikes remain
Known barriers and observations
The barrier and fish observation inventories have similar limitations. Fish observation data is based on the results of field sampling and fish collection permits. The distribution and scale of this work has typically been driven by development requirements in the resource industries. As such, the model is based on unevenly distributed data points. Areas that have seen more development and resource extraction tend to have a higher density of sampling and observation points while other, less developed areas tend to have sparse or non-existent fish observation and barrier data.
Even areas with a high density of fish observation data may be problematic. The observation dataset dates back to the early 20th Century and not all observation points are still valid. Some historical records are found in areas that are no longer accessible to fish because natural (landslides) and/or man–made (dams, weirs, etc.) blockages have occurred since the time of observation. Observation points may also have other errors such as incorrect co-ordinates or mis-identification of fish species. Similarly, the obstructions layer is neither exhaustive nor guaranteed to be accurate. These component data limitations may compound one another.
Potential barriers
The model will generally over-estimate the extent of potential barriers (closed bottom structures / culverts). Two errors in particular lead to this over-estimation:
Open bottom structures / bridges exists at a crossings modelled as a closed bottom structures / culverts
No inventory of built stream crossing structures on resource roads is available. In the absence of further information, the model presumes bridges are present only on ‘double line’ streams and 6th order streams and greater, as noted in section 6c. While bridges are commonplace on 5th order and lower streams in many areas, we do not want to miss potential barriers with very high impact by mis-classifiying them as passable open bottom structures.
No structure exists at mapped road/stream crossings
This can be due to a number of different data issues:
DRA mapping can be over-generous in its definition of ‘road’
the road has been deactivated (information regarding deactivation status is sparse)
FTEN road data is tenure data, not ‘as built’ data - many FTEN roads have never been built
a structure may not have been necessary in road construction (insufficent flow in the stream channel)
the stream is incorrectly mapped or the stream channel has moved from where it was mapped
9. Enhancements / improvements
As expert and local knowledge of natural barriers is added to bcfishpass and as more potential barriers are assessed, the quality of modelling outputs will improve.
Submissions of data fixes to improve the fish passage model are welcome, particularly:
locations of natural barriers to fish passage that are not present in the inventories
identification of modelled natural barriers that are not barriers to fish passage
identification of modelled potential barriers that do not exist or are known to be passable
10. Processing and publication / distribution
Model inputs and outputs are refreshed on a scheduled basis, and processed via github actions - the status of these workflows is available here:
The weekly model processing workflow generates a file based data package (as geopackage), posted to object storage for public distribution:
download link: freshwater_fish_habitat_accessibility_MODEL.gpkg.zip
general metadata: BC Data Catalogue
complete data dictionary