NEWS.md

In station-demand-forecasting-tool/sdft: Model to forecast demand for new local railway stations

sdft 0.3.2

Bug Fixes

• issue with passing database password to clusters
• issue with available cores test.

sdft 0.3.1

• rename sdr_submit() to sdft_submit(). Internal functions retain sdr_ prefix (these are still exported for expert use only)
• add PostgreSQL port option to sdft_submit()

Bug fixes

• Abstraction analysis catchments not all being written to output folder.
• with PostGIS v3.0 and pgRouting v3.1 a dimensionality error when using st_collect() on the node geoms prior to using st_concavehull() to generate service areas in sdr_create_service_areas. All the points appear to be 2D, so not clear what is causing this. However, it can be corrected by using ST_Force2D() on the geom prior to the st_collect().

sdft 0.3.0

• rename package to sdft
• converted main.R script into sdr_submit() function that takes database connection details and directory path (for inputs and outputs) as arguments. User must set the database user password using key_set().
• several additional configuration options via config.csv: testing, loglevel and cores
• some additional input checks.
• now check whether the postcode-polygons table is present in the data schema. If it isn't then the catchment will only consist of postcode centroids. This is because the OS postcode polygons dataset is not open data.
• Results are now pulled from postgreSQL into an output folder. Includes forecasts and catchments in GeoJSON format.

Data Issues

• data.stations table updated to amend car parking spaces:
• Apperley Bridge: NRE states 98 spaces. However, at: https://www.wymetro.com/corporate/news/projects/apperley-bridge-new-rail-station/ from West Yorkshire Combined Authority, clearly states that the station opened with 297 spaces. Also confirmed, here: https://www.westyorks-ca.gov.uk/projects/west-yorkshire-plus-transport-fund/rail-park-and-ride-phase-1-apperley-bridge/
• Fishguard and Goodwick. Originally appears to have had 23 spaces (See: http://fishguardtrains.info/?p=381.) Although it does appear to have recently been expanded but no information on current number of spaces. Unclear on Google Earth the exact number.
• Kemble: NRE states 220 spaces. Actual is 339. (See: https://www.apcoa.co.uk/parking-in/kemble/kemble-station/).

stationdemandr 0.2.4

Model changes

• switched to using the calibrated trip-end model with jobs within 2-minute drive time. Equates to 1 mile at 30mph (so approx 20-minute walk distance). More likely to capture nearby workplace zones where the centroid is not near a road edge. (#10)

Fixes

• added new script start.R. This simply calls main.R using source(). start.R can then be run from the terminal using rscript. If main.R is called directly using rscript then stop() calls will not terminate the model.

• problem if category was 'F'. Type was not specified when stations.csv read nor when proposed_stations table written to database. When category was 'F' type Boolean was assumed by default. Must be type text.

• pre-flight checks

• allow space in station name

• allow car park spaces to be zero

• the location_geom column in proposed_stations is used by the SQL function to measure distances from postcode centroids to the proposed station. This should therefore be generated from the proposed station's access NGR. However, this was instead being generated (in main.R) using the station's physical location NGR. This has been corrected.

• bug in sd_generate_choicesets. When sdr_crs_pc_nearest_stationswithpoints() is called crs is passed as the third argument of the function. This parameter is used in sdr_crs_pc_nearest_stationswithpoints() to avoid expensive distance lookups when the proposed or at_risk station is not in the set of stations to be measured. This is fine for generating a choiceset for a proposed station, or for a before abstraction analysis choiceset. However, this is the wrong parameter if an after abstraction analysis is being carried out as in that case crs contains the proposed station CRS not the at-risk station. This has been corrected to use pc_crs instead.

• expanded station service areas to include 2, 3 and 4km. Should improve performance of choiceset generation process.

Data Issues

• In the table data.hhsize there were trailing spaces in some of the area_code values. This requires correcting by running: update data.hhsize set area_code = trim(area_code)

• additional service areas (2, 3, and 4km) added to data.stations table

• data.stations table updated with car parking spaces for Great Northern stations. NRE shows these stations as having a car park but no spaces information. These car parks are run by Saba and details of spaces was obtained from https://www.sabaparking.co.uk/great-northern-railway.

• data.stations table updated with date station opened (where recorded).

• data.stations table updated to amend car parking spaces:

• Ilkeston NRE states a car park but gives no spaces. Appears that there are 150 spaces, in two car parks either side of the tracks. Source: https://anonw.com/tag/ilkeston-station/. Photo of one of the car parks suggests 150 is probably about right:photo
• Kenilworth - NRE states a car park but gives no spaces. Managed by Saba. Their website states 71 spaces. See: https://www.sabaparking.co.uk/car-park/kenilworth-station-car-park

• Pye Corner, Newport - NRE states car par but doesn't give spaces. According to BBC article there are 70 spaces: https://www.bbc.co.uk/news/uk-wales-south-east-wales-22550481. Google Maps satellite view indicate 60-65.

• Bus services. The following stations change from NA to TRUE:

• Kirkstall Forge
• Lea Bridge
• Bermuda Park
• Oxford Parkway
• Apperley Bridge

• Pye Corner

• Bus services. The following stations change from NA to FALSE:

• James Cook University Hospital

stationdemandr 0.2.3

Performance improvements

• Modified sdr_create_json_catchment() to improve catchment display performance. Probability is now rounded to 1 decimal places and then st_union is used to dissolve the postcode polygons, grouped by probability. The GeoJSON is then generated. This means that any catchment will now have a maximum feature count of 10. Also defined maxdecimaldigits for st_asgeojson, set to 5 rather than the default of 15.

stationdemandr 0.2.2

Fixes

• Fixed issue with the BEFORE abstraction analysis. Frequency group adjustments were not applied to the before situation - there was a single shared probability table for each at_risk station. This appeared to make sense when the decision was made. However, if one of the at-risk stations is also subject to an upward frequency adjustment then its weighted population will increase in the after situation (because of the higher frequency variable). This has been amended so that a separate probablity table is created for each proposed_station:at_risk station pair (the same as for the after analysis) and frequency group adjustments are made as required.

stationdemandr 0.2.1

Fixes

• Fixed bug with main script not able to handle empty data frames resulting from import of empty exogenous.csv and freqgroups.csv input files (which are optional).

stationdemandr 0.2.0

Fixes

• Fixed bug with parallel logging
• Fixed bug in sdr_create_service_area(). ST_ConcaveHull can return a point or line in some circumstances. Need to apply a buffer to ensure sa is a polygon and can be written to table.
• Fixed bug in SQL function sdr_crs_pc_nearest_stationswithpoints() - in the select case 60km and 80km service areas were incorrectly specified as 30km and 40km.
• Regexp for job_id pre-flight check amended to allow 20 character length.

New/amended features

• Import Boolean variables as logical - ensures they are all TRUE or FALSE which are also valid Boolean PostgreSQL values.
• Argument columns added to sdr_create_service_areas() (logical) to indicate whether the service area columns should be created. Increases flexibility of function.
• Column name changed from dailyfrequency_2017_all to frequency in stations table and sdr_generate_probability_table() function amended.
• Column name changed from entsexits2017 to entsexits in stations table and main script amended appropriately.
• Additional pre-flight check for region field.
• sapply replaced with vapply throughout. Preferred as sapply is not type safe.

Performance improvements

• Views previously used in sdr_generate_choicesets() replaced with materialized views.
• Various indexes (incl. spatial) now created for the materialized views.

Data issues

• Removed any edges and nodes disconnected from the main network graph from the OS OpenRoads tables. The affected nodes were identified using the pgRouting function pgr_connectedComponents() and were then used to delete affected edges from openroads.roadlinks and nodes from openroads.roadnodes. This deals with potential problem of the nearest edge to a postcode or station not being connected to the rest of the road network (a potential issue for routing and building station service areas). Centroidnodes and roadlink virtual node tables regenerated. Example disconnected network (in green) shown below:

  

• Full stations data created (up-to-date with latest station openings and closures).

• Service frequency updated from latest timetable data.
• Services and facilities updated from NRE KB.
• Full postcode centroid and population data.
• Full workplace zone centroid and population data. Also includes workplace zones for Scotland (previously not available from Census Scotland - postcode centroids previously used).

stationdemandr 0.1.0

Initial release

station-demand-forecasting-tool/sdft documentation built on July 11, 2021, 4:23 a.m.