Closed popovs closed 4 days ago
Results with buffer years added around retirement
0 | 5 | 10 | 15 | |
---|---|---|---|---|
raw | 50% | 52% | 52% | 52% |
legacy v | 76% | 76% | 76% | 76% |
So it seems the error seems to stabilize around 5 years.
Investigating the 25% that don't line up with historical verifications and 48% that don't line up with raw value. The majority are < 100 m difference.
VDIFF
> summary(investigate$dist_road_VDIFF)
Min. 1st Qu. Median Mean 3rd Qu. Max.
6.00 13.00 23.00 60.28 63.00 940.00
RAWDIFF
> summary(investigate$dist_road_RAWDIFF)
Min. 1st Qu. Median Mean 3rd Qu. Max.
4.00 14.25 36.00 140.61 135.75 1943.00
Investigated dens where the discrepancy between new GIS verification and legacy verification is >25m. N = 29.
Investigated dens where the discrepancy between new GIS verification and raw value is > 25m. N = 65, 23 of which were the ones that also tripped the legacy verification.
Out of the discrepancies, errors in the field data ("Field error") + mismatches between which roads to measure to between field & verification data ("Unclear road data") account for the greatest share. After that, it's almost all caused by issues in the GIS road data that can't really be fixed by filtering - e.g. ghost roads that don't exist, presumably awarded but never built, and overgrown roads with no/incorrect retirement date.
Currently, for all road GIS verifications, road sections are being filtered by award date and retirement date. However, roads are often still driveable many years after being retired on paper. Add 5, 10, and 15 year 'buffers' to the retirement date of the road to include them in the distance to road calculations and see which ones line up best with field estimates of distance to road. Mary is guessing 10 years is our best bet.