buf_points.md

No more buffered points?

Yichuan Shi

The problem

Not possible

to accurately calculate total area with points

Not impossible

to estimate total area with reasonable confidence (make do with data!)

Zero knowledge Ideas

• The ceiling (assuming no overlap):

sum(rep_area)

• The floor (assuming all overlap):

0

• The true value must be in-between

[0, sum(rep_area)]

An attempt

• learn: current relationship between MPAs' overlap based known boundary
• predict: overlap of MPAs with points based on the learnt relationship

Why it cannot be done

• the relationship cannot be modelled using non-spatial attributes alone
• permutation of boundaries is technically not possible

model fitting and cross-validation

The proposed solution

point estimate

• those not possible to overlap (by search radius based on rep_area)

sum(rep_area1)

• those possible to overlap

sum(rep_area2) * percentage

where percentage is calculated using MPAs with polygon boundary

percentage = total_dissolved_area / sum(rep_area_gis_polygon)

interval estimate

• The high confidence total

TOTAL = total_gis_dissolve_polygon + sum(rep_area1)

• The ceiling consider all those possible to overlap does not overlap

TOTAL + sum(rep_area2)

• The floor consider all those possible to overlap overlap

TOTAL + 0

search radius

Circular buffer is still the best guess for unknown boundary

Implementation

# calculate search radius based on rep_area
search_radius = sqrt(rep_area / PI)
# if a point overlaps with any polygon
# possible (must) to overlap
if point_overlap_with_polygon:
	overlap = True
# if a point's search radius is greater than distance to nearest polygon
# possible (very likely) to overlap
elif search_radius > distance_to_nearest_polygon:
	overlap = True
# if a point's search radius is shorter than distance to nearest polygon
# no overlap (very unlikely)
else:
	overlap = False

Discussions