Geospatial rolling average from geohashes on BigQuery

Question

Geohashes form a grid, much like a matrix.

If I have a table with a geohash7 and a value, how could I calculate say the average of all the adjacentcells on BigQuery?

Sample source table

geohash7	value
r1q6fh3	7
r1q6fh6	2
r1q6fh7	5
r1q6fh1	7
r1q6fh4	2
r1q6fh5	6
r1q6f5c	4
r1q6f5f	2
r1q6f5g	7

Sample output table

geohash7	value
r1q6fh4	4.6
...	...

Ideally in a scalable way as I want to run this over a large landmass.

Answer 1

First you'll need a function that returns neighbors. It is probably easier to grab one of javascript geohash libraries and use it as UDF, rather than do bit fiddling in SQL.

Once you have a function that returns neighbors, say

GeoHashNeibors(center STRING) RETURNS ARRAY<STRING>

You can then join original table with the flattened list of neighbors and compute average. Something like

WITH with_neighbors AS (
  select geohash, value, neighbor
  from data d, unnest(GeoHashNeibors(d.geohash)) neighbor
),
joined AS (
  select t1.geohash, t1.value, t1.neighbor, t2.value as n_value
  from with_neighbors t1 left join data t2
  on t1.neighbor = t2.geohash
)
SELECT geohash, ANY_VALUE(value) value, AVG(n_value) AS avg_neighbor
FROM joined 
GROUP BY 1