Accelerating distances calculations using GPU
2019-06-26, 09:00–09:30, Room 1

The core concept of geospatial analysis algorithms is taking into account distances between features. The approach presented significantly increases the speed of calculations, as well as covers all cases of features locations.

Computation of distances on Earth's surface is not trivial by itself due to Earth’s shape and edge cases near poles and 180° meridian. If you have huge volumes of data, things get more complicated: an amount of this complex calculations growth exponentially.

I will go through different approaches for finding the nearest distances in Python and focus on the fastest one — GPU calculations.
This method in combination with spacial indexing can provide a huge boost to processing speed. Also, this approach can be used for any areas on Earth with no need to switch to local projection.

Presentation slides

vincenty-cuda-nns package