To perform binning on observed catalog for variable sized spatial grids.

[khawajasim]

I need to perform gridding on observed catalog. The origin coordinates of every grid are available. Now there should be a way to plug such observed catalog into PyCSEP

[wsavran]

right now the gridding is optimized for regions that have regularly spaced bins. this works because you can directly compute the offset into the spatial region by something resembling. floor(value / dh).

for a variable spaced grid you would need to preform a binary search to find the insertion points into the irregularly spaced region. this functionality must be added.

can you share the QuadTree library you are using for the binning, maybe there are existing methods and we can wrap them.

[khawajasim]

1. PyGeoTile: https://pypi.org/project/pyGeoTile/
2. Mercantile: https://pypi.org/project/mercantile/
   First I used the 1st one, then I found out the 2nd one better and more accurate. So I shifted to 2nd now.

[wsavran]

great! im looking at it right now and there are a couple of modifications that will need to be made in order to support these types of regions.

these are items on the list that need to be solved, so this is a good opportunity to make that happen.

(1) the binning routines need to be routed through the Region classes, right now they are hard-coded assuming that grids are regular. for example, see AbstractBaseCatalog.spatial_counts() or AbstractBaseCatalog.spatial_magnitude_counts().

(2) we need to implement a QuadTreeRegion class that shares the same methods as the other region classes to perform (1).

this is not a major change, but a refactoring nonetheless. i think the first step is to make the QuadTreeRegion and a mock catalog that implements binning using those methods, along with appropriate tests. then we can implement this into the main catalog class. on a side note, this might be a good opportunity to create a single catalog class instead of the multiple different ones that currently exist.

[khawajasim]

You are right. It would be nice exciting feature to add to PyCSEP

[wsavran]

@khawajasim i am bumping this thread, because it seems timely with the issue you found in the current gridding. we can use this as a place to brainstorm how we are going to implement the quad-tree regions in pycsep.

[khawajasim]

Hi,
I was thinking a lot integrating quadtree gridding approach with pyCSEP. Here is what I thought. Lets start this with simple steps. We can take following steps as start. Right?

In order to first acquire quadtree grid, then evaluate forecast using quadtree grid, I should distribute all the relevant functions in 2 or 3 classes.

1. One class should deal with generating and acquiring all the grid related information based on available training catalog. Lets call it "GlobalQuadtreeGrid" . This piece of code for this class should go in file: regions.py. This class is only responsible for providing a spatial grid to the modelers and all the information associated with spatial grid.

2. Second class should deal with loading/reading forecast based on quadtree grid. This class should be totally independent of the previous class. Lets call this class "QuadtreeGridForecast". Once a modelers has a forecast for certain quadtree grid. He may use this class to load in the grid for which forecast was generated and forecast data. Then this class is responsible for interfacing forecast with CSEP1 tests.

3. Third class can deal with observed catalog (test catalog), lets call it QuadtreeGriddedCatalog.This can provide gridded catalog data for testing. However, we can also choose to merge these steps in "QuadtreeGridForecast" instead of making a new class.

Looking forward to hear your opinion.

[wsavran]

fixed by #115

[wsavran]

implemented in v0.6.0 release