xagg 0.0.5

Aggregating raster data over polygons

xagg

A package (eventually) to aggregated gridded data in xarray to polygons in geopandas using area-weighting from the relative area overlaps between pixels and polygons. Check out the binder link above for a sample code run!

Intro

Science often happens on grids - gridded weather products, interpolated pollution data, night time lights, remote sensing all approximate the continuous real world for reasons of data resolution, processing time, or ease of calculation.

However, living things don't live on grids, and rarely play, act, or observe data on grids either. Instead, humans tend to work on the county, state, township, okrug, or city level; birds tend to fly along complex migratory corridors; and rain- and watersheds follow valleys and mountains.

So, whenever we need to work with both gridded and geographic data products, we need ways of getting them to match up. We may be interested for example what the average temperature over a county is, or the average rainfall rate over a watershed.

Enter xagg.

xagg provides an easy-to-use (2 lines!), standardized way of aggregating raster data to polygons. All you need is some gridded data in an xarray Dataset or DataArray and some polygon data in a geopandas GeoDataFrame. Both of these are easy to use for the purposes of xagg - for example, all you need to use a shapefile is to open it:

import xarray as xr
import geopandas as gpd

# Gridded data file (netcdf/climate data)
ds = xr.open_dataset('file.nc')

# Shapefile
gdf = gpd.open_dataset('file.shp')

xagg will then figure out the geographic grid (lat/lon) in ds, create polygons for each pixel, and then generate intersects between every polygon in the shapefile and every pixel. For each polygon in the shapefile, the relative area of each covering pixel is calculated - so, for example, if a polygon (say, a US county) is the size and shape of a grid pixel, but is split halfway between two pixels, the weight for each pixel will be 0.5, and the value of the gridded variables on that polygon will just be the average of both [TO-DO: add visual example of this].

I know you often need to weight your data by more than just its relative area overlap with a polygon (for example, do you want to weight pixels with more population more?). xagg has a built-in support for adding an additional weight grid (another xarray DataArray) into xagg.pixel_overlaps().

Finally, xagg allows for direct exporting of the aggregated data in several commonly used data formats (please open issues if you'd like support for something else!):

• netcdf
• csv for STATA, R
• shp for QGIS, further spatial processing

Best of all, xagg is flexible. Multiple variables in your dataset? xagg will aggregate them all, as long as they have at least lat/lon dimensions. Fields in your shapefile that you'd like to keep? xagg keeps all fields (for example FIPS codes from county datasets) all the way through the final export. Weird dimension names? xagg is trained to recognize all versions of "lat", "Latitude", "Y", "nav_lat", "Latitude_1"... etc. that the author has run into over the years of working with climate data; and this list is easily expandable as a keyword argumnet if needed.

Please contribute - let me know what works and what doesn't, whether you think this is useful, and if so - please share!

Use Cases

Climate econometrics

Many climate econometrics studies use societal data (mortality, crop yields, etc.) at a political or administrative level (for example, counties) but climate and weather data on grids. Oftentimes, further weighting by population or agricultural density is needed.

Area-weighting of pixels onto polygons ensures that aggregating weather and climate data onto polygons occurs in a robust way. Consider a (somewhat contrived) example: an administrative region is in a relatively flat lowlands, but a pixel that slightly overlaps the polygon primarily covers a wholly different climate (mountainous, desert, etc.). Using a simple mask would weight that pixel the same, though its information is not necessarily relevant to the climate of the region. Population-weighting may not always be sufficient either; consider Los Angeles, which has multiple significantly different climates, all with high densities.

xagg allows a simple population and area-averaging, in addition to export functions that will turn the aggregated data into output easily used in STATA or R for further calculations.

Left to do

• Make this a package. It's currently a somewhat stable module, but a goal is to have this be a fully-built and tested package.
• Make the outputs to the two main wrapper functinos (xagg.pixel_overlaps and xagg.aggregate) their own classes with the export functions as bound methods of the xagg.aggregate output class.
• Just testing, bug fixes, stability checks, etc.
• Share widely! I hope this will be helpful to a wide group of natural and social scientists who have to work with both gridded and polygon data!