j40-cejst-2/data/data-pipeline/data_pipeline/etl/sources/tribal_overlap/etl.py
Lucas Merrill Brown aca226165c
Issue 1900: Tribal overlap with Census tracts (#1903)
* working notebook

* updating notebook

* wip

* fixing broken tests

* adding tribal overlap files

* WIP

* WIP

* WIP, calculated count and names

* working

* partial cleanup

* partial cleanup

* updating field names

* fixing bug

* removing pyogrio

* removing unused imports

* updating test fixtures to be more realistic

* cleaning up notebook

* fixing black

* fixing flake8 errors

* adding tox instructions

* updating etl_score

* suppressing warning

* Use projected CRSes, ignore geom types (#1900)

I looked into this a bit, and in general the geometry type mismatch
changes very little about the calculation; we have a mix of
multipolygons and polygons. The fastest thing to do is just not keep
geom type; I did some runs with it set to both True and False, and
they're the same within 9 digits of precision. Logically we just want to
overlaps, regardless of how the actual geometries are encoded between
the frames, so we can in this case ignore the geom types and feel OKAY.

I also moved to projected CRSes, since we are actually trying to do area
calculations and so like, we should. Again, the change is small in
magnitude but logically more sound.

* Readd CDC dataset config (#1900)

* adding comments to fips code

* delete unnecessary loggers

Co-authored-by: matt bowen <matthew.r.bowen@omb.eop.gov>
2022-09-20 14:53:12 -04:00

208 lines
7.6 KiB
Python

import geopandas as gpd
import numpy as np
import pandas as pd
from data_pipeline.etl.base import ExtractTransformLoad, ValidGeoLevel
from data_pipeline.etl.sources.geo_utils import (
add_tracts_for_geometries,
get_tribal_geojson,
get_tract_geojson,
)
from data_pipeline.score import field_names
from data_pipeline.utils import get_module_logger
logger = get_module_logger(__name__)
class TribalOverlapETL(ExtractTransformLoad):
"""Calculates the overlap between Census tracts and Tribal boundaries."""
# Metadata for the baseclass
NAME = "tribal_overlap"
GEO_LEVEL = ValidGeoLevel.CENSUS_TRACT
PUERTO_RICO_EXPECTED_IN_DATA = False
ALASKA_AND_HAWAII_EXPECTED_IN_DATA = True
EXPECTED_MISSING_STATES = [
# 15 is Hawaii, which has Hawaiian Home Lands, but they are not included in
# this dataset.
"15",
# The following states do not have any federally recognized Tribes in this
# dataset.
"10",
"11",
"13",
"17",
"18",
"21",
"24",
"33",
"34",
"39",
"50",
"51",
"54",
]
# A Tribal area that requires some special processing.
ANNETTE_ISLAND_TRIBAL_NAME = "Annette Island LAR"
CRS_INTEGER = 3857
# Define these for easy code completion
def __init__(self):
self.COLUMNS_TO_KEEP = [
self.GEOID_TRACT_FIELD_NAME,
field_names.COUNT_OF_TRIBAL_AREAS_IN_TRACT,
field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT,
field_names.NAMES_OF_TRIBAL_AREAS_IN_TRACT,
]
self.output_df: pd.DataFrame
self.census_tract_gdf: gpd.GeoDataFrame
self.tribal_gdf: gpd.GeoDataFrame
@staticmethod
def _create_string_from_list(series: pd.Series) -> str:
"""Helper method that creates a sorted string list (for tribal names)."""
str_list = series.tolist()
str_list = sorted(str_list)
return ", ".join(str_list)
def extract(self) -> None:
self.census_tract_gdf = get_tract_geojson()
self.tribal_gdf = get_tribal_geojson()
def transform(self) -> None:
logger.info("Starting tribal overlap transforms.")
# First, calculate whether tracts include any areas from the Tribal areas,
# for both the points in AK and the polygons in the continental US (CONUS).
tribal_overlap_with_tracts = add_tracts_for_geometries(
df=self.tribal_gdf, tract_data=self.census_tract_gdf
)
tribal_overlap_with_tracts = tribal_overlap_with_tracts.groupby(
[self.GEOID_TRACT_FIELD_NAME]
).agg(
{
field_names.TRIBAL_ID: "count",
field_names.TRIBAL_LAND_AREA_NAME: self._create_string_from_list,
}
)
tribal_overlap_with_tracts = tribal_overlap_with_tracts.reset_index()
tribal_overlap_with_tracts = tribal_overlap_with_tracts.rename(
columns={
field_names.TRIBAL_ID: field_names.COUNT_OF_TRIBAL_AREAS_IN_TRACT,
field_names.TRIBAL_LAND_AREA_NAME: field_names.NAMES_OF_TRIBAL_AREAS_IN_TRACT,
}
)
# Second, calculate percentage overlap.
# Drop the points from the Tribal data (because these cannot be joined to a
# (Multi)Polygon tract data frame)
tribal_gdf_without_points = self.tribal_gdf[
self.tribal_gdf.geom_type.isin(["Polygon", "MultiPolygon"])
]
# Switch from geographic to projected CRSes
# because logically that's right
self.census_tract_gdf = self.census_tract_gdf.to_crs(crs=self.CRS_INTEGER)
tribal_gdf_without_points = tribal_gdf_without_points.to_crs(crs=self.CRS_INTEGER)
# Create a measure for the entire census tract area
self.census_tract_gdf["area_tract"] = self.census_tract_gdf.area
# Performing overlay funcion
# We have a mix of polygons and multipolygons, and we just want the overlaps
# without caring a ton about the specific types, so we ignore geom type.
# Realistically, this changes almost nothing in the calculation; True and False
# are the same within 9 digits of precision
gdf_joined = gpd.overlay(
self.census_tract_gdf,
tribal_gdf_without_points,
how="intersection",
keep_geom_type=False,
)
# Calculating the areas of the newly-created overlapping geometries
gdf_joined["area_joined"] = gdf_joined.area
# Calculating the areas of the newly-created geometries in relation
# to the original tract geometries
gdf_joined[field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT] = (
gdf_joined["area_joined"] / gdf_joined["area_tract"]
)
# Aggregate the results
percentage_results = gdf_joined.groupby(
[self.GEOID_TRACT_FIELD_NAME]
).agg({field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT: "sum"})
percentage_results = percentage_results.reset_index()
# Merge the two results.
merged_output_df = tribal_overlap_with_tracts.merge(
right=percentage_results,
how="outer",
on=self.GEOID_TRACT_FIELD_NAME,
)
# Finally, fix one unique error.
# There is one unique Tribal area (self.ANNETTE_ISLAND_TRIBAL_NAME) that is a polygon in
# Alaska. All other Tribal areas in Alaska are points.
# For tracts that *only* contain that Tribal area, leave percentage as is.
# For tracts that include that Tribal area AND Alaska Native villages,
# null the percentage, because we cannot calculate the percent of the tract
# this is within Tribal areas.
# Create state FIPS codes.
merged_output_df_state_fips_code = merged_output_df[
self.GEOID_TRACT_FIELD_NAME
].str[0:2]
# Start by testing for Annette Island exception, to make sure data is as
# expected
alaskan_non_annette_matches = (
# Data from Alaska
(merged_output_df_state_fips_code == "02")
# Where the Tribal areas do *not* include Annette
& (
~merged_output_df[
field_names.NAMES_OF_TRIBAL_AREAS_IN_TRACT
].str.contains(self.ANNETTE_ISLAND_TRIBAL_NAME)
)
# But somehow percentage is greater than zero.
& (
merged_output_df[field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT]
> 0
)
)
# There should be none of these matches.
if sum(alaskan_non_annette_matches) > 0:
raise ValueError(
"Data has changed. More than one Alaskan Tribal Area has polygon "
"boundaries. You'll need to refactor this ETL. \n"
f"Data:\n{merged_output_df[alaskan_non_annette_matches]}"
)
# Now, fix the exception that is already known.
merged_output_df[
field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT
] = np.where(
# For tracts inside Alaska
(merged_output_df_state_fips_code == "02")
# That are not only represented by Annette Island
& (
merged_output_df[field_names.NAMES_OF_TRIBAL_AREAS_IN_TRACT]
!= self.ANNETTE_ISLAND_TRIBAL_NAME
),
# Set the value to `None` for tracts with more than just Annette.
None,
# Otherwise, set the value to what it was.
merged_output_df[field_names.PERCENT_OF_TRIBAL_AREA_IN_TRACT],
)
self.output_df = merged_output_df