compute_lumi: Compute land-use mix indicators on a spatial grid

In cnefetools: Access and Analysis of Brazilian CNEFE Address Data

Compute land-use mix indicators on a spatial grid

Description

compute_lumi() reads CNEFE records for a given municipality, assigns each address point to spatial units (either H3 hexagonal cells or user-provided polygons), and computes the residential proportion (p_res) and land-use mix indices, such as the Entropy Index (ei), the Herfindahl-Hirschman Index (hhi), the Balance Index (bal), the Index of Concentration at Extremes (ice), the adapted HHI (hhi_adp), and the Bidirectional Global-centered Balance Index (bgbi), following the methodology proposed in Pedreira Junior et al. (2025, 2026). The 2026 article introduces the BGBI, and the adapted HHI is documented in the 2025 preprint.

Usage

compute_lumi(
  code_muni,
  year = 2022,
  polygon_type = lifecycle::deprecated(),
  polygon = NULL,
  crs_output = NULL,
  h3_resolution = 9,
  verbose = TRUE,
  cache = TRUE,
  cache_dir = NULL,
  backend = c("duckdb", "r")
)

Arguments

code_muni	Integer. Seven-digit IBGE municipality code.
year	Integer. The CNEFE data year. Currently only 2022 is supported. Defaults to 2022.
polygon_type	The aggregation mode is now inferred from polygon: leave it NULL for an H3 grid, or pass an sf::sf object for user polygons. Passing polygon_type still works and warns.
polygon	An sf::sf object with polygon geometries. Supplying it switches the output from an H3 grid to these polygons. A warning is issued reporting the percentage of CNEFE points covered by the polygon area. If no CNEFE points fall within the polygon, an error is raised.
crs_output	The CRS for the output object. Only used when polygon is supplied. Default is NULL, which uses the original CRS of the polygon argument. Can be an EPSG code (e.g., 4326, 31983) or any CRS object accepted by sf::st_transform().
h3_resolution	Integer. H3 grid resolution (default: 9). Only used for the H3 grid, so it is ignored when polygon is supplied.
verbose	Logical; if TRUE, prints messages and timing information.
cache	Logical. If TRUE (default), the downloaded data is stored as a gzipped CSV in the user cache directory and reused in future calls. If FALSE, a temporary file is used and deleted after the call.
cache_dir	Character. Directory to use for cached downloads. If NULL (default), the CNEFETOOLS_CACHE_DIR environment variable is used when it is set, otherwise tools::R_user_dir() with which = "cache". Use this to point large downloads at a secondary drive or a shared volume.
backend	Character. "duckdb" (default) uses DuckDB with the H3 extension, reading the cached gzipped CSV directly. "r" computes H3 in R using h3jsr instead, and needs no DuckDB extension. "r" exists for environments where DuckDB extensions cannot be installed, such as some restricted computing clusters. It is not the lighter option: it materialises the filtered address table in R memory, so its footprint grows with the municipality, while DuckDB aggregates in a streaming fashion and stays nearly flat. On São Paulo (5.7 million addresses) the measured peak is about 8.4 GB under "r" against 0.6 GB under "duckdb", alongside being roughly 15 times slower. If the constraint is memory rather than installability, keep the DuckDB backend and cap it with the cnefetools.duckdb_config option instead. See ?cnefetools for that option, and the benchmark article at https://pedreirajr.github.io/cnefetools/articles/bench_duckdb.html for the measurements.

Details

Binary land-use classification

The indices computed here rest on a binary split. An address is counted as residential when COD_ESPECIE == 1 (private household), and as non-residential otherwise. This follows the formulation of the indices as published in Pedreira Junior et al. (2026), where the measures are defined and empirically validated on that two-category basis.

Exclusion of buildings under construction

compute_lumi() drops records with COD_ESPECIE == 7 (building under construction or renovation), because such records describe a transitional state rather than a realised land use. Note that cnefe_counts() does not apply this exclusion and reports these records as addr_type7.

The citywide baseline P

Two indices use a citywide residential share P: the bgbi index, which is referenced against it, and the Balance Index (bal), which uses it through r = P / (1 - P). The other indices are computed entirely within each spatial unit. Two properties of P are worth stating.

First, P is computed from CNEFE address-type counts rather than from census population, so it describes the distribution of address types and not the distribution of residents.

Second, P is always computed over the full municipality, including when polygon is supplied, so it does not adapt to the area the supplied polygons happen to cover. This is intended, as P describes the context the addresses sit in, which is the municipality, and a sub-area of a city is still part of that wider context. A baseline recomputed over the sub-area would measure something different, namely mix relative to the sub-area itself rather than relative to the city.

Value

An sf::sf object containing:

When polygon is NULL (H3 grid):

• id_hex: H3 cell identifier

• p_res, ei, hhi, bal, ice, hhi_adp, bgbi: land-use mix indicators

• geometry: hexagon geometry (CRS 4326)

When polygon is supplied:

• Original columns from polygon

• p_res, ei, hhi, bal, ice, hhi_adp, bgbi: land-use mix indicators

• geometry: polygon geometry (in the original or crs_output CRS)

References

Pedreira Junior, J. U.; Louro, T. V.; Assis, L. B. M.; Brito, P. L.; Bomfim, F. G. (2026). BGBI: A citywide-referenced and bidirectional land use mix index for planning and policy evaluation. Land Use Policy, 169, 108135. https://doi.org/10.1016/j.landusepol.2026.108135

Pedreira Junior, J. U.; Louro, T. V.; Assis, L. B. M.; Brito, P. L. (2025). Measuring land use mix with address-level census data. engrXiv preprint. https://engrxiv.org/preprint/view/5975 (where the adapted HHI (hhi_adp) is documented)

Massey, D. S. (2001). The prodigal paradigm returns: ecology comes back to sociology. In A. Booth & A. C. Crouter (Eds.), Does It Take a Village? Community Effects on Children, Adolescents, and Families. Lawrence Erlbaum.

Song, Y.; Merlin, L.; Rodriguez, D. (2013). Comparing measures of urban land use mix. Computers, Environment and Urban Systems, 42, 1–13. https://doi.org/10.1016/j.compenvurbsys.2013.08.001

Examples

# Compute land-use mix indices on H3 hexagons
lumi <- compute_lumi(code_muni = 2929057, cache = FALSE)

# Compute land-use mix indices on user-provided polygons (neighborhoods of Lauro de Freitas-BA)
# Using geobr to download neighborhood boundaries
library(geobr)
nei_ldf <- subset(
  read_neighborhood(year = 2022),
  code_muni == 2919207
)
lumi_poly <- compute_lumi(
  code_muni = 2919207,
  polygon = nei_ldf,
  cache = FALSE
)

cnefetools documentation built on Oct. 2, 2026, 1:08 a.m.