Skip to contents

Build a pre-1962 land series measured on each year's own borders

Source: R/historical_land_areas.R
build_historical_land_areas.Rd

Emit the land table the pre-1962 production back-cast consumes – year, area_code, Cropland, Pasture and agriland, all in Mha – with the hectares summed from gridded LUH2 inside the polygon of the polity that area_code resolved to in that year, instead of inside present-day borders.

The cell-by-polity intersection is not measured here: it is read from the polycell support table (read_polycell_support()), which is that intersection measured geodesically on s2 and keyed on each polity's validity interval.

A cell's land is shared among the polities whose territory covers it, in proportion to that territory renormalised to one per cell, which is the rule build_cell_polity_fraction() already uses. Renormalising matters: LUH2's state fractions are fractions of the whole cell and already discount open water, so weighting them by a raw coastal cell's land share would discount it twice and lose 12-15% of the land of an island or heavily coastal country.

fill_proxy_growth() consumes only this series' year-on-year ratios, so a change of territory can only reach the back-cast as a ratio. What that ratio should be is a real choice, and boundary_step makes it:

  • "relink" (default) re-measures the previous year inside the incoming polity's polygon before taking the ratio, so only within-territory growth is ever used and annexing a province never moves the back-cast. On Ethiopia in 1952, when Eritrea joins, that is +1.9% instead of +8.0%.

  • "level_step" takes the ratio between the two polygons as measured, so the territorial change passes through as a level step and the 1850 row is scaled to the smaller empire it is labelled with. That is the reframing the whole method exists for; it is also the option most exposed to a bad polygon, because an artefact of the polity database then compounds down the back-cast exactly as a real annexation would.

Measured over 1850-1961 against the present-day series, 18.0% of back-cast crop tonnage at 1850 sits between the two rules, falling to 0.07% by 1960.

This reads gridded LUH2 for every requested year and is minutes-to-tens-of- minutes of work, so it belongs in a data-raw/ materialisation step, not in a test or an example.

Usage

build_historical_land_areas(
  years = 1850:1961,
  boundary_step = c("level_step", "relink"),
  data = NULL,
  example = FALSE
)

Arguments

years

Integer vector of calendar years to measure. Defaults to 1850:1961, the span the back-cast uses.

boundary_step

How a year-on-year ratio is taken across a change of territory: "level_step" (default) or "relink". They answer different questions and differ by up to 18% of back-cast tonnage, so the choice is the method, not a tuning knob – see the description.

"level_step" lets the series step when the territory changes, because a different polity is a different thing being measured. That is what a per-polity series means, and it is why this function exists: on Ethiopia it puts 1850 cropland at 1.52 Mha against the present-day 3.22, dropping the land Menelik annexed in the 1880s-90s that the area never held in 1850.

"relink" re-measures the previous year inside the incoming polity's polygon so a change of territory never appears as growth. That suits a FIXED-territory series, where the step is an artefact. It is NOT the conservative choice here: because fill_proxy_growth() consumes only ratios, suppressing that channel also suppresses the correction, and Ethiopia's 1850 comes back to 3.24 Mha – within 0.6% of the present-day figure this method exists to replace (whep#761).

data

Named list of pre-loaded inputs bypassing the readers, for tests: polity_areas (year, area_code, polity_code), support (a read_polycell_support() table), cover (polity_code, lon, lat, frac, which is support already reduced to one weight per cell) and cell_areas (year, lon, lat, land_use, area_ha). Each falls back to its reader when absent.

example

If TRUE, return a small fixture instead of reading remote data. Defaults to FALSE.

Value

A tibble with columns year, area_code, polity_code, Cropland, Pasture and agriland. area_code is the polity_area_code aggregation bucket, the same key .read_land_areas() emits, so the result is a drop-in for it at the back-cast seam. polity_code names the territory each year was measured on, and is semicolon-separated where a bucket holds more than one polity in a year.

Examples

build_historical_land_areas(example = TRUE)
#> # A tibble: 10 × 6
#>     year area_code polity_code    Cropland Pasture agriland
#>    <int>     <int> <chr>             <dbl>   <dbl>    <dbl>
#>  1  1961       255 BEL-1831-2025      1.02   0.718     1.73
#>  2  1961        51 F51-1947-1993      5.35   1.81      7.16
#>  3  1900       203 ESP-1800-2025     16.2    8.20     24.4 
#>  4  1961       228 F228-1945-1991   238.   332.      570.  
#>  5  1850       238 ETH-1800-1889      1.52   1.88      3.40
#>  6  1900       238 ETH-1897-1902      6.00  13.6      19.6 
#>  7  1951       238 ETH-1941-1952      9.45  23.0      32.4 
#>  8  1952       238 ETH-1952-1993     10.2   30.0      40.2 
#>  9  1961       238 ETH-1952-1993     12.0   29.6      41.5 
#> 10  1961       248 F248-1947-1991     8.40   6.46     14.9