Input data set: Land Transformation (ISIMIP4)

Description:

Annual land-use transitions per grid cell, derived from the CMIP7 input4MIPs UofMD Land-Use Harmonization (LUH3) dataset and delivered on the ISIMIP 0.25° and 0.5° grids. Each variable is the fraction of the grid cell that changes from one land-use class to another during the year.

Twelve land-use classes are tracked: primary forest, primary non-forest, secondary forest, secondary non-forest, managed pasture, rangeland, urban land, and the five crop functional types (C3 annual, C4 annual, C3 nitrogen-fixing, C3 perennial, C4 perennial). Every ordered pair of distinct classes is published, giving 108 transition variables.

A transition labelled year Y is the flux out of year Y into year Y+1. Transitions are therefore published for one year fewer than the corresponding land-use states.

This dataset carries transitions only. The land-use states themselves — how much of each class a cell holds in a given year — are the separate landuse datasets, documented under Land Use (ISIMIP4).

Derived from the CMIP7 input4MIPs UofMD Land-Use Harmonization (LUH3) dataset, multiple-transitions product. Project page: https://luh.umd.edu/

Historical: UofMD-landState-3-1-2
SSP1-VL: UofMD-landState-vl-3-1-1
SSP3-H: UofMD-landState-h-3-1-1

All three were taken at input4MIPs publication version v20260408.

Chini, L. and Hurtt, G.: Land-Use Harmonization 3 v1.2 for CMIP7 Historical (LUH3 v1.2), source_id UofMD-landState-3-1-2, https://doi.org/10.5281/zenodo.19261724 — used for histsoc, 1850soc, 1901soc and 2021soc

Chini, L. and Hurtt, G.: Land-Use Harmonization 3 Future Scenario VL v1.1 for ScenarioMIP (LUH3-VL v1.1), source_id UofMD-landState-vl-3-1-1, https://doi.org/10.5281/zenodo.19353397 — used for ssp1vlsoc-noadapt

Chini, L. and Hurtt, G.: Land-Use Harmonization 3 Future Scenario H v1.1 for ScenarioMIP (LUH3-H v1.1), source_id UofMD-landState-h-3-1-1, https://doi.org/10.5281/zenodo.19354835 — used for ssp3hsoc-noadapt

References for the LUH methodology:
Hurtt et al. (2020), Geoscientific Model Development, https://doi.org/10.5194/gmd-13-5425-2020
Chini et al. (2021), Earth System Science Data, https://doi.org/10.5194/essd-13-4175-2021

A transition labelled year Y is the flux out of Y into Y+1. The future scenarios therefore end in 2099 rather than 2100, and the ISIMIP4a files end in 2024 rather than 2025, one year short of the corresponding land-use state datasets. Under ISIMIP4a the 2024 field is zero by definition: the source covers transitions through 2023, which produce the observed 2024 state, and the 2025 state repeats 2024, so the 2024-to-2025 flux is no change rather than a repeat of the last observed one.

Twenty of the 108 variables are identically zero in every year and every scenario. This is the source carried through, and they are published so the transition matrix stays structurally complete.
Two groups:
Four cross-overs between the forest and non-forest secondary classes — primf_to_secdn, primn_to_secdf, secdf_to_secdn, secdn_to_secdf. LUH's forest / non-forest distinction is a property of the land's potential vegetation, which is static, so a parcel cannot move between them.
Sixteen crop-type transitions, including every annual-to-perennial and perennial-to-annual pair, which LUH3 does not resolve as direct fluxes. A change between two crop types appears in the state datasets but not as a transition.

Plantation transitions are not published. The LUH3 pltns state is all-zero in this release, so its transitions are excluded.

Wood-harvest area and biomass are not included here. They are published separately as the wood_harvesting dataset.

The two resolutions do not give the same global totals. The 0.5° product is conservatively remapped from the native 0.25° grid and renormalized over the land area of each coarser cell, so a global area total computed from a 30arcmin file exceeds the same total from the 15arcmin file by about 1 %, concentrated in coastal cells. Use the 15arcmin files when the global total matters.

Annual values are interpolated. The LUH3 reconstruction is anchored on decadal to centennial steps for much of the historical period. Smooth year-to-year behaviour in the early record is a property of that interpolation, not observed interannual variability.

2021soc means different things in the two rounds. Under ISIMIP4b it holds the 2021 transition field fixed across the 2022–2099 future window; under ISIMIP4a it holds the same field fixed across the full 1850–2024 historical period, as a present-day baseline.