Impact model: BuRNN

Statistical fire model using the mean of 5 LSTMs at each location. Trained on GFEd5.1. BuRNN v1.1 uses gpp, lai and cveg from VISIT for all simulations.

ISIMIP3b
ISIMIP3a

Person responsible for model simulations in this simulation round

Seppe Lampe: seppe.lampe@vub.be, 0000-0002-7907-4496, Vrije Universiteit Brussel (Belgium)

Basic information

Model Version: 1.1

Person responsible for model simulations in this simulation round

Seppe Lampe: seppe.lampe@vub.be, 0000-0002-7907-4496, Vrije Universiteit Brussel (Belgium)

Additional persons involved: Wim Thiery

Basic information

Model Version: 1.1

Model Homepage: https://github.com/VUB-HYDR/BuRNN

Model License: CC BY 4.0

Reference Paper: Main Reference: Lampe S, Gudmundsson L, Kraft B, Hantson S, Kelley D, Humphrey V, Le Saux B, Chuvieco E, Thiery W et al. BuRNN (v1.0): a data-driven fire model. Geoscientific Model Development,19,955-988,2026

Resolution

Spatial aggregation: regular grid

Horizontal resolution: 0.5°x0.5°

Vertically resolved: No

Temporal resolution of input data: climate variables: monthly

Temporal resolution of input data: land use/land cover: annual

Input data

Observed atmospheric climate data sets used: GSWP3-W5E5 (ISIMIP3a)

Land use data sets used: Historical, gridded land use

Other human influences data sets used: Lakes area fraction

Other data sets used: Lightning, Land-sea mask

Climate variables: hurs, sfcWind, tasmax, pr

Spin-up

Was a spin-up performed?: Yes

Spin-up design: Repeated the 1901-1903 in front of the 1901-2019 input dataset to spin up the memory of the LSTM (similar procedure as during model traiing).

Natural Vegetation

Natural vegetation partition: We applied the CLM pft structure and then regrouped the original pfts into 11 pfts (see model description paper).

Natural vegetation dynamics: It is not, for gpp, lai and cveg we used the values available on the ISIMIP repository from VISIT (which was not coupled to BuRNN).