Extended Data Fig. 8: Modeling the relationships between changes in summer total burned area (ΔBA) and changes in summer fire weather variables using linear and cubic functions.
From: Amplified Arctic–boreal fire regimes from permafrost thaw feedbacks
Fire weather variables include FWI and its annual 95th percentile (FWI95), as well as ISI and its annual 95th percentile (ISI95). a-b, Schematic of liner (a) and cubic fittings (b). R2 and mean absolute error (MAE) were used to evaluate model performance. Δ indicates the year-to-year variation (that is, the difference between two consecutive years). The green circles include long time series data within a 3×3 spatial moving window. The yellow circles represent the mean ΔBA values under 20 bins of ΔFWI. The blue line indicates the fitted linear or cubic regression. c-d, Spatial patterns of model accuracy for the linear (c) and cubic models (d), with labels indicating regional average R2 and MAE values. The cubic function was selected to model the relationship between ΔBA and ΔFWI, as it yielded the highest R2 (0.53; red label). Maps created with ArcGIS (v.10.8) using basemap data from the IPCC-WG1 Atlas GitHub repository80.
