Fig. 3: Performance of the climate-burned area model with observations and seasonal predictions, presented by region and season.
From: Enhancing seasonal fire predictions with hybrid dynamical and random forest models
Panels a–d show the distribution of AUC (area under the ROC curve) values for the model based on observations (obs, in white) and predictions (pred, in orange) across different seasons: Panel a December–January–February (DJF), panel b March–April–May (MAM), panel c June–July–August (JJA), and panel d September–October–November (SON). The median is shown as a solid line, the squares indicate the interquartile range (25–75th percentile), and the ends of the whiskers represent the 5–95th percentile range. The percentages in blue indicate the proportion of the surface area with significant predictive skill (p-value < 0.01) in each region (Supplementary Fig. 3). The bars at the bottom of the graphs show the percentage of the significant domain. This figure was generated using R software, version 4.3.1.
