Extended Data Fig. 3: Description of the weighting.
From: Global burned area increasingly explained by climate change
First, the burned area (BA) observations and simulations are transformed to relative anomalies. Then, we calculate the climatological RMSE and total NME of between the observational RA (monthly) and the factual simulated RA (monthly). From the RMSE, we generate random noise and add that to the simulated values. We repeat this process 1000 times, the bottom right plot is a visualization of the aggregation of these 1000 series (using yearly data instead of monthly for simplification), showing the median value for each model for each timestep along with the P2.5-P97.5. We then combine these 1000 series with the NME and the kneedle algorithm; to find the optimal σD and the according weights. This results in 1000 sets of weights (box shows the inter-quantile range (IQR) centred around the median, while the whiskers extend from the box by 1.5 times the IQR and the dots represent outliers), which are used in our analysis.
