Note
Click here to download the full example code
Brier Decomposition Plot¶
What it's for: The Brier Score Decomposition plot visualizes the three components of the Brier Score: Reliability, Resolution, and Uncertainty. The total Brier Score (BS) is calculated as: BS = Reliability - Resolution + Uncertainty.
When to use: Use this to gain a deeper understanding of why a probabilistic forecast system is performing a certain way. It allows you to distinguish between a model that is poorly calibrated (Reliability) and one that lacks the ability to distinguish different outcomes (Resolution).
How to read: * Reliability (Lower is better): Measures the weighted average of the squared differences between forecast probabilities and the relative frequencies of observed events. * Resolution (Higher is better): Measures how much the frequencies of events for specific forecast categories differ from the overall climatological frequency. * Uncertainty (Fixed for a given set of observations): Represents the inherent variability of the events being forecast. * Interpretation: A perfect model would have a Reliability of 0 and a Resolution equal to the Uncertainty, resulting in a Brier Score of 0.
import matplotlib.pyplot as plt
import pandas as pd
from monet_plots.plots.brier_decomposition import BrierScoreDecompositionPlot
data = pd.DataFrame(
{
"model": ["Model A", "Model B", "Model C"],
"reliability": [0.02, 0.05, 0.01],
"resolution": [0.15, 0.10, 0.18],
"uncertainty": [0.25, 0.25, 0.25],
}
)
plot = BrierScoreDecompositionPlot(figsize=(10, 7))
plot.plot(
data,
reliability_col="reliability",
resolution_col="resolution",
uncertainty_col="uncertainty",
label_col="model",
title="Brier Score Decomposition Comparison",
)
plt.tight_layout()
plt.show()
Total running time of the script: ( 0 minutes 0.340 seconds)
Download Python source code: plot_brier_decomposition.py