Note
Click here to download the full example code
Conditional Bias¶
What it's for: A Conditional Bias plot examines how the mean bias of a model varies as a function of the observed or modeled value itself.
When to use: Use this plot to identify systematic errors that are dependent on the magnitude of the variable. For example, it can reveal if a model consistently over-predicts low concentrations but under-predicts high concentrations (a common issue in many environmental models).
How to read: * X-axis: The reference value (typically the observed value), often grouped into bins. * Y-axis: The mean bias (Model - Observation) calculated for each bin. * Zero Line: A horizontal line at zero represents no bias. * Interpretation: Points above the zero line indicate over-prediction for that range, while points below indicate under-prediction. The trend of the points shows how the model's systematic error changes across the data range.
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from monet_plots.plots.conditional_bias import ConditionalBiasPlot
# 1. Prepare sample data
np.random.seed(42) # for reproducibility
n_samples = 500
# Simulate observations
observations = np.random.normal(loc=10, scale=3, size=n_samples)
# Simulate forecasts with a bias that depends on the observation value:
# - For lower observations, forecast is slightly higher (positive bias)
# - For higher observations, forecast is slightly lower (negative bias)
# - Add some random noise
forecasts = (
observations
+ (0.5 - 0.1 * observations)
+ np.random.normal(loc=0, scale=0.5, size=n_samples)
)
# Ensure data is in a DataFrame
df = pd.DataFrame({"observations": observations, "forecasts": forecasts})
# 2. Initialize and create the plot
plot = ConditionalBiasPlot(figsize=(10, 6))
plot.plot(df, obs_col="observations", fcst_col="forecasts", n_bins=15)
# 3. Add titles and labels
plot.ax.set_title("Conditional Bias Plot (Forecast vs. Observation)")
plot.ax.set_xlabel("Observed Value")
plot.ax.set_ylabel("Mean Bias (Forecast - Observation)")
plt.tight_layout()
plt.show()
Total running time of the script: ( 0 minutes 0.450 seconds)
Download Python source code: plot_conditional_bias.py