monet-stats

Getting Started with Monet Stats

This guide will help you get up and running with Monet Stats quickly. You’ll learn how to import the library, load your data, and perform common statistical analyses used in atmospheric sciences.

Prerequisites

Before starting, ensure you have Monet Stats installed:

pip install monet-stats

First Steps

Basic Import and Setup

import numpy as np
import monet_stats as ms

print(f"Monet Stats version: {ms.__version__}")

Simple Example: Temperature Forecast Evaluation

Let’s start with a basic example evaluating temperature forecasts:

import numpy as np
import monet_stats as ms

# Sample observed temperatures (°C)
observed_temps = np.array([
    15.2, 16.8, 14.5, 13.9, 17.2, 18.5, 19.1, 20.3, 18.7, 16.9,
    15.8, 14.2, 13.6, 12.8, 15.4, 17.9, 19.5, 21.2, 20.1, 18.4
])

# Corresponding model predictions
model_temps = np.array([
    15.8, 16.2, 15.1, 13.4, 17.8, 18.9, 19.8, 20.1, 18.3, 17.2,
    15.1, 14.8, 12.9, 13.3, 15.9, 18.2, 20.1, 21.5, 19.8, 18.9
])

# Basic error metrics
rmse = ms.RMSE(observed_temps, model_temps)
mae = ms.MAE(observed_temps, model_temps)
mb = ms.MB(observed_temps, model_temps)

print(f"RMSE: {rmse:.3f}°C")
print(f"MAE: {mae:.3f}°C")
print(f"Mean Bias: {mb:.3f}°C")

Common Statistical Analyses

1. Error Metrics Analysis

# Comprehensive error analysis
error_metrics = {
    'RMSE': ms.RMSE(observed_temps, model_temps),
    'MAE': ms.MAE(observed_temps, model_temps),
    'Mean Bias': ms.MB(observed_temps, model_temps),
    'Normalized Mean Bias': ms.NMB(observed_temps, model_temps),
    'Normalized Mean Error': ms.NME(observed_temps, model_temps)
}

print("Error Metrics:")
for metric, value in error_metrics.items():
    print(f"  {metric}: {value:.3f}")

2. Correlation Analysis

# Correlation metrics
correlation_metrics = {
    'R²': ms.R2(observed_temps, model_temps),
    'Pearson Correlation': ms.pearsonr(observed_temps, model_temps)[0],
    'Index of Agreement': ms.IOA(observed_temps, model_temps),
    'Kling-Gupta Efficiency': ms.KGE(observed_temps, model_temps)
}

print("\nCorrelation Metrics:")
for metric, value in correlation_metrics.items():
    print(f"  {metric}: {value:.3f}")

3. Skill Score Analysis

# Skill scores relative to climatology
skill_scores = {
    'Nash-Sutcliffe Efficiency': ms.NSE(observed_temps, model_temps),
    'Modified NSE': ms.NSEm(observed_temps, model_temps),
    'Log NSE': ms.NSElog(observed_temps, model_temps),
    'Mean Absolute Percentage Error': ms.MAPE(observed_temps, model_temps)
}

print("\nSkill Scores:")
for metric, value in skill_scores.items():
    print(f"  {metric}: {value:.3f}")

Working with Different Data Types

NumPy Arrays

# Using NumPy arrays (most common)
obs = np.random.normal(20, 5, 1000)  # Normal distribution
mod = obs + np.random.normal(0, 2, 1000)  # Add noise

# All metrics work with NumPy arrays
r2 = ms.R2(obs, mod)
rmse = ms.RMSE(obs, mod)
print(f"NumPy arrays - R²: {r2:.3f}, RMSE: {rmse:.3f}")

Pandas DataFrames

import pandas as pd

# Using pandas DataFrames
data = pd.DataFrame({
    'observed': observed_temps,
    'modeled': model_temps,
    'station_id': ['STN1'] * 10 + ['STN2'] * 10,
    'time': pd.date_range('2023-01-01', periods=20, freq='H')
})

# Calculate metrics by station
results = data.groupby('station_id').apply(
    lambda x: pd.Series({
        'RMSE': ms.RMSE(x['observed'], x['modeled']),
        'R2': ms.R2(x['observed'], x['modeled']),
        'MB': ms.MB(x['observed'], x['modeled'])
    })
)

print(results)

XArray DataArrays

import xarray as xr

# Using xarray DataArrays
obs_da = xr.DataArray(
    observed_temps,
    dims=['time'],
    coords={'time': pd.date_range('2023-01-01', periods=20, freq='H')},
    attrs={'units': '°C', 'long_name': 'Observed Temperature'}
)

mod_da = xr.DataArray(
    model_temps,
    dims=['time'],
    coords={'time': pd.date_range('2023-01-01', periods=20, freq='H')},
    attrs={'units': '°C', 'long_name': 'Modeled Temperature'}
)

# Metrics return DataArray with coordinates and history metadata
r2_da = ms.R2(obs_da, mod_da)
print(f"XArray result with coordinates:\n{r2_da}")
print(f"Operation history:\n{r2_da.attrs['history']}")

Xarray Accessor (Pangeo Integration)

For the best experience in the Pangeo ecosystem, use the .monet_stats accessor. This allows for cleaner code and method chaining while automatically handling metadata and laziness.

# Compute climatology directly on the DataArray
climo = obs_da.monet_stats.climatology(freq="month")

# Calculate summary statistics on a Dataset
ds = xr.Dataset({"Obs": obs_da, "Mod": mod_da})
stats_dict = ds.monet_stats.stats()

# Chain operations
spatial_mean = obs_da.monet_stats.resample_data(freq="D").monet_stats.weighted_spatial_mean()

# Perform verification directly on the DataArray
rmse = mod_da.monet_stats.rmse(obs_da, dim="time")

# Get a comprehensive bundle of metrics as a Dataset
metrics = mod_da.monet_stats.verify(obs_da, dim="time")
print(metrics.MAE.values)

Lazy Evaluation with Dask

For datasets larger than RAM, monet-stats leverages Dask for lazy evaluation and parallel processing.

import xarray as xr
import monet_stats as ms

# Open large dataset with chunks
ds = xr.open_dataset("large_model_output.nc", chunks={"time": 100, "lat": 100, "lon": 100})
obs = xr.open_dataset("large_observations.nc", chunks={"time": 100, "lat": 100, "lon": 100})

# Compute metric lazily (returns a Dask-backed DataArray)
rmse_lazy = ms.RMSE(obs.temperature, ds.temperature, axis="time")

# The computation only happens when you explicitly call .compute()
rmse_map = rmse_lazy.compute()

Categorical Event Analysis

Precipitation Verification

# Binary precipitation data (1 = rain, 0 = no rain)
obs_precip = np.array([0, 1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 0, 1])
mod_precip = np.array([0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 1])

# Contingency table metrics
pod = ms.POD(obs_precip, mod_precip, threshold=0.1)  # Probability of Detection
far = ms.FAR(obs_precip, mod_precip, threshold=0.1)  # False Alarm Rate
csi = ms.CSI(obs_precip, mod_precip, threshold=0.1)  # Critical Success Index
hss = ms.HSS(obs_precip, mod_precip, threshold=0.1)  # Heidke Skill Score

print("\nPrecipitation Verification:")
print(f"  POD: {pod:.3f}")
print(f"  FAR: {far:.3f}")
print(f"  CSI: {csi:.3f}")
print(f"  HSS: {hss:.3f}")

Probabilistic Forecast Evaluation

# Ensemble forecast example
n_ensemble = 50
n_times = 100

# Generate ensemble forecasts
ensemble_forecasts = np.random.normal(20, 3, (n_ensemble, n_times))
observed_values = np.random.normal(20, 2, n_times)

# Calculate ensemble statistics
ensemble_mean = ms.ensemble_mean(ensemble_forecasts)
ensemble_std = ms.ensemble_std(ensemble_forecasts)

# Probabilistic metrics
crps = ms.CRPS(ensemble_forecasts, observed_values)
bss = ms.BSS(observed_values > 20, ensemble_mean > 20, threshold=0.5)

print(f"\nEnsemble Statistics:")
print(f"  Ensemble Mean RMSE: {ms.RMSE(ensemble_mean, observed_values):.3f}")
print(f"  Continuous Rank Probability Score: {crps:.3f}")
print(f"  Brier Skill Score: {bss:.3f}")

Spatial Analysis

2D Field Verification

# Create sample 2D fields (e.g., precipitation or temperature)
obs_field = np.random.exponential(5, (20, 20))
mod_field = obs_field + np.random.normal(0, 1, (20, 20))

# Ensure non-negative values
obs_field = np.maximum(obs_field, 0)
mod_field = np.maximum(mod_field, 0)

# Spatial verification metrics
fss = ms.FSS(obs_field, mod_field, window=5, threshold=10)
sal_s, sal_a, sal_l = ms.SAL(obs_field, mod_field, threshold=5)

print(f"\nSpatial Verification:")
print(f"  Fractions Skill Score: {fss:.3f}")
print(f"  SAL - Structure: {sal_s:.3f}, Amplitude: {sal_a:.3f}, Location: {sal_l:.3f}")

Advanced Analysis Methods

1. Kolmogorov-Zurbenko (KZ) Filter

The KZ filter is a low-pass filter widely used in air quality analysis to separate different time scales.

# Separate long-term trend from daily variations
filtered_data = ms.kz_filter(observed_temps, m=5, k=3)

2. Air Quality Metrics (MDA8)

Maximum Daily 8-hour Average (MDA8) is a standard regulatory metric for ozone.

# Assuming hourly ozone data
mda8_ozone = ms.mda8(hourly_ozone_da)

3. Climatology and Diurnal Cycle

# Compute monthly climatology
monthly_climo = ms.climatology(obs_da, freq="month")

# Compute average hourly profile
diurnal = ms.diurnal_cycle(obs_da)

4. Weighted Spatial Mean

For global or regional lat/lon grids, area-weighting (cos(lat)) is crucial.

spatial_mean = ms.weighted_spatial_mean(da_2d, lat_dim="lat", lon_dim="lon")

5. FFT Analysis

Fast Fourier Transform (FFT) analysis for periodicities and spectral decomposition.

# Compute Power Spectral Density
psd = ms.fft_analysis(obs_da, dim="time", output="psd")

6. Power Spectrum (Welch’s Method)

For smoother spectral estimates, use Welch’s method.

# Compute Power Spectrum using Welch's method
psd_smooth = ms.power_spectrum(obs_da, dim="time", fs=1.0, nperseg=256)

7. Anomalies

Compute anomalies by subtracting the climatology (monthly, seasonal, or daily).

# Compute monthly anomalies directly using the accessor
monthly_anom = obs_da.monet_stats.anomalies(freq="month")

8. Detrending

Remove linear or constant trends from your data.

# Remove a linear trend from a DataArray
detrended_data = obs_da.monet_stats.detrend(method="linear")

Wind Direction Analysis

Specialized Wind Metrics

# Wind direction data (degrees)
obs_wind_dir = np.array([45, 90, 180, 270, 355])
mod_wind_dir = np.array([50, 95, 175, 265, 5])

# Wind direction error metrics
wind_rmse = ms.RMSE(obs_wind_dir, mod_wind_dir)
wind_bias = ms.MB(obs_wind_dir, mod_wind_dir)

# Circular bias for wind direction
circular_bias = ms.circlebias(wind_bias)

print(f"\nWind Direction Analysis:")
print(f"  RMSE: {wind_rmse:.1f}°")
print(f"  Bias: {wind_bias:.1f}°")
print(f"  Circular Bias: {circular_bias:.1f}°")

Batch Processing

Processing Multiple Metrics

def comprehensive_analysis(obs, mod):
    """Perform comprehensive statistical analysis"""

    metrics = {
        # Error metrics
        'RMSE': ms.RMSE(obs, mod),
        'MAE': ms.MAE(obs, mod),
        'Mean Bias': ms.MB(obs, mod),
        'Normalized Mean Bias': ms.NMB(obs, mod),

        # Skill scores
        'R²': ms.R2(obs, mod),
        'NSE': ms.NSE(obs, mod),
        'Index of Agreement': ms.IOA(obs, mod),
        'Kling-Gupta Efficiency': ms.KGE(obs, mod),

        # Relative metrics
        'Mean Percentage Error': ms.MPE(obs, mod),
        'Normalized Mean Error': ms.NME(obs, mod)
    }

    return metrics

# Run comprehensive analysis
results = comprehensive_analysis(observed_temps, model_temps)

print("\nComprehensive Analysis Results:")
for metric, value in results.items():
    print(f"  {metric}: {value:.3f}")

Working with Missing Data

Handling NaN Values

# Data with missing values
obs_with_nan = np.array([15.2, np.nan, 14.5, 13.9, np.nan, 18.5, 19.1])
mod_with_nan = np.array([15.8, 16.2, np.nan, 13.4, 17.8, np.nan, 19.8])

# Monet Stats automatically handles NaN values
rmse_clean = ms.RMSE(obs_with_nan, mod_with_nan)
print(f"\nRMSE with NaN handling: {rmse_clean:.3f}")

Next Steps

Now that you’ve learned the basics:

1. Explore Advanced Features: Check out the Climate Data Analysis Workflows
2. Dive into API: Browse the complete API Reference
3. Learn Spatial Methods: Discover Spatial & Ensemble Metrics
4. Ensemble Analysis: Understand Ensemble Analysis
5. View Examples: See practical Examples

Tips for Best Practices

Data Preparation

• Always check for data quality and missing values
• Ensure temporal/spatial alignment between observations and model data
• Consider data preprocessing (normalization, transformation)

Metric Selection

• Use multiple complementary metrics for comprehensive evaluation
• Consider the specific application when choosing metrics
• Skill scores are often more interpretable than raw error metrics

Performance Optimization

• For large datasets, consider chunked processing
• Use NumPy arrays for best performance
• Leverage xarray for multidimensional data

Documentation

• Always document your analysis methods
• Include units and metadata with your data
• Consider reproducibility when setting up analysis pipelines

Getting Help

If you need assistance:

• 📖 Full Documentation
• 🐛 Report Issues
• 💬 Community Discussions
• 📧 Email: arl.webmaster@noaa.gov

This site is open source. Improve this page.