Examples¶

This page provides practical examples for using the physics_informed_ml_odw package.

Basic Usage¶

Predicting ODW from a Satellite Image¶

from sensingpy import reader
from physics_informed_ml_odw import predict_2d

# Load your satellite image
image = reader.open('path/to/satellite_image.tif')

# Predict ODW using the default ML model
odw = predict_2d(image, model='ML')

Comparing Models¶

Using Both ML and CS_ML Models¶

This example demonstrates how to compare predictions from both available models.

from sensingpy import reader, plot
from physics_informed_ml_odw import predict_2d

# Load the satellite image
image = reader.open('data/formosa_2018.tif')

# Predict using both models
image['ODW_ML'] = predict_2d(image, 'ML')
image['ODW_CS_ML'] = predict_2d(image, 'CS_ML')

Visualization¶

Creating a Comparison Plot¶

Visualize the original image alongside ODW predictions using sensingpy's plotting utilities.

from sensingpy import reader, plot
from physics_informed_ml_odw import predict_2d
import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap, BoundaryNorm
from matplotlib.cm import ScalarMappable

# Load image and generate predictions
image = reader.open('data/formosa_2018.tif')
image['ODW_ML'] = predict_2d(image, 'ML')
image['ODW_CS_ML'] = predict_2d(image, 'CS_ML')

# Create a figure with 3 subplots
fig, axs = plot.get_geofigure(image.crs, 1, 3, figsize=(6 * 3, 6))

# Plot true color RGB composite
plot.plot_rgb(image, 'Rrs_B4', 'Rrs_B3', 'Rrs_B2', axs[0], brightness=30)

# Plot ODW predictions
plot.plot_band(image, 'ODW_ML', axs[1])
plot.plot_band(image, 'ODW_CS_ML', axs[2])

# Add titles
axs[0].set_title('True Color')
axs[1].set_title('ODW ML')
axs[2].set_title('ODW CS + ML')

# Add single discrete colorbar on the right
cmap = ListedColormap(['#440154', '#fde725'])  # viridis endpoints
bounds = [0, 0.5, 1]
norm = BoundaryNorm(bounds, cmap.N)
sm = ScalarMappable(cmap=cmap, norm=norm)

cbar = fig.colorbar(sm, ax=axs.tolist(), ticks=[0.25, 0.75], location='right', shrink=0.6)
cbar.ax.set_yticklabels(['ODW', 'OSW'])

ODW Comparison

Working with Models Directly¶

Loading and Inspecting a Model¶

from physics_informed_ml_odw import load_model

# Load the ML model
pipeline = load_model('ML')

# Inspect the model's expected features
print("Expected features:", pipeline.feature_names_in_)

# Access pipeline steps
for name, step in pipeline.steps:
    print(f"Step: {name}, Type: {type(step).__name__}")

Custom Predictions with Loaded Model¶

import pandas as pd
import numpy as np
from physics_informed_ml_odw import load_model

# Load model
pipeline = load_model('CS_ML')

# Prepare your own data
data = pd.DataFrame({
    'Rrs_B1': [...],
    'Rrs_B2': [...],
    'Rrs_B3': [...],
    'Rrs_B4': [...],
    # Include all required bands
})

# Make predictions
predictions = pipeline.predict(data)

Error Handling¶

Handling Invalid Model Names¶

from physics_informed_ml_odw import load_model

try:
    model = load_model('invalid_model')
except ValueError as e:
    print(f"Error: {e}")
    # Output: Error: Model must be one of ['ML', 'CS_ML'], got 'invalid_model'

Integration with Workflows¶

Batch Processing Multiple Images¶

from pathlib import Path
from sensingpy import reader
from physics_informed_ml_odw import predict_2d
import numpy as np

# Directory containing satellite images
image_dir = Path('data/images/')

# Process all GeoTIFF files
for image_path in image_dir.glob('*.tif'):
    print(f"Processing: {image_path.name}")

    # Load and predict
    image = reader.open(str(image_path))
    odw = predict_2d(image, model='ML')

    # Save results
    output_path = image_path.with_suffix('.odw.npy')
    np.save(output_path, odw)

    print(f"Saved: {output_path.name}")