Field Preprocessing

The wrangler.preproc.field module provides functionality for preprocessing satellite image fields through a series of customizable operations. This document describes the available preprocessing steps, their parameters, and usage.

Main Preprocessing Function

main(field, mask, **kwargs)

Preprocesses an input field image with a series of optional steps.

Parameters:

• field (numpy.ndarray) – Input field array

• mask (numpy.ndarray) – Data mask (True = masked values)

• kwargs – Optional preprocessing parameters

Returns:

Tuple of (preprocessed_field, metadata_dict)

Return type:

tuple

Preprocessing Steps

The preprocessing steps are applied in the following order if enabled through parameters:

1. Inpainting

   Fill masked regions using a biharmonic inpainting algorithm.

   param inpaint:

   Enable inpainting

   type inpaint:

   bool

   param only_inpaint:

   Return after inpainting

   type only_inpaint:

   bool

2. Resizing

   Resize the field to a specified size.

   param resize:

   Enable resizing

   type resize:

   bool

   param field_size:

   Target size for the field

   type field_size:

   int

3. Noise Addition

   Add Gaussian noise to the field.

   param noise:

   Standard deviation of noise to add

   type noise:

   float

4. Median Filtering

   Apply a median filter to reduce noise while preserving edges.

   param median:

   Enable median filtering

   type median:

   bool

   param med_size:

   Size of the median filter window (tuple)

   type med_size:

   tuple

5. Downscaling

   Reduce the field size by local mean downscaling.

   param downscale:

   Enable downscaling

   type downscale:

   bool

   param dscale_size:

   Downscaling factors (tuple)

   type dscale_size:

   tuple

6. Mean Removal

   Normalize the field by removing the mean value.

   param de_mean:

   Enable mean removal

   type de_mean:

   bool

   param min_mean:

   Minimum required mean value

   type min_mean:

   float

7. Sigmoid Transformation

   Apply error function (erf) transformation to field values.

   param sigmoid:

   Enable sigmoid transformation

   type sigmoid:

   bool

8. Scaling

   Apply a multiplicative scaling factor to field values.

   param scale:

   Scaling factor

   type scale:

   float

9. Exponentiation

   Raise field values to a power, preserving sign.

   param expon:

   Exponent value

   type expon:

   float

10. Gradient Enhancement

    Apply a Sobel filter to enhance edges and gradients.

    param gradient:

    Enable gradient enhancement

    type gradient:

    bool

11. Logarithmic Scaling

    Apply logarithmic scaling to gradient values.

    param log_scale:

    Enable logarithmic scaling

    type log_scale:

    bool

Metadata

The function returns a metadata dictionary containing statistics about the field:

• Tmax: Maximum temperature value

• Tmin: Minimum temperature value

• T10: 10th percentile temperature

• T90: 90th percentile temperature

• mu: Mean field value

If gradient enhancement is enabled, additional metadata is included:

• G10: 10th percentile gradient value

• G90: 90th percentile gradient value

• Gmax: Maximum gradient value

Usage Examples

Basic Inpainting

from wrangler.preproc.field import main as process_field

# Load field and mask data
# ...

# Just inpaint the field
inpainted_field, _ = process_field(field, mask, inpaint=True, only_inpaint=True)

Multiple Preprocessing Steps

# Apply several preprocessing steps
processed_field, metadata = process_field(
    field,
    mask,
    inpaint=True,
    median=True,
    med_size=(3, 1),
    downscale=True,
    dscale_size=(2, 2),
    de_mean=True,
    scale=0.1
)

print(f"Mean value: {metadata['mu']}")
print(f"Temperature range: {metadata['Tmin']} to {metadata['Tmax']}")

Gradient Enhancement Pipeline

# Create a gradient-enhanced version
gradient_field, metadata = process_field(
    field,
    mask,
    inpaint=True,
    de_mean=True,
    gradient=True,
    log_scale=True
)

print(f"Gradient range: {metadata['G10']} to {metadata['G90']}")

Important Notes

• Not all combinations of parameters are valid. For example, log_scale=True requires gradient=True.

• The function will return None, None if any of these conditions occur: - Inpainting fails - The field contains NaN values after processing - The mean is below the min_mean threshold

• For resizing, the field is resized to match the specified field_size

• When using expon, the sign of values is preserved (negative values remain negative)