Remote Sensing Data Processing

• Digital Image Processing is the 8th lecture in the module "GIS and Remote Sensing for Agriculture."

Image Capture and Initial Processing

• Data is transmitted directly from the satellite or via a communication satellite to a receiving station.
• The Receiving station performs:
  • System corrections with coarse geometric adjustments for orbit variations and earth curvature.
  • Quick-looks.
  • Precision correction to a map reference system.
  • Data distribution to agents/users.

Pre-processing Steps

• Radiometric correction: DN (Digital Number) converted to Radiance, then to Reflectance.
• Atmospheric correction.
• Topographic correction.
• Geometric correction, including georeferencing.

Subsequent Image Data Processing

• Image enhancement: Colour composite, contrast enhancement, multispectral transformations, and vegetation indices.
• Filtering.
• Image classification.
• Quantitative analysis of image properties.
• Post-processing.

DN to At-Sensor Radiance Conversion

• Two methods exist, depending on the metadata.
• Method 1: Uses "Gain" and "Bias" in the formula: Lλ = gain × DN + bias, where Lλ is at-sensor radiance in W sr⁻¹ m⁻² µm⁻¹.
• Method 2: Is lengthier, using more parameters.

DN to At-Sensor Radiance: Example Calculation

• For Landsat 8 band 5, if DN = 15300, the ARad can be calculated using provided scaling values. Lb5 Formula: Lb5 = 5.9206 × 10⁻³ × 15300 + (-29.60312) = 60.98 W sr⁻¹ m⁻² µm⁻¹.

DN to At-Sensor Radiance: Alternative Method

• An alternative formula incorporating more parameters is used: .
• LMAXλ = spectral radiance scales to QCALMAX.
• LMINλ = spectral radiance scales to QCALMIN.
• QCALMAX = the maximum quantized calibrated pixel value.
• QCALMIN = the minimum quantized calibrated pixel value.

At-Sensor Radiance to At-Sensor Reflectance

• Assumes a Lambertian surface.
• Considers the zenith angle variation with location and time.
• Uses Astronomical Units (AU) to measure the Sun-Earth distance.

Reflectance Formula

• ρλ = (π × Lλ × d²) / (Eexo × cos θ)
• ρλ = At-sensor reflectance (unitless).
• π = Mathematical constant ~3.14159 (unitless).
• Lλ = Spectral radiance at the sensor (W sr⁻¹ m⁻² µm⁻¹).
• θ = Solar zenith angle (degree).
• δ = Solar elevation angle (degree).
• d = Distance Earth-Sun (AU).
• Eexo = Exoatmospheric irradiance (W sr⁻¹ m⁻² µm⁻¹).

Application of Reflectance Formula

• Landsat 8 Collection 1 DN values can be directly converted to ARef values.

Surface Reflectance (SRef)

• Three methods to correct for atmospheric effects:
  • Empirical correction methods (Flat-field, IARR, Empirical line).
  • Use of radiative transfer codes, modeling atmospheric optical behavior (e.g., ATREM, ATCOR, FLAASH).
  • In-flight calibration, which detects incoming irradiance and reflected radiance simultaneously.
• Empirical approaches output reflectivity relative to a standard scene target (relative reflectance).
• Other methods output absolute reflectance values.

Topographic Correction

• Topography strongly influences the spectral signal in rugged terrain.
• Problems arise in subsequent scene classification.

Geometric Distortions

• Can occur due to:
  • Earth rotation
  • Viewing angle
  • Earth curvature
  • Scanning time
  • Platform position
  • Non-linear minor speed
  • Topography

Geometric Corrections

• Utilize geometric models of satellite/orbit/earth.
• Use control points and polynomials for georeferencing.

Georeferencing Process

• Two transformation equations are needed to transform pixels in a distorted image to correct map coordinates.

Georeferencing Equations

• Equations are calculated with statistical models.
• i = f₁(x,y) and j = f₂(x,y)
• GCP positions are determined in the image and geometrically correct reference.
• Data pairs are used for transformation.

Quality Assessment of Georeferencing

• Root Mean Square Error (RMSE) measures the accordance between initial and corrected images.

Georeferencing Best Practices

• More Ground Control Points (GCP) improve accuracy.
• GCP properties include:
  • High contrast in the image.
  • Small size.
  • Temporal Stability.
  • Minimal height differences

Georeferencing Implementation

• Artificial elements like street crossings and building corners are the most reliable.
• GCPs should be spread to even out local distortions.
• Interpolation of new pixel values is achieved through resampling.

Radiometric Interpolation

• Geometric errors cause shifts in pixel values, influencing radiometric information.
• Transformation of pixels requires radiometric interpolation.

Methods of Radiometric Interpolation (Resampling)

• Includes Nearest Neighbor, Bilinear, and Cubic methods.

Image Enhancement

• Contrast stretching enhances the visual differences in an image.

Colour Composite Imagery

• Satellite images are usually multi-band.,
• Computers can display only 3 bands at once, corresponding to red, green, and blue colour channels.
• Different band combinations can enhance specific features.

Types of Colour Composites

• True Colour Composite (TCC): Simulates natural colours by assigning red, green, and blue satellite bands to the corresponding computer channels
• False Colour Composite (FCC): Uses different band assignments, like assigning the near-infrared (NIR) band to the red channel.

Spatial Feature Manipulation

• The aim of Spatial filtering is to emphasize or deemphasize image data frequencies.
• Spatial filtering is based on values of the neighboring pixels
• Moving windows (kernels) are used.
• Window sizes, such as e.g. 3x3 or 5x5, are used to modify pixel values when highlighting unique characteristics.

Spatial Filtering Application

• Values inside the window differ and are used to enhance unique qualities.
• Convolution filters.
• Emphasize specific areas.