Satellite Data Analysis

1) Electro-Magnetic Radiation

Remote sensing

It means to:

• observe matter characteristics and physical quantities
• use faraway sensors (satellite-based)

EM Radiation

The radiation emitted from the sun is reflected in a particular way based on what it hits.

EM radiation is made of an electric (E) and magnetic (M) fields.

wavelength: λ = _c/_nf    c = 3∙10⁸ km/h    n = refraction index

The atmosphere is transparent only at some ranges of λ:

• near the visible light (OPT - optical sensing);
• radio waves (MW - microwave sensing).

Spectral reflectance: an object creates a particular signature of light reflected.

EM Spectrum:

• ultraviolet    λ < 0,4 μm
• visible    0,4 μm < λ < 0,7 μm
• infrared    λ > 0,7 μm up to a few μm
• microwaves    λ > than mm

The most important colors to keep in mind are:

• BLUE > 0,45 μm
• GREEN > 0,55 μm
• RED > 0,65 μm

Peaks of water absorption:

• 1,5 μm
• 1,9 μm

Measuring the EM radiation

Energy = capability to perform work (Joules J).

Radiant energy = energy of an EM wave

Flux of radiant energy = energy carried by an EM wave in a time unit.

Power of the EM wave (Watts W).

Density of radiant flux = radiant flux emitted or absorbed per surface unit (W/m²).

Spectral reflectance: β(λ) = _{Power leaving}/^{Power landing}

The reflectance is computed at a particular wavelength.

Irradiance = RF of an incoming EM wave.

Emittance = RF of an outgoing EM wave.

Directionality

The measurement happens at a specific angle, so it doesn't really intercept the whole emittance. The spreading of the radiation could be uneven, so the direction of measuring should be taken into account.

Radiance = density of radiant flux emitted by a surface unit and seen through a solid angle.

Diffuse reflection = the radiation is reflected evenly (ideal scenario).

Specular reflection = the reflection happens only at a specific angle.

The real-world case is a mix of the two reflections. The less rough it is, the more the reflection is concentrated in a narrow cone.

Rayleigh Criterion

The reflection type depends on the roughness of the surface compared to the wavelength.

Critical height: hc = _λ/^{8 cos θ}

Reflection Type: h > hc → rough → diffuseh < hc → smooth → mirror

In the case of visible light data acquisition, the material should be smooth at μm scale to have specular reflection (it is the case for water).

Earth Observation Platforms

Sensor = sensing device that measures radiation and produces data.

Platform = supports the sensor and puts it in a particular position.

HAPS = High Altitude Platforms. At a certain distance, the atmosphere is too thin to allow flight.

Airborne remote sensing employs drones, planes and balloons.

Spaceborne RS uses satellites outside the atmosphere at 400-36000 Km.

With respect to airborne RS, spaceborne costs more upfront, but costs less to maintain, and there will be much more data which is produced (low per-sqm cost).

The cost also changes with respect to the resolution needed.

Spaceborne RS has two parts involved:

• a space segment: acquires and downlinks data (potentially to relays);
• a ground segment: communicates with the satellite and processes data. It can also upload new software or make other corrections.

Satellite orbits

Satellites have a geocentric orbit, with a revolution that is slower the farther away the satellite is from Earth:

• Low Earth orbits → 400-2000 Km (used for Earth observation)
• Mid Earth orbits → 2000-35780 Km
• High Earth orbits → ≥35780 Km

To keep looking at the same point for a long time, more satellites will be needed. At a certain height, the orbit becomes geosynchronous, having the same speed as the Earth turning. It is usually used for weather observations (details in the order of ~1 Km).

Period = time to complete an orbit (~hour).

Repeat cycle = After some orbits, the satellite replaces its path (~weeks).

Revisit frequency = how often the satellite views the same point with same conditions.

Coverage frequency = how often the satellite views the same point with any conditions.

Characterising data

Resolution

The smallest change in input that gives a change in the output of the measuring instrument.

Types of resolution in RS:

• wavelength → spectral resolution
• location → spatial resolution
• intensity → radiometric resolution
• time → temporal resolution

Spectral resolution

It's the capability to discriminate wavelengths → Δλ It's the parameter that distinguishes multispectral and hyperspectral imaging.

Higher spectral resolution is not always better because:

• there is more data to be handled
• it's more costly and less available
• Hughes phenomenon: the number of samples required to maintain statistical confidence and functionality in hyperspectral data for classification purposes grows exponentially.

In general, whenever possible it is more practical to use multispectral data.

Geometric resolution

Spatial information: Data related to the dimension and (relative) position of an object. With low geometric resolution, some spatial information might be lost.

Another way of analysing spatial information is not through pixels, but with the object-based image analysis paradigm.

High resolution data will cover smaller areas, be more costly, won't be available for many areas, and will be relatively recent.

For spectral data, even today the data is in a broad spectrum of resolution, because low-quality data is still useful. For spatial data, the quality of available data is all of high quality (panchromatic data optimized for spatial resolution).

Digital models

DEM (Digital Elevation Model) = generic term for a 3D model of the Earth's surface.

The data has (x,y,z) and the elevation is usually interpolated to generate a smoother surface.

A DEM could be based on a TIN (Triangulated Irregular Network), which is a series of contiguous triangles, so it is a vector representation. The points are placed only in meaningful positions, so near mountains they will be more dense.

DSM (Digital Surface Model) has buildings and vegetation.

DTM (Digital Terrain Model) = no buildings.

There are algorithms which can transform DSM acquired data into DTM.

Visualisation of data

Representing a subset of the data as an image can help human interpretation.

To represent data we need a conversion criterion to images.

For single-band data, a good criterion is to use brighter shades of gray for increasing values.

Multi-band data could be broken down into multiple single-band images, but information is scattered across them. A better approach is leveraging colors to put more information in a single image.

The tristimulus system is composed of the red, green and blue colors perceived by the human eye (almost) independently.

Multi-band data can be represented by assigning one band to each of the 3 colors.

When selecting the bands, it has to be taken into account that the human eye is less sensitive to blue so it shouldn't be used to represent fine details.

If RGB are the actual bands, it's the color representation; otherwise, it's called false-color.

Pseudocolor: Color images do not mean that multivariate data is being represented. Color can be also convenient for scalar data, for example when representing temperatures.

HSI space: Hue, Saturation, Intensity. Color information is separate from brightness information (the eye sees better brightness than color differences).