2. Image properties I

Learning goals

What is a ‘raster’?
Recap the “remote sensing resolutions”
Get introduced to image/raster visualization in QGIS
Explore how different land cover types behave in “true-” and “false-colour” visualizations

What is raster data?

A quick warm-up of your GIS knowledge. The two most popular spatial data formats are vector and raster data. Digital images - including remote sensing imagery - are stored in rasters.

A raster is a matrix of cells (pixels) organized into rows and columns.

We describe raster data along certain image properties:

Images recorded with our normal cameras and phones are organized in the same way as shown above. However, some properties (spatial extent, pixel resolution, …) are unqiue to spatial raster data (since it is georeferenced data).

You are familiar with common image formats such as .jpg or .png. Generally, there are hundreds of different image formats (see gdal.org), most common image formats regarding digital remote sensing imagery are illustrated in the table below:

Format name	Extension	Description
GeoTIFF	.tif, .tiff, .gtiff	TIFF + geospatial reference
ENVI	, .bsq, .bil, .bip, .dat	generic, often used in imaging spectroscopy community; Header file (.hdr) with meta data!
JPEG2000	.jp2, .j2k	used by many data providers; usually for integer values only
HDF4, HDF5	.hdf, .h4, .hdf4, .h5, .hdf5	hierarchical data format, version 4 or 5; multi-resolution raster
netCDF		Network Common Data Format; multi-resolution raster
SAVE		Standard Archive Format for Europe e.g. Sentinel-1 and Sentinel-2

The by far most common format used is the geospatial Tagged Image File Format or GeoTIFF. What is important for data formats when dealing with geospatial raster data is the ability to store spatial metadata and (usually) to save the data lossless to disc, i.e. the original recorded data is not manipulated or lost due to heavy compression (e.g. .jpg is therefore not a suitable format for remotely sensed data).

Many of the properties mentioned above are determined or influenced by the sensor system we record the data with. In the context of remote sensing this largely relates to the major resolutions of the sensor system.

Remote sensing resolutions

Recall from the lecture, that remote sensing sensor systems can be broadly characterised by four resolutions:

The four Remote Sensing resolutions that define the image data.

The spatial resolution describes the length of one side of a single image pixel. Often, the term ground sample distance (GSD) is used interchangeably, which describes the distance between two adjacent pixel centers measured on the ground.
The temporal resolution is the interval between two observations of the same spot on Earth. Note: There is a trade-off between spatial and temporal resolution as such that finer spatial resolution leads to lower observation density
The radiometric resolution describes the range and discernible number of discrete gray values a sensor can distinguish. The radiometric resolution is measured in bit (binary digit). The radiometric resolution governs the datatype that needs to be used (we take a closer look at this in the next session)
With spectral resolution we often refer to the width and number of spectral bands in the sensor system. Technically, it refers to the width of the wavelength interval of individual bands.

Unique sensor resolutions may then directly result in specific image properties as shown above. For instance, the ‘OLI’ sensor onboard the Landsat-8 satellite measures large parts of the electromagnetic spectrum with a GSD of 30 meters. Accordingly, the default pixel width and height of the projected raster data is 30 meters.

Exercise

It is time for our first hands-on on some raster data. We will use QGIS for this. Navigate to the sub-folder Landsat-8 inside the data folder for today’s session. Inside this folder are two files. Both share the obscure name LC08_L1TP_193023_20190726_20200827_02_T1_int16. Do not worry about the name for now. All that matters is that this is data recorded by Landsat-8. Chances are very high you have looked at Landsat data before when using Google Earth.

We are searching for the image file.

In which data format is the image stored?
Right-click the file and open the properties. Is it possible to retrieve metadata such as the spatial resolution from the file?

Next, open QGIS and read in the Landsat-8 image, either by clicking in the top-left panel or simply by draggin-and-dropping the file in the central map-window of QGIS.

Where on earth are we?
How does the image appear (colour)? Recall from the lecture that we can differentiate between true- and false-colur image visualization. What is your guess we are looking at by default? Hint: Take a look at the roof tops and the fields surrounding Berlin.

Now that we have clarified that remote sensing images are stored in rasters which come with specific properties and are stored in certain data formats, let us recap from the lecture how our eyes and digital photography sense light.

Sensing light

The human eye

In many ways, imaging sensors resemble the human eye - we can describe it as a ‘sensor with three bands’.

Spectral response of the human eye (Source: Wikimedia Commons).

The human eye is sensitive for electromagnetic radiation in the visible domain, meaning the blue, green and red regions of the electromagnetic spectrum. We can say the human eye is sensitive between ~400-700 nm, which is defined as the spectral range in the context of remote sensing. The human eye consists of three receptors cells (cones) for color vision, which respond to light of shorter (S), medium (M) and longer (L) wavelengths regions. The perception of colors depends on the stimulation of these cones, e.g. red is perceived when the L cone is stimulated more than the other cones. The width of the response curve of each receptor at half of the maximum sensitivity (peak) is defined as the spectral resolution, the intervals between the maximum sensitivity between receptors corresponds to the spectral sampling interval in the context of remote sensing.

Digital photography

Photos taken in the visible domain and displayed as we perceive it with our eyes are called true-colour images. The recording of light in the different wavelength regions (blue, green, red) is realized separately, i.e. the magnitude of reflected light (0-100%) is captured and stored separately in an individual image (or band). Standard digital cameras therefore represent imaging sensors with three bands:

Grayscale and RGB visualization of a digital photo of the Geography Department

Above a photograph of the Geography Department as RGB visualization and separated in its red, green, and blue band. Every band can be visualized on a grayscale from black (low values = low light intensity) to white (high values = high light intensity). You may notice that the blue sky appears brightest in the blue band, while the red facade of the old building appears brightest in the red band. Accordingly, when we combine the three single-band images into one RGB multi-band band image as shown above, the combined relative magnitude of each band consitutes to the overall colour of the objects.

Optical remote sensing

Now to optical remote sensing systems, or often also called imaging spectrometers. In essence, these are simply fancy cameras. Imaging spectrometers are capable to record light beyond the human eye’s sensitivity. Depending upon the wavelength region of the electromagnetic spectrum the sensor records, we differentiate between three major domains of remote sensing:

Optical remote sensing observes the visible light (VIS; ~380 - 700 nm), the near infrared (nIR; 0.7 - 1.3 µm) and short-wave infrared (swIR; 1.3 - 3 µm)
Thermal remote sensing observes thermal infrared radiation (tIR; 5 - 15 µm)
Radar remote sensing observes microwave radiation (1 mm - 1 m)

The electromagnetic spectrum (Source: Wikimedia Commons).

In this course we are focusing on optical remote sensing. Images obtained from sensors in this domain therefore usually have bands in the visible VIS, near-infrared (nIR) and shortwave-infrared (swIR).

We can use the visible bands acquired in the blue, green and red domains for a grayscale or a true-colour RGB visualization, just like we did with standard digital photography. However, since our sensor also records light outside our eye’s capabilities, visualizing other bands is also possible. In order to do this - say look at the near-infrared reflectance band of an image - we simply display it as a grayscale image or as false-colour RGB visualization.

True-color, false-color and grayscale visualization of optical remote sensing data

RGB visualization

The basic principle behind visualizing colors and color gradients on our computer monitor is the additive color model. Any color we see is a mix of the three primary colors, (R) red, (G) green, and (B) blue, we therefore talk about the RGB visualization. We can scale the intensity of each color. In technical terms, the intensity can be expressed as values ranging from 0 (lowest intensity) to 255 (maximum intensity) in each of the three colors. Your computer screen will adjust the illumination of R, G, B according to these values. For example, R=255, G=0, B=0 will result in an intense red color. This principle allows us to visualize 256³, or 16.8 million different colors. The above numbers are an example using an 8-bit color depth, but we can produce even more colors using for example 16 or 32-bit.

Additive RGB color model (left) and a close-up of an LCD screen displaying the letters R in red, G in green, and B in blue against a white background (Source: Wikimedia Commons).

Remote sensing images can be presented differently using the RGB visualization capabilities of the monitor. Arbitrary bands can be displayed on screen in different RGB combinations, we therefore talk about RGB-composites. Sometimes we want to look at a single band only. In this case, we can visualize a grayscale image. Internally, this is realised by assigning all channels of your screen display (RGB) the same image band (e.g. red). This then leads to a grayscale gradient from black to white.

Satellite images as single band gray, RGB-composite (true colour), RGB-composite (false colour)

Exercise

With the LC08_L1TP_193023_20190726_20200827_02_T1_int16 file opened in QGIS, take a look at the Layers panel (by default: bottom-left window in QGIS). Open the visualization setting by right-clicking on the raster object and selecting Properties or via double-click on the raster object. Navigate to Symbology to display the visualization settings:

What is the default “Render type”?
What is the current visualization, i.e. which bands correspond to which display channel (RGB)?
Based on your previous observation, are we looking at a true- or false-colour image?
How many bands are available for this image file?

Next, try to navigate to your place where you live. Why is this difficult even though you might have used Google Maps in satellite view for years?

Last, visualize the red band as single-band greyscale image (Symbology > Render type > Singleband gray).

How rich in contrast is the image? Can you still differentiate major spatial patterns (e.g. large roads, Spree)?
What is minimum and maximum digital number (DN) that is assigned to completely black and white, respectively?
How does the image representation change when you set the maximum value to 1500?
Can you determine what the pixel values (e.g. 1500) measure? Hint: It is a ratio of energy, but the values were multiplied by factor of 10,000. Accordingly, the 1500 are 0.15 on the original scale. But 0.15 what? Bananen, Birnen, Äpfel?

It appears the visualization of the single-band red band has very little contrast by default. We will now learn how to deal with low contrast appearance.

Contrast enhancement

If not specified otherwise, visualization software often assumes images contain pixel values of very low and very high intensity, althought this is often not the case with remote sensing imagery. Due to surface, recording and sensor conditions, data sets (or bands) often contain only lie between a limited range of obtainable gray values / DNs. Why? Because in theory they are designed to be able to measure objects with very little reflectance (e.g. deep water surfaces) to highly reflective sands in the Sahara desert.

Take a look at the histogram (the frequency distribution) of values measured in the red band of our image:

The red band of a satellite image with corresponding image histogram and assigned graylevel

You can see that the vast majority of values lies below 10% reflectance. Standard computer screens each have a RGB color depth of 8-bit (equivalent to 256 graylevels). The transfer of pixel values to a monitor may occur initially 1:1 (compare the x-axes). In this setting, the images are often displayed in low-contrast to a degree where differences are virtually impossible to discern. Now QGIS, for instance, at least uses the calculated minimum and maximum value from the image to assign these values to 0 and 255 on the screen. Still, as you have seen in the exercise above, this may result in poor contrast:

Schematic visualization of contrast enhancement.

We can see that our image mostly contains values at the lower end. The majority of pixels however has values below 20% reflectance, and we can see a peak of very low values of at ~3% reflectance. If we now choose to visualize the image without accounting for the image-specific distribution, we will see a dark image with poor contrast. Instead, we would like to utilize the full gray value range optimally.

Contrast enhancement describes a function that is used to transfer pixel values into on-screen gray values. There different functions we can go on about this:

Schematic representation of linear histogram stretch between min/max (top) or between +/- one standard deviation from the mean (Source: Jensen, 2011)

Often, linear contrast enhancement is used where the increase in gray value per increase in pixel value remains the same for the most relevant value ranges (top-row in figure above). We can achieve this by modifying the linear strech applied to the image, e.g. linear contrast stretch adjusted to the specific image statistics. We may calculate for example minimum and maximum value and apply a linear stretch between those values. As the extreme values (min & max) are very much influenced by outliers, we may prefer using specific percentiles (e.g. 2% and 98%), or stretch linearly between +/- two standard deviations from the mean. This may be called percentile stretch and helps to maximize the contrast within the most frequently occurring image values (bottom-row).

As an example, here we apply a linear stretch starting at the 2% percentile and ending at the 98% percentile. Compare the image contrast with the graph above:

The red band of a satellite image with corresponding image histogram and the corresponding graylevel. We applied a linear stretch between the lowest and highest 2%, highlighted by red lines.

There are more ways of applying contrast enhancement to an image. Keep in mind that all these techniques are only used to visualize images on screen. By applying them, only the screen visualization changes, the actual data remains the same (e.g. a measured 15% reflectance remains 15%).

Linear contrast enhancement describes a function that is used to transfer pixel values into on-screen gray values to make use of the full value range that can be displayed. For this, a minimum and a maximum value in the original data is defined/calculated which then represent 0 (black) and 255 (white) on our computer screen. All pixels that fall between the minimum and maximum pixel values are linearily interpolated.

Exercise

With the above explanation of contrast enhancement in mind, switch to the visualization settings and adjust the contrast (Properties > Symbology > Min / Max Value Settings). Switch between Min/Max and Cumulative count cut (leaving the deault percentile values). Which is greater in contrast and why?

Next visualize the image as true-colour multi-band. By default the Cumulative count cut should be active. Navigate to the Grunewald. Can you easily discern water from the forest? Try to set the Statistics extent (Min / Max Value Settings) from Whole raster to Updated canvas. What is (also given the wording) likely to happen here?

Assignment

The objective of this assignment is to get familiar with different RGB visualization of optical remote sensing images and to learn how different land cover types are represented in true- and false-color RGB-composites.

True- & false color RGB-composites

Create a true and false color (R = nIR, G = red, B = green) representation of the Landsat-8 image and answer the following questions:
- What are the main visual differences between the true and false color representations?
- For what phenomena and/or surfaces does the nIR channel seem to be particularly sensitive?

Visualizing surface types

In order to easy the navigation in the relatively coarse Landsat imagery, you can add Web Map Services (WMS) such as the Google Satellite layer or OpenStreetMap to QGIS. This can help you with identifying objects in greater spatial detail. To do this, simply navigate to XYZ Tiles section in the Browser window (default: center-right), right-click and select New Connection, name it Google Satellite and put https://www.google.cn/maps/vt?lyrs=s@189&gl=cn&x={x}&y={y}&z={z} into the URL box. Then, simply double-click onto the new connection to add it to your map view.

Find one example area for each of the following surfaces:
- Deciduous forest
- Coniferous forest
- Grass
- Artificial turf
Provide a screenshot of each example area in true- and false-color RGB-composite, answer the following questions:
- Are there visible differences between deciduous and coniferous forest?
- What are the differences between grass and artificial turf?

Submission

Document your answers to the questions above with bullet points and the help of your screenshots in PowerPoint, Word or another software package. Please use the following layout suggestion:
Upload your documentation as a PDF file on Moodle.
General submission notes: Submission deadline for the weekly assignment is always the following Monday at 10am. Please use the naming convention indicating session number and family name of all students in the respective team, e.g. ‘s01_surname1_surname2_surname3_surname4.pdf’. Each team member has to upload the assignment individually. Provide single file submissions, in case you have to submit multiple files, create a *.zip archive.