Remote Sensing Notes

Oferim servicii de instalare, configurare si monitorizare servere linux (router, firewall, dns, web, email, baze de date, aplicatii, server de backup, domain controller, share de retea) de la 50 eur / instalare. Pentru detalii accesati site-ul BluePink.

General Remote Sensing Principles - Part 1

Introduction

Remote sensing is the science and art of obtaining information about an object, area or phenomenon through the analysis of data acquired by a device that is not in contact with it. This device can be a camera or a bank of sensors operated from an airplane or a satellite. A brief outline of basic remote sensing methodology follows.

Radiant flux from the sun, which is the primary source of energy, is modified by the earth's atmosphere and surface through the processes of scattering, reflection and absorption. Radiant flux absorption of an object leads to a thermal energy buildup, a rise in internal temperature, and the emittance of radiation. Every object on the earth reflects back part of the energy it receives from the sun and at the same time it emits its own radiation at some portion of the electromagnetic spectrum. Airborne or spaceborne sensors record the intensity of this radiation either photographically or electronically.

Different objects receiving the same radiation absorb and reflect different wavelengths of the electromagnetic spectrum, depending on their physical and chemical properties. Two identical objects at unequal temperatures and moisture levels respond differently. Every object has a repeatable characteristic reflectance pattern which can be used to represent it and is called its spectral signature. These signatures are used to identify objects.

Radiation intensity values are recorded by a sensor. Then they are processed and analyzed, using various viewing and interpretation devices. Digital data are converted to image data and then are analyzed. Inventory data about the resources being studied are used when and where available to assist in the data analysis. With the aid of this extra data, the analyst extracts information about the type, extent, location and condition of the various resources over which the sensor collected reflectance value data.

Finally, the information is presented to users who apply it to their decision making process.

Technical Information

To this date most remote sensing research has been done in the visible and near-infrared part of the spectrum (400 nm to 1um). Note that the visible spectrum occupies only a small portion of the entire electromagnetic spectrum. Microwave and radar remote sensing systems operate in the wavelength range of 1 mm to 1 m.

As mentioned earlier, the detection of electromagnetic energy can be performed either photographically or electronically. The process of photography uses chemical reactions on the surface of a light sensitive film to detect energy variations within a scene. On the other hand, electronic sensors generate an electrical signal that corresponds to variations in radiant energy received from the scene under study.

One of the most widespread techniques for data interpretation involves the transformation of the captured data into an image and the visual or computer-assisted analysis of that image. The term image is used for any pictorial representation of the data. There are two types of images: a photograph and a digital image. Although a digital image appears to be a continuous tone photograph, it is actually composed of a two-dimensional array of discrete picture elements called pixels. The intensity of each pixel corresponds to the average radiance measured electronically over the ground area corresponding to each pixel. This intensity is represented in digital numbers (DN). Digital numbers are positive integers (usually 0-255) that result from the analog-to-digital conversion of the original electrical signal recorded from the sensors.

Although this analog-to-digital conversion is extremely fast, (less than 1/100,000 second) compared to the satellite's speed it is quite slow. During this conversion time the ground distance covered by the constantly-moving satellite is the resulting spatial resolution.

LANDSAT Spatial/Spectral Resolution

Spatial resolution is described by pixel size and represents the smallest area on the ground that a sensor is able to capture its reflectance properties. Not all sensors have the same analog-to-digital translation speed. So different types of sensors have different spatial resolutions. For example, Thematic Mapper (TM) sensors have a resolution of 30m x 30m, while the previous generation of sensors, like MSS, have a resolution of 79m x 79m.

Another type of resolution is the spectral resolution. It is defined as the number of spectral intervals (channels) that are sampled from a pixel. The bigger the spectral resolution, the better the identification of the image.

		Resolution
Satellite	Sensor Type	Spatial	Spectral
Landsat 1, 2, 3 Landsat 4, 5	MSS RGV TM MSS	79m x 79m 79m x 79m 30m x 30m 79m x 79m	5 channels 4 channels 7 channels 4 channels

There are two types of remote sensing: active and passive. With active remote sensing the sensors supply their own source of energy to illuminate features of interest. Radar is an example of an active microwave sensor. It is capable of penetrating the atmosphere under virtually all conditions. On the other hand, with passive remote sensing the sensors detect the naturally available energy. For them the sun is the primary source of energy.

A 1/4 LANDSAT scene (all seven bands) has 2983 scanning lines by 4320 pixels/line for each band and covers an area on the ground 89 x 130 kilometers. The data for such a scene (all seven bands) requires 95 MB of hard disk space.

A seven band LANDSAT Thematic Mapper (TM) image with 30m by 30m spatial resolution in all bands except band 6 which has 120 by 120m. The characteristics of these bands are listed below (Lillesand and Kiefer 1987).

Band	Wavelength	Spectrum
	in micro m
1 2 3 4 5 6 7	0.45 - 0.52 0.52 - 0.60 0.63 - 0.69 0.76 - 0.90 1.55 - 1.75 10.4 - 12.5 2.08 - 2.35	Blue-green Green Red Near-IR IR Thermal Far-IR

SATELLITE REMOTELY SENSED DATA

Landsat 5 images are purchased by path and row number, acquisition date and percent cloudiness. A full scene covers an area of 170 km by 185 km. A 7-band Thematic Mapper (TM) image has 30 x 30 m spatial resolution in bands 1, 2, 3, 4, 5 and 7, and 120 x 120m in band 6.

Extraction of meaningful information from a stream of satellite data requires appropriate processing of the data through a computer so that they can become easier to interpret (Jensen 1986). The techniques used for this purpose (as well as for manipulation of other forms of digital image data) is known collectively as digital image processing. The processing of remotely sensed data can be grouped into four classes: enhancement, classification (supervised and unsupervised), geometric and radiometric correction. Enhancements can be spectral, spatial or multichannel. A commonly used process to enhance digital images involves algebra, the ratioing of two or more bands.

Digital image processing requires powerful computers and computer programs. ERDAS IMAGINE (ERDAS Inc. 2000) is a popular image processing software package that can produce files that can be imported into a GIS such as ArcView, MAPINFO and ArcInfo.

Digital image processing of satellite remotely sensed data is usually done in two phases: preprocessing and processing, although various authors differ as to the image manipulations they include in each phase.

IMAGE PREPROCESSING

Image preprocessing is the initial processing of the raw data in order to correct geometric distortions, to calibrate the data radiometrically and to remove noise (if any) from the image due to failure of the sensors, and limitations during the digitization or data recording process. As a result the need for preprocessing is case specific, depending strongly on the sensors' characteristics and the quality of the acquired image (Schowengerdt 1983, Lillesand and Kiefer 1987). The resulting corrected image is ready for visual interpretation and further manipulation and analysis.

Raw digital images usually have such significant geometric distortions that they cannot be used as maps, nor be compared with maps or to each other. These distortions are due to the Earth, the satellite, the orbit and the image projection. The contribution of the Earth comes from its rotation, oblateness, and curvature. The satellite causes image distortion by its variation in velocity, attitude, and altitude. The projection of the earth's spherical surface on a flat image and the scan skew of the sensor are also responsible for significant geometric errors (Lillesand and Kiefer 1987, Nguyen and Ho 1988). The purpose of geometric correction is to compensate for the distortions introduced by these factors, so that the image will have the properties of a map. When the image achieves the geometric integrity of a map meaningful image-to-image comparisons can be made, as well as comparisons between images acquired at different times and by different sources. Also in applications which require precise geographical positioning of ground characteristics such as cartographic mapping or analysis of certain features in specific locations, these images must be geometrically corrected in order to perform image-to-map registration (Nguyen and Ho 1988). The most widely used projection system in remote sensing is Universal Transverse Mercator (UTM). This involves projecting the earth surface on cylinders touching the earth along its meridians. This projection is well suited for Landsat imagery except in the polar regions where there is considerable distortion (Nguyen and Ho 1988).

Using the ERDAS "GEOMETRIC CORRECTION" option a LANDSAT image can be geometrically corrected and registered. A number of clearly visible small ground features, well-distributed across the original map sheet, must be located on the image. These control points are often road and/or stream intersections. Each control point is located on the displayed image and the equivalent UTM coordinates are keyed in. Three (3) different types of geometric correction (data resampling) techniques are available as follows:

The nearest neighbor resampling technique is often used because it offers the advantage of computational simplicity and does not alter the original input pixel values (Lillesand and Kiefer 1987).

Correction For Atmospheric Scattering (Haze Removal) -- Solar radiation is basically unaffected as it travels through the vacuum of space. However, during the transmission through the earth's atmosphere it is selectively scattered and absorbed (Jensen, 1986). Atmospheric scattering is the result of multiple interactions between light rays and gases (such as oxygen, nitrogen and carbon dioxide) and particles of the atmosphere (such as smoke and dust) (Sabins 1987). Atmospheric scattering produces haze, which causes the atmosphere to have a radiance of its own and results in an image with low contrast. There are three types of scattering:

depending on the size of particles and gas molecules in the atmosphere that cause it (Lillesand and Kiefer 1987). Rayleigh scattering is the primary cause of haze in the image. The effect of Rayleigh scatter is inversely proportional to the fourth power of wavelength so it strongly influences the short wavelengths in the visible region (0.4-0.7 ćm) while the longwave region ( >0.7 ćm) is practically free from scattering (Jensen, 1986, Lillesand and Kiefer 1987).

Haze is an unwanted effect, creating problems during the image interpretation process. It causes the visible bands to have falsely higher reflectance values thus reducing the detection capability of the image. This problem can be eliminated by applying a correction technique.

The correction technique is based on the fact that Thematic Mapper (TM) band 7 is essentially free from atmospheric effects. Upon examining an area in the image that is in deep shadow or a body of homogeneous deep nonturbid water, the resulting reflectance value in band 7 is either 0 or 1. A histogram of the reflectivity values in band 7 for this area starts from 0 or 1. On the contrary, a histogram of the reflectivity values in bands 1, 2 and 3 for the same area starts from much higher values as a result of haze. This offset, characteristic for each one of the three bands is subtracted from the initial reflectance values and the result is a haze corrected image (Sabins 1987).

Correction For Topographic Effects--The topographic effect is indicated on Landsat images of rugged terrain by the visual impression of relief. It is caused by the variation in spectral radiance due to surface slope and aspect variations. The difference in radiance between a horizontal and sloping surface of the same cover type provides a measure of the topographic effect (Holben and Justice 1981). Holben and Justice (1979) also measured it and showed that the effect is most extreme in areas of rugged terrain and especially for slopes in the principal plane of the sun and at low solar elevations.

Many studies have shown that topographic effect complicates greatly the correlation of stand characteristics with reflectance characteristics so they propose removal or reduction of this unwanted effect before further data processing and especially before classification (Walsh 1987, Justice et al. 1981, Woodcock et al. 1980, Spanner et al. in press).

The most widely used correction technique is band ratioing. With this technique a new variable is created by dividing the radiance value in one channel by the corresponding radiance value in a second channel (Lillesand and Kiefer 1987, Holben and Justice 1981). The rationale behind this technique is that the spectral radiance received by the sensor is equal to a multiplicative term plus an additive term. The multiplicative term is direct irradiance which is the product of direct spectral irradiance impinging the target at time t, target reflectance at time t and atmospheric transmittance at time t, and as such it is quantitative information. The additive term is basically haze which is added to the multiplicative term (Holben and Justice 1981).

The fundamental assumption of band ratioing is that the undesirable effects in the data are multiplicative and that the additive factors influencing the data are small in magnitude so they can be ignored. Ratioing of two bands cancels out the common multiplicative terms, removing the undesirable effects (Woodcock 1982).

The popularity of band ratioing is due to its simplicity and to the fact that it does not require any additional data sources. On the other hand this technique reduces the number of available image variables and often removes valuable information because it suppresses the difference in albedo; surfaces that have different albedos but similar slopes in their spectral curves may have the same ratio results. Holben and Justice (1981) showed that band ratioing reduces but does not entirely eliminate the topographic effect. They specifically showed that ratioing is not effective for reducing the topographic effect on shaded surfaces which are illuminated solely by scattered light. They concluded that the unexplained residual effects are related to scattering from surrounding terrain, path radiance, and the inappropriateness of the implicit assumption that the undesirable multiplicative effects in the data are independent of wavelength.