EnvirSt 371 Midterm Exam (copy)

Theory of Remote Sensing

Basics

Definitions:

• Remote Sensing: The science of deriving information about the Earth from satellite images and aerial photographs - class of images taken from an overhead perspective by means of reflected or emitted electromagnetic energy, including those based on non-visible radiation.

• Aerial Photographs: Photographs taken from an airborne perspective, such as from a kite, plane, or balloon.

• Aerial Photography: the field concerned with taking aerial photographs

• Image Interpretation: deriving information from images using visual elements

• Satellite Imagery: photos from orbiting satellites - cover large areas of Earth’s surface in multiple spectra with modest detail. Consistent return times and paths allow for data collection over time and space as well as across spectra.

  • Landsat: first earth-orbit satellite designed for observing land. Made spectral, spatial, temporal data of Earth’s surface available and analyzeable on a new scale

    • Spurred innovation of digital analysis tools now that regular, useful digital images were being produced

  • MSS: Multispectral Scanner System - became the primary sensor for Landsat 1 after its primary sensor failed. Produced reliable, high-quality, multispectral imagery that became the model for later Landsats.

• UAV: unmanned aerial vehicles

  • Uses: aerial mapping, aerial photography, wildlife counts, infrastructure monitoring,

• Photo-Interpretation: the field of analyzing photographs to derive information. precursor to remote sensing, which includes readings from nonphotographic sensors.

• Photogrammetry: the practice of making accurate measurements from photographs

Scales of Remote Sensing:

• Neighborhood Scale: high res, 4m pixels, detailed but small footprint

• City Scale: Medium res, 30m pixels

• Regional/Global Scale: coarse res, 500-1000m pixels, loss of detail but large footprints, daily coverage

History of Remote Sensing:

• Foundation: the partnership of aerial photography with the airplane! Previously, aerial photography using kites and balloons, but the airplane allowed for more controlled, more expansive coverage of the earth.

• WW1 marked the use of aerial photography systematically, and cameras designed for aerial use. Aerial photography was used for military reconnaissance.

• Military applications drove rapid development. Photogrammetric instruments were developed, and aerial photography was used for civil applications as well.

• WWII: use of the EM spectrum extends beyond the visible range - IR and microwave radiation becomes known

• Cold War: photo interpretation of aerial photography, stereoscope viewing

• Robert Colwell applies color infrared film, previously used to detect camoflauged enemies, to aerial agriculture and forestry. His research explored applying the NIR spectrum to monitor vegetation health!

Remote Sensing Imagery vs Maps: Maps are man-made, inherently annotated and made to be accessible to the viewer. Remote Sensing Imagery is raw sensor data that needs to be processed, annotated, corrected, etc before achieving the same level of accessibility

Energy

Energy Definition: the ability to do work. cannot be created or destroyed. can be absorbed, reflected, emitted

Radiation: transfer of energy in the form of EM waves or particles

Types:

• Conduction: body transfers kinetic neergy to another body through collision

• Convection: transfer of energy by circulation through gas or liquid

• EM Radiation: emitted by all objects. energy generated by changes in E levels of electrons, radioactive decay, thermal motions of atoms and molecules, etc.

  • Nuclear reactions in the sun produce a full spectrum of EM radiation

  • Consists of an electrical field and a magnetic field at right angles to each other, perpendicular to the directrion of the radiation’s movement.

The EM Spectrum

Wavelength: Distance from one wave crest to the next

Frequency: The number of crests passing a fixed point in a given period of time - often measured in hertz, which = 1 cycle per second, and multiples of hertz.

Amplitude: The height of each peak. Often measured as E levels (spectral irradiance) in watts per square meter per micrometer

Phase: the extent to which the peaks of one waveform align with another. “in phase” = aligned perfectly.

Quanta: A quantum is the smallest unit of energy

Photons: Quantums of light - tiny units of light energy

Theorys of EMR:

• Wave Theory: dependent on the frequency of radiation

  • velocity of emr - speed of light

• Particle Theory: energy as a stream of particles (photons), moving in a straight line. size of particle proportional to radiation frequency. particles striking metal cause e- emission, generating electrical current.

  • Q (radiant E) = h (planck constant) * v (frequency)

EMR Spectrum and Properties/Importance: (Raging Martians Invaded Venus Using Xray Guns)

• Radio: see below

• Micro: especially significant in active remote sensing using radar, and aircraft+satellite sensors. penetrates haze, rain, smoke, clouds, etc.

  • active or passive sensing modes

  • two regions:

    • radar: up to 100cm

    • radio: longer

• Infrared: hotter objects emit more IR. lower frequency, longer wavelengths

  • Thermal IR: conveys temperature information

  • SWIR: not subject to atmos. scattering, effective in mapping minerals, fires, crop health, and surface moisture

  • NIR: Not subject to atmospheric scattering, effective in detecting and monitoring vegetation (high reflection by leaf structure). Behaves analagous to visible light.

• Visible: range defined by the human visual system. Films, sensors, and imagery built around this. Percieved colors reflect an object’s physical characteristics and circumstances.

  • objects only emit visible light at VERY high temperatures

  • higher frequency, shorter wavelengths

• UV: almost completely absorbed by ozone in atmos. sun observed in extreme UV. In special cases, plants can absorb UV radiation and emit visible light.

• X-Rays: completely absorbed by atmos. sense earth’s aurora (charged particles from the sun energize e-s in earth’s magnetosphere, give off x-rays)

• Gamma: smallest, most energy, highest frequency. Completely absorbed by the upper atmosphere, so not available for terrestrial sensing, but can be used for sensing celestial bodies, stars, planets, galaxies.

History of Vis. Wavelengths:

• Newton conducts prism study, finds white light split into wavelengths, finds that red light alone can’t become white again, but that rainbow can be condensed back to white - white light as a combination of colors.

Laws:

• All objects above absolute zero have a temperature and emit energy. As temperature increases, E emitted increases, wavelength becomes shorter.

• c = λv : speed of light = wavelength * frequency

  • c is only constant in a vacuum, changes by medium

Radiant Flux: Rate at which photons strike a surface, in Watts. Measure of E delivered to surface in a given time.

Irradiance: radiant flux per unit area (Watts/square meter) Measure of radiation striking a surface.

Radiant Exitance: Rate at which radiation is emitted from a unit area, also Watts/sq meter

Atmospheric Transmittance: amount of radiation that reaches earth’s surface. transmitted rad/incedent rad

Atmospheric Windows: wavelength ranges in the atmosphere where energy is allowed to pass through

Water Absorption: highest at “dips” on spectrum at 1.4, 2, 2.6 in vegetation (SWIR)

Atmospheric Constituents of Bands:

• Ozone: absorbs short wavelengths, UV

• CO2: absorbs mid and infrared wavelengths

• Water Vapor: absorbs 80% in two main regions: 5.5-7, 27+

Atmosphere-Surface Interactions:

• Refraction: incoming (incedent) radiation is bent from its trajectory as it travels from one atmospheric layer to another.

  • Index of Refraction = c(vacuum)/c(medium)

• Absorption: atmosphere prevents transmission of radiation or its E thru the atmosphere - constituents absorb radiation and E.

• Reflection: photons “bounce off” surface - actually simultaneously absorbed and re-emitted

  • Specular: reflection off a flat surface, all reflected photons leave at the same angle

  • Diffuse: reflection off a complex surface (like a tree), photons leave at all different angles.

• Transmission: radiation passes through a substance without significant alteration.

  • ability to transmit energy = transmittance = transmitted rad/incident rad.

  • transmittance of films, filters, extremely important in remote sensing to capture EMR accurately.

  • plant leaves tend to reflect visible radiation but transmit significant amounts of IR.

  • a yellow object may reflect OR transmit green and red light while absorbing blue.

Atmospheric Scatter: Blue and UV regions most impacted

• Scattering: EM field is disturbed by atmospheric constituents which redirect EM energy and alter the spectral distribution of energy in the beam.

• Rayleigh: EM wave interacts with very small particles - occurs in pure air, caused by even fine dust or large gas molecules. particles are smaller than the wavelength of incoming flux. shorter wavelengths are scattered much more than longer - more energy = hit molecules emit more photons

• Mie: EM wave interacts with particles a similar size to incedent flux wavelength. ie aerosols, suspended salts, small dust particles. Influences longer wavelengths than rayleigh. Produces hazy white light seen in mist and fog. stronger effect than rayleigh.

• Non-selective scatter: particles very large compared to wavelength of incedent energy. ice crystals, water drops, large dust. all wavelengths impacted equally.

Scatter Effect on Image:

• Brighter

• Less contrast

• lower resolution

• cooler colors

• Non-Selective Active vs Passive Remote Sensing:

  • Passive Remote Sensing: detects radiation reflected by objects due to naturally generated energy sources, like the sun. Also detects radiation emitted by objects in the thermal IR spectrum (warm vs cool objects)

  • Active Remote Sensing: detects radiation reflected by objects due to artificially generated energy sources, like radar, sonar

Remote Sensing of Vegetation

Levels of Organization: Interaction of Light at Each Level:

• Pigments: molecules that absorb radiation in visible wavelengths

  • pigment color = visible emr not absorbed

  • absorbed emr causes e-s in pigments to jump to higher E levels, emitting IR radiation

• Chlorophyll Absorption and Reflection (Chloroplasts):

  • Chlorophyll is a green pigment, located in the chloroplasts.

  • they recieve light passing into the leaf and use it to fuel reactions within the plant

  • chloroplasts also contain “accessory pigments” like carotenoids which also absorb light, and pass its energy to chlorophyll.

  • chlorophyll a: most important photosynthetic agent in most green plants

  • chlorophyll b: slightly different structure, found in green leaves, algae, and bacteria

  • chlorophyll preferentially absorbs blue and red light, and reflects green, as well as emitting heat and light.

• Photosynthesis: light absorbed by chloroplasts, used to synthesize carbs from CO2 and H2O

• Fluorescence: object illuminated with radiation at one wavelength emits radiation at another. Some sulfide minerals and chlorophyll emit visible radiation when illuminated with UV. In plants, fluorescence can be a measure of photosynthesis and indicate health.

• Leaves: green and NIR scattered by chloroplasts/spongy mesophyll

  • Palisade Parenchyma: contains the largest chloroplasts - specialized for photosynthesis. amt of chlorophyll affects absorption/reflectance of visible light

  • Spongy Mesophyll Parenchyma: irregularly shaped cells surrounded by openings, with a large surface area. impact amount of NIR reflected or absorbed

  • Epidermis: cells close together without gaps. upper epidermis covered by cuticle, a waxy layer preventing water loss. The lower epidermis holds stomata which regulate movement of co2 into the leaf, and water out, which alters the thermal signature of the plant.

  • Stomata: regulate co2 and water levels

• Reflectance Controlled By

  • Leaf Pigments: pigments reflect visible light

  • Cell Structure: palisades contain high chlorophyll, mesophyll reflects NIR, stomata regulates water content and CO2

  • Water Content: drier plants have higher reflectance in IR, especially SWIR where water absorption bands lie

• Canopy: treetops. IR reflectance increases with leaf area, while visible light reflectance decreases.

  • this decrease in the visible may be due to the shadowing of lower leaves by upper leaves in the canopy.

  • the same is not true for NIR, which can be transmitted through upper layers, reflected by lower leaves, and retransmitted back through the upper leaves, enhancing NIR signal

    • absorption is directly tied to leaf area, or LAI

• Albedo: reflectance of surfaces. varies depending on species, canopy structure. Snow has a very high albedo. Grass has a relatively high albedo for vegetation.

• Leaf Area Index: LAI - ratio of leaf area to ground area. NIR reflectance is very sensitive to LAI, because leaves have high NIR transmission.

Spectral Vegetation Curve:

• Red Edge: the contrast in vegetation reflectance between low red (chlorophyll absorption) and high NIR (mesophyll scattering) creates an "edge”

• Red Shift: as crops approach maturity, the chlorophyll absorption edge shifts towards longer wavelengths due to an increase in chlorophyll a

Vegetation Reflectance Factors:

• Water: Lowers SWIR reflectance

• Soils: lower moisture content and lower organic matter content = higher reflectance

• Phenology: the study of periodic biological events as influenced by the environment - ie sprouting, flowering. changes in phenology can indicate changing climates.

• Health: a stressed plant will lose IR reflectance due to deterioration of cell walls.

Soil Reflectance Factors:

• most energy incident on soils is reflected or absorbed - low transmittance.

• main controlling factors:

  • moisture content (#1!)

  • organic content (#2!)

  • soil texture

  • soil structure

  • iron oxide content

Applications of Vegetative Sensing:

• tracking health of wild and agricultural vegetation

• determining ripeness of crops

• determining soil moisture and organic content

• assessing LAI of forested areas

• distinguishing between species on a broad scale, biodiversity surveys.

Methods of Remote Sensing

Principles of Air Photos

Resolution: the ability to separate two adjacent objects in an image spatially

• Grain: film resolution. smaller grain size, higher res

• Pixel: digital resolution. smaller pixel size/more pixels, higher res

Vertical:

• Advantages: use shadows to measure height, consistent scale

• Disadvantages: Not intuitive, unfamiliar perspective, no view of sides of structures

Oblique:

• Low: cannot see the horizon

• High: can see the horizon

• Advantages: 3-dimensional view, can easily see relative heights and sides of features, “intuitive perspctive”

• Disadvantages: harder to calculate objective height and position

Air Photo Components:

• Fiducial Marks: index marks attached to camera so that they are recorded on each photograph. where they intersect is the principal point.

• Principal Point: optical center of the image - where a line perpendicular to the center of the camera lens would intersect the focal plane

• Conjugate Principal Point: the location of one image’s principal point in an adjacent image in a stereopair.

• Nadir:

  • Ground Nadir: point vertically below the center of the camera lens

  • Photographic Nadir: the intersection betweeen the line between the ground nadir and lens center on the actual image.

Flight Characteristics/Flight Lines:

• Flight Line: line running through centers of aerial photos that marks the vehicle’s flight path

• Altitude: “H” higher altitude allows for larger area captured, less detail

• Overlap: overlap between images along the direction the plane is moving. typically 60%

• Sidelap: overlap between images along parallel paths of the plane. 20-30%.

  • Drone lap is greater when there is greater distance between drone and ground.

• Laps: avoid missing data between images, help align images together into a mosaic

• Mission Planning: consider safety of site, evaluate site beforehand, assign points and flight path of drone, establish ground control station.

• Camera-Ground Relationship: higher altitude = wider IFOV

• Lens of Angle View: wider angle = wider IFOV

• Focal Length: “f,” the optical distance from point where light rays converge (in lens) to the digital sensor (back of camera)

  • a shorter focal length means a more zoomed-out image, a wider angle

  • a longer focal length means a zoomed-in image, a narrow angle

Scale: relationship between size of image and area of earth surface portrayed

• Small Scale: large area covered, low detail

• Large Scale: small area covered, high detail

Representative Fraction: map scale/true scale

• small RF = small scale image

• large RF (closer to 1) = large scale image

• Photo Scale Equation: RF = f/H

Calculating Scale: RF = PD (distance in photo between objects)/GD (distance on ground between objects)

Viewing Perspective: perspective of camera

Central Projection: top-down, map style

Orthographic Projection: 3-d perspective

Orthophotos and Orthomaps:

• Orthophoto: corrected form of an aerial photograph, without errors of tilt and relief displacement

• Orthomap: maps made from orthophotos, with place names, symbols, and geographic coordinates. can be made up more quickly and cheaply than maps

Stereovision:

• Stereoscopy: deriving distance information (and/or height information) from two images of the same scene slightly offset.

• Principles:

  • the stereoscope holds two low-power lenses that create parallel lines of sight to allow each eye a different image of the scene

  • stereoscopic photos are taken in sequences to get overlapping views of the terrain.

  • the photos are aligned so that the flight path passes from left to right

  • the interpreter identifies a destinctive feature present in the zone of overlap, and position the images so that the feature is 2.5 inches (the distance between the eyes) apart.

  • images should merge into a single 3d image

• Advantages: pick up on subtle features not noticeable in a single image. view 2d images in 3 dimensions, allowing an enhanced interpretation of distance, texture, pattern.

• Stereoscopic Parallax: differences in the appearance of objects from one eye to the other (when a scene is viewed with only the right eye, it looks slightly different than when viewed with only the left, or both).

Scale:

• Small Scale: larger area per pixel, less detail, smaller RF

• Large Scale: smaller area per pixel, more detail, RF closer to 1

• 3 Ways To Represent Scale:

• Representative Fraction: the map scale over the true scale on the ground.

• 3 Photo Scale Equations for Representative Fractions:

  • RF = focal length/altitude

  • RF = photo distance between objects/ground distance between objects

  • RF = 1/(map distance between objects)(map scale denominator)/photo distance between two objects

Photo-Interpretation

Recognition Elements:

• Tone/Color: an object’s ability to reflect radiation depends on its materials, wetness, internal structure, pigments

  • black and white: tonal contrasts

  • color: hue, chroma, value

• Size:

  • Absolute: calculate using photo scale

  • Relative: unknown object compared to known object

• Shape: cultural objects (buildings, crop circles) vs natural objects (lakes, rivers) can have distinctive shapes

• Texture: apparent roughness or smoothness of region - pattern of highlighted and shadowed areas

• Pattern: arrangement of individual objects into distinctive recurring forms - orchard-planted trees, street grid, etc

• Shadow: can reveal aspects of size and shape. best when image is taken while sun is at a low angle, to maximize shadow. also marks objects that may otherwise be too small or camouflaged to be seen in their environment.

• Site: topographic position. ie, sewage treatment plants positioned near flowing water, orchards near large water bodies.

• Association: associating features with a larger context - a tractor implies farm work is being done nearby, etc

Context: it’s important to understand the perspective and orientation of images to avoid confusion, such as percieving images as inverted when lit from an unexpected direction, as well as the larger cultural, temporal, and topographical context the image is taken in.

Collateral Information: also ancillary information; non-image info used to assist in image intepretation. may be maps, field data, as well as intuitive knowledge of the interpreter.

Image Interpretation Keys: reference materials designed for rapid ID of features in aerial images. Consist of a collection of captioned images or stereograms, and a graphical or text description. Used especially in botany and zoology, such as the key of tree species crowns from lab 2.

Natural-Color (True-Color): Red, green, and blue bands represent the image as it is seen in the visible spectrum.

Color-Infrared (False-Color): NIR, red, and green bands, with NIR signal being represented in the red band. This allows a visual representation of how NIR signal varies across the image’s features, as well as red and green signal. Especially useful in observing vegetation and soil reflectance.

Additive and Subtractive Color Theory:

• Additive: mixing light. intermediate colors formed when two or more primaries are reflected by an object, creating “yellow” (red+green) or “purple” (red+blue), etc.

  • relevant in film exposures, matters concerning radiant E

  • equal portions of all 3 make white light (complete reflection)

  • additive primaries: red/blue/green

• Subtractive: mixing light absorbing substances. objects with subtractive primaries each absorb 1/3 of the visible spectrum, ie, yellow absorbs blue and reflects red and green, cyan absorbs red and reflects blue and green, magenta absorb green and reflects red and blue.

  • representations of colors in films and paintings formed by subtractive. matters concerning pigments, dyes, reproductions of color.

  • equal portions of all 3 make black (complete absorption)

  • Subtractive primaries: yellow/cyan/magenta

Use in Multiband Intepretation (Color Mixing for Interpretation):

• understand what wavelengths are being absorbed or reflected based on image color and brightness

Photogrammetry:

Definition: The practice of making accurate measurements from photographs

Basics:

Example measurements, tasks:

Factors that Impact Interpretation and Measurements:

• Perspective:

• Projection:

• Distortion:

  • Film/print shrinkage

  • atmospheric refraction

  • haze, clouds

  • lens imperfections

  • Aircraft movements:

    • Roll: turning around nose-tail axis. (tilting head)

      Tilt: displacement of the focal plane (plane behind camera lens that image is projected onto) so that one side is higher than another.

    • Yaw: turning around top-bottom axis (shaking head)

    • pitch: turning around left-right axis (nodding)

    • Crab: failure to orient camera to flight line - camera is pointed skewed from direction of flight line.

    • Drift: shift of aircraft from flight line due to wind, etc. camera is still pointed with flight line

• Radial Displacement: objects appear to be leaning away from focal point, especially taller objects

• Relief Displacement:

  • if surface isn’t flat, photo scale will not be uniform

    • objects above nadir elevation displaced outward

    • objects below nadir elevation displaced inward!

  • distance between where object appears in photo and where it is on the ground

• Vertical Exaggeration:

  • type of relief displacement

  • worse as flying height gets lower

  • stretches image vertically

Measuring Height in Air Photos:

• Shadow Length:

  • if we know solar altitude angle at time of image

  • h = I (shadow length) * tan(a) (solar altitude angle)

• Displacement:

  • h = altitude * displacement / distance between object and principal point

Image Correction:

DN = Digital Number

Radiometric Correction: corrects for distortion related to EMR: atmospheric effects, sensor noise, etc

Geometric Correction: corrects for spatial distortions, aligning image to a map or coordinate grid

How To Tell Which Is Needed:

Geometric Preprocessing:

• 3 Conceptual Approaches

Georectification (concept, method):

Orthorectification (concept, method):

Statistics For Digital Data:

Histograms:

• What are they?

  • figures that tell you the frequency of each value in a data set

• Why do we use them?

  • can tell you the frequency of pixel values in an image, across bands - show you the darkness/lightness of an image

Look Up Tables:

• What are they?

• What is interpolation?

Important Statistics:

• Max: highest value

• Min: lowest value

• Mean: average value

• Standard Dev: square root of the average of squares of deviations about mean of dataset — tells you how much variation there is from the mean

Enhancement Techniques:

Image correction vs enhancement:

• Correction:

  • Fix imperfections

  • remove sensor error, atmospheric correction, geometric (spatial) corrections

• Enhancement:

  • Enhance contrast, make information clearer

  • spatial, spectral, temporal

Types of Stretches:

• Linear: new minimum and maximum applied to the data, old intermediate values scaled proportionally to new values. distribute clustered values further.

• Linear 2%: trims lowest and highest 2% from histogram

• Linear 5% (aka saturation linear): trims lowest and highest 5%

• Preconcieved Notion Stretches:

• Gaussian: outputs normal distribution, bell curve. enhances contrast by distributing pixels more evenly. mean set to 128, 66% of data within ± stand dev, at 86 and 170

• Histogram Equalization: 10% of histogram within 0 - 25.5, etc, divided into 10% bins from 0 to 255.

• Piece-Wise: analyst chooses breakpoints, sets segments of histogram to chosen values. can be used to deal with difficult characteristics, match images for display purposes.

Bit Depth: more bits, more gradient between black and white

Density Slicing: dividing band into intervals, assigning each interval a color, applying to original image - emphasizes features. creates crude classification.

Threshholding: density slicing with one cutoff point - all above x is black, all below white, etc. single threshhold.

Color Mapping: color density slicing. ie, DN 10 - 85 = red, DN 85 - 255 = blue. more crude classification

Masking: remove information from image by setting portions of image to 0 or 1, etc. can mask water, clouds, out of image to focus on other features. removes variance in selected portions, making them uniform.

• Building:

• Applying:

Image Transforms - Band Arithmetic:

• What are they? perform mathematical operations on bands

• Why do we use them? enhance particular features, remove radiometric distortion, reduce illumination effects from topography

• Applications:

  • Addition: reduce noise, increase signal to noise ratio. normalization, averaging, smoothing.

  • Subtraction: change detection - before/after events

  • Division: image ratios! depict dominance of certain features

Vegetation Indices:

What are they?

What is the theory behind them?

Why do they work so well?

Common Slope-Based Indices:

• Simple Ratio: Red/Nir or Nir/Red (mostly)

• NDVI: nir - red/nir + red

• Applications, examples of uses

• Compare/contrast with other transforms, indices

Why are so many vegetation indices available?

Kauth-Thomas Tasseled Cap Transform:

Orthogonal Transforms:

Theoretical explanation, migration of pixel through growing season:

Brightness (relative components, intepretation, for all):

Greenness:

Wetness:

What it can tell you, why it’s used:

Compare to NDVI and vegetation indices:

Classification:

Classification: assigns labels to pixels, or, a semi-automated method to categorize data into specific classes or themes

Pattern Recognition:

• Spatial: who are the pixels in your neigborhood? group pixels by their proximity in shapes like rectangles (buildings, fields).

  • Examples:

• Spectral: where pixels fall on scatter plots comparing bands; finding pixels with similar spectral signatures in the same bands and grouping them. ie, green vegetation will have low red and high NIR

  • Finds the mean spectral signatures across classes, then determines key spectral bands to use for classifcation

    • Examples: vegetation high in NIR, low in red. water low in both. soil and human activity high in both, soil higher in red.

• Temporal: compare spectral signatures across readings from multiple times - multitemporal

  • Examples: find when different crops ripen - distinguish crop cover by comparing signals at time when one has ripe signals and another doesn’t

• Classification Steps:

• Preprocessing:

  • What is it? correcting, enhancing, transforming initial image

  • Common types:

  • Choose classification scheme: could be ecology-based (many vegetation classes, one class for non-vegetated) or urban-oriented (classify residential vs commercial, high-density vs low-density)

  • Collect in situ data:

  • Ancillary data sources: other maps, field data, etc

  • Image Correction: correct for distortions, atmospheric interference

  • Image Enhancements: improve contrast, visual interpretation

  • Image Transforms (as necessary):

• Thematic Classification:

  • Definition:

  • Continuous: each pixel is an individual, not homogenous with its neighbors

  • Discrete: pixels divided into specific classes, values

  • Purpose of classification:

  • First Steps:

    • Goal:

    • Class Scheme:

    • Imagery:

    • Ancillary Data:

  • Types of Classification Schemes:

  • Land Use vs Land Cover: FUNCTION vs PHYSICAL ATTRIBUTES

  • Unsupervised Classification:

    • Method:

      • data aggregated into clusters

      • analyst labels clusters

    • Advantages: good for finding patterns in the data, doesn’t require knowledge of the image initially

    • Disadvantages: algorithm has no intent, can’t focus on a desired subject. requires interpretation after processing

  • Supervised Classification:

    • Method:

      • analyst provides examples of classes - training samples

      • each pixel of the full image is compared to the training samples and labeled based on similarity

    • Advantages: more focused on analyst’s goal

    • Disadvantages: based on prior knowledge, requires image interpretation at beginning

  • Spectral Classes: groups of pixels sharing similar spectral properties

  • Thematic Classes: classes determined by analyst, like forest, grassland

  • Training Samples: examples of thematic classes given by analyst to train supervised classification algorithm

  • Generalization:

  • Types of Response Patterns or Spectral Signatures:

    • Why Are These Important:

  • Unsupervised Classification:

    • k-means algorithm:

      • analyst chooses number of clusters

      • arbitrary points are chosen

      • pixels are assigned to two classes

      • mean is calculated for each group

      • initial classes are reassigned to the means of each group

      • all pixels are reassessed for how they compare to the new means

      • pixels assigned to new groups

      • repeats until means stop moving.

    • euclidean distance algo:

    • clusters vs classes: clusters are the groupings produced by unsupervised classification, while classes are what the analyst wants as their output - clusters will be refined inti classes

      • Interpretation:

      • mixed clusters: grouping of pixels containing 2 or more thematic classes

      • problem resolution: rerun, allowing for more clusters

        • mask out problems

        • constrain classificstion to small area

    • hierarchical unsupervised classification:

      • (masking out some classes and rerunning classifier)

        • masks out some classses and reruns classifier to get more clusters in the same area

        • ex: deermine forest class, then mask out, rerun, tog et more detailed classification of non-forest areas

  • Supervised Classification:

    • supervised classification algorithms:

      • min. distance to means

        • no unclassified pixels, runs fast

      • maximum likelihood classifier:

        • no decision boundaries

        • probability of pixel’s class calculated, assigned to class with highest

        • assumes class samples are normally distributed

        • most accurate, most widely used

        • computarially intense

        • relies on normal distribution

    • statistical techniques:

      • probability:

        • finds means, variance, covariance of training data

        • assumes class samples are normally distributed

        • computes probabilities for each pixel

      • decision rules: pixels classified by comparing pixel value to upper and lower limits such as: min, max for each band, the mean ± std. dev, or user-defined boundaries

      • minimum distance to means: calculates spectral distance between a pixel and the mean for each class, assign to class with lowest distance

Sensors and Data Types

Introduction to Digital Imagery

Platforms

Satellites:

Smallsats:

Aircraft:

UAVs:

Ground Networks:

Air Photos:

• Advantages: closer to subject, control over flight path

• Disadvantages: can be dangerous and time-consuming to cover areas, plan and enact missions

Satellite Images:

• Advantages: consistent return time, always updating

• Disadvantages: expensive to send up and repair, expensive to fix

Orbits (characteristics, altitudes):

• Geosynchronous: match earth’s rotation, same spot. good for weather data. ex: communications satellites

• Near-Polar: satellites cross both polar regions during orbit. cross at 10 and 3 for good useful shadows. low earth orbit.

Scanners:

• Pushbroom (along-track):

• Whiskbroom (cross-track)

• Spectral Sensitivity:

• Signal-to-Noise Ratio:

Digital Data:

• Binary: represents data using 2 symbols, 0 and 1. one byte = 8 bits (0s and 1s)

• ASCII: integers

• Bits: 0, 1 in binary

• Bytes: 8 bit strings

• Data Storage Formats: (raster!) used to store data of multiple bands

  • Bip: band interleaved by pixel - each corresponding pixel from each band is written together. each value for pixel 1, then 2, then 3, etc

  • Bil: band interleaved by line - writes each band’s value side by side in row 1, row 2, etc

  • Bsq: band sequential - stores info one band at a time. all band 1, then all band 2.

• Data Compression: reduces amount of data needed to store info by exploiting redundancies: if values or patterns are repeated in sequence, they can be reduced to a shorthand that can be decompressed later.

Types of Resolution:

• Telemetry: the process of recording and transmitting the readings of an instrument

• Spatial: measure of the smallest separation of two objects that can be identified

  • Instantaneous Field of View (IFOV): the ground covered by the angle through which sensor is sensitive to radiation

  • Mixed Pixels: pixels containing multiple ground covers

  • Spatial Resolution vs Swath Width: better resolution (each pixel represents a smaller area on the ground) means a narrower swath width

• Spectral:

  • 3 Components:

    • Bandwidth: range of wavelengths where a sensor is sensitive

    • Number of Bands:

    • Location of Bands: in the EM spectrum

  • Signal to Noise Ratio: image quality vs sensor quality - useful signal vs background noise

• Panchromatic: single band spanning visible or false-color spectrum

• Multispectral: More than 3 bands

• Hyperspectral: Up to hundreds of bands - more detail, specfic analysis, distinction between closely related materials

• Temporal: frequency of image collection for area

  • Factors affecting temporal resolutiom: altitude, orbit type, swath width, cloud/smoke cover

  • temporal vs spatial resolution: large swath = move quickly

  • Multitemporal vs Hypertemporal:

    • Multitemporal: more than 3 images over time

    • Hypertemporal: many images over time

• Radiometric: sensitivity of sensor to detecting differences in signal strength, smallest brightness value

  • Characteristics: better = more bits, more gradient between white and black

  • Importance:

• Resolution trade-offs, ways to overcome these limitations:

  • spatial vs temporal

  • spatial vs swath width

  • high spectral resolution limits swath size

  • launch multiple identical satellites to make up for low temporal resolution!

• Available Data:

Applications of Remote Sensing

Urban, Industrial Applications

Observe traffic patterns, bottlenecks, hubs of activity

Chart urban growth and human migration

Track industrial spills and pollutants

Survey natural resources for industrial use

Disaster response - find areas of change, areas with worst damage

Urban Planning

Survey land cover, land elevation, moisture, for suitable building grounds

• areas that would need to be drained, leveled, etc

Track population activity - where are more roads needed, where do most people live, eat, work, etc

Agricultural Monitoring, Mapping, Etc, Crop Type Mapping

Monitor crop health and maturity, ripeness

Distinguish between crop types

Forest Transitions

Track deforestation over time, new growth vs old growth, forest fire damage and movement, different species

Deforestation

chimpanzee habitats, eg

Land Use and Land Cover

Land use: what is being done with the land - human activity. Farming, nature preserve, city, factory, etc

Land Cover: physical features of the land, regardless of intent. Concrete, grassland, snow, water