Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
97 Cards in this Set
- Front
- Back
Visible wavelengths |
Blue: 0.4 -0.5um
Green: 0.5-0.6um Red: 0.6-0.7um |
|
Infrared wavelengths and atmospheric windows |
NIR: 0.7-1.0um
SWIR: 1-3um [1.5-1.75um, 2.0-2.4um] MIR: 3-5um FIR: 5um-1mm [7-13um, but large O3 absoprtion at 9, so tend to avoid that] |
|
Explain c = v l |
c = 3 x 10"8 [speed of light]. v and l have an inverse relationship - as wavelength increases, frequency decreases. |
|
What three things is the amount of scattering that occurs dependent on? |
~ particle size and abundance ~ Atmospheric depth [sunsets are red because the light has to pass through more atmosphere b/c the angle, and only longer waves - red - can travel that far]. P.S. shadows should be pitch black, but not b/c atmos provides light. |
|
Rayleigh scattering |
Particles smaller than wavelength, makes blue over-represented by the atmosphere. Gases in the upper atmosphere. 'Clean' atmosphere scattering. Scattering inversely proportional to the fourth power of the wavelength, so the shorter the wavelength, the greater the proportion of scattering. Blue 9.4 times more scattered than red. |
|
Mie scattering |
Particle sizes are roughly equal to wavelength. Water and dust in lower atmos. Clouds scatter all visible wavelengths equally, so appear white. |
|
Non-selective scattering |
Particles much larger than incident radiation. Water droplets and large dust in lower atmos. Causes haze - reflects all wavelengths equally. |
|
Absorption |
Affects wavelengths other than visible. Energy absorbed, then put back out in longer wavelengths - energy loss. Main absorbers: O3, high atmos, UV. |
|
Refraction |
Affects geometric accuracy of RS imagery. The atmosphere is stratified, refraction is when energy moves from one density layer to another. No effect when atmos is stable. Like when putting hand in water - distortion. |
|
Indcident radiaton equation |
Reflected + absorbed + refracted. |
|
What 4 factors is spectral reflectance dependent on? |
Surface smoothness Material type Wavelength Other factors [slope etc.] |
|
What is the relationship between plants and incident radiation? [HINT: wavelengths, and water]. |
Chloropyll-a absorbs the blue wavelength. Green is relatively strongly reflected. Chrolohyll-b absorbs the red wavelength. When a deciduous leaf is becoming senescent, chlorophyll-b is breaking down, thus red is being reflected instead of absorbed. NIR is strongly reflected, this is structure based as the wavelength enters the leaf, is scattered around the air spaces, and reflected back out. As a leaf gets older, structure breaks down and there are more air spaces = higher reflectance. Red edge. Three major water absorption bands: 1.4um, 1.9um, 2.7um. Lower water content = higher reflectance. |
|
Relationship between EMR and soil? |
Soil moisture content [most influential]: wetter = less reflection, b/c water coats particles, so less space for scattering and reflection. Iron oxide: reflects more in red, absorbs more in B, G, NIR. Organic content: Inverse relationship, more organic content = less reflection. Texture: smooth particles reflect more. So finer particles = more reflectance. |
|
Relationship between EMR and water? |
EMR is mostly absorbed or transmitted. Most visible is transmitted [so: clear]. NIR: very strong absorption. |
|
Types of water reflectance |
Surface reflectance: 'sun glint', specular. Volume reflectance: from within the body of water, water quality. Bottom reflectance: floor, dominant in shallow water. |
|
Affects on water reflectance. [HINT: reflectance peak]
|
Depth: volume or bottom? Shallow = little on quality. Roughness: obvo. Material: non-organic: ↑ B, G, R, NIR Tannin [decomposing humus]: ↑ R, ↓ B, - G, NIR Chlorophyll: ↑ G, ↓ R, B, - NIR When non-organic material is present, the increase in reflectance means the reflectance peak occurs at a longer wavelength. What is a ref peak? |
|
Incoming solar radiation [HINT: orange diagram] |
3 absorbed by O3 25 reflected by clouds 19 absorbed by dust/gases in lower atmos 45 absorbed by Earth 8 reflected by Earth |
|
Emitted solar radiation [HINT: red diagram] |
113 radiated straight from ground to atmos 8 through turbulent transfer 22 evaporation 15 through atmospheric windows [yay] 49 radiated to space by atmos 98 back to earth??? I didn't write this down lol |
|
Instantaneous Field of View [IFOV] |
The solid angle/amount of land illuminated at any one instant by a sensor, same as the spatial resolution I think. |
|
Field of View [FOV] |
The total angle/area measured by the instrument in one scan - CCD/pushbroom or optical mechanical/whiskbroom. |
|
Ground Resolution Element [GRE] |
Area measured by the sensor for each IFOV [pixel]. I kind of feel like this is used interchangeably with IFOV. |
|
Swath |
Width of area measured [across/along track length]. |
|
Whiskbroom scanner [and +ve]
|
Across-track. Optical-mechanical scanner. Scan lines perpendicular to direction of platform movement. Uses rotating mirror. Less computationally intensive than the vast amounts of data from CCDs. |
|
Pushbroom scanner [and +ve] |
Along track. Charge-coupled device. Scan lines parallel to platform movement. Use multiple detectors that each record spectral info for their IFOV. More compact, and more efficient in detecting photons. Respond linearly to brightness so relationships are more consistent. |
|
Geostationary platforms |
Around 38,000km above Earth. Observe the same place, useful for weather.
+ve: high temporal resolution. -ve: low spatial and spectral resolution. |
|
Sun-synchronous platforms/polar-orbiting |
Reduce diurnal variations by moving around 1degree East every day, so observes each part of the Earth at the same local time every pass. +ve: full orbit. +ve: higher spatial resolution. -ve: needs a very precise launch. -ve: orbital decay, atmospheric drag = greater eccentricity. |
|
Panchromatic [and what sensors have them] |
Very wide waveband. On board Landsat 7's ETM+. SPOT 1, 2, 3's HRVs [0.51-0.73um] 10m. SPOT 4's HRVIVs [0.61-0.68um red] 10m. SPOT 5's HRGs had two [0.48-0.71] 5m each, combined 2.5m IKONOS 0.45-0.9um, 1m. |
|
Landsat platform |
Landsat 1 was launched in 1972. Continuity, cosistency, large-scale, medium resolution, multispectral. 185km swath. 30m spatial resolution. Orbits every 16 days. |
|
Successive pass Adjacent pass |
Successive: Same day, N-S pass, about 3000m apart. Adjacent pass: different days, day 2's adjacent pass is about 179m from day 1's. They have overlap. |
|
MSS |
80m spatial resolution. 4 wavebands: B, G, R, NIR. LS3 also had a very coarse [120m] FIR band. Landsat 1-5. Poor geospatial at first, it was only experimental. 6 bit resolution. |
|
TM |
30m spatial resolution. 7 wavebands: B, G, R, NIR, MIR, FIR, MIR. FIR band still 120m. Landsat 4-5. Scanned in both directions now, better picture, slower. 8 bit. |
|
ETM+ |
30m spatial resolution. 8 wavebands: same as TM, plus 15m panchromatic. Landsat 7 only. Better calibration, data transmission etc. Continuity with TM. 2005 scan line anomaly failure. 8 bit. |
|
OLI and TIRS |
Operational Land Imager. 30m spatial resolution. 9 wavebands. Landsat 8. Pushbroom instead of whiskbroom! Thermal Infrared Sensor, idk... |
|
Three types of sensor system |
Broad-scale coverage: coarse resolution, large scale, land cover mapping, environmental change. AVHRR, MODIS. Landsat-like systems: medium resolution, rather broad geo coverage, moderate detail. Finer-resolution satellite systems: small regions, maybe urban planning, highwy or pipeline design. Can use with GPS and GIS. IKONOS. |
|
SPOT platform |
Operated by French CNES. 1, 2, 3 HRVs, 4 HRVIRs, 5 HRGs. Sun-synchronous, 26 day intervals. Can point 27degree of nadir, so successive imagery up to 2.5 days. Two identical pushbroom sensor, move independently, can be operated from Earth. Use a large plane mirror to reflect off nadir images to sensors. 60km swath each. 8 bits. Stereo used for digital elevation models. |
|
High Resolution Visible [HRVs] |
SPOT 1, 2 & 3. Two modes: panchromatic - 0.51-0.73um, high spatial resolution - 10m. Multispectral - G, R, NIR - 10m. |
|
High Resolution Visible and Infrared high-Resolution [HRVIRs] |
SPOT 4. Three modes: panchromcatic - basically red 0.61-0.68um, 10m. Multispectral - G, R, NIR, 20m. MIR - 20m. |
|
High Resolution Stereoscopic [HRSs] |
SPOT 5. Three modes: now two panchromcatic - 0.48-0.71um, 5m each, so 2.5m together. Multispectral - G, R, NIR 10m. MIR - 20m. Same bands as HRVIR for continuity. |
|
SPOT VEGETATION |
Broad-scale coverage. 1km spatial resolution. Revisit time 1 day. B, R, NIR, SWIR. Regional to global scale vegetation coverage. |
|
AVHRR |
1.1km spatial resolution at nadir. R, NIR, MIR, FIR, FIR. Originally used for snow, water and atmosphere observations/differentiations, esp for clouds. But then it was also used for vegetation - makes sense w/ R and NIR, NDVI. Can monitor ecosystems: forests, tundra and grasslands. Agriculture assessment, continental snow cover. |
|
MODIS |
Carried on Terra and Aqua. 36 spectral bands, 3 spatial resolutions. B1&2 [R & NIR] are 250m. B8-36 [MIR and thermal] are 1000m. Red and NIR finest - NDVI? 12 bits [mental]. 2^12 = 4096, 8 times as good at 8 bit. |
|
IKONOS |
Fine-resolution. Panchromatic: spatial res 1m, 0.45-0.9um, 11-12 day revisit time. Mutispectral: spatial res, 4m, 4 bands B, G, R, NIR, 3 day revisit time. |
|
Radar shadow |
Areas behind high points invisible. More extreme in the far-range because the angle is more extreme |
|
Radar depression angle |
How far from horizontal the beam is. Shallow look angle: far-range. Steep look angle: near-range. |
|
Radar issues: Radar layover |
Because at a slant, images in the far range are going to be further than near. But because of topography, the radar may reach the top of a mountain first, so thinks that is closer to the near edge, when fore-slope actually is. Reorders to BAC. Also, thinks top of skyscraper is closer than base - actually same geographic space. |
|
Radar issues: Radar foreshortening |
Modest to high relief in mid-far range. On ground AB = BC, but radar lights near slope as brighter and steeper, and far darker and shallower. So in image AB < BC. Dependent on topography and position in image |
|
Antenna length and resolution |
For a given antenna length, a smaller wavelength - better resolution. But longer can penetrate clouds. |
|
Radar interferometry |
Using multiple, stereo images of one area from different angles to create accurate topographic maps. Can establish 'temporal baseline' to reveal changes. |
|
Real aperture systems [and pulse length jfl] |
Oldest, simplest, cheapest. Higher resolution = longer antenna. Beamwidth = wavelength/antenna length. Impracticalities. Pulse length important for both - v long, may class two features in one pulse, loss of detail. |
|
Synthetic Aperture Systems [SAR] |
Much more complex and expensive. The antenna moves, posing as a large one. "The distance the SAR device travels over a target in the time taken for the radar pulses to return to the antenna creates the large 'synthetic' antenna aperture" Longer antenna = higher res. Larger aperture = can collect more of the returned data. |
|
Band Interleaved by Pixel [BIP] |
Each row of data grid contains the DN value for each pixel for each band sequentially. B, G, R B, G, R B, G , R |
|
Band Interleaved by Line [BIL] |
Each row of the grid contains data of DN values for each band, for each image line. Line 1: BBBBBB, |
|
Band Sequential [BSQ] |
DNs for each band are stored in their entire grid, then the next band. BBBBBBBBBB |
|
Image histogram |
Graph showing number of pixels in an image in a function of their brightness values. X-axis: DN. Y-axis: frequency. |
|
Linear contrast stretch |
Finds the darkest pixel and the brightest pixel, assigns them to 0 and 255. |
|
Histogram equalisation |
The most frequent values are spread out so that the output histograms contains more uniform distribution. |
|
Scatter plot |
B1 on x-axis, B2 on y-axis, shows their DNs, pixel relationships between the DNs for both bands. Density slice adds colour to show densest parts. |
|
Cosmetic Operation: Missing Scan Line |
Noise = error. ~ Replace line with corresponding pixels on preceding line. ~ OR replace lines with average of pixels above and below. |
|
Cosmetic operation: Line Striping |
Noise = error. ~ Linear method which replaces with averages on either side ~ OR histogram matching with the other detectors. |
|
Radiometric calibration |
Transforms the spectral radiance detected by the sensor to computer compatible DNs. Needed when using radiance values from different sensors or times. |
|
Atmospheric correction |
Removes atmospheric effects of scattering. Dark Object Subtraction method. Identify low DNs, not 0, subtract that from image. Simple, but atmos can be varied, also is 0 really expected? Also Empirical Line method, and Radiative Transfer models. |
|
Atmospheric correction SONG ET AL. [2001] |
Pearl River Delta China, when classifying single image, not needed bc relative. When classifying two images from diff dates seperately, then comparing post classification, not needed. Up to 50% of NDVI data can actually be due to atmospheric effects over thin/broken vegetation cover. DOS method is good, there are many more complex ones, that try to estimate aerosol components, but these just introduce more error. |
|
Geometric correction |
Transforming the x-y dimensins of a RS image to have the same scale and properties as a selected map projection. Image coords in rows and lines, map coords in northings and eastings, cannot combine. |
|
Geometric correction: ground control points |
Locations that are found in both the image and the map, match up. Minimum of 20, and evenly distributed. |
|
Geometric correction: Image resampling: Nearest neighbour |
Assigning a DN to the point on the output matrix, based on the nearest pixel. +ve: retains original values. -ve: duplication, omission. |
|
Geometric correction: Image resampling: Bilinear interpolation |
Uses weighted average of nearest 4 pixels to output point. |
|
Geometric correction: Image resampling: Cubic convolution |
Weighted average of nearest 16 pixels. +ve: smooths out noise. |
|
Optical spectrum |
UV [0.30um] to part of thermal [15um]. Wavelengths that can be reflected and refracted using mirrors and lenses. |
|
Reflective spectrum |
0.38-3.0um [Blue to SWIR] |
|
Euclidean Distance |
Works out whether the pixel is closer to A or B, here based on the Red and NIR wavelengths. sqrt: [aRED - bRED]2 + [aNIR - bNIR]2 Simplest method. |
|
Unsupervised Classification |
Has no prior knowledge [no training sites] so just examines and groups pixes, user usually specifies number of classes. +ves: no knowledge needed. +ves: reduces human error. +ves: unique classes recognised that analyst may have missed. |
|
Supervised Classification |
Identifying 'known' targets then uses the spectral signatures to group other pixels too. Three classification method decision rules: Parellelpiped [BOX]. -ves: based on forced training sites. -ves: training data may not be representative. |
|
Supervised Classification: Parallelpiped [BOX] |
A number of stdevs are set for each training class, this creates boundaries. Pixels that fall into the boundaries are grouped there. Overlapping boundaries are the main issue. |
|
Supervised Classification: Minimum Distance to Means |
Calculate mean for each class, uses Euclidean Distance to group. Set a limit beyond which the pixel remains unclassified. Slower than BOX. Insensitive to variance within spectral values of the training data. |
|
Supervised Classification: Maximum Likelihood [ML] |
The probability of classification is calculated for each class, then pixel goes to class with highest probability. Theoretically the best, but very computationally expensive and slow. |
|
Accuracy assessment |
Collect ground reference data to check how accurate the land cover map really is. Issues: ~ choosing a reference source RS imagery? Ancillary? |
|
Classification accuracy equation |
+[Classified correctly]/100 x 100= % accuracy |
|
Optimistic and conservative bias |
Thinking more accurate than it is. |
|
Users accuracy |
Error of inclusion. Ref data/classified data |
|
Producers accuracy |
Error of omission. [# of pixels correctly classified in a class/ # of pixels in ref data in that class] x 100 = % Classified data/ref data |
|
Land cover classifications: mutually exclusive |
There is no taxonomic overlap of any classes. |
|
Land cover classifications: exhaustive |
Classifies everything |
|
Land cover classifications: hierarchal |
Sub-level classes. Residental: 1] single parent families, 2] multiple parent families. |
|
UK Land Cover Map [2007] |
Habitat based classes. Summer and Winter satellite data from Landsat TM, SPOT etc. Validation with nearly 10,000 field data points |
|
USGS land use/land cover system |
Based on Anderson's classification scheme: 4 levels, Level 1 has 9 classes. |
|
Song et al [2001] |
All of the atmos correction, SC, NDVI stuff [50%]. Also different techniques of correction. Pearl River Delta, China, used maximum likelihood, gained ground truthing from field visit. Originally had three temporal stages of agriculture but had to condense to avoid confusion. |
|
NDVI |
Normalised Difference Vegetation Index. [NIR-red]/[NIR+red]. -ves: saturates at high biomass. -ves: very sensitive to background variation. |
|
Franklin and Wulder [2002] |
Large scale mapping has been driven by deforestation, ecological disasters, urban development, and climate change. Main seasonal issues with classification: 1) bu break 2) chlorophyll absorption 3) plant moisture 4) leaf biomass 5) understory conditions. Idea of creating land cover classes coarse enough to be recognised by AVHRR, MODIS, SPOT VEG, but which could be subdivided at finer res. |
|
De Grandi et al. [2000] |
Used SAR imagery for study of tropical Africa in the Global Rainforest Mapping Project. Cut through clouds v important. Training sites were drawn on smaller scale from maps and textural info, then scaled up to whole mosaic. Accuracy assessment with TM maps was 68-70%. |
|
Homer et al [1997] |
Used 24 TM images, 14 base, 10 cloud patching. Used SC, training sites based on topographic DEM info, or ancillary land use. Post-classification modelling important b/c difficult to scale local spectral info to a whole mosaic. 36 cover types with 75% accuracy. |
|
Fuller et al. [1998] |
Land cover map of Britain used mosaiced TM data, Summer and Winter imagery combined to exploit seasonal differences. Recognised the many issues concerning scale and generalisation seriously affect detail - with mapping a of Britain. |
|
Henderson and Dawson [2008] |
Used AVHRR NDVI data from different times to evaluate how much damage an invasive species - feral goats - have caused to a Galapagos island. The 1.1km resolutions is teh most cost-effective method of veg analysis in remote areas where field ovservations would not be feasible. Suggest issue with NDVI: if a patch of endemic, endangered plants are replaced with a patch of weeds, NDVI same, ecology vastly changed. |
|
Boyd et al. [2006] |
AVHRR ENSO rainforest drought. AVHRR NDVI to study biomass and phenology. Long term dataset, unequalled in monitoring terrestrial land cover. AVHRR can use temperature in FIR and moisture detection in MIR to evaluate drought, and it's vegetation impacts. |
|
Jackson et al [2004] |
Used TM and ETM+ imagery to map vegetation water content for corn and soybeans. Wanted high spatial resolution, so opted for Landsat instead of AVHRR or MODIS etc. Supervised classification. Suggested MODIS should be used in the next step as they encountered temporal issues that would not have arisen using broad-scale |
|
Marsik et al [2011] |
Used Landsat imagery to monitor deforestation in Bolivian Amazon. Deforestation was occurring near roads from Brazil, would not have been clearly seen if using broad-scale. |
|
Foody [2002] |
Accuracy. A map is a simplification of reality, of course things won't always match up perfectly. Accuracy errors: Type of error - poor calibration, mixed pixels, unnoticed noise. In Britain land cover map, 48% of testing sites were so badly locate dthey had to spatially adjust them. Accuracy of GCPs - just another classification. Spatial distribution of error - it's not uniform, much worse at borders. Error of magnitude - all misallocations are equally weighted. Classing a grassland as shrubland is the same as classing a grassland as water. Accuracy assessments throw everything into doubt, but are they ever questioned? |
|
Mayaux et al [2004] |
Mapping the entire continent of Africa for 2000. Land cover controls flows of energy, water, gases and nutrients. Landsat was not temporally sufficient enough. Used SPOT's VEG - 1km [SWIR good for moisture], and ERS and JERS which are radar - 100m. Most provided by VEG, cloud gaps filled by radar [sensitivity to water proved invaluable]. Contested some African government on how much rainforest they claim to have left. |