2009 10 CEGEG 046 GEOG 3051 Principles Practice

2009 -10 CEGEG 046 / GEOG 3051 Principles & Practice of Remote Sensing (PPRS) 6: ground segment, pre-processing & scanning Dr. Mathias (Mat) Disney UCL Geography Office: 113, Pearson Building Tel: 7670 0592 Email: mdisney@ucl. geog. ac. uk www. geog. ucl. ac. uk/~mdisney

Recap • Last week – orbits and swaths – Temporal & angular sampling/resolution + radiometric resolution • This week – data size, storage & transmission – pre-processing stages (transform raw data to “products”) – sensor scanning mechanisms 2

Data volume? • Size of digital image data easy (ish) to calculate – size = (n. Rows * n. Columns * n. Bands * n. Bits. Per. Pixel) bits – in bytes = size / n. Bits. Per. Byte – typical file has header information (giving rows, cols, bands, date etc. ) (0, 0) n. Columns (0, 0) n. Bands n. Rows n. Bands n. Columns (r, c) Time 3

Aside • Several ways to arrange data in binary image file – Band sequential (BSQ) – Band interleaved by line (BIL) – Band interleaved by pixel (BIP) From http: //www. profc. udec. cl/~gabriel/tutoriales/rsnote/cp 6 -4. htm 4

Data volume: examples • Landsat ETM+ image? Bands 1 -5, 7 (vis/NIR) – size of raw binary data (no header info) in bytes? – 6000 rows (or lines) * 6600 cols (or samples) * 6 bands * 1 byte per pixel = 237600000 bytes ~ 237 MB • actually 226. 59 MB as 1 MB 1 x 106 bytes, 1 MB actually 220 bytes = 1048576 bytes • see http: //www. matisse. net/mcgi-bin/bits. cgi – Landsat 7 has 375 GB on-board storage (~1500 images) Details from http: //ltpwww. gsfc. nasa. gov/IAS/handbook_htmls/chapter 6. htm 5

Data volume: examples • MODIS reflectance 500 m tile (not raw swath. . )? – 2400 rows (or lines) * 2400 cols (or samples) * 7 bands * 2 bytes per pixel (i. e. 16 -bit data) = 80640000 bytes = 77 MB – Actual file also contains 1 32 -bit QC (quality control) band & 2 8 -bit bands containing other info. • BUT 44 MODIS products, raw radiance in 36 bands at 250 m • Roughly 4800 * 36 * 2 ~ 1. 6 GB per tile, so 100 s GB data volume per day! Details from http: //edcdaac. usgs. gov/modis/mod 09 a 1. asp and http: //edcdaac. usgs. gov/modis/mod 09 ghk. asp 6

Transmission, storage and processing • Ground segment – receiving stations capture digital data transmitted by satellite • A: direct if Ground Receiving Station (GRS) visible • B: storage on board for later transmission • C: broadcast to another satellite (typically geostationary telecomms) known as Tracking and Data Relay Satellite System (TDRSS) From http: //www. ccrs. nrcan. gc. ca/ccrs/learn/tutorials/fundam/chapter 2_15_e. html 7

Transmission, storage and processing • Ground receiving station – – dish to receive raw data (typically broadcast in wave) data storage and archiving facilities possibly processing occurs at station (maybe later) dissemination to end users From http: //www. ccrs. nrcan. gc. ca/ccrs/learn/tutorials/fundam/chapter 2_15_e. html 8

Transmission, storage and processing • Ground receiving station, Kiruna, Sweden From http: //www. esa. int/SPECIALS/ESOC/SEMZEEW 4 QWD_1. html#subhead 1 9

Transmission, storage and processing • Scale? – can be very small-scale these days – dish or aerial for METEOSAT-type data – desktop PC and some disk space 10

E. g. MODIS direct broadcast (DB) • MODIS DB – ideal for smaller organisations, developing nations etc. – Only need 3 m dish and some hardware • Pre-processing stage can be VERY complex! Before you let users loose. . • From http: //daac. gsfc. nasa. gov/DAAC_DOCS/direct_broadcast/ 11

(Pre)Processing chain • Task of turning raw top-of-atmosphere (TOA) radiance values (raw DN) into useful information • geophysical variables, products etc. DERIVED from radiance – Can be very complex, time- (and space) consuming – BUT pre-processing determines quality of final products • e. g. reflectance, albedo, surface temperature, NDVI, leaf area index (LAI), suspended organic matter (SOM) content etc. – typically require ancillary information, models etc. – combined into algorithm for turning raw data into information 12

(Pre? ) Processing chain • Typically: – – radiometric calibration radiometric correction atmospheric correction geometric correction/registration 13

Radiometric calibration • Account for sensor response – cannot assume sensor response is linear – account for non-linearities via pre-launch and/or in-orbit calibration • On-board black body (A/ATSR), stable targets (AVHRR), inter-sensor comparisons etc. DNout DNin 14

Processing chain • Typically: – – radiometric calibration radiometric correction atmospheric correction geometric correction/registration 15

Radiometric correction • Remove radiometric artifacts – dropped lines • detectors in CCD may have failed – fix by interpolating DNs either side? – Automate? • Topographic effects? CHRIS-PROBA image over Harwood Forest, Northumberland, UK, 9/5/2004 See http: //www. chris-proba. org. uk 16

Radiometric correction • Remove radiometric artifacts – striping • deterioration of detectors with time (& non-linearities) • Filter in Fourier domain to remove periodic striping From http: //visibleearth. nasa. gov/cgi-bin/viewrecord? 7386 17

Fourier domain filtering • Filter periodic noise/aretfacts Fourier transform (to freq. domain) Convolve with Fourier domain filter Apply inverse FT From http: //homepages. inf. ed. ac. uk/rbf/HIPR 2/freqfilt. htm 18

Processing chain • Typically: – – radiometric calibration radiometric correction atmospheric correction geometric correction/registration 19

Remember? Interactions with the atmosphere R 4 R 1 target R 2 target R 3 target • Notice that target reflectance is a function of • Atmospheric irradiance (path radiance: R 1) • Reflectance outside target scattered into path (R 2) • Diffuse atmospheric irradiance (scattered onto target: R 3) • Multiple-scattered surface-atmosphere interactions (R 4) From: http: //www. geog. ucl. ac. uk/~mdisney/phd. bak/final_version/final_pdf/chapter 2 a. pdf 20

Atmospheric correction: simple • So. . need to remove impact of atmosphere on signal i. e. turn raw TOA DN into at-ground reflectance • Simple methods? – Convert DN to apparent radiance Lapp – sensor dynamic range – Convert Lapp to apparent reflectance (knowing response of sensor) – Convert to intrinsic surface property - at-ground reflectance in this case, by accounting for atmosphere 21

Atmospheric correction: simple • Simple methods – e. g. empirical line correction (ELC) method – Use target of “known”, low and high reflectance targets in one channel e. g. nonturbid water & desert, or dense dark vegetation & snow – Assuming linear detector response, radiance, L = gain * DN + offset – e. g. L = DN(Lmax - Lmin)/255 + Lmin Radiance, L Offset assumed to be atmospheric path radiance (plus dark current signal) Lmax Regression line L = G*DN + O (+ ) Target DN values DN Lmin 22

Atmospheric correction: simple • Drawbacks – require assumptions of: • Lambertian surface (ignore angular effects) • Large, homogeneous area (ignore adjacency effects) • Stability (ignore temporal effects) – Also, per-band not per pixel so assumes • atmospheric effects invariant across image • illumination invariant across image • ok for narrow swath (e. g. airborne) but no good for wide swath 23

Example: airborne data Haze due to scan angle of instruments Airborne Thematic Mapper (ATM) data over Harwood Forest, Northumberland, UK, 13/7/2003 See: http: //www. nerc. ac. uk/arsf Compact Airborne Spectrographic Imager (CASI) data over Harwood Forest, Northumberland, UK, 13/7/2003 24

Atmospheric correction: complex • Atmospheric radiative transfer modelling – use detailed scattering models of atmosphere including gas and aerosols • Second Simulation of Satellite Signal in Solar Spectrum (6 s) Vermote et al. (1997) • MODTRAN/LOWTRAN (Berk et al. 1998) • SMAC Rahman and Dedieu (1994) • FLAASH, ACORN, ATREM etc. http: //www-loa. univ-lille 1. fr/Msixs/msixs_gb. html http: //geosci. uchicago. edu/~archer/cgimodels/radiation. html 25

Atmospheric correction: complex • 6 S radiative transfer model: calculate upward and Direct + diffuse reflectance from target (we want) + downward direct and diffuse fluxes surroundings TOA reflectance, Transmitted, Path i. e. what we radiance, measure Direct & diffuse from sun Diffuse (mscatt) between ground atmos ρ* (θs, θv, Δϕ) = Top-of-atmosphere spectral reflectance, as a function of view and sun zenith θs, v and relative azimuth, Δϕ; tg = total gaseous transmission i. e. solar radiation to surface, then escaping on the way up; ρa = atmospheric reflectance, function of molecular aerosols optical properties; τ = atmos. optical depth (e-t/μs and e-t/μv = direct transmittance in sun & view directions, where μs, μv are cos(θs) and cos(θv) respectively; td(θs), td(θv) = diffuse transmittance in sun & view directions; ρc = reflectance of target (what we want); ρe = reflectance of surrounding area; S = spherical (direct + diffuse) albedo of the atmosphere i. e. 1 -ρe. S accounts for multiple 26 scattering between ground (outside target) and atmosphere…. .

Atmospheric correction: complex • Radiative transfer models such as 6 S require: – Geometrical conditions (view/illum. angles) – Atmospheric model for gaseous components (Rayleigh scattering) • H 2 O, O 3, aerosol optical depth, (opacity) – Aerosol model (type and concentration) (Mie scattering) • Dust, soot, salt etc. – Spectral condition • bands and bandwidths – Ground reflectance (type and spectral variation) • surface BRDF (default is to assume Lambertian…. ) • If no info. use default values (Standard Atmosphere) From: http: //www. geog. ucl. ac. uk/~mdisney/phd. bak/final_version/final_pdf/chapter 2 a. pdf 27

Atmospheric correction • Can measure from ground and/or use multiangle viewing to obtain different path lengths through atmos e. g. MISR, CHRIS – infer optical depth and path radiance AND aerosols – so use data themselves to infer atmos. scattering From: http: //visibleearth. nasa. gov/cgi-bin/viewrecord? 129 28

Atmospheric correction: summary • Convert TOA radiance to at-ground reflectance • VERY important to get right (can totally dominate signal) • Simple methods – e. g. ELC but rough and ready and require many assumptions • Complex methods – e. g. 6 S but require much ancillary assumptions – BUT can use multi-angle measurements to correct – i. e. treat atmosphere as PART of surface parameter retrieval problem • different view angles give different PATH LENGTH 29

Processing chain • Typically: – – radiometric calibration radiometric correction atmospheric correction geometric correction/registration 30

Geometric correction • Account for distortion in image due to motion of platform and scanner mechanism – Particular problem for airborne data: distortion due to roll, pitch, yaw From: http: //liftoff. msfc. nasa. gov/academy/rocket_sci/shuttle/attitude/pyr. html 31

Geometric correction • Airborne data over Barton Bendish, Norfolk, 1997 • Resample using ground control points – various warping and resampling methods – nearest neighbour, bilinear or bicubic interpolation. . – Resample to new grid (map) 32

BRDF effects? • Multi-temporal observations have varying sun/view angles • To compare images from different dates, need same view/illum. conditions i. e. account for BRDF effects – fit BRDF model & use to normalise reflectance e. g. to nadir view/illum. • e. g. MODIS NBAR nadir BRDF-adjusted reflectance (http: //geography. bu. edu/brdf/userguide/nbar. html) AVHRR bands 1 & 2 uncorrected Corrected to sza = 45° vza = 0 ° From: http: //www. ccrs. nrcan. gc. ca/ccrs/rd/apps/landcov/corr/brdf_e. html 33

BRDF effects? • Field measurements of BRDF: goniometer e. g. European Goniometric Facility (EGO) at JRC, & FIGO in CH – http: //www. geo. unizh. ch/rsl/research/Spectro. Lab/goniometry/index. shtml Movable sensor head: alter view zen. angle Azimuthal rail: alter view azimuth angle ASIDE: Chapter (12) in Liang (2004) book on validation, sampling; Also Jensen chapter (11) 34

Pre-processing: summary • Convert raw DN to useful information – – calibrate instrument response and remove radiometric blunders remove atmospheric effects remove BRDF effects? resample onto grid • Results in more fundamental property e. g. surface reflectance, emissivity etc. – NOW apply scientific algorithm to convert reflectance to LAI, f. APAR, albedo, ocean colour etc. 35

Sensor scanning characteristics • Range of scanning mechanisms to build up images • Different applications, different image characteristics and pros/cons for each type – scanning mechanisms: electromechanical • discrete detectors • whiskbroom scanners • pushbroom scanners – digital frame cameras 36

Discrete detectors • Mirror can rotate or scan – individual detectors record signal in different bands – How do we split signal into separate bands? • Dichroic mirror or prism Separate bands Lens Scan mirror Sensor path Dichroic mirrors Adapted from Jensen, 2000, p. 184 37

Scanning mechanisms: across track • 3 main types of electromechanical (detectors, optics plus mechanical scanning) mechanisms – across track or “whiskbroom” scanner (mechanical) – linear detectors array (electronic) – beam splitter / dichroic / prism / filters splits incoming signal into separate wavelength regions Dichroic lens/prism From Jensen, J. (2000) Remote sensing: and Earth resource perspective, p. 184 Sensor motion 38

Scanning mechanisms: across track • Whiskbroom scanner – Mirror either rotates fully, or oscillates – Oscillation can have delays at either end of scan (vibration? ) – Restricted “dwell time” requires tradeoff with no. of bands to give acceptable SNR – motion of platform and mirror causes image distortion • Diameter of IFOV on surface H – H = flying height; = nominal angular IFOV in radians – e. g. For 2. 5 mrad IFOV, H = 3000 m, D = 2. 5 x 103 x 3000 = 7. 5 m – Typically. 5 to 5 mrad - tradeoff of spatial resolution v SNR IFOV sweeps surface Adapted from Lillesand, Kiefer and Chipman, 2004 p. 332 Examples: Landsat MSS, TM and ETM, AVHRR, (MODIS) See Jensen Chapter 7 39

Scanning mechanisms: along track • Pushbroom scanner – pixels recorded line by line, using forward motion of sensor – less distortion across track but overlap to avoid gaps – No moving parts so less to go wrong and longer “dwell time” – BUT needs v. good calibration to avoid striping – Ground-sampled distance (GSD) in x-track direction fixed by CCD element size – GSD along-track fixed by detector sampling interval ( T) used for AD conversion Examples: SPOT HRVIR and Vegetation, MISR, IKONOS, Quick. Bird Sensor motion See Jensen Chapter 7 From: http: //ceos. cnes. fr: 8100/cdrom/ceos 1/irsd/pages/datacq 4. htm & J. Jensen (2000) 40

Scanning mechanisms • Central perspective / digital frame camera area arrays – Multitple CCD arrays – Silicon (vis/NIR), Hg. Cd. Te (SWIR/LWIR)? – Similar image distortion to film camera Sensor motion • distortion increases radially away from focal point From: http: //ceos. cnes. fr: 8100/cdrom/ceos 1/irsd/pages/datacq 4. htm & Jensen (2000) 41

Aside: CCD • Charge Couple Device From http: //www. na. astro. it/datoz-bin/corsi? l 1 a 42

Aside: CCD • Photons arrive (through optics and filters) and generate free electrons in CCD elements (few x 106 on a CCD) • More photons == more electrons collected • Charge coupling: CCD design allows all packets of charged electrons to be moved one row at a time by varying voltage of adjacent rows across CCD - cascade effect • i. e. Count is done at one point (lower corner) – so delay due to read time • http: //electronics. howstuffworks. com/digital-camera 2. htm • http: //www. oceanoptics. com/Products/howccddetectorworks. asp 43

Aside: CCD • Si (Silicon) CCD – vis/NIR up to ~ 1. 1 m • In. Ga. As (Indium Gallium Arsenide) – IR (~0. 9 - 1. 6 m) • In. Sb (Indium Antimonide) – mid-IR ~3. 5 - 4 m • Hg. Cd. Te (Mercury Cadmium Telluride) – IR (~10 - 12 m) 44

Summary • Ground receiving – transfer data from sensor to ground station (storage v. transmission? ) – can be small-scale these days e. g. MSG, MODIS DB etc. • Pre-processing chain – atmospheric, geometric correction, radiometric correction and calibration • can obtain raw data (level 0 product), some pre-processing (level 1) or fully processed to reflectance, radiance etc. (level 1 b/2/3 etc. ) – then REAL work begins! • Scanning mechanisms – various depending on application – have pros/cons - usual tradeoff of reliability, spatial res. V SNR and geometric distortions (see Lillesand, Kiefer, Chipman section 5. 9) – Reading – Rahman and Dedieu (1994); Vermote et al. (1997) 45