TODO(instructor): the course plan suggests moving the essential remote-sensing material (spectrum, bands, reflected near-infrared) ahead of the NDVI lab so students meet it before Lab 2 rather than after. That is an instructor decision and has not been made.

Concepts lecture. Everything here is about how a sensor turns energy into numbers, and what those numbers let you measure. The lab that applied it was Lab 2, NDVI, which they have already done; this hour is the physics underneath it. The thread through the whole hour is the spectrum: every section comes back to it. Section 2 is a camera as a three-band sensor, section 3 is bands as stacked grids, section 4 is hundreds of narrow slices of the same spectrum, section 5 pulls real images apart band by band, and section 6 puts the sensor in orbit. The speaker note attached to this slide in the source PowerPoint was a leftover ModelBuilder workshop abstract from another deck (ArcGIS 9 era) and had nothing to do with remote sensing. It was removed during conversion.

stamp:begin

stamp:end

Set expectations: this is a "how the data get made" lecture. Nothing here is software-specific, but it is what makes the band math in Lab 2 mean something.

The point to land: visible light is a sliver of a very wide spectrum, and a remote sensor is simply an instrument built to measure some other part of it. Walk the labels left to right: short wavelengths carry more energy per photon, and the atmosphere absorbs most of the ultraviolet and everything shorter, so Earth imaging lives from the visible rightward. Near-infrared sits just past 700 nm and is reflected sunlight, the band NDVI used in Lab 2. Thermal infrared is further out (roughly 8 to 14 µm for Earth imaging) and is heat the surface emits. Radar is different again: the satellite supplies its own energy, which is where next week's active-sensor story starts. The figure is a still from a short MonkeySee explainer video. The video link that used to sit on this slide is dead and was removed on 2026-09-23; the labels were added in its place.

Worth saying out loud, because it is the single most common confusion in this material: *reflected near-infrared* is sunlight bouncing off a surface, exactly like visible light — it is not heat. *Thermal infrared*, further to the right on this chart, is energy the surface emits because of its own temperature. Vegetation looks bright in the near-infrared because leaves reflect it, not because plants are warm. Diagram credit: Louis E. Keiner, Coastal Carolina University.

This slide joins the last two to the physics. The strip across the top uses the same names as the spectrum slides: visible, then near-infrared, then shortwave infrared. The shaded column is the visible, and it is exactly where the sun's curve peaks: our eyes evolved to use the part of the spectrum where there is the most light and the atmosphere lets it through. Then the dashed lines: at 1.4 and 1.9 µm water vapor absorbs essentially all of the sunlight, so no satellite puts a band there. Ask why nobody builds a satellite sensor at 1.4 µm before you say it. This is why band choices on real satellites look arbitrary until you put this curve behind them. Thermal infrared, around 10 µm, is off the right edge of this chart; that energy is emitted by the Earth, not reflected from the sun.

The bridge from section 1: a digital camera is a remote sensor with three bands, all inside the visible slice. Each band window gets its own number for every pixel. The windows drawn here are schematic, not a particular camera's filter curves. Everything in this section is those three numbers.

Digital photos are raster images. Each pixel has a different value from the one next to it, representing a different color. Raster works really well for digital photos. Zoom far enough into any photo and the picture stops being a picture and becomes a grid of numbers: one for the red window, one for green, one for blue.

Two hex digits per channel, and the dashed boxes show which pair belongs to which band. Walk through #FF0000, #00FF00, #0000FF on the board if the class has not seen hex before.

0 through 255 in decimal, 00 through FF in hex. Same number, different base. Cover the swatches and have the class predict each color from its three pairs before you reveal it: purple is equal red and blue with no green; orange is full red, about two-thirds green, no blue; pure red is one band full and the other two empty. Black is all three bands at zero: no light at all.

Answer: yes, 24-bit color is comfortably more than the eye can resolve, by more than a factor of ten. The grid on the right is a slice of the hex color space — neighboring swatches differ by one step in one channel, and most of those steps are invisible.

In the source deck this said "open an image in ArcGIS" — ArcMap-era wording, updated to ArcGIS Pro. In ArcGIS Pro, a multiband raster comes into the Contents pane as one layer with a band list; the Symbology pane is where you choose which band drives red, green, and blue.

Read it left to right. Each sheet of grid paper is one band: the same 4 by 4 grid of pixels, but each sheet holds only one number per cell, how much red, green, or blue light that pixel sent back. Stack the three and every pixel now has three numbers, which is exactly the hex color from the last section: the roof pixel is R 200, G 60, B 50, or #C83C32. Run it backwards too, because that is the point for the rest of the hour: any full-color image can be pulled apart into its bands, and a satellite image is the same idea with more sheets. The pixel values are illustrative, chosen to look like water, grass, a red roof, and pavement. The figure is generated by tools/week04_remote_sensing_figures.py.

Three slides, one scene, three bands, all drawn in ArcGIS Pro with the SAME grayscale stretch, so a brighter pixel really did reflect more light. Set the pattern here: a "band" is one wavelength window, stored as its own grid of numbers, one sheet from the last slide. The scene is the Lab 2 scene: Landsat 8 OLI, LC08_L2SP_038032_20250712_02_T1 (path 38, row 32, 12 July 2025), Collection 2 Level-2 surface reflectance, clipped to Utah Lake, Provo, and Orem. The circles at the lower left are center-pivot fields west of the lake. Band 4 is the red half of the NDVI they computed in Lab 2. The labeled numbers are surface reflectance (fraction of sunlight reflected, 0 to 1), sampled from the band rasters: a center-pivot field 0.06, bare ground on the west side 0.17. The stretch is digital number 7,700 to 18,800 (reflectance about 0.01 to 0.32) on all three visible bands. Band images rendered by ArcGIS Pro from C:\Ames\Week04\RemoteSensing.aprx (tools/week04_arcgis_project.py and tools/week04_arcgis_render.py).

Ask the class what changed before you say anything. Only the labels that changed are on this slide: Utah Lake at 0.16 (0.10 in red) and the pivot field at 0.08 (0.06 in red). Utah Lake is shallow and turbid, and in July it carries algae, which is why the water is brighter in green than in any other band. Plants reflect a little more green than red or blue, which is why they look green, but far less than they reflect in the near-infrared (two slides on).

Same scene again, blue band, same stretch. Bare ground falls to 0.10 (0.17 in red), and Provo Bay, the darker lobe on the east side of the lake, is the darkest water at 0.03. Blue is also the band the atmosphere scatters most, which is why it is the band most affected by haze and smoke; that comes back in the wildfire slide. The next slide is the band that makes vegetation obvious.

The pivot field that was 0.06 in red is 0.48 here: healthy leaves scatter near-infrared strongly, which is exactly what NDVI = (NIR - Red) / (NIR + Red) measures. The lake is 0.01. Caution on the stretch: this band is drawn with its own stretch (digital number 7,000 to 26,300, reflectance about 0.0 to 0.52). On the stretch the visible bands share, every field and hillside saturated to white. Compare the labeled reflectance numbers across slides, not the brightness.

Same stack of sheets as the RGB slide. Left figure: a multispectral sensor samples the spectrum at a few chosen bands (the bar chart), a hyperspectral sensor samples it so finely that the result is a continuous curve. That is why hyperspectral imaging is also called imaging spectroscopy: every pixel is a spectrometer reading. Lower right: drill down through the cube at one pixel and you get that pixel's spectrum; soil, vegetation, and water each have a shape you can match against a library. The trade-off, per the Edmund Optics note: hyperspectral is best when the difference you care about is subtle and lies along a continuous spectrum; multispectral carries less data, so capture, processing, and analysis are faster, and its bands can be placed exactly where an application needs them (Landsat's red and near-infrared for NDVI). Worth one sentence if there is time: sensors build the cube by scanning, one point at a time (whiskbroom), one line at a time (pushbroom, as on Landsat 8 and 9), one wavelength at a time (area scanning), or all at once (snapshot). The original slide said "Can't open in ArcGIS…". That is no longer true. ArcGIS Pro's supported raster formats list (checked 2026-09-23) reads ENVI header files (.hdr, with .dat, .img, .raw, or .bsq data), AVIRIS (.hdr), HDF4 and HDF5, netCDF, EMIT (.nc), and Hyperion (.tif); PRISMA, EnMAP, and DESIS are not on the list. If a student asks: yes, ArcGIS Pro opens a hyperspectral cube as one many-band raster. Sources. Left figure: Edmund Optics application note (Figure 4), cropped above its caption; captured 2026-09-24. Cube figure: Peg Shippert, "Introduction to Hyperspectral Image Analysis," Online Journal of Space Communication, vol. 2, issue 3 (Figure 3, caption cropped). The deck used to cite it at satjournal.tcom.ohiou.edu/pdf/shippert.pdf, which is dead; the journal now lives at Ohio Open Library (the archived PDF was checked to be the same article).

This is the cube, unfolded along the wavelength axis from the spectrum slides. Top row: your eye, three wide overlapping bands, all in the visible. Middle row: Landsat 8/9's Operational Land Imager, eight bands in this range, each a few tens of nanometers wide; band 4 (red) and band 5 (near-infrared) are the pair you used for NDVI in Lab 2. Band 9 sits deliberately inside the 1.38 µm water-vapor notch: it sees only high cirrus cloud, because the atmosphere hides the ground. Bottom row: a hyperspectral sensor with 224 contiguous bands of about 10 nm, the AVIRIS layout. The gray columns are the same water-vapor notches as the sunlight-at-the-ground slide. Landsat's thermal bands (10 and 11, near 11 and 12 µm) are far off the right edge. The figure is generated by tools/week04_remote_sensing_figures.py from the USGS band designations.

Same move as the grid-paper slide, on real pixels. The tower's shadow is also how you get height out of a straight-down image. The band images are ArcGIS Pro renderings: the JPEG added to a map as three single-band layers (Band_1, Band_2, Band_3), each symbolized as Stretch, Minimum Maximum, with Edit min/max values set to 0 and 255 and a white-to-red, white-to-green, or white-to-blue color scheme. White-to-color was chosen because it reads better on a white slide; black-to-color is the physically literal version (0 means no light), and it is the one whose three bands add back up to the photo, which the next slide shows in the Symbology pane. Source of the photo noted in the original deck: llll20.wordpress.com/2007/06/09/very-cool-google-satellite-maps/

Optional: skip this slide if time is short, or do it live instead (about two minutes, project C:\Ames\Week04\RemoteSensing.aprx, map Eiffel, saved with the three layers blended). Why the recombination is exact: each layer has light in only one channel, and Screen combines layers channel by channel, so red + green + blue layers give back R, G, B. Checked in ArcGIS Pro 3.7.1: the two-color ramp is set in the Color Scheme Editor (Color scheme dropdown → Color scheme properties…, Continuous Color Scheme, one stop black and one stop red). Layer Blend on the Raster Layer tab (Effects group) offers Normal, the lightening modes Screen, Color Dodge, Lighten, and Linear Dodge, and darkening, comparison, and divergent modes. Linear Dodge (add) gives the same result here. This slide deliberately uses black-to-color, not the white-to-color of the band slides: with a black ramp, 0 adds nothing, so three layers sum back to the photo. White-to-color reads better on a slide, but white is full light in every channel, so those layers do not recombine.

A winter storm over Maryland and the mid-Atlantic, January 2009. The lesson is that the answer to "snow or cloud?" is on the band ruler, not in the picture: you need a band outside the visible. That is what the Normalized Difference Snow Index is built on. The state-line bullet was corrected on 2026-09-24: it used to say the lines vanish from the blue band, which is backwards. Measured on this image, the yellow lines average about R 229, G 236, B 87: as bright as snow in red and green, so they vanish there, and low in blue, so the blue band is where they stand out. Source noted in the original deck: weblogs.marylandweather.com, January 2009. Band images rendered in ArcGIS Pro, set up as on the Eiffel Tower slide.

Numbers from this image: open desert averages about R 198, G 171, B 142; smoke over open water about R 88, G 112, B 123, against clear water at R 8, G 18, B 28. The smoke adds more to the blue band than to the red. Point back at the blue-band slide. Source noted in the original deck: andrewlias.blogspot.com, October 2007. Band images rendered in ArcGIS Pro; the pixel values in ArcGIS Pro match these numbers.

Night lights are used as a proxy for population and for economic activity. The orange of older European street lighting comes from sodium-vapor lamps, which put nearly all their light near 589 nm, between green and red on the spectrum; the switch to white LEDs is adding blue to night imagery. Caution: this is a rendered composite (a day side and a night side on one globe), not a single satellite frame, so the colors are the renderer's choice. Use it for the band logic, not as a measurement. Source noted in the original deck: strangetravel.com. Band images rendered in ArcGIS Pro.

Open discussion. Draw out that "false color" means someone chose which band drives red, green, and blue: the same three sheets, but one of them is from outside the visible. TODO(instructor): this image came into the deck from a "cool satellite photos" link and its subject, sensor, and location are not recorded anywhere in the source. Identify and attribute it, or replace it, before using it as a discussion prompt.

The whole lecture on one diagram; walk the numbers. (1) Sunlight arrives, peaking in the visible: the dashed black-body curve. (2) The atmosphere absorbs and scatters it on the way down and again on the way back up, which is why bands sit in the windows. (3) Each surface reflects each band differently: that difference is the signal, and it is what NDVI exploits. (4) Warm surfaces also emit thermal infrared, day and night. (5) The sensor counts the energy in each band window and writes one grid of numbers per band, the sheets from section 3. All of this is passive: the sun supplies the energy. Next week LiDAR supplies its own. Radar satellites such as Sentinel-1 do the same with microwaves, which is why they see through cloud and work at night.

Low and close means sharp pixels but a long wait to come back; high and far means you never look away but each pixel covers kilometers. Landsat flies at about 705 km, Sentinel-2 at about 786 km. Weather satellites like GOES are geostationary, which is why the Katrina infrared loop a few slides on could be refreshed every few minutes.

No satellite wins on all four. Sharp pixels cost coverage and revisit; frequent revisit costs pixel size. Radiometric resolution ties back to the hex slides: an 8-bit band has 256 levels, while Landsat 8 records 12 bits (4,096 levels) and Landsat 9 records 14 bits. Landsat has an unbroken record back to 1972, and Landsat and Sentinel-2 data are free, which is why most engineering work starts there. MODIS status (checked 2026-09-23, NASA Earthdata "Transition from MODIS to VIIRS"): Terra and Aqua are still delivering but drifting from their orbits, and NASA says both platforms begin shutting down in late 2026 or early 2027. VIIRS takes over daily global imaging; Suomi NPP data delivery ends 1 November 2026, so the table names NOAA-20 and NOAA-21. There is no planned successor for Terra's morning overpass. VIIRS: 16 moderate bands at 750 m, 5 imagery bands at 375 m, and a day/night band (the source of most "Earth at night" imagery). The WorldView-3 band count and revisit are summarized loosely on purpose; the commercial lineup changes quickly.

Hurricane Katrina, August 2005. The true-color view shows the whole storm in the Gulf. The high-resolution commercial image of the Superdome after the storm shows the roof membrane torn away: the detail an insurance adjuster or engineer needs, and useless for tracking the storm. The enhanced infrared view genuinely IS about emitted heat: colors are a temperature enhancement, and the coldest cloud tops are the tallest, most vigorous convection, which is why the eyewall lights up. Timestamp on the image: 17:25Z, 29 August 2005. Sources noted in the original deck: sapphireeventsnola.com (true color); satimagingcorp.com/galleryimages/hurricane-katrina-superdome-picture.jpg (Superdome). TODO(instructor): provenance of the infrared image. Its only recorded source is a NOAA URL (www.srh.noaa.gov/images/hun/stormsurveys/katrina/katrina_IRsat_29_1725Z.png) that the September 2026 audit found no longer resolves. Re-source or re-attribute it from a current NOAA archive.

Yuriage in Natori, and Yagawahama: 2007–2008 imagery on the left, 12 March 2011 on the right. This is what temporal resolution buys: a satellite that passes regularly already has the "before" image on file. Before-and-after pairs are the single most common emergency-response product. Ask what you would have to do to these two images before you could difference them, which is exactly the georectification problem from Tuesday. Source noted in the original deck: boingboing.net, March 2011.

Two minutes of joining up. Say "reflected near-infrared", not "heat": the near-infrared in NDVI is reflected sunlight that leaf structure bounces back, and nothing in NDVI measures temperature. Then hand off to Week 5: everything today was passive, measuring sunlight that something else emitted; LiDAR is active, and that difference is where Tuesday starts.

Four or five minutes, then a show of hands on two items. The false-color item (3) is the one to discuss: any color ramp on any single band is a false-color display, with no need for three bands. The near-infrared item (2) splits the room every time: near-infrared in a vegetation image is reflected sunlight bouncing off leaf structure, and a good half of the class will pick "leaves are warmer than the ground" because the word infrared sounds like heat. Draw the distinction on the board one more time - reflected near-infrared is sunlight, thermal infrared is what a surface emits because of its own temperature - and point back at the Katrina infrared slide, which is the one image today that really is heat. If the room has no signal, put the URL on the board; the items read aloud just as well.

Split notes (2026-09-10). This was "Remote Sensing and 3D Imaging", 43 slides, at slug remote-sensing-3d-imaging. Its last nine slides were LiDAR and they moved to Week 5, into slides/week-05/elevation-data-lidar.md, at the instructor's decision: 43 slides is too many for a 75-minute session, and a bare-earth LiDAR surface is a DEM, so LiDAR belongs beside terrain analysis rather than at the end of a spectrum-and-bands lecture. What changed here beyond the removal: the deck is renamed "Remote Sensing"; the title slide's background was a LiDAR city render that left with the LiDAR block, replaced with the false-color terrain image; the LiDAR goal became a false-color goal; "Where we are going" lost its sixth item; and the closing "Where today shows up" now hands off to Week 5 on the passive-versus-active distinction rather than pointing at LiDAR slides that are no longer in this deck. Old slug remote-sensing-3d-imaging.html is now a 404. Learning Suite carries it on Thu Sep 24.

Revision notes (2026-09-23), instructor review the day before the lecture. Removed the "Where we are going" contents slide. Removed the dead YouTube link from the spectrum slide and labeled the figure instead. The irradiance slide and the Digital Images divider now tie back to the spectrum. Added dashed pair boxes to the hex slides, and four decoded colors to the counting slide. Added the grid-paper RGB stack slide and a band-ruler slide after the hyperspectral cube. The old gallery is gone: Eiffel, snow, smoke and night lights are now band splits (section 5), the inauguration, Dubai and second Japan slides are dropped, and Katrina's three images are one slide in a new Satellites section (section 6). New figures: tools/week04_remote_sensing_figures.py. ArcGIS Pro captures still to make: tools/week04-arcgis-capture-plan.md.

Lab 3 is due this Saturday. Next week is elevation: LiDAR on Tuesday, terrain analysis on Thursday, and Lab 4 needs a DEM, so Tuesday is the hour that tells them where to get one.