Back to the on-screen lesson ·
Pixel size, revisit time and clouds limit what satellites see, and every classification needs checking on the ground.
Paper packet. Every task here also exists on screen, where it is checked automatically; answers written on paper are not assessed by Nydus. When you are back at a device, enter your answers there.
By the end of this lesson you will be able to judge whether a sensor can show a feature, estimate how many usable images a period gives, and explain why ground checks matter.
You can compute NDVI from spectral bands and measure area from pixel counts. This lesson asks what satellite data cannot do: which features are too small, which days are hidden by cloud, and why every map made from images needs checking on the ground.
| Term | What it means |
|---|---|
| Spatial resolution | The ground width of one pixel; the finer it is, the smaller the features shown. |
| Temporal resolution | How often a sensor images the same place, its revisit time. |
| Revisit time | The days between a satellite's passes over one place. |
| Mixed pixel | A pixel covering more than one surface, given a single blended value. |
| Ground truthing | Checking image classifications against observations made on site. |
| Radar | An active sensor that sends its own microwaves and sees through cloud. |
Every sensor trades detail, frequency and coverage.
Another way: picture
Picture a mosaic made of tiles the size of a bathroom floor, trying to show a garden hose. The hose is narrower than a tile, so it cannot appear as a line; at best it tints the tiles it crosses. Now picture being allowed to look at the garden only every two weeks, and only on clear days.
Another way: steps
Provenance is the chain from collection to the file you use: producer, observation dates, method, transformations, edition and permission to reuse it. A screenshot can show a striking pattern while concealing all of that chain. Resolution describes sampling detail; positional accuracy describes agreement with a reference location. Neither is guaranteed by the number of pixels on the screen. An authoritative producer can supply excellent regional data that are unsuitable for locating a narrow footpath.
In an invented vegetation survey, a 20 m raster has documented positional uncertainty of 8 m. A 3 m hedgerow might influence a cell's value, but the grid cannot reliably delineate its edges. Resampling to 1 m cells gives more output cells without new observations. Before combining images, compare acquisition seasons, masks, classification definitions and CRS as well as nominal cell width. If those differ, an apparent change may be methodological. Mark cloud-obscured or unobserved cells as missing, not as bare ground.
Imagine publishing a trail-maintenance map with nearby household complaints. A dated raster with reuse permission supports a regional land-cover summary; it does not authorize disclosure of addresses. Keep identifiable records restricted, and consider a coarser public summary with small counts suppressed. Check whether another public map could still reveal an individual through overlapping summaries. Aggregation reduces some risks but is not a guarantee of anonymity. Explain the missing detail and invite affected communities to review the representation. The public product should name its source edition, collection period, cell width and known uncertainty so that readers can decide what conclusions remain supportable.
| Sensor | Pixel width | Revisit |
|---|---|---|
| MODIS | 250 m | daily |
| Landsat 8 and 9 | 30 m | 16 days each, 8 together |
| Sentinel-2 | 10 m | about 5 days |
| NAIP aerial images | about 1 m or finer | every two or three years |
Fine pixels show small features but cover less ground or come less often; coarse pixels see the whole country every day but miss fields and streams.
A feature exactly one pixel wide rarely lines up with the grid, so it spreads across two pixels, each only partly filled. As a rule of thumb, a feature needs to span two or three pixels to be detected and more to be recognized by its shape.
A sixty-meter river is six pixels wide at ten meters, two at thirty, and a quarter of a pixel at two hundred fifty, where it vanishes into mixed pixels.
A thirty-meter pixel at the edge of a lake holds both water and shore. Its value is a blend, and a classifier must choose one label. Across a large map, edge pixels add up to real errors in area.
Finer pixels shrink the problem but never remove it; there is always an edge.
A satellite in a sun-synchronous orbit passes over the same place at the same local time every few days. Landsat 8 and 9, offset by half a cycle, together image each place every eight days.
For slow changes, such as city growth, that is plenty. For fast ones, such as floods or crop stress in a heat wave, eight days may miss the event.
Optical sensors record reflected sunlight, and clouds block it. In humid regions, most passes can be cloudy for weeks. A satellite passing every five days, with a third of passes clear, gives about six clear images in ninety days.
Radar satellites send their own microwaves, which pass through cloud and work at night, but record surface roughness and moisture rather than color.
A classification made from images is a hypothesis. Analysts check it by visiting sample sites, or examining finer images, and comparing the labels.
The result is an accuracy table: of the pixels labeled forest, how many really were? Published land cover maps report overall accuracy, often around eighty to ninety percent, and which classes are confused most.
Images show what is on the surface, not why. Low NDVI may mean drought, pests or a fallow field; a bright roof may be a warehouse or a church. Interpreting images needs knowledge of the place.
Images also cannot see beneath tree canopy, inside buildings, or underground, where much of what matters happens.
Checking an answer. Pixels across must be fewer than pixels in an area. Clear images can never exceed passes.
Dividing a width by the pixel width is allowed because pixels tile the ground in equal steps. Dividing days by revisit time is allowed because passes come at a regular interval.
Multiplying by the clear share is an estimate, allowed when cloud is roughly random from pass to pass; in a monsoon it is not, and the real count may be lower.
Commercial satellites now sell images with pixels under half a meter, enough to see cars and backyard pools. Aerial surveys are finer still.
That detail helps assessors, insurers and disaster teams, and raises questions about who may watch whom, which the ethics lesson takes up.
A sensor cannot have everything. Finer pixels mean more data per square kilometer, so a fine sensor usually covers a narrower strip, called its swath, and returns less often. Landsat images a swath about 185 kilometers wide at thirty meters; MODIS covers a swath over two thousand kilometers wide at two hundred fifty meters, and sees nearly the whole Earth every day.
Spectral resolution trades too: more, narrower bands separate surfaces better but collect less light in each. Choosing data means choosing among these: a daily coarse view for a fast-moving drought, a detailed but less frequent view for mapping new roads, or aerial images for a single town. Naming the trade-off is part of reporting what the data can show. A study that names its sensor, dates and pixel size lets readers judge whether it could have seen what it claims.
The most common slip is dividing the pixel width by the feature width, which gives a fraction. Another is squaring when only pixels across are wanted.
A third is counting every pass as a usable image, forgetting clouds. A fourth is trusting a classification without asking how it was checked.
Knowing a sensor's limits is part of using its data. The next unit turns from sensors to people in the field, where sampling design sets the same kind of limits on what measurements can show.
When Hurricane Harvey stalled over Houston in August 2017, it dropped record rain for days under thick cloud. Optical satellites, including Landsat, saw mostly cloud tops while neighborhoods flooded below.
Radar satellites, which send microwaves through cloud, mapped standing water across the region, since smooth water reflects radar away from the sensor and shows up dark. Emergency managers combined those maps with aerial photographs taken from aircraft flying under the clouds, and with reports from residents.
The flood showed each tool's limits: radar struggles in dense city blocks, where buildings scatter the signal, and aircraft cannot cover everything at once. No single sensor gave the full picture.
Many American cities set goals for tree canopy, the share of land shaded by trees, because trees cool streets, soak up rain and clean the air. Measuring canopy needs images fine enough to see individual crowns.
The U.S. Department of Agriculture's National Agriculture Imagery Program flies over the country every two or three years in the growing season, collecting aerial images with pixels of a meter or finer. A ten-meter crown covers about seventy-eight such pixels, enough to outline it; in a thirty-meter Landsat pixel it would be a tenth of a pixel.
Cities combine those images with laser scans of tree height and sample checks by foresters on the ground, and publish canopy maps by neighborhood, which often show that poorer districts have far fewer trees.
It is natural to imagine satellites watching every place, all the time, in any detail. But each sensor has a pixel size that hides smaller features, a revisit time that leaves days unseen, and optical sensors see nothing through cloud.
Even what they do see must be interpreted and checked on the ground: a classification is only as good as its accuracy, and images show what is there, not why.
A highway is $40$ m wide, and pixels are $10$ m. Divide the widths.
$40 \div 10 = 4$
Pixels across.
Judge whether it shows.
$\text{yes, clearly}$
Several pixels.
Try thirty-meter pixels.
$40 \div 30 \approx 1.3$
Barely.
Say what edge pixels do.
$\text{blend road and verge}$
Mixed values.
A season is $160$ days; revisit is $8$ days. Count the passes.
$160 \div 8 = 20$
Passes.
Suppose $40$ percent are clear. Take that share.
$20 \times 0.4 = 8$
Clear images.
Name the slip of counting every pass.
$20$
Clouds ignored.
Add a second satellite.
$16$
Twice the passes.
Say what radar adds.
$\text{images through cloud}$
Different information.
A crown is $10$ m across. Find its radius.
$5\ \text{m}$
Half the width.
Find its area.
$\pi \times 5^2 \approx 78.5\ \text{m}^2$
A circle.
Divide by one-meter pixels.
$78.5 \div 1 = 78.5$
Pixels on the crown.
Divide by thirty-meter pixels.
$78.5 \div 900 \approx 0.09$
A tenth of a pixel.
Say which image maps trees.
$\text{the aerial one}$
Dozens of pixels per tree.
Say what the coarse image still shows.
$\text{canopy share by pixel}$
Blended greenness.
Count along and across.
$600 \div 30 = 20,\ 240 \div 30 = 8$
Rows and columns.
Multiply the counts.
$160$
Pixels in all.
Say which pixels are least reliable.
A feature on the ground is $90$ m wide, and a satellite image has pixels $30$ m wide. How many pixels span the feature's width?
Complete the worked solution: a rectangular field is $450$ m long and $90$ m wide, and an image has pixels $30$ m wide. Find the pixels along its length, across its width, and covering it in all.
Count the pixels along.
$\text{length} \div \text{pixel width} =$ a
One row.
Count the pixels across.
$\text{width} \div \text{pixel width} =$ b
One column.
Count the pixels in all.
$\text{along} \times \text{across} =$ n
Rows times columns.
Say what edge pixels do.
$\text{mix field and road}$
Borders are blurred.
Match each limit of remote sensing to what it constrains.
| the smallest feature an image can show | how often a place is imaged | whether an optical sensor sees the ground at all | how accurate a classification is, checked on site | |
|---|---|---|---|---|
| spatial resolution | ||||
| revisit time | ||||
| cloud cover | ||||
| ground truthing |
A river is $60$ m wide. Fill in how many pixels span it in images with 10 m, 30 m and 250 m pixels, to two decimals where needed.
| pixels | |
|---|---|
| 10 m pixels across | |
| 30 m pixels across | |
| 250 m pixels across |
In a cloudy region, $60$ percent of a satellite's passes are clear. Write the number of clear images per year, taking a year as $365$ days, as a function of the revisit time $x$ in days.
Answer:
A growing season lasts $160$ days. A satellite revisits every $8$ days, and $40$ percent of its passes are clear. About how many clear images of a field does the season give?
Answer: images
A city maps its street trees from aerial images with pixels $0.6$ m wide. Suppose a tree's round crown is $6$ m across. About how many pixels cover the crown?
Answer: pixels
Mark every supported sentence needed for a fictional publication audit, covering fitness, provenance and disclosure. Paths are 3 m wide. An agency raster has a date, 30 m cells and 15 m positional uncertainty. A screenshot lacks date, CRS and license. Household address coordinates are identifiable.
This task has no paper form; do it on a device.
Lesson test: one question per skill, one attempt each, no hints. Your answers are checked when you submit.
In a cloudy region, $40$ percent of a satellite's passes are clear. Write the number of clear images per year, taking a year as $365$ days, as a function of the revisit time $x$ in days.
Answer:
You can reason about remote-sensing limits. Explain why a sixty-meter river can vanish in a coarse satellite image.
26. Your turn: a field is $600$ m by $240$ m, with thirty-meter pixels. How many pixels cover it?, step 3
$\text{the edges}$
Mixed with neighbors.