[27] Extra: Gaps from Held-out Tests

Where is average methane concentration higher, Barrow (Alaska) or the South Pole, and why?

Barrow, by roughly 100 ppb or more, because most methane sources (wetlands, fossil fuels, agriculture) are in the Northern Hemisphere.

Why does methane at high northern sites like Barrow show a larger seasonal cycle than at the South Pole?

Barrow sits near large Northern sources that vary by season, and summer sunlight there makes strong OH that destroys methane. South Pole air is remote and well mixed, so it swings less.

In which season is Northern Hemisphere methane lowest, and why?

Late summer, when strong sunlight makes the most OH (hydroxyl), methane's main sink.

If warming melts ice and Earth settles at a new equilibrium with lower albedo, how does global average OLR compare with today?

Higher, because more sunlight is absorbed and at equilibrium OLR must equal absorbed sunlight.

On a CERES OLR map of a much warmer world, where would OLR rise most, and why?

Over the Arctic and other high latitudes, which warm fastest (Arctic amplification, sea ice loss), and a warmer surface emits more longwave (σT⁴).

What does InSAR (interferometric synthetic aperture radar) measure?

Ground deformation as small as millimeters (subsidence, earthquake slip, volcano swelling, glacier flow), from the phase difference between two SAR images of the same spot.

Which radar technique maps millimeter-scale uplift or sinking of the ground over a wide area?

InSAR (interferometric SAR), which compares the phase of two radar passes.

Why can InSAR detect ground motion far smaller than its pixel size?

It measures the phase of the returned wave, not just its timing. Each full color fringe means about half a wavelength of motion toward or away from the radar (a few cm for C-band).

What causes speckle, the grainy salt-and-pepper look of radar images?

Constructive and destructive interference of echoes from the many scatterers inside each pixel.

Why do radar images show speckle while ordinary sunlit optical images do not?

Radar sends coherent waves (one frequency, fixed phase), so echoes inside a pixel interfere randomly. Sunlight is incoherent, so its reflections simply add.

How is radar speckle usually reduced, and what is the cost?

Multi-look averaging or spatial filters (median, Lee filter). It smooths the image but lowers spatial resolution.

Converting top-of-atmosphere (TOA) reflectance to surface reflectance removes what?

The atmosphere's effects: scattering (haze, path radiance) and absorption by gases, leaving what the ground itself reflects.

Over dark targets such as water, why is TOA reflectance in the blue band higher than the true surface reflectance?

Rayleigh scattering by air adds path radiance (scattered light that never touched the surface) on the way to the sensor.

In image classification, what is the key difference between supervised and unsupervised methods?

Supervised: the analyst first gives training areas of known classes (e.g. maximum likelihood). Unsupervised: the computer clusters pixels first and the analyst names the clusters afterward.

ISODATA and k-means are algorithms for which type of image classification?

Unsupervised classification (clustering pixels by spectral similarity, no training data).

Which classification type needs no training data from the analyst before it runs?

Unsupervised classification (for example ISODATA or k-means).

In a confusion (error) matrix, what does producer's accuracy measure?

How well a real ground class was captured on the map: correct pixels of that class divided by all reference pixels of that class (1 minus it is omission error).

In a confusion matrix, what does user's accuracy measure?

How often a class shown on the map is really that class on the ground: correct divided by all pixels mapped as that class (1 minus it is commission error).

A map user wants to know how often a pixel labeled "forest" is truly forest. Producer's or user's accuracy?

User's accuracy. Producer's accuracy answers how much of the real forest was mapped as forest.

How is overall accuracy computed from a confusion matrix?

The sum of the diagonal (correctly classified pixels) divided by the total number of reference pixels.

What does the kappa coefficient correct overall accuracy for?

Agreement expected by chance. Kappa = 0 means no better than random, 1 means perfect agreement.

Post-classification change detection vs image differencing: what does each compare?

Post-classification compares two separately classified maps. Image differencing subtracts the pixel values of two co-registered dates.

Which change detection method directly gives "from-to" changes, such as forest to cropland?

Post-classification comparison, because each date already has class labels.

How wide is one UTM (Universal Transverse Mercator) zone?

6° of longitude (60 zones around the globe).

WGS84 and NAD83 are examples of what?

Geodetic datums (a reference ellipsoid and origin for coordinates). GPS uses WGS84.

Datum vs map projection: what does each define?

A datum defines Earth's shape and origin so coordinates have meaning. A projection flattens that curved surface onto a flat map.

How do you turn a map scale written as 1:N into a ground distance?

One unit on the map equals N of the same units on the ground. At 1:50,000, 1 cm = 50,000 cm = 500