Nidhi

1What is remote sensing?

Remote sensing is the science of learning about an object, surface or gas without touching it, by measuring the electromagnetic radiation it reflects, emits or scatters. Satellites do this over the whole globe, repeatedly, which no network of ground stations can match.

2The physics: spectrum, windows and signatures

Every material interacts with light differently at different wavelengths. Gases absorb at narrow, characteristic wavelengths (their "fingerprints"), while surfaces have broad reflectance patterns.

• Absorption, scattering, emission: the signal reaching the sensor is the surface signal modified by the atmosphere (Rayleigh and aerosol scattering, gas absorption, thermal emission).
• Beer–Lambert law: absorption by a gas grows with its amount along the light path. This is the basis for retrieving trace gas columns such as NO₂, O₃, HCHO and CH₄.
• Radiance vs reflectance vs brightness temperature: radiance is what is measured; reflectance normalises it by the Sun, brightness temperature is the equivalent blackbody temperature in the thermal IR.

3Platforms and orbits

Where the sensor sits decides what it can see and how often.

Trade-off in one line: low orbits give fine detail and global coverage but infrequent looks; geostationary orbits give frequent looks but a coarser, fixed view that degrades towards high latitudes. Aircraft, balloons and drones fill the gap with high-detail, targeted measurements.

4Sensors and the four resolutions

A sensor is described by what it can resolve in space, wavelength, time and brightness.

5Key missions and datasets

A starter map, not an exhaustive list. Always check the mission page for current status, resolution and versions.

Aircraft campaigns such as CAFE measure the same gases in situ, giving the detailed truth against which satellite columns and model fields can be compared.

6From light to numbers: retrievals

A satellite measures radiance, not NO₂ or temperature. A retrieval (inverse problem) turns one into the other using a radiative-transfer model.

• Forward model: predicts the radiance for an assumed atmosphere and surface; the retrieval adjusts the unknowns until prediction matches measurement.
• Common methods: DOAS (differential absorption, typical for UV–Vis trace gases), optimal estimation (profiles, thermal IR), look-up-table and dark-target/deep-blue algorithms (aerosols), machine-learning retrievals.
• Columns vs profiles: most UV–Vis sensors give a vertical column (molecules per area); sounders and lidars give coarse profiles. A column is dominated by the layer where the gas is most abundant.
• Averaging kernels and a priori: describe how sensitive the retrieval is at each altitude and how much it leans on prior information. Use them when comparing satellite data with models or in-situ data.
• Quality flags: every Level-2 product carries cloud fraction, solar zenith angle and a quality value (e.g. qa_value ≥ 0.75 for TROPOMI NO₂). Filter before analysing.

7Typical processing workflow

Your own analysis is the last link of the chain above:

8Applications

9Where to get data and which tools to use

10Validation and uncertainty

A satellite number is an estimate with an error bar. Always ask how it was validated.

• Ground networks: AERONET (aerosols), Pandora and MAX-DOAS (NO₂, HCHO columns), TCCON (CO₂/CH₄), ozonesondes, air-quality monitors.
• Aircraft and ship campaigns: detailed in-situ profiles for process understanding and for testing retrievals.
• Error sources: clouds, surface reflectance, aerosols, viewing geometry, a priori profiles, instrument degradation and sampling (clear-sky bias, overpass time).
• Representativeness: a 5 km pixel is not a point measurement; compare like with like using averaging kernels and matching times.

11Common pitfalls for newcomers

• Forgetting quality flags and cloud filters, then interpreting noise as signal.
• Mixing product versions or sensors with different overpass times in one trend.
• Comparing a column with a surface measurement without considering the vertical distribution.
• Averaging in the wrong space (average radiances or columns, not ratios or indices computed on raw counts).
• Ignoring the pixel size, swath position and projection when regridding.
• Not reading the product user guide and known-issues page before using the data.

12Glossary