Remote Sensing Fundamentals: A Beginner's Guide

What is remote sensing?

Remote sensing is the process of collecting information about an object or area without physically being in contact with it. Satellites, aircraft and drones can all be used for it. Instead of visiting every field, forest or lake and measuring it manually, we can collect information about large areas from above, which is how a satellite image can show where the forests, water bodies, agricultural fields and cities are.

How does remote sensing work?

The Sun sends electromagnetic energy toward the Earth, and different surfaces interact with that energy differently. For example:

• Vegetation absorbs a lot of visible red light.
• Healthy vegetation reflects strongly in the near-infrared.
• Water absorbs much of the near-infrared and shortwave infrared energy.
• Bare soil has its own reflectance characteristics.
• Buildings and roads also have distinct spectral responses.

A satellite sensor measures some of this energy and records it as digital data. We can then process and analyze that data to understand what is happening on the ground.

What is a sensor?

A sensor is the instrument that detects electromagnetic energy. For example, satellites such as Landsat 8/9 and Sentinel-2 carry sensors designed to observe the Earth's surface in multiple parts of the electromagnetic spectrum.

A normal photograph shows only a limited portion of what is happening. Satellite sensors can detect wavelengths that our eyes cannot see.

Platforms

A sensor is carried on a platform, which can be a:

• Satellite
• Aircraft
• Drone

A drone can capture very high-resolution imagery, while a satellite can cover an entire region or country. The choice of platform depends on the problem you are trying to solve.

Sensor types worth knowing

The platform carries the sensor, and the sensor decides what is recorded. These are the main types you will meet:

What can remote sensing be used for?

Remote sensing and GIS

Remote sensing and GIS are often used together. Remote sensing gives us information collected from sensors. GIS lets us organize, analyze and visualize that information along with other geographic datasets.

For example, you can calculate NDVI from satellite imagery, identify stressed vegetation, overlay agricultural field boundaries and calculate the affected area.

The electromagnetic spectrum

Electromagnetic energy travels as waves. The distance between one wave peak and the next is the wavelength, and the full range of wavelengths is called the electromagnetic spectrum.

Our eyes can only see a small slice of it, called visible light, which spans roughly 400 to 700 nm. Remote sensing sensors can record much more: the near-infrared, shortwave infrared, thermal infrared and microwave regions are all used to observe the Earth.

Regions used in remote sensing

Atmospheric windows

The atmosphere absorbs or blocks some wavelengths. Sensors are designed to record in the ranges where the atmosphere is reasonably transparent, which are called atmospheric windows. This is one reason the bands on a satellite sit where they do.

Spectral signatures

A full chapter with measured Bengaluru values is in Spectral signatures.

Different materials reflect and absorb energy differently at different wavelengths. If you plot how much energy a surface reflects across wavelengths, you get a curve called its spectral signature. Each kind of surface has a characteristic shape, and that is what lets us tell them apart from space.

How to read the chart

Passive vs active remote sensing

Remote sensing systems are classified as passive or active, and the difference comes down to one question: does the sensor provide its own energy, or does it rely on energy from an external source?

Passive remote sensing

A passive sensor detects naturally occurring electromagnetic energy. For Earth observation, the most important natural source is usually the Sun. The Sun illuminates the Earth's surface, different materials reflect different amounts of energy, and the sensor detects that reflected energy.

Examples include many optical satellite sensors, such as Landsat and Sentinel-2.

Passive sensing has two limitations:

• If you rely on sunlight to illuminate the surface, there is not much reflected sunlight to measure at night.
• Clouds can block the sensor from seeing the surface.

Active remote sensing

Active sensors do not have to wait for the Sun. They send their own energy toward the Earth's surface and measure the signal that comes back.

Radar is a common example. SAR (Synthetic Aperture Radar) is an important active remote-sensing technology. The satellite sends microwave pulses toward the Earth and measures the returned signal.

Why is radar so useful?

Microwave signals can often penetrate clouds. This means SAR can collect useful observations even when optical imagery is blocked by cloud cover. SAR is particularly useful for:

• Flood mapping
• Monitoring land deformation
• Agriculture
• Forest monitoring
• Surface change detection

One example is Sentinel-1, a SAR satellite.

In many real-world projects, active and passive data complement each other.

Understanding spectral bands

Satellite imagery does not always come as one image. It often comes as separate files, or bands, such as Blue, Green, Red, Near Infrared and SWIR.

What is a spectral band?

A spectral band is a specific range of wavelengths of electromagnetic radiation that a sensor records. Instead of measuring everything at once, a satellite sensor divides the electromagnetic spectrum into different wavelength ranges. Each band captures different information about the Earth's surface.

Why do we need different bands?

Different materials interact with electromagnetic energy differently. A healthy green crop looks green to our eyes because it reflects more green light than some other visible wavelengths. Healthy vegetation also strongly reflects near-infrared radiation, which is not visible to our eyes but can be detected by a satellite sensor. That is one reason near-infrared is so useful for vegetation analysis.

Common satellite bands

Thermal infrared: measuring heat

Most bands record sunlight that the surface has reflected. Thermal infrared is different: it records heat energy that the surface emits. That makes it useful for studying surface temperature and urban heat, which is one of the urban applications listed earlier.

Thermal bands usually have coarser spatial resolution than the optical bands. For example, Landsat 8 and 9 collect their thermal bands at 100 m, compared with 30 m for the optical bands.

Why band combinations matter

Bands can be combined into indices. For example:

• Red + NIR → NDVI
• Green + NIR → NDWI (in one commonly used formulation)
• NIR + SWIR → NDMI

The exact index formula matters, but the main point is that different wavelengths give different clues about what is happening on the ground.

The four types of resolution

When people say a satellite has "high resolution", they usually mean spatial resolution. In remote sensing, resolution has four dimensions: spatial, spectral, radiometric and temporal. Understanding all four makes satellite data much easier to compare.

Spatial resolution

Spatial resolution describes the size of the area represented by each pixel. For example, 10 m resolution means each pixel represents approximately a 10 m × 10 m area on the ground for that band. Smaller pixels generally mean more spatial detail, so 10 m imagery can show smaller features than 30 m imagery.

Spectral resolution

Spectral resolution describes how finely a sensor separates different wavelength ranges. A sensor with many relatively narrow spectral bands can provide detailed information about how surfaces behave across the electromagnetic spectrum. This is particularly useful for identifying or studying different materials.

Radiometric resolution

Radiometric resolution describes how well a sensor can distinguish between different levels of energy, which is to say how sensitive it is to differences in signal intensity.

Think of a brightness scale. An 8-bit image can represent 2⁸ = 256 possible digital values (0 to 255). A 12-bit sensor can represent 2¹² = 4096 possible values (0 to 4095). The 12-bit sensor can represent many more levels of intensity.

This matters because two surfaces may have very similar reflectance. A sensor with higher radiometric resolution may be better able to distinguish the subtle difference between them, for example in vegetation, surface properties, water, soil or thermal conditions.

Temporal resolution

Temporal resolution describes how frequently a satellite revisits the same area. If a satellite can image a location every few days under suitable conditions, it has a relatively high revisit frequency for that application.

Temporal resolution becomes particularly important when studying things that change quickly, such as:

• Flooding
• Crop growth
• Forest fires
• Construction
• Seasonal water levels

Why all four matter

When monitoring crops, you might want high spatial resolution to see individual fields, useful spectral resolution to distinguish vegetation characteristics, good radiometric resolution to detect subtle differences, and high temporal resolution to follow changes through the growing season.

So rather than asking which satellite has the highest resolution, ask which type of resolution matters most for your analysis.

Landsat vs Sentinel-2

For orbits, swaths, revisit times and other missions, see Earth observation satellites.

Landsat and Sentinel-2 are two of the most widely used sources of satellite imagery. Both are useful for Earth observation, but they are not identical.

Landsat

The Landsat program has been observing the Earth's surface for decades, and its long historical record is a major strength. Landsat 8 and Landsat 9 provide multispectral imagery at different spatial resolutions, including 30 m multispectral data for many of the commonly used bands.

This makes Landsat particularly useful when your project looks at change over a long period. For example: How has urban expansion changed since 2000? A long historical archive matters for questions like this.

Sentinel-2

Sentinel-2 is part of the European Union's Copernicus Earth observation program. It provides multispectral imagery with spatial resolutions from 10 m to 60 m, depending on the band. Several important visible and near-infrared bands are available at 10 m resolution, which can provide more spatial detail for many applications.

Side by side

A practical comparison

Suppose you are studying urban expansion from 1990 to 2026. Landsat may be particularly useful because of its long historical archive. Now suppose you are analyzing the condition of individual agricultural fields today. Sentinel-2's higher spatial resolution for several bands may be more useful.

Choose the dataset that suits your study area, time period and analysis.

Spectral indices: NDVI, NDWI and NDMI

The chapter Spectral indices gives formulas for Sentinel-2 and Landsat 8 and real values from Bengaluru.

Satellite imagery holds more information than a normal image shows. Spectral indices are one way to extract it: instead of interpreting several bands one by one, an index combines spectral information into a single value that makes a particular feature easier to analyze. Three you will encounter frequently are NDVI, NDWI and NDMI.

NDVI: vegetation

NDVI is commonly used to analyze vegetation. The formula is:

NDVI = (NIR − Red) / (NIR + Red)

These two bands are used because healthy vegetation tends to absorb red light and strongly reflect near-infrared, so it often produces relatively high NDVI values. Lower values can indicate bare soil, built-up surfaces, water or sparse vegetation.

NDVI with a worked example

Here are two pixels.

A pixel of healthy vegetation, with NIR = 0.45 and Red = 0.05:

NDVI = (0.45 − 0.05) / (0.45 + 0.05)
NDVI = 0.40 / 0.50
NDVI = 0.80

A pixel of mostly bare soil, with NIR = 0.30 and Red = 0.25:

NDVI = (0.30 − 0.25) / (0.30 + 0.25)
NDVI = 0.05 / 0.55
NDVI ≈ 0.09

The vegetation pixel produces a much stronger NDVI signal.

What do NDVI values mean?

Very broadly:

These are rough guides, not universal rules. NDVI values also depend on the sensor and surface characteristics.

NDWI: water

NDWI refers to the Normalized Difference Water Index, but there are multiple formulations of NDWI in the remote sensing literature. One common formulation uses Green and NIR and can help highlight open water. Another commonly used water-related index uses different bands, such as NIR and SWIR, so check which formula a workflow uses before comparing results.

NDMI: moisture

NDMI stands for Normalized Difference Moisture Index. A common formulation uses NIR and SWIR. It is often used to examine vegetation or surface moisture conditions.

NBR: burned areas

NBR stands for Normalized Burn Ratio. It is used to map burned areas and to judge how severe a fire was. It uses NIR and the longer SWIR band (around 2.2 µm).

Healthy vegetation has high NIR and low SWIR, so NBR is high. After a fire the plants are gone, so NIR drops, and the dry, bare surface reflects more SWIR. NBR falls, often to negative values.

Analysts usually compare NBR from before and after a fire. The difference, called dNBR, shows how much burned and how severely.

More spectral indices

There are many indices, each designed to answer a particular question. These are some of the most common.

Notes on these indices

• EVI: the constants in the formula assume reflectance values from 0 to 1. It is designed to stay sensitive in dense vegetation, where NDVI tends to level off.
• SAVI: L is a soil adjustment factor. A value of 0.5 is commonly used.
• NDRE: it needs a red-edge band. Sentinel-2 has red-edge bands (B5, B6, B7 and B8A), but Landsat 8/9 does not.
• MNDWI: it is often better than NDWI at separating water from built-up areas, which can confuse the Green and NIR version.
• NDBI: bare soil can look similar to built-up areas, so check the result.
• NDSI: clouds can also look bright, so snow results need care.
• dNBR: it compares two dates, so the two images should be as comparable as possible.

Which bands to use

The same index uses different band numbers on different satellites. This table shows the usual bands for Landsat 8/9 and Sentinel-2.

Why are indices useful?

They turn imagery into something easier to read: a vegetation index shows vegetation patterns, a water index shows water extent, and a moisture index shows moisture-related patterns.

True color vs false color composites

Vegetation that appears bright red in a satellite image is usually a false color composite (FCC), not an error.

True color composite

A true color composite uses the visible bands that correspond approximately to what our eyes see: red, green and blue. For Sentinel-2, a common true-color combination is B4 – B3 – B2, where B4 is red, B3 is green and B2 is blue.

The result looks relatively similar to a normal photograph. Vegetation appears green, water appears dark or blue, and buildings appear in colors closer to what we would expect.

False color composite

A false color composite uses bands in combinations that do not correspond to normal human vision. A very common vegetation-focused combination is NIR – Red – Green. For Sentinel-2, that is B8 – B4 – B3.

When this combination is displayed as RGB, healthy vegetation appears bright red, because it strongly reflects near-infrared radiation and the NIR signal is assigned to the red display channel.

Why make an image look "wrong"?

Our eyes cannot see near-infrared. By assigning invisible wavelengths to visible colors, we can make certain features much easier to identify. FCC can help show:

• Vegetation
• Water
• Agricultural patterns
• Vegetation stress
• Land-cover differences

See it for yourself

How to read the colors in that image

True vs false color

What is atmospheric correction?

A building looks different depending on whether you photograph it on a clear day, through haze, through dust, or at sunset. Satellite sensors face a similar problem.

Before sunlight reaches the Earth's surface and returns to a satellite sensor, it has to travel through the atmosphere. Along the way, gases, aerosols, water vapour and other atmospheric components can affect the signal, which can change what the sensor records.

What does atmospheric correction do?

Atmospheric correction attempts to reduce or compensate for atmospheric effects, so that the imagery represents the Earth's surface more accurately. It tries to separate the signal coming from the surface from effects introduced by the atmosphere.

The atmosphere can cause:

• Scattering
• Absorption
• Haze
• Aerosol effects

Why is it important?

Suppose you want to compare imagery from June 2025 with imagery from June 2026. If atmospheric conditions were very different on the two dates, some differences in pixel values may be caused by the atmosphere rather than by real change on the ground. Atmospheric correction can help make comparisons more meaningful.

Does every project need it?

Not necessarily. It depends on:

• Which satellite product you are using
• What processing level it already has
• Your analysis
• Whether you are comparing dates
• The accuracy you need

For example, Sentinel-2 provides products that have already undergone different levels of processing, so you should not automatically assume you need to apply every atmospheric correction step yourself.

Cloud masking and other preprocessing

Raw satellite data usually needs some preparation before it can be analyzed. Atmospheric correction is one part of that. These are the main ideas.

Supervised vs unsupervised classification

Identifying water, forest, agriculture, built-up areas and bare soil across a large satellite image by looking at every pixel would take a very long time. Image classification automates this. Two common approaches are supervised and unsupervised classification.

Supervised classification

In supervised classification, you provide training information to the algorithm. You essentially tell it, "These pixels represent water", "These pixels represent vegetation", and "These pixels represent buildings." The algorithm learns the characteristics of those classes and uses them to classify the rest of the image. It is similar to teaching a model with examples.

Unsupervised classification

In unsupervised classification, you do not provide predefined training samples. Instead, the algorithm groups pixels based on their spectral similarity. It may discover several clusters in the image, and you then examine those clusters and decide what they represent. One common example is K-means clustering.

Which should you use?

Supervised classification is useful when:

• You know the classes you are interested in
• You have reliable training samples
• You want specific land-cover categories

Unsupervised classification can be useful when:

• You do not have training data
• You are exploring an unfamiliar area
• You want to identify natural spectral groupings

A simple analogy

Consider sorting fruit.

• Supervised: you show someone an apple, an orange and a banana as examples, then ask them to classify the rest.
• Unsupervised: you give them a basket and say, "Group similar things together." They create the groups first, and you decide what the groups represent afterward.

Land use and land cover (LULC)

A satellite image contains millions of pixels, and they are not very meaningful until we interpret them. One common application is Land Use/Land Cover (LULC) classification. Land use and land cover are related but not the same.

In many GIS projects the two are used together as LULC.

Checking the result

A classification map is not automatically correct just because the software produced it. You need to compare the classified result against reliable reference data. Common accuracy measures include:

• Overall accuracy
• Producer's accuracy
• User's accuracy
• Confusion matrix
• Kappa coefficient, where appropriate

Why is LULC useful?

Once you have a classified map, you can ask:

• How much forest exists?
• How much agricultural land exists?
• How much has the built-up area expanded?
• Where has vegetation been lost?
• How has land cover changed over time?

A classified map turns imagery into measurable geographic information.

Change detection

One satellite image is a snapshot. Change detection compares two or more images taken at different times to identify differences in the Earth's surface.

For example, comparing an image from 2018 with one from 2026, you might want to know:

• Where did forests disappear?
• Where did cities expand?
• Which agricultural areas changed?
• Did a water body shrink?
• Which areas were affected by a flood?

Multi-temporal imagery

Multi-temporal simply means data collected at multiple points in time, such as January 2020, January 2022, January 2024 and January 2026. By comparing them, you can examine how the landscape evolved.

For example, if you study a city with imagery from 2015 and 2025, you might discover that agricultural land around the city has gradually been replaced by residential development. That is far more useful than saying "the city looks bigger", because you can measure the change spatially.

The biggest challenge

Differences can also come from:

• Different seasons
• Cloud or haze
• Atmospheric conditions
• Different sensor characteristics
• Shadows
• Differences in preprocessing
• Soil moisture

This is why selecting comparable images matters. When comparing two dates, keep as many conditions consistent as possible. For example, January 2020 vs January 2026 is generally more meaningful for seasonal vegetation analysis than January 2020 vs July 2026, because seasonal vegetation differences could dominate the result.

Questions change detection answers

Change detection is more than "Image A − Image B". The useful questions are what changed, where it changed, how much changed, and what might explain the change.

Popular remote sensing tools and data sources

You do not need all of these to start. They are where beginners commonly find imagery and software.

Glossary

Test yourself

Try to answer each question before opening the answer.

Quick quiz

Pick an answer for each question to see if you got it right.

Key takeaways