GIS and remote sensing

In short

A Geographic Information System (GIS) is computer software plus a database that stores, displays, queries and analyses data tied to position on the Earth, while remote sensing is the science of obtaining information about the surface without touching it, using sensors on satellites, aircraft or drones. Together they form the data-capture and data-analysis halves of modern spatial engineering: remote sensing supplies imagery and DEMs, GIS converts them into maps, areas, routes and decisions. In Indian universities this topic appears in Surveying-II / Advanced Surveying and in GATE Civil (Geomatics Engineering section), examined through resolution types, electromagnetic spectrum, data models, projections, NDVI and classification. Employers who test it include NRSC/ISRO, Survey of India, RITES, L&T, AECOM, WAPCOS, state remote sensing application centres, smart-city SPVs, and GIS firms such as Genesys, RMSI and ESRI India.

Explained

What GIS is: components, data models and functions

• Definition: a system of hardware, software, spatial data, people and methods that captures, stores, manages, analyses and displays geographically referenced information.
• Five components: hardware (workstation, GNSS, plotter), software (ArcGIS, QGIS, GRASS, ERDAS, Google Earth Engine), data (the costliest part, typically 60-80% of project cost), people (analysts, surveyors), methods (standard workflows, metadata standards).
• Spatial vs attribute data: spatial data answers *where* (coordinates, shape); attribute data answers *what* (pipe diameter, land use code, owner name). They are linked by a common identifier in a relational table.
• Vector model: real-world features as points (borewell, benchmark), lines/arcs (road, pipeline), polygons (plot, water body). Compact, accurate boundaries, good for discrete features and network analysis.
• Raster model: the area is divided into a grid of pixels/cells, each holding one value (elevation, reflectance, land-use code). Good for continuous surfaces and imagery; file size grows as 1/(cell size)².
• Topology: the stored relationships of connectivity, adjacency and containment — essential for network tracing and for detecting sliver polygons, dangles, undershoots and overshoots.
• TIN: Triangulated Irregular Network, a vector surface of non-overlapping triangles (Delaunay triangulation) used for terrain where breaklines matter.
• Core GIS functions: data input (digitising, scanning, GNSS import, imagery), storage (geodatabase, shapefile, GeoTIFF, KML), query (attribute and spatial, SQL-like), analysis (overlay, buffer, network, surface), and output (maps, reports, web services).
• Metadata: data about data — source, date, scale, accuracy, projection, lineage. India follows the NSDI metadata standard; map series and their release rules follow the National Map Policy 2005.

GIS analysis operations

1. Buffer: a zone of given width around a feature — e.g. 100 m from a river bank for a flood-plain or riverfront building restriction.
2. Overlay (vector): *intersect* (common area only), *union* (all areas), *clip* (cookie-cutter), *erase*, *symmetric difference*, *identity*. Intersect, union and identity outputs carry attributes from both inputs; clip and erase keep only the input layer's attributes.
3. Map algebra (raster overlay): cell-by-cell arithmetic or logical combination; used in weighted-overlay suitability models (e.g. landfill siting, groundwater potential).
4. Reclassification: grouping values into classes (slope 0-5°, 5-10° …) before weighting.
5. Network analysis: shortest path, service area, allocation, tracing — needs connected, topologically clean lines with impedance (length, time, cost).
6. Interpolation: generating a continuous surface from point samples — IDW (inverse distance weighting), kriging, spline, natural neighbour; used for rainfall, groundwater table, contaminant plumes.
7. Terrain analysis on a DEM: slope, aspect, hillshade, flow direction and accumulation, watershed delineation, viewshed, cut-and-fill volumes, profile and cross-sections.
8. Geocoding and address matching: converting addresses to coordinates; the reverse is reverse geocoding.
9. Digitising errors to fix: sliver polygons, dangling nodes, duplicate arcs, unclosed polygons — cleaned by snapping tolerance and topology rules.

Coordinate systems, datums and projections

• Geoid: the equipotential surface of mean sea level; ellipsoid/spheroid: the mathematical best-fit figure; datum: an ellipsoid fixed in position and orientation (WGS84 global, Everest 1830 for older Indian work, WGS84/ITRF-based for current Survey of India products).
• Geographic coordinates: latitude, longitude in degrees on the ellipsoid; projected coordinates: eastings and northings in metres on a plane.
• Projection distortions: no flat map preserves area, shape, distance and direction together. Conformal preserves shape locally (Mercator, Transverse Mercator); equal-area preserves area (Albers, Lambert azimuthal); equidistant preserves distance along chosen lines.
• UTM: Transverse Mercator in 6°-wide zones, scale factor 0.9996 at the central meridian, false easting 500,000 m, false northing 0 (N) / 10,000,000 m (S). India spans UTM zones 42N to 47N, from Lakshadweep and Gujarat in the west to Arunachal Pradesh in the east.
• Survey of India sheets: modern Open Series Maps (OSM) use the UTM projection on the WGS84 datum; older India toposheets used the Polyconic projection on Everest 1830. Lambert Conformal Conic is the usual conformal choice for east-west elongated regions.
• Datum transformation: a shift between Everest 1830 and WGS84 of the order of a few hundred metres — mixing datums is the single commonest blunder in Indian GIS projects.
• Georeferencing: assigning real-world coordinates to a scanned map or raw image using ground control points (GCPs), then resampling (nearest neighbour for class data, bilinear or cubic convolution for continuous data). Quality is judged by RMSE of residuals at the GCPs.
• Map scale: 1:50,000 toposheets are the Indian standard medium-scale series; 1:25,000 and 1:10,000 for detailed planning. Rule of thumb: smallest mappable detail ≈ 0.25 mm × scale denominator (≈12.5 m at 1:50,000).

Remote sensing: physics and platforms

• Definition and process chain: energy source → interaction with atmosphere → interaction with target → sensor recording → transmission and pre-processing → interpretation and analysis → application.
• Electromagnetic spectrum regions used: visible (0.4-0.7 µm), near infrared NIR (0.7-1.3 µm), shortwave infrared SWIR (1.3-3 µm), thermal infrared TIR (3-14 µm), microwave (1 mm-1 m, radar bands X, C, L).
• Atmospheric windows: wavelength bands where absorption by H₂O, CO₂ and O₃ is low, so sensing is possible. Scattering: Rayleigh (molecules, ∝ λ⁻⁴, makes the sky blue and hazes blue bands), Mie (aerosols), non-selective (cloud droplets).
• Interaction with the target: incident energy is reflected, absorbed or transmitted. Spectral signature = reflectance plotted against wavelength; it is what makes classification possible.
  • Healthy vegetation: low red (chlorophyll absorption), very high NIR (leaf mesophyll scattering).
  • Water: absorbs NIR strongly → almost black in NIR; useful for delineating water bodies.
  • Soil/built-up: reflectance rises gradually with wavelength; concrete and bare soil can be confused.
• Passive vs active: passive sensors record reflected sunlight or emitted thermal energy (cannot see at night in the optical range, blocked by cloud); active sensors supply their own energy — SAR (radar) and LiDAR — and work at night, SAR also through cloud, vital for flood and monsoon mapping in India.
• Platforms: ground-based, airborne (aircraft, drones), spaceborne. Orbits: sun-synchronous polar (near-polar, fixed local solar time, used by Resourcesat, Cartosat, Landsat, Sentinel-2) and geostationary at ~35,786 km (INSAT-3D/3DR, meteorology).
• Scanning geometry: whiskbroom (cross-track oscillating mirror) and pushbroom (linear CCD array, as in Indian LISS and Cartosat sensors).
• Swath, revisit and stereo: swath is the ground width imaged; revisit/temporal resolution is the repeat interval; stereo pairs (fore-aft tilt, as on Cartosat-1) generate DEMs.

Resolutions, sensors and Indian missions

• Spatial resolution: ground size of one pixel (GSD). Spectral resolution: number and width of bands — panchromatic (1 broad band), multispectral (a few bands), hyperspectral (hundreds of narrow bands). Radiometric resolution: bits per pixel (8-bit = 256 grey levels, 12-bit = 4,096). Temporal resolution: revisit period.
• Trade-off: higher spatial resolution usually means smaller swath, larger data volume and lower revisit; a sharper spectral resolution means less energy per band, so more noise.
• Indian programme (ISRO/NRSC): IRS series — Resourcesat-1/2/2A carrying LISS-III (23.5 m, 4 bands), LISS-IV (5.8 m), AWiFS (56 m, wide swath for crop and drought); Cartosat-1 (2.5 m stereo), Cartosat-2 series (sub-metre class), Cartosat-3 (about 0.25-0.3 m panchromatic); RISAT C-band and X-band SAR; Oceansat (OCM, scatterometer); INSAT-3D/3DR (imager and sounder); EOS series designations now used for newer missions. NISAR, the NASA-ISRO L/S-band SAR mission, launched in 2025.
• Foreign data commonly used in India: Landsat 8/9 (OLI, 30 m, free), Sentinel-2 (10/20/60 m, 5-day revisit, free), Sentinel-1 SAR, MODIS (coarse, daily), SRTM 30 m and ASTER 30 m DEMs; CartoDEM is the Indian national DEM from Cartosat-1.
• Indian portals and agencies: Bhuvan (NRSC geoportal, free imagery and thematic layers), Bhukosh (GSI), Bhoonidhi (NRSC's EO data portal for open and priced IRS data), Survey of India OSM sheets and the CORS network, NSDI, NNRMS (National Natural Resources Management System). State centres: KSRSAC, APSAC, MPCST/MPRSAC, etc.; BISAG-N (Gandhinagar) is now a national institute under MeitY, elevated from Gujarat's state centre.
• Policy and law to quote: National Geospatial Policy 2022 (liberalised, targets 1:500 scale urban mapping and high-resolution national DEM, aims at a national geospatial data infrastructure); Geospatial Guidelines, February 2021 (Indian entities freed from prior approval for acquiring and processing geospatial data, with threshold values for fine-resolution data and negative-list restrictions for foreign entities); Remote Sensing Data Policy (RSDP-2011) was superseded by the 2021 guidelines and the Indian Space Policy 2023; drone work falls under the Drone Rules 2021 and the Digital Sky platform. Threshold values and licensing details have been revised more than once — verify against the current notification before quoting in practice.

Image processing and classification

• Pre-processing: radiometric correction (sensor calibration, destriping, de-noising), atmospheric correction (dark-object subtraction, FLAASH/6S, conversion to surface reflectance), geometric correction and orthorectification using a DEM, cloud masking, mosaicking.
• Enhancement: contrast stretching (linear, histogram equalisation), band ratios, filtering (low-pass for smoothing, high-pass/edge detection), principal component analysis to reduce correlated bands, pan-sharpening (fusing a 5.8 m LISS-IV multispectral with a sharper panchromatic band).
• False colour composite (FCC): the standard is NIR→red, red→green, green→blue, so healthy vegetation appears bright red, water near-black, built-up cyan/grey.
• Visual interpretation elements: tone, texture, pattern, shape, size, shadow, site, association, resolution.
• Classification:
  • Unsupervised — the computer clusters pixels (K-means, ISODATA); the analyst labels clusters afterwards. Needs no training data.
  • Supervised — the analyst draws training sites of known classes; algorithms include minimum distance to mean, parallelepiped, maximum likelihood (assumes normally distributed classes, most used classically), and now machine learning (random forest, SVM, CNN).
  • Per-pixel vs object-based (OBIA): OBIA segments the image into homogeneous objects first and uses shape and texture — better for high-resolution imagery where single pixels are noisy.
  • Hard vs soft (fuzzy): hard gives one class per pixel; soft gives class proportions, which handles mixed pixels at boundaries.
• Accuracy assessment: build an error (confusion) matrix against reference points; compute producer's accuracy (1 − omission error), user's accuracy (1 − commission error), overall accuracy and the kappa coefficient (agreement corrected for chance). Reporting classification without an error matrix loses marks in exams and credibility on the job.

GIS-RS integration and civil engineering workflows

• Typical integrated workflow: acquire imagery → pre-process → classify or digitise → bring into GIS with survey/GNSS data → analyse (overlay, network, hydrology) → validate in the field → publish map or web service.
• Why they need each other: remote sensing gives up-to-date, large-area, repetitive coverage but no ownership, names or network connectivity; GIS supplies the database, topology and legacy survey data but cannot itself observe change.
• Accuracy hierarchy: total station / DGPS (cm) > high-resolution satellite ortho (sub-metre to a few metres) > medium-resolution imagery (tens of metres). Never claim an output is more accurate than its worst input, and never mix scales silently.
• Limitations to state in exams: cloud cover in the Indian monsoon, mixed pixels, cost of very-high-resolution data, spectral confusion between similar classes, need for ground truth, datum/projection errors, data currency, restricted-area and licensing constraints.

Key points

• GIS - hardware, software, data, people and methods that store and analyse georeferenced data; data is the costliest component.
• Vector vs raster - vector stores discrete features with exact boundaries; raster stores a grid of cells and suits continuous surfaces and imagery.
• Topology - stored connectivity, adjacency and containment; without it, network and overlay analysis is unreliable.
• Four resolutions - spatial (pixel size), spectral (number/width of bands), radiometric (bit depth), temporal (revisit).
• Spectral signature - reflectance versus wavelength; vegetation is low in red and high in NIR, water absorbs NIR.
• Atmospheric window - a wavelength band of low atmospheric absorption through which sensing is possible.
• Passive vs active - passive uses the Sun or emitted heat; active (SAR, LiDAR) carries its own source and works at night, and SAR also images through cloud.
• Orbits - sun-synchronous polar for Earth-resources satellites; geostationary at ~35,786 km for INSAT meteorological satellites.
• FCC - false colour composite with NIR in red; healthy vegetation appears bright red.
• NDVI - (NIR − Red)/(NIR + Red), ranging −1 to +1; a vegetation vigour index.
• Supervised vs unsupervised - supervised needs training sites and class knowledge; unsupervised clusters first and labels later.
• Kappa coefficient - classification agreement corrected for chance agreement; reported with the error matrix.
• UTM - 6° zones, scale factor 0.9996, false easting 500,000 m; India lies in zones 42N-47N.
• Datum - an ellipsoid fixed in space; Everest 1830 and WGS84 differ by hundreds of metres, so never mix them.
• Georeferencing - fitting an image or scanned map to coordinates using GCPs; quality measured by RMSE of residuals.
• Indian sensors - LISS-III 23.5 m, LISS-IV 5.8 m, AWiFS 56 m, Cartosat-1 2.5 m stereo, Cartosat-3 ~0.25-0.3 m, RISAT SAR.
• Bhuvan and NRSC - the national free geoportal and data archive; Survey of India supplies toposheets, OSM sheets and CORS.
• Policy - National Geospatial Policy 2022 and the February 2021 Geospatial Guidelines liberalised Indian geospatial data use.

Formulas and examples

Example 1 — NDVI from LISS-III reflectance. A pixel gives Red = 0.08 and NIR = 0.46 reflectance; a second pixel gives Red = 0.22, NIR = 0.26.

• Pixel 1: NDVI = (0.46 − 0.08)/(0.46 + 0.08) = 0.38/0.54 = 0.704 → dense healthy vegetation.
• Pixel 2: NDVI = (0.26 − 0.22)/(0.26 + 0.22) = 0.04/0.48 = 0.083 → bare soil or sparse cover.
• A third pixel with Red = 0.05, NIR = 0.02: NDVI = (0.02 − 0.05)/(0.02 + 0.05) = −0.03/0.07 = −0.429 → water body.

Example 2 — Pixel count, area and file size. A LISS-III scene (23.5 m resolution, 4 bands, 10-bit stored in 16-bit) covers a 141 km × 141 km swath.

• Pixels across = 141,000/23.5 = 6,000; down = 6,000 → 36,000,000 pixels per band.
• Area of one pixel = 23.5² = 552.25 m² = 0.055225 ha; scene area = 141 × 141 = 19,881 km².
• File size = 36,000,000 × 4 bands × 2 bytes = 288,000,000 bytes ≈ 274.7 MiB (288 MB decimal).
• At LISS-IV 5.8 m over the same area, pixels across = 141,000/5.8 ≈ 24,310, giving about 5.91 × 10⁸ pixels per band — roughly 16.4 times more data (because (23.5/5.8)² = 16.4).

Example 3 — Area of a classified land-use class. A supervised classification on a 23.5 m raster labels 48,200 pixels as "built-up".

• Area = 48,200 × 552.25 m² = 26,618,450 m² = 26.62 km² = 2,661.8 ha.
• If the same region five years earlier had 31,500 built-up pixels (= 17,395,875 m² = 17.40 km²), the increase is 9.22 km², i.e. 53.0% growth, an annual compound rate of 100 × [(26.618/17.396)^1/5 − 1] = 8.9% per year.

Example 4 — Georeferencing RMSE. Four GCPs give residuals (Δx, Δy) in metres: (1.2, −0.8), (−1.5, 0.6), (0.9, 1.1), (−0.4, −1.3).

• Squared errors: 1.44+0.64 = 2.08; 2.25+0.36 = 2.61; 0.81+1.21 = 2.02; 0.16+1.69 = 1.85. Sum = 8.56.
• RMSE = √(8.56/4) = √2.14 = 1.463 m — acceptable for a 5.8 m pixel (well under one pixel), but not for mapping at 1:1,000 where sub-0.5 m control would be expected.

Example 5 — Classification accuracy and kappa. A 3-class error matrix (rows = classification, columns = reference), N = 200:

• Overall accuracy = (45 + 68 + 58)/200 = 171/200 = 85.5%.
• Producer's accuracy, built-up = 58/70 = 82.9%; user's accuracy, built-up = 58/70 = 82.9% (equal here only because row and column totals coincide).
• User's accuracy, vegetation = 68/80 = 85.0%; producer's accuracy, vegetation = 68/80 = 85.0%.
• Chance agreement pₑ = (50×50 + 80×80 + 70×70)/200² = (2,500 + 6,400 + 4,900)/40,000 = 13,800/40,000 = 0.345.
• κ = (0.855 − 0.345)/(1 − 0.345) = 0.510/0.655 = 0.779 — substantial agreement.

Comparison tables.

Where it is used

• Infrastructure and transport planning: highway and railway alignment studies, DPRs and feasibility reports (NHAI, RITES, L&T, AECOM) using CartoDEM/LiDAR terrain, slope and land-use layers to compare corridors and estimate land acquisition.
• Urban local bodies and smart cities: GIS-based property tax surveys, utility and asset mapping (water, sewer, streetlight), base maps under AMRUT/National Urban Digital Mission, and ward-level layers for master plans.
• Water resources and irrigation: watershed delineation, command-area and canal network mapping, reservoir sedimentation, groundwater potential maps (CGWB, state WRDs), flood inundation mapping with SAR during Bihar, Assam and Kerala floods.
• Disaster management: NRSC/NDMA damage assessment, cyclone and landslide hazard zonation, Bhuvan-based flood and forest-fire alerts; state agencies such as GSDMA.
• Land records and agriculture: digitisation of cadastral maps under DILRMP/SVAMITVA (drone-based village mapping), crop acreage and yield forecasting (FASAL/CCE), PMFBY crop insurance verification.
• Job roles: GIS analyst/engineer, remote sensing analyst, geospatial data technician, photogrammetrist, spatial database administrator and web-GIS developer; typical tool stack ArcGIS Pro, QGIS, ERDAS Imagine, Google Earth Engine, PostGIS, GeoServer, Python with GDAL/rasterio.

Interview questions

• Q: What exactly does a GIS do that a CAD drawing cannot? A: A GIS ties every feature to real-world coordinates and to an attribute database, so it can be queried and analysed (buffer, overlay, network, statistics); a CAD drawing stores geometry for drafting without a spatial database or topology.
• Q: Differentiate raster and vector data with one civil engineering example each. A: Raster is a cell grid, best for continuous data such as a DEM used for slope analysis; vector stores coordinate geometry, best for discrete features such as a pipeline network used for leak tracing.
• Q: Name the four types of resolution in remote sensing. A: Spatial (pixel ground size), spectral (number and width of bands), radiometric (bit depth, hence number of grey levels) and temporal (revisit interval).
• Q: Why does healthy vegetation look bright red on a standard FCC? A: The composite assigns NIR to the red display channel, and healthy leaves reflect NIR strongly while absorbing red light through chlorophyll.
• Q: Why is SAR preferred for flood mapping in India? A: It is active and microwave, so it penetrates cloud and works at night — the only reliable option during the monsoon when floods actually occur, and smooth water gives a distinctly dark return.
• Q: What is a mixed pixel and how do you deal with it? A: A pixel covering more than one land cover, so its signature is a blend; it is handled by finer resolution, soft/fuzzy classification, spectral unmixing or object-based classification.
• Q: Difference between supervised and unsupervised classification? A: Supervised uses analyst-defined training sites of known classes to train the classifier; unsupervised clusters pixels statistically first and the analyst assigns labels to the clusters afterwards.
• Q: What does a kappa of 0.78 mean? A: The classification agrees with reference data substantially better than chance — about 78% of the agreement that was achievable beyond random agreement — and it is a stricter measure than overall accuracy.
• Q: What is the difference between geoid, ellipsoid and datum? A: The geoid is the mean-sea-level equipotential surface, the ellipsoid is a mathematical approximation of it, and a datum is an ellipsoid fixed in position and orientation for a region.
• Q: Which UTM zones cover India and what is the scale factor? A: Mainly zones 42N to 47N (6° wide each), with a central-meridian scale factor of 0.9996 and a false easting of 500,000 m.
• Q: How do you check whether a georeferenced image is good enough for 1:2,000 mapping? A: Compare the GCP RMSE with the permissible planimetric error — roughly 0.25 mm at map scale, i.e. about 0.5 m at 1:2,000 — and verify against independent check points, not the GCPs used in the fit.
• Q: What has the National Geospatial Policy 2022 changed for practice? A: It liberalised the collection, processing and sharing of geospatial data by Indian entities, set goals for high-resolution national base maps and a national DEM, and built on the February 2021 Geospatial Guidelines that removed prior-approval requirements; specific threshold values have been revised, so the current notification should be checked.

Common mistakes

• Mixing datums or projections between layers (Everest 1830 with WGS84, or degrees with metres), producing shifts of hundreds of metres that look like real misalignment.
• Confusing spatial with spectral or radiometric resolution — writing "high resolution" without saying which resolution.
• Computing NDVI on raw DN values after a contrast stretch, or on an uncorrected image, then comparing dates as if the values were calibrated reflectance.
• Reporting a classification without an error matrix, or computing accuracy using the same pixels that trained the classifier.
• Forgetting that a map's usable accuracy is limited by its worst input: overlaying a 1:50,000 land-use layer on a cadastral plan and reading plot boundaries from it.
• Assuming satellite data is always available — ignoring monsoon cloud cover, revisit gaps and licensing or restricted-area constraints.
• Treating pixel count as area without multiplying by (cell size)², or dropping the m²-to-hectare conversion.

Related topics

Photogrammetry and drone surveys · Theodolite, total station and GPS · Surveying: chain, compass and levelling · Transportation engineering (alignment and route surveys) · Hydrology and watershed management · Urban planning and town planning.