Short, plain-language definitions of the core terms in GIS and remote sensing — each with the original source reference and, where one exists, a link to a fuller explanation. This list grows over time; if a term you're looking for isn't here yet, it likely will be soon.
A measure of vegetation health calculated from how strongly a surface reflects red versus near-infrared light. Healthy vegetation reflects a lot of near-infrared and absorbs red light, producing values from -1 to 1 — higher values generally mean denser, healthier vegetation.
ReferenceRouse, J.W., Haas, R.H., Schell, J.A., & Deering, D.W. (1974). Monitoring vegetation systems in the Great Plains with ERTS. NASA Special Publication, 351, 309–317.
Calculate NDVI for your study area →A vegetation index designed to improve on NDVI in two of its weak spots: it stays sensitive in very dense canopy (where NDVI saturates) and corrects for some atmospheric and soil-background effects, using the blue band alongside red and near-infrared. Often preferred over NDVI in tropical forests and high-biomass regions.
ReferenceHuete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., & Ferreira, L.G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83(1–2), 195–213.
An NDVI variant that adds a soil-brightness correction factor (L) to reduce the influence of bare soil showing through sparse vegetation. Useful in drylands and early-season cropland, where thin canopy over bright or dark soil would otherwise distort a plain NDVI reading.
ReferenceHuete, A.R. (1988). A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25(3), 295–309.
A normalized index built to make open water stand out, using the green and near-infrared bands — water reflects green and strongly absorbs NIR, so it returns high NDWI values while vegetation and soil return low ones. Widely used for delineating lakes, rivers, and flood extent. (A separate NIR/SWIR formulation by Gao is used for vegetation water content.)
ReferenceMcFeeters, S.K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425–1432.
Compute NDWI in the Spectral Index tool →A classification of the Earth's surface into categories like Forest, Urban, Water, or Cropland, usually derived from satellite imagery using a trained classifier. "Land cover" describes the physical material on the ground; "land use" describes how humans are using it — the two are related but not identical.
Classify land cover for your boundary →The temperature of the ground and rooftops as measured by a satellite's thermal infrared bands — distinct from the air temperature a weather station reports. LST is the basis for mapping urban heat islands, since dark, dense surfaces like asphalt and concrete absorb and re-radiate heat differently from vegetation or water.
How LST reveals urban heat islands →
ReferenceVoogt, J.A., & Oke, T.R. (2003). Thermal remote sensing of urban climates. Remote Sensing of Environment, 86(3), 370–384.
Map surface temperature for your area →An active radar sensor that emits its own microwave pulses and measures what bounces back, rather than relying on sunlight. Because those wavelengths pass through cloud and work day or night, SAR (such as Sentinel-1) is invaluable for flood mapping during storms and monsoons, when optical satellites see only cloud. The returned signal — "backscatter" — also responds to surface roughness and moisture.
ReferenceTorres, R., Snoeij, P., Geudtner, D., et al. (2012). GMES Sentinel-1 mission. Remote Sensing of Environment, 120, 9–24.
A measure of classification accuracy that accounts for the accuracy you'd expect purely by chance. A raw accuracy of 80% sounds good, but if one class covers 80% of the map, a classifier that just guesses that class everywhere would also hit 80% — Kappa corrects for this, so a high Kappa is a stronger signal than a high raw accuracy alone.
ReferenceCohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20(1), 37–46.
Compute Kappa with the calculator →Two per-class accuracy measures read off a confusion matrix. Producer's Accuracy asks: of the pixels that truly belong to a class, how many did the classifier catch? (a measure of omission). User's Accuracy asks: of the pixels labelled as a class, how many actually belong to it? (a measure of commission, and the one that tells a map user how much to trust a given class). A class can score well on one and poorly on the other, which is why overall accuracy alone can hide real problems.
ReferenceCongalton, R.G., & Green, K. (2019). Assessing the Accuracy of Remotely Sensed Data: Principles and Practices (3rd ed.). CRC Press.
Summarizing raster values (like temperature or vegetation index) within the boundary of a vector polygon (like a district or a farm plot) — producing a mean, min, max, or standard deviation for that specific zone, rather than for an entire image.
Run zonal statistics in your browser →A table comparing a classifier's predicted classes against the true classes (usually from held-out reference points), showing exactly which classes get confused with which — e.g. how often "Shrubland" gets misclassified as "Grassland." Producer's and User's Accuracy, and Kappa, are all derived from this table.
A non-parametric statistical test that checks whether an apparent trend over time (e.g. rising NDVI over 10 years) is statistically significant, or could plausibly be explained by random year-to-year noise. It doesn't assume the data follows a normal distribution, which makes it well-suited to noisy environmental time series.
ReferencesMann, H.B. (1945). Nonparametric tests against trend. Econometrica, 13(3), 245–259. · Kendall, M.G. (1975). Rank Correlation Methods (4th ed.). Griffin, London.
Run a Mann-Kendall trend test →A machine learning classifier that builds many individual decision trees on random subsets of the training data, then has them vote on the final classification. It tends to be more robust to noisy or limited training data than a single decision tree (CART), at the cost of being somewhat less interpretable.
ReferenceBreiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32.
Train a Random Forest classifier →Two ways to turn imagery into a land-cover map. Supervised classification learns from labelled training points you provide ("this pixel is water, that one is forest") and applies the learned pattern everywhere — Random Forest, CART, and SVM are all supervised. Unsupervised classification (e.g. k-means clustering) instead groups pixels by spectral similarity with no labels, leaving you to interpret what each cluster represents afterward. Supervised methods dominate applied land-cover work because they map directly to the classes you care about.
ReferenceRichards, J.A. (2013). Remote Sensing Digital Image Analysis: An Introduction (5th ed.). Springer.
Try a supervised classification →Collecting reference data — from field visits, high-resolution imagery, or existing verified maps — that represents the actual, known condition of the land. Used both to train a classifier and, critically, to validate its accuracy on samples it never saw during training.
Label training points on a live map →The ground area represented by a single pixel — "30m resolution" means each pixel covers a 30×30m patch of Earth. Finer resolution (a smaller number) resolves smaller features but usually comes with a shorter historical archive or a narrower swath. It's a core trade-off in choosing a sensor: Sentinel-2's 10m pixels resolve individual fields that Landsat's 30m pixels blur together, while Landsat reaches back to 1984.
Resolution vs. history, compared →
ReferenceCampbell, J.B., & Wynne, R.H. (2011). Introduction to Remote Sensing (5th ed.). Guilford Press.