‹ back to GIS Glossary

Remote Sensing Workflows

Definition

A remote sensing workflow is the complete, repeatable pipeline that turns raw imagery from satellites, aircraft, or drones into decision ready information. It begins with planning the acquisition window and choosing sensors that match the spatial, spectral, radiometric, and temporal requirements. Raw scenes are then corrected for radiometry and atmosphere, orthorectified to remove terrain and sensor geometry effects, and mosaicked while masking clouds and shadows. Next comes feature engineering such as vegetation indices, texture, and spectral transforms, followed by supervised or unsupervised classification, object based image analysis, or change detection. Results are validated with accuracy assessments and uncertainty estimates, documented with metadata and lineage, and finally published to web services or automated monitoring dashboards so the same steps can run at scale across time and sensors.

Application

Well designed workflows enable crop monitoring, deforestation alerts, wildfire mapping, shoreline change, illegal mining detection, and rapid damage assessment after storms or earthquakes. Municipalities use them to map impervious surfaces, heat islands, and roof suitability for solar. Conservation teams monitor habitat condition and human encroachment. Utilities watch right of way vegetation. Insurers estimate exposure and post event losses. Because the workflow is scripted and versioned, teams can reproduce historical baselines, compare sensors like Sentinel‑2, Landsat 8–9, and commercial constellations, and harmonize drone surveys with satellite time series. The same pattern also supports model deployment, for example shipping a trained land cover classifier so it runs whenever a new scene arrives.

FAQ

Why is atmospheric correction essential in a remote sensing workflow?

Atmospheric gases, aerosols, and water vapor distort the signal that the sensor records, so top of atmosphere reflectance varies with day and viewing geometry. Atmospheric correction estimates surface reflectance that is comparable across dates, sites, and sensors. Without it, indices like NDVI drift, change detection produces false positives, and models trained at one time fail when applied later.

How do you design a reproducible workflow that spans multiple sensors and years?

Use open specifications for inputs and outputs, lock versions of algorithms and coefficients, and script every step from discovery to publication. Harmonize bandpasses and resolutions, resample consistently, and store parameters in configuration files so the same code can target Landsat, Sentinel, or a drone. Capture provenance and quality flags so analysts understand what scenes and masks produced each pixel.

Which accuracy and uncertainty checks make results trustworthy for decisions?

Split reference data into train and test sets, report confusion matrices, kappa or F1 scores, and spatial cross validation to reduce spatial autocorrelation bias. For continuous outputs, provide MAE and RMSE with prediction intervals. Map per pixel confidence or class probability and track sample frame coverage so end users know where results are strong or weak.

When should processing run on the edge, on premises, or in the cloud?

Edge processing fits drone missions that need immediate results like search and rescue or precision spraying. On premises works when data sovereignty rules or very high bandwidth sensors make upload impractical. Cloud pipelines shine for time series stacks, scalable training, and multi user access, especially when compute can be scheduled near public archives to avoid egress costs.