Satellite Change Detection for Beginners: Spot Floods, Fires and Urban Growth
- Primary Signal: Satellite Change Detection for Beginners: Spot Floods, Fires and Urban Growth
- Overview: Comparing two satellite images reveals floods, fires, and urban sprawl. Learn the three main change detection methods, which free data to use, and how to build your first change ma...
- Verification: Analyzed and compiled by FreeSky GeoData editorial monitoring desk.
A single satellite image tells you what the Earth looks like. Two images of the same place, taken months or years apart, tell you what changed, and that is often far more useful. Change detection is how analysts map floods, track wildfires, measure urban sprawl, and monitor deforestation, and it is one of the most practical skills in remote sensing. The methods range from simple visual comparison to automated algorithms, and the data is free. This guide explains the three main approaches, which data to use, and how to build your first change map in QGIS.
The simplest change detection method is also the most intuitive: put two images side by side, or use a swipe tool to drag a boundary across them, and look for differences. This works well for dramatic changes such as a new housing estate, a burned forest, or a flooded plain, and it requires no special software beyond QGIS. Load both images as layers, use the same styling for both, and compare them systematically, ideally with a checklist of features to look for. The limitation is subjectivity: two analysts can disagree about what changed, and subtle changes are easy to miss. Visual comparison is best used as a first step, to understand the area and identify the obvious changes, before moving to a quantitative method that can measure the change and map it precisely.
Image differencing subtracts one image from another, pixel by pixel, to highlight where values changed. The most useful version for vegetation is NDVI differencing: calculate NDVI for both dates, subtract the earlier NDVI from the later one, and map the result. Large negative values indicate vegetation loss, such as a fire or deforestation, while large positive values indicate growth. In QGIS, the process is straightforward: calculate NDVI for each date using the Raster Calculator, then subtract one from the other, and style the result with a diverging colour ramp. Set a threshold, for example a change of more than 0.2, to separate real change from noise. This method is quantitative, repeatable, and easy to explain, which makes it the best starting point for most beginners. Its weakness is that it is sensitive to differences in season, cloud cover, and atmospheric conditions between the two dates.
The most rigorous method is to classify both images independently, for example into forest, cropland, urban, and water, and then compare the two classifications to produce a change map. This gives you not just where change happened but what changed into what: forest to cropland, cropland to urban, and so on. The method requires a classification step, which can be done in QGIS with the Semi-Automatic Classification Plugin or in Google Earth Engine with a supervised classifier. The accuracy of the change map depends on the accuracy of both classifications, and errors compound: if each classification is 90 percent accurate, the change map may only be 80 percent accurate. For that reason, always validate the classification with ground truth or high-resolution imagery before trusting the change map. Classification comparison is the standard method for land cover change studies and for reporting deforestation to policymakers.
The biggest trap in change detection is false change: differences between images that look like real change but are not. Seasonal differences are the most common cause: a field that is green in spring and brown in summer will show a large NDVI change, but nothing has actually changed. Always compare images from the same season, ideally the same month, in different years. Cloud shadows and haze cause false change too, so filter for cloud-free images. Differences in sun angle and atmospheric conditions add noise, which is why using atmospherically corrected surface reflectance data, Sentinel-2 Level-2A or Landsat Collection 2, is important. Finally, check your results against high-resolution imagery or local knowledge before reporting them. A change map that claims a forest was cleared when it was actually a seasonal crop is worse than no map at all.
What is satellite change detection? Comparing satellite images from different dates to identify and measure changes on the ground. What is the easiest change detection method? NDVI differencing in QGIS: calculate NDVI for two dates and subtract one from the other. Which satellite data should I use? Sentinel-2 for recent change at 10 metres, Landsat for change over decades. How do I avoid false change? Compare images from the same season, use cloud-free scenes, and use atmospherically corrected data. How accurate is change detection? It depends on the method and the data, but classification-based change maps are typically 80 to 90 percent accurate after validation.