Satellite Data Fusion
Definition
Satellite data fusion combines observations from different sensors or times to produce a product that is richer or more accurate than any single source. Approaches include spatial fusion that injects fine detail from a high resolution image into a lower resolution product, spectral fusion that merges bands from different instruments, and temporal fusion that predicts daily values from weekly or monthly time series. The methods range from Bayesian models to machine learning and physics based energy balance. A successful fusion workflow documents sensor characteristics, aligns geometry, handles uncertainty, and produces outputs with clear error estimates.
Application
Common applications include sharpening vegetation and temperature products, filling gaps from clouds, harmonizing historic sensors with new missions, and fusing radar with optical data to map floods in all weather. Agriculture gains daily crop condition at field scale. Urban studies combine night lights with thermal and land cover to map heat exposure. Conservation projects blend optical and radar to monitor forest disturbance even during rainy seasons.
FAQ
What are the main risks when fusing sensors with very different spatial resolutions?
Mixed pixels and scale mismatch can invent detail that is not physically present. Use downscaling methods that conserve energy and validate against independent high resolution references. Report uncertainty so users understand where small features are reliable and where they are not.
How do you handle geolocation differences before attempting fusion?
Co register images using ground control points, tie points, or sensor models, and resample to a common grid. Check for parallax in high relief areas and correct with a digital elevation model. Even small misalignments can create false change in a fused product.
When is model based temporal fusion preferable to simple compositing?
When the goal is to estimate daily or hourly dynamics such as evapotranspiration or photosynthesis, models that use physics or learned relationships fill cloudy gaps more faithfully. Simple compositing only selects the best clear pixel and cannot represent rapid changes or diurnal cycles.
How should fused products present confidence to end users?
Publish per pixel uncertainty or probability layers, provide validation statistics by land cover class, and describe conditions where error increases. Offer both the fused surface and a quality mask so analysts can filter or weight results in later models.
SUPPORT
© 2025 GISCARTA