Satellite fusion maps wetland flooding daily at 6-meter scale

Researchers at North Carolina State University built a daily, high-resolution wetland inundation map by fusing radar and optical satellite classifications across eastern North Carolina. The approach captures short-lived floods that often slip past standard products, which could improve methane and carbon estimates and support broader wetland monitoring.

Why it matters: - Small, short-lived wetland floods are a major methane source, but they are hard to detect with standard satellite products. - Better flood tracking could improve methane and carbon models by capturing the brief inundation events that change wetland greenhouse-gas exchange. - The method also has potential for monitoring sea-level rise, saltwater intrusion and broader hydrologic shifts.

What happened: - Researchers at North Carolina State University’s Center for Geospatial Analytics and colleagues published the study in Journal of Remote Sensing on 15 October 2026. - The team combined classifications from Sentinel-1, Sentinel-2 and PlanetScope to build a daily wetland inundation record. - The study focused on eastern North Carolina’s Albemarle-Pamlico Peninsula, including Alligator National Wildlife Refuge. - The full report is available here.

The details: - The study classified three surface states: dry land, inundated vegetation and open water. - Sentinel-2 was the most accurate sensor overall at 93.9%. - PlanetScope followed at 88.4% accuracy. - Sentinel-1 radar scored 84.6% accuracy. - Sentinel-2 detected inundated vegetation best. - PlanetScope captured fine open-water features. - Sentinel-1 added cloud-free observations, but it also overpredicted inundation in fields. - Fusion balanced those sensor biases and produced daily inundation probabilities at 6 m resolution. - The fused product captured short flood pulses that individual sensors missed. - The fused map showed more wetland inundation than the most conservative Landsat Dynamic Surface Water Extent class at a moderate threshold. - The gap narrowed when the Landsat threshold became stricter. - The team used five years of imagery from the three satellites. - Training and validation points came from selected dates and were checked against high-resolution reference imagery. - Random forest and extreme gradient boosting models used spectral indices, radar backscatter, terrain, canopy cover and recent precipitation. - The best model for each sensor was applied through time. - Fixed-rank kriging treated each classification as a binary observation, assigned error based on sensor accuracy and predicted a latent inundation process. - A simulation test showed that fixed-rank kriging could recover a known process from irregular observations. - The fusion output included 90% credible intervals. - The maps were compared with water-table, precipitation and lake-elevation records, plus Landsat DSWE. - Imagery was processed with Google Earth Engine and atmospheric correction tools, with strict cloud and cloud-shadow masking. - Predictors included spectral water and vegetation indices, radar backscatter combinations, slope, topographic wetness, canopy cover and 7-day and 30-day precipitation. - The team used stratified sampling, accuracy assessment and the FRK R package. - Benchmarks included Landsat DSWE and last-observation-carried-forward gap filling. - Threshold sensitivity tests ran from 30% to 90%.

Between the lines: - The study shows that optical sensors and radar are complementary rather than interchangeable. - Optical imagery works best in vegetated wetlands, while radar helps when clouds block optical views. - Fusion turns disagreement among sensors into probabilities and uncertainty instead of forcing a single yes-or-no map. - Dense canopy cover and cloud-mask errors still limit performance. - The framework is designed to be transferable to new sensors and existing products.

What’s next: - The authors say the approach could feed methane and carbon models by carrying inundation uncertainty into ensembles. - Future versions could incorporate L-band radar or water-level observations. - The main barriers are computational cost and difficulty detecting water beneath dense canopies. - The authors say multiresolution designs, near-real-time filtering, stricter cloud screening and denser field networks could improve operational use. - The broader goal is a more continuous record of brief wetland floods that shape greenhouse-gas emissions.

The bottom line: - Daily satellite fusion can recover wetland flooding events that single sensors miss, which may tighten estimates of methane emissions and improve long-term wetland monitoring.