A multisensor, multitemporal approach for monitoring herbaceous vegetation growth in the Amazon floodplain

Abstract

The Amazon River floodplain is an important source of atmospheric CO2 and CH4. Aquatic herbaceous vegetation (macrophytes) have been shown to contribute significantly to floodplain net primary productivity (NPP) and methane emission in the region. Their fast growth rates under both flooded and dry conditions make herbaceous vegetation the most variable element in the Amazon floodplain NPP budget, and the most susceptible to environmental changes. The present study combines multitemporal Radarsat-1 and MODIS images to monitor spatial and temporal changes in herbaceous vegetation cover in the Amazon floodplain. Radarsat-1 images were acquired from Dec/2003 to Oct/2005, and MODIS daily surface reflectance products were acquired for the two cloud-free dates closest to each Radarsat-1 acquisition. An object-based, hierarchical algorithm was developed using the temporal SAR information to discriminate Permanent Open Water (OW), Floodplain (FP) and Upland (UL) classes at Level 1, and then subdivide the FP class into Woody Vegetation (WV) and Possible Macrophytes (PM) at Level 2. At Level 3, optical and SAR information were combined to discriminate actual herbaceous cover at each date. The resulting maps had accuracies ranging from 80% to 90% for Level 1 and 2 classifications, and from 60% to 70% for Level 3 classifications, with kappa values ranging between 0.7 and 0.9 for Levels 1 and 2 and between 0.5 and 0.6 for Level 3. All study sites had noticeable variations in the extent of herbaceous cover throughout the hydrological year, with maximum areas up to four times larger than minimum areas. The proposed classification method was able to capture the spatial pattern of macrophyte growth and development in the studied area, and the multitemporal information was essential for both separating vegetation cover types and assessing monthly variation in herbaceous cover extent.