We propose to conduct a multi-temporal, multi-scale, multi-sensor analysis of
inundation and wetland vegetation in the Amazon basin that will be linked to
biogeochemical measurement and modeling activities of LBA.
Our proposed remote sensing analyses will include optical (Landsat, AVHRR and EOS
sensors), passive microwave (SMMR/SSMI) and active microwave (SIRC, JERS, ERS, and
Radarsat) data to determine the temporally varying extent of inundation and associated
vegetation. We will (1) provide synoptic, seasonal mapping of inundation and wetland
vegetation structure for the Amazon basin; (2) incorporate the inundation and vegetation
data into a GIS-based database; and (3) apply results from our analyses of wetland
vegetation and inundation to related LBA studies of hydrological, ecological, and
The expected time periods for recording an image useable for inundation analysis with
the different remote sensing instruments we propose to employ varies from days to months.
Based on an analysis of the spatial and temporal resolutions of the satellite sensors, we
have determined that data fusion amongst the instruments will be critical to insure
sufficient temporal coverage at the appropriate spatial scales. We anticipate that there
will be an inundation-mapping limit for typical rivers with contributing drainage basins
on the order of 1000-10,000 km2 when the flood conditions occur only for a week to a
month. Individual sites may be mappable above and below this limit, depending on the local
geomorphology and inundation hydrology.
We anticipate results of our analyses to be important for LBA activities associated
with (1) methane and other trace gas emissions, (2) carbon dynamics of flooded forests,
(3) land use on flood plains, (4) regional hydrologic modeling, and (6) detection of
seasonal and inter-annual climate variability.
Implementation of our activities will be done as follows:
Optical sensing - The most effective technique for tracking the zone of river-water
influence, in contrast to local-water influence, on wetland inundation is through the use
of optical data of sufficiently fine resolution. We will modify our current method for
suspended sediment analysis to incorporate data from the optical instruments expected
during LBA, i.e., Landsat 7, MODIS, and AVHRR. The optical image analysis will be limited
by access to sufficiently cloud and smoke-free data.
Active microwave sensors - We plan to use a multi-stage, hierarchical, rules-based
approach to delineate floodplain inundation and vegetation using a decision-tree model
which constructs a binary classification tree by recursively partitioning the training
data into increasingly homogeneous subsets. Inundation mapping at a fine scale (12.5 m to
100 m pixel spacing) will be carried out using a combination of JERS (LHH), Radarsat
(CHH), and ERS-2 (CVV) data. All these data will offer multi-temporal coverage of selected
regions; only Radarsat's ScanSAR is expected to provide multi-temporal coverage of the
whole Amazon basin during LBA.
Passive microwave sensors - Low-resolution sensors, such as the Scanning
Multi-channel Microwave Radiometer (SMMR) and the Special Sensor Microwave Imager (SSMI),
afford a synoptic view of the Amazon basin that complements finer-resolution SAR and
optical data. We have developed linear mixing models that incorporate the observed
microwave signature's major end members to estimate fractional inundation area. Data will
be obtained as 0.25° x 0.25° grid cells, and modal values for each 2-week period will be
used to determine inundation area
Field studies - Field surveys will be required at key LBA sites and selected
wetlands to validate our classifications and maps. These surveys will entail low altitude
videography and surface inspections from land and water. Geo-location of flight lines and
surface sites will be done with portable GPS units. Further, we will use information
obtained from the many hours of low altitude videography and field observations that we
have collected previously, as well as from published literature and from personal
contacts. To improve correlation between water levels and inundation extent in wetlands
distant from major rivers or gauging stations on rivers, we will install automatic water
During the first year, we will emphasize extension and validation of our microwave and
optical classification algorithms and assembly of mutli-temporal data sets from ERS,
Radarsat, JERS, Landsat and SSMI acquisitions. During the second and third years, we will
emphasize production and distribution of inundation and wetland vegetation maps to
relevant LBA projects and for our own complementary analyses.