Dual-season mapping of wetland inundation and vegetation for the central Amazon basin
Introduction
Riverine floodplains and other wetlands are common features of the Amazon basin, where they alter flood waves (Richey, Mertes, et al., 1989), store sediments (Dunne, Mertes, Meade, Richey, & Forsberg, 1998), and provide important ecological habitats (Junk, 1997). Measurements in floodplains have shown the significance of these environments to regional carbon biogeochemistry Melack & Forsberg, 2001, Richey et al., 1990. For example, outgassing of CO2 from water to the atmosphere, extrapolated over the whole basin, is at least 10 times the fluvial export of organic carbon to the ocean (Richey, Melack, Aufdenkampe, Ballester, & Hess, 2002). The outgassing is fueled primarily by organic carbon from riparian and flooded forests and floating macrophytes and fluctuates seasonally with changes in inundation. Emission of CH4 fluctuates seasonally as well, and accurate estimation of methane emissions from Amazonian wetlands requires knowledge of seasonal changes in vegetation (Devol, Richey, Forsberg, & Martinelli, 1990) and inundation (Rosenqvist, Forsberg, Pimentel, Rauste, & Richey, 2002). In addition to biogeochemical applications, information on the seasonal extent of floodplain habitats is also required for effective management of Amazon fisheries, since many key fish species harvested in the Amazon basin obtain nutrition in flooded forests (Goulding, 1980) or from organic matter derived from floodplain algae (Forsberg, Araujo-Lima, Martinelli, Victoria, & Bonassi, 1993).
Mapping of wetlands extent and inundation is thus essential to the Large-Scale Biosphere–Atmosphere Experiment in Amazônia (LBA), which focuses on understanding the ecological functioning of the region with particular emphasis on the carbon cycle (Avissar & Nobre, 2002). Although estimates of wetland area in the Amazon basin exceed 1 million km2 (Junk, 1997), this value is based on actual measurements in few places. The quantitative analysis of inundation and vegetation dynamics of wetlands in the Amazon basin requires appropriate remotely sensed data, acquired basin-wide. Optical sensors such as Landsat Thematic Mapper have been used to map wetland vegetation in the Amazon Mertes et al., 1995, Novo & Shimabukuro, 1997, but are limited by the fact that vegetation often covers underlying waters, and clouds or smoke frequently obscure the ground. Global or continental maps of land cover derived from coarse-resolution optical data have emphasized nonwetland cover: they have either omitted wetlands as a class Hansen et al., 2000, Stone et al., 1994 or have significantly underestimated wetland area (Loveland et al., 2000).
Passive and active microwave sensors, which are much less influenced by clouds and smoke and can penetrate vegetation at some wavelengths, have also been employed for wetlands mapping Melack & Hess, 1998, Prigent et al., 2001, Sippel et al., 1994. Many studies (reviewed by Lewis, 1998) have successfully used synthetic aperture radar (SAR) sensors to map inundation and wetlands vegetation. When specular reflections from an underlying water surface interact with vegetation via double-bounce or multiple scattering, SAR backscattering is enhanced (Hess, Melack, & Simonett, 1990). Although C-HH sensors (C-band, horizontal send polarization, horizontal receive polarization) such as RADARSAT are able to delineate sub-canopy inundation in some types of floodplain forests (Townsend, 2001), the longer wavelength of L-HH sensors such as the Japanese Earth Resource Satellite-1 (JERS-1) maximizes canopy penetration and discrimination between flooded and nonflooded forest (Hess, Melack, Filoso, & Wang, 1995). Flooding in forested wetlands of the Amazon has been successfully mapped using L-HH imagery from JERS-1 Melack & Wang, 1998, Rosenqvist et al., 2002, Saatchi et al., 2000 and Shuttle Imaging Radar-C (SIR-C) (Hess, 1999). The combination of JERS-1 L-band and RADARSAT C-band data is preferable to either sensor alone for studying biomass and species composition of aquatic macrophyte communities of Amazon lakes and reservoirs Costa et al., 2002, Novo et al., 2002. However, the sites for which near-simultaneous JERS-1 and RADARSAT data have been acquired are few.
SAR images of the Amazon basin acquired by JERS-1 during low- and high-water periods (Rosenqvist et al., 2000) provide the first data set suitable for regional mapping of Amazon wetlands. We present here our approaches for mapping wetland vegetation and inundation for the central Amazon at approximately 100-m resolution, and for validating these products using high-resolution digital videography. In addition to producing the first high-resolution wetlands map for the region, our study addresses general questions relevant to wetlands mapping and to land cover mapping of large regions at high resolution, such as:
- •
How useful is image segmentation for delineating the complex spatial patterns of riverine wetlands, and for increasing the ease with which they can be distinguished from spectrally similar nonwetland land cover types?
- •
Is it practical to apply image segmentation to large data sets having both complex land cover patterns and significant radiometric uncertainties?
- •
Given the inherent limitations of using a single wavelength and polarization to map a broad range of vegetation types, how can dual-season data, timed to high- and low-water periods, be used to best advantage in a classifier?
- •
In remote, cloud-covered regions, what is an appropriate validation method for high-resolution maps of dynamic processes such as flooding?
Section snippets
SAR mosaics
The Amazon wetlands mapping was carried out using mosaics of SAR images acquired by JERS-1 as part of the Global Rain Forest Mapping (GRFM) Project. Begun in 1995, the GRFM project is an ongoing effort led by the National Space Development Agency of Japan (NASDA) to produce spatially and temporally contiguous SAR data sets over Earth's tropical regions (Rosenqvist et al., 2000). Since the primary scientific objective for the Amazon basin acquisitions was to determine seasonal patterns of
Wetlands mask
The wetlands mask for the central Amazon is shown in Fig. 4a. Seventeen percent of the quadrat, approximately 300,000 km2, is occupied by wetlands. This figure represents the maximum inundatable area, not the area actually flooded on the high-water mosaics. Because of regional differences in the timing of inundation, the entire wetland area would not be flooded on a single date. Moreover, the highest elevations on floodplains may not flood every year. Channels and seasonally inundated
Mapping methodology
Although L-HH SAR is the best single-band sensor for mapping central Amazon wetlands, the backscattering response from nonflooded forest—the most common nonwetland cover—overlaps those of common wetland vegetation such as forest and aquatic macrophyte. The difficulty of class discrimination is compounded by residual radiometric anomalies in the mosaics. A classification based on backscattering alone would thus result in misclassified flooded forest or macrophyte pixels scattered throughout the
Summary
Using the Global Rain Forest Mapping Project dual-season mosaics of JERS-1 SAR data, we have mapped wetland habitats at 100 m resolution for an 18×8° quadrat of the central Amazon region. Wetlands were defined for this purpose as inundatable areas. A mask showing wetland extent was created by a procedure of image segmentation and clustering, followed by human interpretation and editing of clustered polygons. Within the wetlands portion of the study area, the radar mosaics were classified into
Acknowledgements
We thank Dana Slaymaker, Ake Rosenqvist, Bruce Chapman, Andre Monteiro, and Ryan Ashker for their contributions in acquiring and processing the radar and videographic data sets. Support was provided by NASA grants to the GRFM-JAMMS project at UCSB and to LBA-ECO investigation LC-07. Acquisition of the JERS-1 imagery was made possible by NASDA's Global Rain Forest Mapping Project.
References (73)
- et al.
SPRING: Integrating remote sensing and GIS by object-oriented data modeling
Computer Graphics
(1996) Flood tolerance and tree distribution in central Amazonian floodplains
- et al.
Spatial patterns of hydrology, geomorphology, and vegetation on the floodplain of the Amazon River in Brazil from a remote sensing perspective
Geomorphology
(1995) Classifier performance and map accuracy
Remote Sensing of Environment
(1996)- et al.
Mapping deforestation and secondary growth in Rondonia, Brazil, using imaging radar and Thematic Mapper data
Remote Sensing of Environment
(1997) - et al.
Determination of inundation area in the Amazon River floodplain using the SMMR 37 GHz polarization difference
Remote Sensing of Environment
(1994) - et al.
Selective logging along the middle Solimões River
- et al.
Preface to special issue on the Large-Scale Biosphere–Atmosphere Experiment in Amazonia (LBA)
Journal of Geophysical Research
(2002) As Matas de Várzea do Mamirauá
(1993)- et al.
Mapping Amazon Basin wetlands through region-growing segmentation and segmented-based classification of JERS-1 data
Um método de classificação não-seu pervisionada por regiões
Satellite imagery segmentation: A region growing approach
Quantitative comparison of the performance of SAR segmentation algorithms
IEEE Transactions on Geoscience and Remote Sensing
The JERS Amazon Multi-Season Mapping Study (JAMMS): Observation strategies and data characteristics
International Journal of Remote Sensing
Assessing the accuracy of remotely sensed data: Principles and practices
Biophysical properties and mapping of aquatic vegetation during the hydrological cycle of the Amazon floodplain using JERS-1 and Radarsat
International Journal of Remote Sensing
Seasonal dynamics in methane emissions from the Amazon River floodplain to the troposphere
Journal of Geophysical Research
Segmentation and classification of vegetated areas using polarimetric SAR image data
IEEE Transactions on Geoscience and Remote Sensing
Exchanges of sediment between the flood plain and channel of the Amazon River in Brazil
Geological Society of America Bulletin
National vegetation classification standard
Autotrophic carbon sources for fish of the central Amazon
Ecology
The JERS-1 Amazon Multi-Season Mapping Study (JAMMS): Science objectives and implications for future missions
International Journal of Remote Sensing
Testing land-use map accuracy: Another look
Photogrammetric Engineering and Remote Sensing
The fishes and the forest
Prospects for arable farming in the floodplains of the central Amazon
Global land cover classification at 1 km resolution using a classification tree approach
International Journal of Remote Sensing
On structural complexity and fish diversity in an Amazonian floodplain
Delineation of inundated area and vegetation along the Amazon floodplain with the SIR-C synthetic aperture radar
IEEE Transactions on Geoscience and Remote Sensing
Radar detection of flooding beneath the forest canopy: A review
International Journal of Remote Sensing
Geocoded digital videography for validation of land cover mapping in the Amazon basin
International Journal of Remote Sensing
Sedimentation and sediments of Amazonian rivers and evolution of the Amazonian landscape since Pliocene times
Herbaceous plants of the Amazon floodplain near Manaus: Species diversity and adaptations to the flood pulse
Amazoniana
Plant life in the floodplain with special reference to herbaceous plants
Geomorphic and hydrologic applications of active microwave remote sensing
Cited by (487)
The use of satellite remote sensing for exploring river meander migration
2023, Earth-Science ReviewsWeekly mapping of surface water extent in the intertropical wetlands using spaceborne GNSS reflectometry
2023, Journal of HydrologyExpressive fluxes over Amazon floodplain revealed by 2D hydrodynamic modelling
2023, Journal of HydrologyQuantification of wetland vegetation communities features with airborne AVIRIS-NG, UAVSAR, and UAV LiDAR data in Peace-Athabasca Delta
2023, Remote Sensing of EnvironmentContribution of automatically generated radar altimetry water levels from unsupervised classification to study hydrological connectivity within Amazon floodplains
2023, Journal of Hydrology: Regional Studies