Global Wetlands Mapping Using ALOS PALSAR data
Bruce Chapman(1), Kyle McDonald(1) and Laura Hess(2)
(1) Jet Propulsion Laboratory, 4800 Oak Grove Dr., Pasadena CA 91109, United States
(2) UC Santa Barbara, ICESS, Santa Barbara CA 93106, United States
*** Submitted to the "ALOS K&C Special Session" ***
ALOS PALSAR, an orbiting L-band SAR launched by the Japanese Aerospace and Exploration Agency (JAXA) in 2006, has been pursuing a global observation strategy through its ALOS Kyoto and Carbon Initiative (ALOS KC). As a component of this task, large collections of ScanSAR data are being acquired over wetland areas around the globe. Through the NASA MEASURES program, JPL will be leading an effort to utilize this data to produce a global inundated wetlands product. One of the first steps will be to produce continental-scale mosaics of SAR imagery at maximum and minimum inundation extent.
The heritage of this project is the Global Rain Forest Mapping Project (GRFM), led by the Japanese space agency, in which three agencies (the Jet Propulsion Laboratory, the EC Joint Research Center, and the Earth Observation Research Center of the Japanese space agency) independently used various methods to construct continental-scale mosaics of tropical forest regions (the Amazon River Basin, the Congo River Basin, and South East Asia). This was followed by the Global Boreal Forest Mapping project (GBFM), for which boreal forest regions (in particular, the North American Boreal forests in Alaska and Canada) were mosaicked into seamless data products. These activities occurred in the late 1990s and early 2000s.
In assembling the ALOS SAR mosaics for the global inundated wetlands product, in most cases the mosaics will be orthorectified to the SRTM DEM. The regions mosaicked will be more diverse than that of the GRFM and GBFM and will include most globally significant wetland regions, due to the comprehensive nature of the ALOS KC acquisition plan. The images to be mosaicked will be image ‘strips’, often thousands of kilometers along track, rather than image frames which are roughly as long along track as the cross track dimension. Over certain regions, the ScanSAR acquisition plan of wetland areas will result in dense multi-temporal coverage. For these regions, there is greater flexibility in terms of image selection, with the objective being to capture maximum and minimum inundation extent in the final mosaics. In addition, by examining pixel-by-pixel of the multi-temporal coverage, it will be possible to estimate the duration of flooding for each pixel.
There are three key areas critical to the success of the mosaicking process. The first is accurate geolocation of the data, the second is accurate calibration of the data, and the third is the scientific order of preference for each image’s appearance in the mosaic. Once the data is accurately geolocated and calibrated, the actual mosaicking of the data is a relatively trivial step. Rather than throw out the substantial temporal overlap for a particular region, we will implement a flexible mosaicking option for user-defined temporal selection of image scenes to be included in the mosaic. This can be accomplished in an automatic fashion using a web interface and other tools for selecting the imagery, spawning the stitching of the data, and ftp retrieval of the completed mosaic.
This paper was partially written at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.