The Advanced Spaceborne Thermal Emission and Reflection radiometer (ASTER) is an advanced multispectral imaging sensor that was launched on board the Terra spacecraft in December, 1999 [1
]. ASTER has an along-track stereoscopic viewing capability in its visible and near-infrared (VNIR) bands at 15-m spatial resolution with a base-to-height ratio of 0.6. Because of ASTER’s excellent satellite ephemeris and instrument parameters [3
], this along-track stereoscopic viewing capability makes it possible to generate excellent digital elevation model (DEM) data products from ASTER data without referring to ground control points (GCPs) for individual scenes [5
After nearly a decade of ASTER data acquisition, sufficient cloud-free data had been acquired such that it was possible to create a global DEM from ASTER data (ASTER GDEM). Versions 1 and 2 of the ASTER GDEM, based on 1.2 and 1.5 million scene-based ASTER DEMs, respectively, were released jointly by the Ministry of Economy, Trade, and Industry (METI) of Japan and the U.S. National Aeronautics and Space Administration (NASA) in 2009 and 2011 [6
]. ASTER GDEM Version 3, which was derived from about 1.9 million scene-based DEMs, will be released to the public sometime in 2018.
Waterbody detection is an essential part of DEM generation, because image matching is not directly possible for waterbodies. For ASTER GDEM Version 2, waterbody detection included a methodology for separating land water bodies from the rest of the land surface and then assigning them a flattened elevation value [5
]. The methodology applied was valid only for waterbodies contained in the same 1° latitude-by-1° longitude tile area. Lakes that cross tile boundaries and are situated in two adjacent 1° latitude-by-1° longitude tiles may have slightly different elevations in the two adjacent tiles. Another shortcoming of the ASTER GDEM Version 2 approach to waterbody detection and correction was that river elevations did not uniformly step-down from upstream to downstream. No global water data base was released to the public with ASTER GDEM Version 2.
In recent years, many attempts have been made to create global waterbody databases because of their importance in studying global biogeochemical cycles [7
]. Such databases still have shortcomings related to nonglobal coverage, spatial resolution, and public availability. Although the Shuttle Radar Topography Mission (SRTM) Waterbody Data product (SWBD) satisfies spatial resolution and public availability requirements, the coverage is not global. Rather, data from that mission were collected only between 56° south 60° north latitudes. ASTER data and the ASTER GDEM cover land surface areas between 83° south 83° north latitudes, an important attribute in the generation of a global waterbody database. This paper describes the methodology applied in the production and improvement of a global water database (GWBD) from ASTER data (ASTER GWBD).
In spite of its shortcomings, the SWBD was still useful in the creation of the ASTER GWBD. The SWBD’s ESRI Shapefile format was converted to a raster format for comparison with ASTER GWBD. Another dataset useful in creating the ASTER GWBD was the GeoCover2000 [15
], which was produced from Landsat 7 data. The original GeoCover2000 dataset, covering the Earth with 14.25 m spatial resolution and UTM coordinates, was converted to the same spatial resolution and coordinates as the ASTER GWBD i.e., to geographic latitude/longitude coordinates with 1 arcsecond postings, and 1° latitude-by-1° longitude tile size. This conversion facilitates accurate comparison with the ASTER GWBD.
ASTER GWBD generation consists of two parts: separation of waterbodies from land areas and classification of detected waterbodies into three categories: sea, river, and lake. The separation process was carried out during scene-based DEM generation using an algorithm described in our previous paper [5
]. However, many aspects of ASTER GWBD generation and enhancement required manual intervention, including visual feature identification. Such work was accomplished using our support tool which utilizes ‘region of interest’ (roi) and ‘masking’ functions of ‘ENVI’ image analysis software by Harris Geospatial Solutions.
As mentioned previously, the tile-based water body data are generated from the scene-based waterbody data simultaneously with ASTER GDEM generation, but they were not publicly released as ASTER GWBD with ASTER GDEM Version 2 because of the imperfections previously noted. The new ASTER GWBD was developed in conjunction with ASTER GDEM Version 3 to incorporate the improved water body data into ASTER GDEM Version 3. Important improvements were made to the ASTER GWBD:
Waterbodies are classified into three categories: sea, lake, and river waterbodies based on their features.
Sea-waterbodies have zero elevation.
Lake-waterbodies have flattened (uniform) elevations.
River-waterbody elevations step down monotonically from upstream to downstream.
This paper describes how these improvements to the ASTER GWBD were accomplished. The new ASTER GWBD product consists of a global set of 1° latitude-by-1° longitude tiles that contain water body attribute and elevation data files in geographic latitude and longitude coordinates and with one arc-second postings. Consequently, each tile contains 3601-by-3601 data points, including one common column and one common row with its neighboring tiles. Section 2
describes the basic configuration the GWBD product. Section 3
describes the processing algorithm for sea-waterbody. The major part of the algorithm is zero elevation setting and sea ice removal. Section 4
describes the processing algorithm for lake-waterbodies. The major part of the algorithm is the unique elevation value regardless of the size. Section 5
describes the processing algorithm for river-waterbodies. The major of the algorithm is step down elevation from upstream to downstream. Section 6
describes the processing algorithm how to incorporate the improved waterbody elevation data into GDEM to reflect the improved results. In addition to ASTER GDEM V3, the improved ASTER GWBD also will be released to public sometime in 2018.
In order to set the elevation of sea-waterbodies to zero, they first must be separated from inland lakes and rivers. This separation was carried out for scene-based DEM generation using the global sea-waterbody database that was created using GTOPO30 [5
]. If the sea-waterbody area is larger than 80% of the sea-waterbody GTOPO30 database, this area is identified as a sea-waterbody. The 80% criterion was adopted to compensate for the inaccuracy of the database. The land–sea interface (sea shoreline) is determined during ASTER GDEM generation by calculating the ratio of the number of stacked-sea-waterbody data to the total number of stacked-pixel data. If the ratio is larger than 0.5, the pixel is assigned as a part of the sea waterbody. Otherwise, the pixel is considered to be land. The 50% criterion was adopted in consideration of tidal effects to present an average delineation [6
In high latitude areas, another obstacle to accurate delineation of the sea shoreline is the presence of sea ice, whose effects must be removed if sea shorelines are to be accurately delineated in the ASTER GWBD and GDEM. Target areas for sea ice removal were selected using the global coarse mosaic image that was generated from original ASTER GDEM data. A sea ice removal process was carried out for the following high latitude target areas.
Latitudes of 60 degrees north and further north areas
Latitudes of 60 degrees south and further south areas
Extreme south of Greenland
Sea of Okhotsk
shows the algorithm flow for sea ice removal. It is difficult to delineate sea ice that occurs near sea shorelines using DEM data alone, because sea ice elevations frequently are similar to land elevations near the sea shoreline. Most sea ice exhibits elevations lower than 30 m, but land topography near sea shorelines often does not exceed 30 m, thus the possibility for confusion. Consequently, sea ice removal for the ASTER GWDB utilized ancillary data wherever useful data exist. Two such useful datasets were (1) the Canadian Digital Elevation Data (CDED) [17
], which covers all of Canada with postings every 3 arc-seconds for latitude and every 3, 6, or 12 arc-seconds for longitude, depending on latitude and (2) Alaska Digital Elevation Data [18
] that covers the State of Alaska with postings every 2 arc-seconds of latitude and longitude. The processing steps employed in sea ice removal are summarized as follows.
Generate a 2° latitude-by-2° longitude tile mosaic from unimproved Version 3 ASTER GDEM data.
If the mosaic area includes any sea shoreline, continue the processing.
If ancillary reference data exist in the mosaic area, delineate by comparing the mosaic data with the reference data by visual identification under the support tool. If reference data are not available, delineate by using brightness contrast under the support tool. In this case, the baseline technique is to designate all areas less than 30 m as sea ice, and then exclude land areas from the sea ice areas by comparing to Google Earth and GeoCover images and/or by visual judgments, as necessary and possible.
The improved sea-waterbody data were incorporated into the ASTER GWBD.
Repeat step (1) to step (4) for all sea ice removal target areas.
presents typical results from the sea ice removal process. Several examples are shown of original and corrected DEM images. Most, but not all, of the gray scale image areas shown in each first image of Figure 3
represent sea ice with elevations lower than 30 m. Some, however, are land areas that had to be identified and retained by manual intervention.
A waterbody detection technique is an essential part of DEM generation to delineate land–water boundaries and to set flattened elevations. This paper described the technical methodology for improving the initial tile-based waterbody data that are created during generation of the ASTER GDEM, but which are not suitable for incorporating into new ASTER GDEM Version 3. Waterbodies are classified into three categories: sea, lake, and river.
Sea-waterbodies were separated from inland waterbodies, and their elevations were set to zero. The effects of sea ice were removed to better delineate sea shorelines in high latitude areas, because sea ice prevents accurate delineation of sea shorelines. This process was enhanced by reference to ancillary data, specifically Google Earth and GeoCover images.
Lake waterbodies are classified into three groups based on size. Group1 lakes are much larger than scene DEMs, which thus do not include enough land area to define the lake or calculate its elevation. For Group1 lakes the corresponding SWBD attribute image was used to define the ASTER GWBD area, and the nominal elevation value was used to assign the lake elevation. Group2 lakes have a size larger than a 2° latitude-by-2° longitude tiles mosaic image of ASTER GDEM data and do not belong to group1 lakes. Group3 lakes are all other lakes. For group2 and group3 lakes, the elevation for each lake was calculated from the perimeter elevation data using the mosaic image that covers entire area of the lake.
River elevations are not constant but gradually decline from upstream to downstream. Rivers were separated from lakes by visual inspection, because there is no automated way to discriminate between rivers and lakes. A stepwise elevation assignment was carried out for rivers using manual or automated methods, depending on the situation under support program.
All improved waterbody elevation data were incorporated into the ASTER GDEM Version 3 to reflect the improved results. At the same time, the waterbody perimeter elevations were edited such that those were at least one meter higher than the waterbody elevation.