Streams and Catchments

TDX-Hydro

GEOGLOWS V2 uses a modified subset of the TDX-Hydro streams dataset. TDX-Hydro is a hydrography dataset derived from a 12m Airbus DEM created by the National Geospatial-Intelligence Agency (NGA). It is a public dataset first released in summer 2023. Production of TDX-Hydro is based on a delineation using the TauDEM software followed by extensive modifications. You can download the full dataset and review the full technical description document describing its creation at https://earth-info.nga.mil/ under the "Geosciences" tab. The full TDX-Hydro dataset has about 16 million river segments and covers the full globe. We have excluded select regions of the world farthest north and performed modifications to the streams to reduce the total number of rivers to a total of 7 million rivers.

Modifications Made to TDX-Hydro

All the modifications we performed on each TDX-Hydro region are recorded in Excel format. That Excel table can be viewed here. A brief overview of these changes are listed below.

The regions that were excluded include those that are farther north and some of the smaller islands, where runoff datasets may not be as accurate and there is less interest. Future GEOGLOWS versions may include some of these regions. Additionally, we corrected errors found in the V1 TDX-Hydro dataset, which include: 

1. Streams that have no length and no upstream/downstream segments, i.e. streams where there are only two points and both points are the same location. These, along with any associated catchments, were removed. 

2. Streams that have no length with upstream or downstream segments. These were removed along with any associated catchments, and the attributes of the upstream and/or downstream segments were modified to refer to each other and preserve the stream network's connectivity. 

3. Catchments with a stock identifier of '0' never had an associated stream. These were deleted. 

All of the above errors and methods for correction were corresponded back to the NGA.

For most, but not all, of the regions, the headwaters streams were dissolved with the downstream segments, up to and including the downstream segment with a Strahler stream order of either 2 or 3. It was decided that regions that were largely coastal (Japan, Carribean islands, Indonesia) were more sensitive to changes in their stream networks, and so these regions did not have their headwaters modified. Other areas, such as the Saharan desert, were delineated to the same resolution as the rest of the world -- often too much resolution, creating "streams" that in reality don't exist. In these areas, more features could be dissolved without significantly altering the river routing answer. Thus, the headwaters and downstream streams were dissolved into one feature along with their associated catchments, and relevant attributes such as length and slope were recomputed. The stream order to which the headwaters would be dissolved was also chosen based on these considerations.

Small watersheds up to 200 square kilometers were removed from the TDX-Hydro dataset for all regions. This was done for similar reasons as the differing headwater stream dissolving. The more coastal regions had watersheds between up to 25 and 75 square kilometers dropped. Other areas, like the Saharan desert or northern Canada, had watersheds of 200 square kilometers dropped. In these flatter regions, the high resolution of the delineation creates little "pools" or small collections of streams that do not drain to the ocean and do not represent flowing streams. They often collectively have an area of less than 200 square kilometers. For the less coastal and flatter/drier regions, bigger watersheds were dropped. 

Headwater streams that led directly into a stream with a Strahler stream order of two or greater were dissolved with the immediate downstream segment for most, but not all, of the regions. The decision to prune these streams are the same as above. 

TDX-Hydro Processing Files

The files used to dictate the way in which we modified the TDX-Hydro dataset can be found here. There are three files: processing_options.xlsx (which is the spreadsheet mentioned and explained above), tdx_header_numbers.json, and terminal_node_vpu_list.csv. The TDX header number JSON file maps every TDX-Hydro region number to a unique 2-digit number, with the first digit being the first digit of the region number, and the second digit corresponding to the index of the sorted order of all the regions that share the first digit. The terminal node vpu CSV matches every terminal node (the id associated with the outlet of a watershed) with a VPU number.

Python scripts have been generated to process the TDX-Hydro dataset using these files. Those python scripts can be found here.