Tsangpo River longitudinal profiles

We have used HydroSHEDS to calculate stream slopes for most of the Brahmaputra Basin. and have extended the ASTER GDEM work from Bhutan to the same area. We have not done much yet to evaluate our numbers.
Data will be ready for restricted download soon.

Grids v.Lattices

In GIS, a raster dataset can be seen as a grid or a lattice. If a dataset represents discrete data such as landcover codes, a value represents an area, say one square arc-second, with a uniform value. A land use code of 141 (multi-family, one to three stories high) cannot be averaged with 299 (commercial land under development) to yield 220 (retial sales and services). This is a grid. If a dataset represents elevation, your best guess for the point halfway between your 141-meter and 299-meter is indeed 220 meters. This is a lattice. In practice, both data types are stored in exactly the same way, and they are sometimes processed with the same tools.

We sometimes confront a grid/lattice duality that causes confusion with location. For example, SRTM and Global ASTER DEM lattices are laigned with parallels and meridians. A one-degree block is 3601×3601 values. Some versions of the 3" DEM have 1200×1200 grid cells in a square degree, so values have been shifted. In other words, the corner of a lattice disgnated the center of the corner cell, while the corner of a grid designates the corner of the corner cell.

steps in calculation

basics of DEM-based hydrology

Water does not flow well over the DEM. Sampling issues—as well as inaccuracy—can create "sinks". Even a meter-deep, one-cell sink can swallow up the Amazon River. Therefore, a "filled DEM" is created with no sinks. No-data "holes" are first placed in the bottoms of know undrained basins to prevent inappropriate filling. Then a grid of coded flow directions is created. Next accumulated flow contribution—optionally weighted by cell area or rainfall—are calculated for every cell in the landscape. Those cells with high flows define the rivers. Major drawbacks with this scheme included

Filled "sinks" are often vast, and water is modeled to the shortest distance along these surfaces, ignoring topographic clues that have been filled in.
All the water in a cell is assumed to flow into one of its eight neighbors. Braided channels and deltas cannot be modeled. All rivercourses are one cell wide. (Among the various alternative schemes is the "fuzzy flow accumulation" that Harvey implemented for Montgomery and Dietrich to model groundwater flows on hillslopes. Although it has successfully modeled streams disappearing into swamps, it is generally inappropriate for watercourses.)

Nobody believes that filling sinks yields the best fix for a DEM. If you imagine perfect sampling of the landscape, rivercourses can fall between sampling points. Thus a hydrological modeler will trust the low points and distrust cells that block flows. For a long time, we used proffix2.aml to trace stream courses on DEMs and to create longitudinal profiles. It traces upstream from pour points, then follows downstream on the unfilled DEM. When it reaches the bottom of a sink, it finds it's way to the pour point through a combination of local topography and the path which had been traced on the filled DEM. It then assumes that any uphill portions of the river profile are bogus, and it shaves them away. This aml program can recursively cover entire drainage systems. However, it has two major drawbacks:

It assumes that there are no erroneous low elevations.
It is very slow.
It can get confused and fall into infinite loops
Wait, that's three.
This site is curated by Harvey Greenberg, though others can ask for passwords to upload files. Puppetlab users can see it as z:/www/areas/Tsangpo and samba access will be set up soon.