My study area is too large to be processed in its entireity by any hydrological modelling software I have found yet. I will have to break it up in to smaller areas, but there will be flow across these tiles. My outcome needs to include the total flow for each stream. Are there any tips or best practices for approaching this?

EDIT: My study area is 3740 square km and the DEM is 1 metre resolution (from lidar). It can be split into two major watersheds of roughly equal size. I'd assumed these would be too large to process on their own as I have trouble even loading them into most software.

  • What kind of approach are you considering and what hydrological medelling software have you reviewed? Flow accumultation might refer to a raster output from a terrain model analysis, which can be done through very large areas, but hydrological modelling is something very different, and you do not need to work necessary on the whole area at once or with small scale data. Have you checked the HYPE, they work with very large rivers
    – Marco
    Commented May 3, 2018 at 15:01
  • 1
    Have you checked GRASS command r.watershed? "the upper limit for the seg version is 9 billion-billion (2^63 - 1 = 9.223372e+18) cells"
    – Marco
    Commented May 3, 2018 at 15:02
  • @Marco The approach is the standard 1)hydroconditioning the dem, filling sinks, breaching road crossings 2) flow direction 3) flow accumulation 4) generate stream network. Having the accumulated flow as an attribute to each stream segment is a required output. Currently I am working in Whitebox GAT on a 580 km2 subset of my data. Open to trying other software.
    – fionag11
    Commented May 3, 2018 at 15:51
  • 1
    I will say that 1m data is not only unnecessary but actually problematic in modeling flow. The extreme localized variation has a notable effect on algorithms that were never intended for this type of variation in the surface. A primary assumption here is that the flow surface is locally homogeneous, representing a 1st order plane that water flows over. In solving for stream networks, it does not make sense modeling flow around individual boulders. I would resample your data to something more relevant for these algorithms say, around 10m. This will also solve your processing issue. Commented May 3, 2018 at 16:01
  • @JeffreyEvans I am considering that but I know of another project using the same data that kept the resolution and I am trying to be consistent with their results. They decided to do this after testing the effects of going to lower resolution and finding issue vis-a-vis the known stream network. The little irrelevant bumps in the terrain should be taken care of by the initial filling and breaching operations. .
    – fionag11
    Commented May 3, 2018 at 16:11

2 Answers 2


If your interest is in flow routing and calculating flow accumulation over a large domain, I would recommend checking out the TauDEM software created by Dave Tarboton and his crew. TauDEM stripes input rasters and uses MPI to take advantage of parallel processing for large grids. This would also open the possibility of calculating both D8 and D-inf flow routing.

Alternatively (and only theoretically for now), you could chunk your input raster approximately by watershed/catchment (in the US, using the NHD or WBD for example). These units are meant to be internally draining, so you could calculate flow routing in each of those areas and then define some sort of topology for how they connect in order to route flow from one catchment to the next. Some flow routing tools, like those in ESRI software, allow you to define all cells with flow originating outside your grid as NoData, in this case essentially defining your stream network.


You could try to convert your DEM to 16 bit integer. That would, depending om your software, probably result in faster processing and reduce filesize. Just multiply your DEM with 10 or 100 before to keep the precision. I should use a larger pixelsize maybe by using minimum elevation value.

  • This is very bad advise and is likely to bring out a terracing or contouring bias in the DEM, as demonstrated in several published studies. Commented May 3, 2018 at 20:19
  • @JeffreyEvans. Realize this is an older post but can you provide the actual published studies on terracing or contour bias? It would be helpful to avoid pitfalls when processing DEMs.
    – jbrengel
    Commented Dec 28, 2018 at 20:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.