【SDR model】input data and model running questions

ABOUT INPUT DATA:the watershed shape file.
My study area is an administrative city located in a plain area, the Hydrological Analysis of Arcgis is not suitable for plain area, therefore I tried to extract the watershed of my study area by Swat model, in order to improve the accuracy of the watershed, I add a Burn In file of the steam, but the watershed I extracted doesn’t match the border of study area, I hope someone could help me with the problem.

As I mentioned above, I did lots of attempts of watershed extracting, but failed :(. So I decided to run the model with the watershed which is not matched with the study border, what made me confused is that the model ran successfully, hope someone could help me, thanks.
InVEST-Sediment-Delivery-Ratio-Model-(SDR)-log-2021-07-14–14_18_34.txt (92.5 KB)

Hi @frank -

Can you explain (perhaps provide a graphic of) what you mean by “the watershed I extracted doesn’t match the border of study area”? Where did the border of your study area come from? It is common that watersheds created from a particular DEM do not exactly match with a study area that was created from some other source, such as a different DEM, political boundaries, a national watershed layer, etc. If the issue is with creating watersheds in SWAT, I don’t think most of us have enough experience with that toolset to help troubleshoot, and I would recommend asking SWAT experts.

The SDR model does not know whether your watershed aligns with your study area or not. As long as the input data all overlaps and is in the same coordinate system, the model will run. However, any areas where there is missing data (NoData pixels) within the watershed will end up with a value of NoData in the results. This means that you’re not modeling the entire watershed, so are missing part of the hydrology, sediment export etc, and that’s obviously not optimal.

We always recommend creating your model input layers so that they cover the entire watershed shapefile that you’re also using as input (plus even a small buffer around the watershed to make sure you’re not missing anything around the edges). If the watershed does not align with your “study area”, that’s an issue that you’ll need to work out separately. Perhaps a different DEM would help. Or perhaps model a larger watershed that does encompass the whole study area. Or, simply acknowledge that DEMs don’t always (don’t usually) represent reality very well, and that is a limitation of the data.

A final word about modeling in plains. It’s just hard to create simple hydrology models that work well in very flat areas, since drainage depends on knowing which pixel is downslope, and if everything is flat, it’s non-trivial to choose the next downslope pixel. Our software team has done a lot of work thinking about this problem, but still it’s difficult to model, so if you look at the streams created by SDR, you’ll probably see that they look unexpectedly large where it’s flat, which may lead to some less-than-realistic resulting maps. It also can be a problem when you care about the land use/land cover right around the streams, but the model is making very wide streams, so some of the land that is riparian or in a floodplain in reality is now considered a stream with NoData in the results. I’m not sure how well SWAT handles these cases, but if you’re also using it for modeling, it would be interesting to see the differences between its output and SDR in the flat areas.

~ Stacie


Hi @swolny , thanks a lot!

What I said “the watershed I extracted doesn’t match the border of study area” means the watershed doesn’t align with the area. The Burn In file is a shape file which is basically created by the satellite map, so it almost could model the real streams perfectly.

I also thought about making larger watershed which include the my entire study region and then make further processing. I’m trying to discuss the problem with some people who are familiar with this process.

I really appreciate your help, this is really a nice forum.

1 Like